Sei sulla pagina 1di 275

Springer Series in


Founded by H.K.V. Lotsch

Editor-in-Chief: W.T. Rhodes, Atlanta

Editorial Board: A. Adibi, Atlanta

T. Asakura, Sapporo
T.W. Hänsch, Garching
T. Kamiya, Tokyo
F. Krausz, Garching
B. Monemar, Linköping
H. Venghaus, Berlin
H. Weber, Berlin
H. Weinfurter, Munich
Springer Series in
The Springer Series in Optical Sciences, under the leadership of Editor-in-Chief William T.
Rhodes, Georgia Institute of Technology, USA, provides an expanding selection of research
monographs in all major areas of optics: lasers and quantum optics, ultrafast phenomena, opti-
cal spectroscopy techniques, optoelectronics, quantum information, information optics, applied
laser technology, industrial applications, and other topics of contemporary interest.
With this broad coverage of topics, the series is of use to all research scientists and engineers who
need up-to-date reference books.
The editors encourage prospective authors to correspond with them in advance of submitting a
manuscript. Submission of manuscripts should be made to the Editor-in-Chief or one of the
Editors. See also

William T. Rhodes Ferenc Krausz
Georgia Institute of Technology Ludwig-Maximilians-Universität München
School of Electrical and Computer Engineering Lehrstuhl für Experimentelle Physik
Atlanta, GA 30332-0250, USA Am Coulombwall 1
E-mail: 85748 Garching, Germany
Max-Planck-Institut für Quantenoptik
Hans-Kopfermann-Strass 1
Editorial Board 85748 Garching, Germany
Ali Adibi
School of Electrical and Computer Engineering Bo Monemar
Van Leer Electrical Engineering Building Department of Physics
Georgia Institute of Technology and Measurement Technology
777 Atlantic Drive NW Materials Science Division
Atlanta, GA 30332-0250, USA Linköping University
E-mail: 58183 Linköping, Sweden
Toshimitsu Asakura
Hokkai-Gakuen University Herbert Venghaus
Faculty of Engineering Heinrich-Hertz-Institut
1-1, Minami-26, Nishi 11, Chuo-ku für Nachrichtentechnik Berlin GmbH
Sapporo, Hokkaido 064-0926, Japan Einsteinufer 37
E-mail: 10587 Berlin, Germany
Theodor W. Hänsch
Max-Planck-Institut für Quantenoptik Horst Weber
Hans-Kopfermann-Strasse 1 Technische Universität Berlin
85748 Garching, Germany Optisches Institut
E-mail: Strasse des 17. Juni 135
10623 Berlin, Germany
Takeshi Kamiya E-mail:
Ministry of Education, Culture, Sports
Science and Technology
Harald Weinfurter
National Institution for Academic Degrees Ludwig-Maximilians-Universität München
3-29-1 Otsuka, Bunkyo-ku Sektion Physik
Tokyo 112-0012, Japan Schellingstrasse 4/III
E-mail: 80799 München, Germany
E-mail: harald.weinfurter@physik.
John Heebner • Rohit Grover • Tarek Ibrahim

Optical Microresonators
Theory, Fabrication, and Applications

John Heebner Tarek Ibrahim
Lawrence Livermore National Laboratory RA3-353
7000 East Ave Intel Corporation
Livermore, CA 94550 5200 NE Elam Young Parkway Hillsboro, OR 97124

Rohit Grover
MS F15-58
Intel Corporation
3585 SW 198th Ave
Aloha, OR 97007

ISBN: 978-0-387-73067-7 e-ISBN: 978-0-387-73068-4

DOI: 10.1007/978-0-387-73068-4

British Library Cataloguing in Publication Data

A catalogue record for this book is available from the British Library

Library of Congress Control Number: 2007940365

c 2008 Springer-Verlag London Limited

Apart from any fair dealing for the purposes of research or private study, or criticism or review, as per-
mitted under the Copyright, Designs and Patents Act 1988, this publication may only be reproduced,
stored or transmitted, in any form or by any means, with the prior permission in writing of the publish-
ers, or in the case of reprographic reproduction in accordance with the terms of licences issued by the
Copyright Licensing Agency. Enquiries concerning reproduction outside those terms should be sent to the
The use of registered names, trademarks, etc., in this publication does not imply, even in the absence of a
specific statement, that such names are exempt from the relevant laws and regulations and therefore free
for general use.
The publisher makes no representation, express or implied, with regard to the accuracy of the information
contained in this book and cannot accept any legal responsibility or liability for any errors or omissions
that may be made.

Printed on acid-free paper

9 8 7 6 5 4 3 2 1

Springer Science+Business Media

In writing this book we sought to describe some of the important as-

pects and applications found in the wonderful world of optical microres-
onators. Of course we tell it from our respective points of view. These
vantage points have been clearly biased by the specific roads we took
during our investigations. We only hope that it does not detract from the
ideas and information collected in this research monologue. We would
never admit to perfection and cannot claim mathematical rigor in the
theoretical chapters nor detailed process recipes in the chapter on fabri-
cation. These circulating fields and their interactions have kept us busy
and entertained both conceptually and in the lab for the better part of a
decade. When we started out in this field, there was no textbook to con-
sult. It is our hope that students and researchers entering this field now
have such a guide.
We dedicate this effort to Erika, Priya, Rhea, Uma, Dalia, and Mariam.

John E. Heebner graduated first in his class with a B.E. in engineering

physics from Stevens Institute of Technology, Hoboken, NJ, in 1996. He
then conducted graduate studies at the Institute of Optics, University
of Rochester, where he received a Ph.D. in optics in 2003. Dr. Heebner’s
doctoral research on the topic of nonlinear optical effects in microring
resonators was carried out under the supervision of Prof. Robert W. Boyd.
Much of the material for this book derives from those investigations
supported by the National Science Foundation (NSF) and the Defense
Advanced Research Projects Agency (DARPA). Dr. Heebner is currently
employed as a Senior Optical Scientist by Lawrence Livermore National
Laboratory in Livermore, CA.
John thanks his graduate advisor and skiing instructor Prof. Robert
W. Boyd. Additionally, much appreciation goes out to all his mentors,
colleagues, and friends both at the Institute and in the greater optics,
photonics, and fabrication community, including Govind Agrawal, Gary
Wicks, Turan Erdogan, Richart Slusher, John Sipe, Mark Bowers, Deborah
Jackson, Young Kwon Yoon, Nick Lepeshkin, Mark Sanson, Brian Soller,
Dawn Gifford, Vincent Wong, Sean Bentley, Ryan Bennink, Matt Bigelow,
Giovanni Piredda, Aaron Schweinsberg, Collin O’Sullivan-Hale, Petros Ze-
rom, Laura Allaire, Alan Heaney, Mike Koch, Alan Bleier, John Treichler,
Philip Chak, Rob Ilic, and Ellen Chang, newfound colleagues and friends
in the LLNL ETD photonics group and NIF. Finally, he thanks his parents
Harry and Lily for their endless support and encouragement.
Rohit Grover completed his B.Tech. in engineering physics from the
Indian Institute of Technology Bombay, Mumbai, in 1997. He then re-
ceived both an M.S. and Ph.D. in electrical and computer engineering
from the University of Maryland, College Park, in 1999 and 2004, respec-
tively. Dr. Grover’s research on semiconductor optical microresonators
was carried out primarily at the Laboratory for Physical Sciences, Col-
lege Park, MD. This book draws, in part, on Rohit’s work while at the
University of Maryland, where he received support from a Graduate Re-
search Assistantship (1998-2000), Distinguished Graduate Research As-
sistantship (2000-2003), and an IEEE-LEOS Graduate Student Fellowship
(2001). Dr. Grover is currently a Senior Process Integration Engineer at
viii Acknowledgments

Intel Corporation, Hillsboro, OR, where he is studying die-package inter-

actions for the 45-nm node.
Rohit thanks the staff and his many coworkers while at the Laboratory
for Physical Sciences, and also the staff of the Cornell NanoScale Facil-
ity. In particular, he thanks his advisor, Prof. Ping-Tong Ho, colleagues
Philippe Absil, Kuldeep Amarnath, John Hryniewicz, Tarek Ibrahim (co-
author for this book), Yongzhang Leng, and Vien Van. He also thanks his
family for their support through his Ph.D. and during his professional
life thus far (and hopefully beyond).
Tarek A. Ibrahim received B.S. and M.S. degrees in electrical engineer-
ing from Cairo University, Giza, Egypt, in 1996 and 1999, respectively. He
then joined the Laboratory for Physical Sciences, University of Maryland,
College Park, as a graduate student and completed his Ph.D. in 2004. His
research while at the University of Maryland was on the topic of all-optical
signal processing using nonlinear semiconductor microring resonators.
Dr. Ibrahim received the University of Maryland Graduate Fellowship in
1999 and the Distinguished Electrical and Computer Engineering Grad-
uate Research Assistantship in 2001. He is currently a Senior Process
Integration Engineer at Intel Corporation.
Tarek would like to thank the staff of the Laboratory for Physical Sci-
ences. In particular, he thanks his advisors, Prof. J. Goldhar and Prof.
Ping-Tong Ho, and his colleagues, Philippe Absil, Kuldeep Amarnath, John
Hryniewicz, Rohit Grover, Vien Van, Ken Ritter, and Marshall Saylors. He
also thanks his family for their support throughout his Ph.D. and during
his professional life. He also thanks his advisor at Cairo University, Prof.
Adel El-Nadi, for his great support and continuous encouragement.

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v

Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Optical Microresonators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Historical Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Putting the “Micro” in “Microring” . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Nonlinear Optics in Microresonators . . . . . . . . . . . . . . . . . . . . . 5
1.5 Book Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2. Optical Dielectric Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.1 Total Internal Reflection and Waveguide Confinement . . . . 9
2.2 The Paraxial Waveguiding Equation . . . . . . . . . . . . . . . . . . . . . . 12
2.3 The Planar Slab Waveguide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.1 TE Mode Profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.2 TM Mode Profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3.3 Planar Waveguide Dispersion Relations . . . . . . . . . . . . 18
2.3.4 Normalized Planar Waveguide
Dispersion Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4 Analysis Methods for Rectangular
Dielectric Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4.1 Marcatili’s Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4.2 Effective Index Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4.3 Goell’s Circular Harmonic Method . . . . . . . . . . . . . . . . . 24
2.4.4 Finite Element Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4.5 Beam Propagation Method . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4.6 Finite-Difference Time-Domain Method . . . . . . . . . . . . 26
2.5 Coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.5.1 Perturbation Method for Deriving
Coupling Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.5.2 Coupling Between Symmetric TE
Planar Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.5.3 Coupled Wave Formalism . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.5.4 The Scattering Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5.5 Optimized Coupling for Waveguides
and Resonators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.6 Bending Loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.7 Whispering Gallery Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.7.1 TM Whispering Gallery Modes . . . . . . . . . . . . . . . . . . . . . 39
2.7.2 TE Whispering Gallery Modes . . . . . . . . . . . . . . . . . . . . . . 41
2.7.3 Radiation Loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

x Contents

2.7.4 WGM Dispersion Relations (Resonance Maps) . . . . . . 43

2.7.5 Normalized WGM Dispersion Relations
(Resonance Maps) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.7.6 Spheres, Rings, and Disks . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.8 Scattering Losses Resulting From Edge Roughness . . . . . . . 50
2.8.1 Volume Current Method Formulation for
Scattering Losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.8.2 Current Density Contributions . . . . . . . . . . . . . . . . . . . . 50
2.8.3 Spectral Density Formulation for
Edge Roughness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.8.4 Far-Field Scattered Power . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.8.5 TM Scattering Losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.8.6 TE Scattering Losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.8.7 Normalized Formulation for Edge
Scattering Losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

3. Optical Microresonator Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

3.1 Resonator Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.1.1 Fabry–Perot Resonators . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.1.2 Gires–Tournois Resonators . . . . . . . . . . . . . . . . . . . . . . . . 73
3.1.3 Ring Resonators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.2 All-Pass Ring Resonators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.2.1 Intensity Buildup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.2.2 Finesse F . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.2.3 Effective Phase Shift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.2.4 Group Delay and Group Delay Dispersion . . . . . . . . . . 79
3.2.5 Attenuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.3 Add–Drop Ring Resonators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.3.1 Intensity Buildup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.3.2 Add–Drop Resonance Width ∆ω or ∆λ . . . . . . . . . . . . 87
3.3.3 Free Spectral Range (FSR) . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.3.4 Finesse F . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.4 More on Concepts Associated with Resonators . . . . . . . . . . . 89
3.4.1 Quality Factor Q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.4.2 Physical Significance of F and Q . . . . . . . . . . . . . . . . . . 91
3.4.3 Phasor Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.4.4 Kramers–Kronig Relations . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.5 Higher Order Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

4. Microring Filters: Experimental Results . . . . . . . . . . . . . . . . . . . . . 105

4.1 Passive Resonators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.1.1 GaAs-AlGaAs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.1.2 GaInAsP-InP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
Contents xi

4.2 Active Resonators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

4.2.1 Electro-Optic Tuning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
4.2.2 Material Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.2.3 Tuning Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

5. Nonlinear Optics with Microresonators . . . . . . . . . . . . . . . . . . . . . 123

5.1 Nonlinear Susceptibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.2 Resonator Enhanced χ (3) Nonlinear Effects . . . . . . . . . . . . . . . 124
5.2.1 Enhanced Nonlinear Phase Shift . . . . . . . . . . . . . . . . . . . 124
5.2.2 Nonlinear Pulsed Excitation . . . . . . . . . . . . . . . . . . . . . . . 125
5.2.3 Kerr Effect in Solid State Materials
below Mid-Gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.2.4 Experimental Enhancement of the Kerr Effect . . . . . . 128
5.2.5 Nonlinear Saturation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.2.6 Multistability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
5.2.7 Fabry–Perot, Add–Drop, and REMZ Switching . . . . . . 137
5.2.8 Reduced Nonlinear Enhancement
via Attenuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
5.2.9 Nonlinear Figures of Merit (FOMs) . . . . . . . . . . . . . . . . . 142
5.2.10 Inverted Effective Nonlinearity . . . . . . . . . . . . . . . . . . . . 144
5.3 Resonator-Enhanced Free Carrier Refraction . . . . . . . . . . . . . 144
5.4 Enhanced Four-Wave Mixing Efficiency
in Microring Resonators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
5.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148

6. All-Optical Switching and Logic using Microresonators . . . . . . 149

6.1 All-Optical Switching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
6.1.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
6.1.2 Linear Device Characteristics . . . . . . . . . . . . . . . . . . . . . . 151
6.1.3 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
6.1.4 Nonlinear Device Characteristics . . . . . . . . . . . . . . . . . . 154
6.2 Thresholding and Pulse Reshaping . . . . . . . . . . . . . . . . . . . . . . . 156
6.3 Time-Division Demultiplexing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
6.3.1 Linear Device Characteristics . . . . . . . . . . . . . . . . . . . . . . 159
6.3.2 Nonlinear Device Characteristics . . . . . . . . . . . . . . . . . . 160
6.4 All-Optical Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
6.4.1 AND/NAND . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
6.4.2 NOR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
6.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173

7. Distributed Microresonator Systems . . . . . . . . . . . . . . . . . . . . . . . . 175

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
7.2 Linear Propagation in Distributed Microresonators . . . . . . . 175
7.2.1 Group Velocity Reduction . . . . . . . . . . . . . . . . . . . . . . . . . 177
xii Contents

7.2.2 Group Velocity Dispersion . . . . . . . . . . . . . . . . . . . . . . . . 178

7.2.3 Higher Order Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . 179
7.3 Nonlinear Propagation in Distributed Microresonators . . . . 179
7.3.1 SCISSOR Solitons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
7.3.2 Induced Self-Steepening . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
7.3.3 Pulse Compression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
7.3.4 Nonlinear Detuning and Multistability . . . . . . . . . . . . . 186
7.3.5 Nonlinear Frequency Mixing . . . . . . . . . . . . . . . . . . . . . . . 187
7.4 Limited Depth of Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
7.5 Attenuation in Distributed Microresonators . . . . . . . . . . . . . . 190
7.6 Slow and Fast Light in SCISSORs . . . . . . . . . . . . . . . . . . . . . . . . . 191
7.6.1 Slow Light . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
7.6.2 Tunable Optical Delay Lines . . . . . . . . . . . . . . . . . . . . . . . 195
7.6.3 Fast Light . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
7.7 Generalized Periodic Resonator Systems . . . . . . . . . . . . . . . . . 197
7.7.1 Bloch-Matrix Formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
7.7.2 Directly Coupled Resonators . . . . . . . . . . . . . . . . . . . . . . 202
7.7.3 Single-Channel SCISSORs . . . . . . . . . . . . . . . . . . . . . . . . . . 204
7.7.4 Double-Channel SCISSORs . . . . . . . . . . . . . . . . . . . . . . . . . 206
7.7.5 Twisted Double-Channel SCISSORs . . . . . . . . . . . . . . . . 208
7.7.6 Bandgap Engineering in Distributed
Feedback Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
7.7.7 Slow Light, Group Velocity Dispersion,
and Nonlinearities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
7.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215

8. Fabrication Techniques for Microresonators . . . . . . . . . . . . . . . . 217

8.1 Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
8.2 III–V Semiconductors for Active and
Passive Microrings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
8.3 Growing a Waveguide Stack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
8.3.1 Polymers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
8.3.2 Glass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
8.3.3 III–V Semiconductors (InP and GaAs) . . . . . . . . . . . . . . 220
8.3.4 Silicon Oxynitride . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
8.4 Feature Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
8.4.1 Polymers, Glass, SiON, SiN . . . . . . . . . . . . . . . . . . . . . . . . . 221
8.4.2 III–V Semiconductors (InP and GaAs) . . . . . . . . . . . . . . 221
8.5 Multilayer Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
8.6 Laterally Coupled III–V Passive Microresonators . . . . . . . . . . 225
8.7 Polymer-Bonded, III–V Vertically Coupled
Passive Microresonators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
8.8 Active III–V Laterally Coupled Microresonators . . . . . . . . . . . 234
8.9 Other Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
Contents xiii

8.10 Polymer Microrings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242

8.11 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
1. Introduction

1.1 Optical Microresonators

Optical microresonators have demonstrated great promise as fundamen-
tal building blocks for a variety of applications in photonics. They can
be implemented for such diverse applications such as lasers, ampli-
fiers, sensors, optical channel dropping filters (OCDFs), optical add/drop
(de)multiplexers (OADMs), switches, routers, logic gates, and artificial me-
dia. For brevity and in keeping with their current usage in the literature
of this field, we specialize the term “microresonators” and generalize
the term “microring resonators.” We use these terms interchangeably in
this book to refer to any of a number of compact geometries that sup-
port cyclically propagating modes that close on themselves in a ring-like
One particular embodiment of a microring resonator consists of an or-
dinary waveguide that channels light in a closed loop. But in general, the
loop can take the form of other closed shapes, such as a disk, racetrack,
or ellipse. In the case of a ring, the microresonator is simply a curved
waveguide closed onto itself forming a resonant cavity that supports both
transverse and longitudinal (here azimuthal) modes. The confinement
and channeling of light in this closed geometry, however, does not require
an inner dielectric boundary. This is evidenced by the existence of opti-
cal “whispering gallery” modes in a microdisk or microsphere resonator.
Placement of a microresonator near one or two waveguides (Fig. 1.1) en-
ables access to modes of the resonant cavity. In this particular arrange-
ment, the resonant modes are accessed through evanescent coupling —
a phenomena analogous to tunneling in solid-state physics. Component
wavelengths of an optical signal channeled in a waveguide are resonant
with the cavity if its (effective) circumference supports an integer number
of wavelengths. For these spectral components of the signal, an increased
circulation of intensity can build up in the resonator. The presence of a
second waveguide coupled to the ring enables extraction of the resonant,
circulating signal. Component wavelengths that do not resonate with the
ring bypass it altogether. Thus, at their most fundamental level, micror-
ing resonators act as a spectral filter and a temporary compressor of
energy density. These properties are not unique to microring resonators.
2 1. Introduction

Fig. 1.1. Schematic of an optical microring resonator add/drop filter.

Rather, they are common to all resonant cavities such as the well known
Fabry–Perot resonator.
Although functionally similar to Fabry–Perots, microring resonators
offer several advantages. First, their planar nature is naturally compati-
ble with monolithic microfabrication technologies. Second, high finesse
operation does not require multilayer or distributed Bragg reflectors but
is rather achieved by increasing the gap widths of evanescent couplers.
Third, because the equivalent injected, transmitted, and reflected waves
occupy spatially distinct channels, the need for costly Faraday circula-
tors is eliminated. Fourth, for the same reason, although there is only one
natural way to sequence arrays of Fabry–Perots (into multilayer stacks)
three altogether new possible arrangements for arrays of resonators are
enabled that differ qualitatively in many ways.
The small scale-size of microresonators currently achievable by state-
of-the-art fabrication methods is important for many reasons of which
we highlight two. First, because the propagation velocity of light is of
the order of a few hundred µm per ps in most optical materials of in-
terest, high bandwidths (GHz to THz) are naturally attainable. Second,
their small dimensions allow the integration of many devices on the
same chip, enabling high–level functionalities such as ultrafast all-optical
signal processing at a heretofore unrealized compact scale. Because of
these inherent advantages, the very large–scale integration (VLSI) of high–
bandwidth photonics may rely on optical microresonators. In the next
section, we offer our perspective on how microresonators came to be
important components in the photonic toolbox.
1.2 Historical Perspective 3

1.2 Historical Perspective

The “whispering gallery” effect was analyzed (with wave approaches) as

early as 1910 by Lord Rayleigh [1]. His analysis of the channeling of
acoustic waves by the dome of St. Paul’s cathedral in London is a pre-
cursor to similar methods applied to electromagnetic waves. Ring and
disk resonators for electromagnetic waves have since been implemented
in microwave applications starting in the early 1960s. In the optical do-
main, integrated ring resonators were proposed in 1969 by Marcatili at
Bell Labs [2].
The first guided optical ring resonator was demonstrated by Weber
and Ulrich in 1971 [3–5]. Weber and Ulrich’s device consisted of a
5-mm-diameter glass rod (n = 1.47) coated with Rhodamine-6G-doped
polyurethane (n = 1.55), for a resonator circumference of 31.4 mm.
Light was coupled in and out of the resonator with a prism. By pumping
the polymer with light from a N2 laser (λ = 337.1 nm), they obtained
laser operation. The next relevant demonstration was by Haavisto and
Pajer in 1980 [6]. Their device was the first to incorporate integrated bus
waveguides made with a doped polymethyl methacrylate (PMMA) film on
quartz substrate. A significant feature of their work was that the device
was fabricated without lithography by using direct-writing with a 325-nm
He–Cd laser. Although they demonstrated low-loss waveguides and rings,
the ring was quite large (circumference 28.3 cm). Nevertheless coupling
to the ring was via evanescent coupling to integrated bus waveguides,
and the basic idea had been established.
In 1982, Stokes, Chodorow, and Shaw [7] demonstrated the first op-
tical glass fiber ring resonator, operating at λ = 632.8 nm. Fibers un-
fortunately, do not lend themselves to compact integrated optics; their
resonator had a circumference of 3 m. Between 1982 and 1990, numerous
groups demonstrated integrated ring resonators based on glass. Early ef-
forts used ion-exchange from AgNO3 , KNO3 , and similar compounds that
modify the index of the glass, to make the waveguide core. Walker and
Wilkinson demonstrated a ring resonator with silver ion-exchanged glass
in 1983 [8] (circumference 3.1 mm, operating at λ = 632.8 nm), as did
Mahapatra and Connors in 1986 [9,10] (circumference 4.1 mm, operating
at λ = 632.8 nm). Honda, Garmire, and Wilson demonstrated a ring res-
onator with potassium ion-exchanged glass in 1984 [11] (circumference
25.1 cm). A related effort by Naumaan and Boyd in 1986 [12] used CVD
phosphosilicate glass films. Other efforts used Ti-exchanged LiNbO3 (Ti-
etgen, 1984 [13]), and proton-exchanged LiNbO3 (Mahapatra and Robin-
son, 1985 [14]). Tietgen’s work is especially significant as it represents
the first demonstration of a tunable ring resonator. Instead of a circular
ring, he used a waveguide loop with two 3-dB couplers. His device used
electro-optic tuning, had a circumference of a little over 24 mm, and op-
erated at λ = 790 nm.
4 1. Introduction

Since the early efforts outlined above, there have been numerous
works in various doped and undoped silica-based glasses [15–25], Si-
(Si3 N4 , SiON, SiO2 ) [26–37], and polymers [38–41] in the past decade. Many
of these studies have reported multiring filters, temperature-insensitive
operation and so on. Oda’s work with TiO2 -doped silica-glass rings rep-
resents the first demonstration of serially cascaded rings, with increased
free spectral range over single-ring devices. Rabiei’s work with polymer
rings represents the first passive and active polymer ring resonator.
Microresonators constructed in III–V semiconductors began “seeing
light” in the early 1990s. Several groups demonstrated optically pumped
microdisk lasers in both GaInAsP-InP and III–Nitrides using the whisper-
ing-gallery; the smallest reported disks had circumferences of ∼15 µm
[42–50]. Most of these early efforts did not incorporate bus waveguides
and relied on fibers to directly collect light from the disk. The first GaAs-
AlGaAs microring resonator laterally coupled to bus waveguides was
demonstrated by Rafizadeh et al. in 1997 at Northwestern University,
Evanston, IL [51,52]. Their smallest ring had a circumference of 32.8 µm.
Since then, members of Ping-Tong Ho’s group at the Laboratory for Phys-
ical Sciences (LPS), College Park, MD, have demonstrated both laterally
and vertically coupled rings in GaAs-AlGaAs acting as multi ring devices,
switches, routers, and mux/demux operation [53–61]. The GaInAsP-InP
material system has proven problematic for passive microrings because
of processing difficulties resulting in high device losses. Nevertheless,
the first vertically coupled passive InP-based rings were demonstrated
by Ho’s group [59, 62, 63]. Other groups have concentrated on disk res-
onators; the group at the University of Southern California, for example,
has demonstrated active and passive vertically coupled microdisk res-
onators [64, 65].

1.3 Putting the “Micro” in “Microring”

Initial efforts toward the fabrication of integrated ring resonators pro-

duced very large devices because the index contrast ∆n between the core
and cladding was small, and operation with large radii of curvature min-
imizes bending losses [2, 66–69]. For example, the device fabricated by
Honda et al. had 0.052 < ∆n < 0.067 and a radius of 4.5 cm [11]. Also,
the device was multi mode, and there was considerable mode mixing,
leading to a large background of non resonant light and the convolution
of resonances from multiple modes.
The construction of microrings 10-µm in diameter or smaller is a chal-
lenging effort often requiring high index contrast, anisotropically etched
pedestal waveguide designs with ultrasmooth sidewalls. In addition, pre-
cise coupling gap widths with extremely tight tolerances are demanded
1.4 Nonlinear Optics in Microresonators 5

to engineer coupling strengths. This is accomplished laterally using high-

resolution lithographically or vertically by material growth. In the past
decade, these technologies have reached maturity. The confluence of the
many mature and maturing technologies has led to the demonstration of
single-mode microring resonators both laterally and vertically coupled
to a bus waveguide in GaAs-AlGaAs [51] and Si-SiO2 [21]. Advanced func-
tions, such as high-order filtering for dense wavelength division multi-
plexing applications, notch filters, and wavelength-selective mirrors, have
since been demonstrated [55, 57, 60].
The lateral-coupling approach in III–V semiconductors demands strict
patterning tolerances typically requiring e-beam lithography followed by
advanced etching techniques. Nevertheless, in 2000, Griffel and cowork-
ers at Sarnoff Corp., Princeton, NJ, and Riverside Research Institute,
New York, NY, demonstrated a laterally coupled ring laser with inte-
grated bus waveguides with stepper lithography by employing a bi-level
etching technique [70]. They were able to demonstrate devices with cir-
cumferences of 1000 µm or more. In 2001, Rabus and coworkers at
the Heinrich-Hertz Institute (HHI), Berlin, Germany, demonstrated multi-
mode interference (MMI) coupled rings in GaInAsP-InP, and in 2002 they
demonstrated active rings by integrating semiconductor optical ampli-
fiers and on-chip platinum heaters [71–73]. The HHI work features very
large ring circumference; their smallest rings had circumferences as
large as 928 µm, which resulted in extremely narrow free spectral range
(around 0.8 nm). Laterally coupled ring resonators in GaInAsP-InP were
also demonstrated by Rommel and coworkers from the University of Illi-
nois, Urbana-Champaign, IL, and Sarnoff in 2002 [74]. They demonstrated
a ring resonator with circumference 274.2 µm and free spectral range a
little over 2 nm. Ho’s group recently demonstrated microring resonator
notch filters, both passive and electro-optically tuned, with the smallest
turning radius to date [75–77]. Their smallest device had a circumference
of just 20.1 µm, with a free spectral range much greater than 30 nm.

1.4 Nonlinear Optics in Microresonators

All optical materials display a change in their optical properties under

high-intensity illumination. Fundamentally, this change results from is
due to a nonlinear response of the material polarization to the applied
electromagnetic field. As a concise and practical definition, nonlinear op-
tics is study of high intensity phenomena in which the optical response
of a medium differs from its low-intensity limits. The narrowing of a
microring’s resonance is accompanied by a proportional increase in the
coherent buildup of intensity circulating within the cavity. Although this
buildup is of no direct consequence to the linear transmission properties,
6 1. Introduction

it has a dramatic effect on the nonlinear (intensity-dependent) transmis-

sion properties.
In 1969, Szöke et al. appreciated this effect and proposed inserting a
nonlinear material (saturable absorber) between the mirrors of a Fabry–
Perot; they also described optical multistability with possible applications
for optical logic [78]. The first optical bistability experiments were later
performed by McCall and Gibbs et al. in a sodium-filled Fabry–Perot [79,
80]. Marburger and Felber later developed the theory of a Fabry–Perot res-
onator containing a medium possessing a nonlinear (intensity-dependent)
refractive index and demonstrated that under certain operating condi-
tions, the device could be implemented as an all-optical switch [81].
Miller etal later performed the first two-beam a.c. amplification experi-
ment [82] and derived the optimum parameters for all-optical switching
implementing refractive bistability in a lossy Fabry–Perot [83].
In the late 1980s and 1990s, ring resonators were used in various
experiments in nonlinear and quantum optics. Shelby et al. observed a
finesse-squared enhancement dependence of squeezed light generation
in a fiber ring resonator [84, 85]. In 1989, Braginsky et al. studied the
nonlinear properties of optical whispering-gallery modes [86]. Braginsky
proposed the application of such modes to the lofty goal of switching
with a single quantum [87]. In 1996, Vernooy and Kimble performed
cavity quantum electrodynamic (QED) experiments that exploited the
small mode volumes and high field strengths associated with the modes
of microcavities [88]. In 1997 Chang et al. demonstrated Q-switching
using enhanced saturable absorption associated with the WGM reso-
nances of microdroplets [89]. In 1997, Blom et al. proposed the de-
velopment of an integrated all-optical switch based upon the high-Q
whispering gallery modes of a nonlinear polymer disk [90, 91]. In 1999,
Rosenberger [92] furthered the study of nonlinear optical effects in
microspheres. More recently, enhanced Raman interactions in glass mi-
crospheres were demonstrated [93, 94]. Cavity QED phenomena [95, 96]
still continue to be explored in microresonators because of their potential
for high quality factors and ultrasmall mode volumes.
Despite all this research, microresonators implemented as nonlinear
optical elements continue to be a tremendous technological challenge in
real applications, because nonlinear optical effects are extremely weak,
particularly with respect to their electronic counterparts. In contrast,
electrons interact with each other very strongly because of Coulombic
interaction. At the most fundamental level, photons do not manipulate
photons without first interacting with electrons in practical nonlinear
optical materials. The result is that photons are extremely adept in the
high-fidelity transmission of information, but they possess an underlying
handicap associated with the manipulation of information — as in active
routing or performing logic.
1.5 Book Overview 7

Much like the field of electronics exploded when nonlinear elements

(the transistor) were integrated at a compact scale, the field of photonics
could be waiting to blossom away from traditional optics when nonlinear
photonic devices at the chip level become commonplace. This promise
has been held back in large part by the lack of an optimal material sys-
tem with fast nonlinear response and direct compatibility with a ma-
ture fabrication technology. Some III–V semiconductors such as AlGaAs
come closest to the mark, but they are nevertheless several orders of
magnitude weaker than comparable nonlinear responses in electronics.
Microresonators enable the augmentation of the intrinsic nonlinear opti-
cal properties. Much like quantum dots and photonic crystals, microres-
onators offer another method for creating engineerable materials with
nonlinear responses appreciable enough to construct photonic devices
with advanced functionalities operating at ultrafast speeds.

1.5 Book Overview

In the rest of this book, we present an overview of the theory, fabrica-

tion, and application of optical microresonators. Along the way, where
relevant, we present the results of our research efforts along with those
from other groups. In chapter 2 we introduce the equations describing
confinement and propagation in optical dielectric waveguides. We then go
on to present methods for analyzing coupling, losses, and whispering gal-
ley effects relevant to microresonators. In chapter 3 we delve into the the-
oretical framework describing the properties associated with resonators
in a generalized manner. Chapter 4 presents the results of passive, lin-
ear optical experiments with fabricated microring devices operating as
spectral filters. In chapter 5, we revisit the theoretical framework with
the inclusion of nonlinear optical phenomena associated with microres-
onators. Chapter 6 follows up with demonstrations of all-optical switch-
ing and logic. In chapter 7, we consider both linear and nonlinear optical
propagation phenomena in sequences of multiple microresonators and
draw comparisons to related photonic crystal systems. Finally, in chap-
ter 8, we discuss fabrication processes and techniques involved in the
construction of microresonators.
2. Optical Dielectric Waveguides

In this chapter, we introduce several analytical and numerical methods

used to model the confinement of light in optical dielectric waveguides.
We then go on to develop methods for calculating the coupling strength
between waveguides and for predicting losses resulting from bending
loss. These losses are then considered in the regime of the optical whis-
pering gallery, and we conclude with a treatment of scattering losses re-
sulting from edge roughness in this regime. The ideas presented in this
chapter are of fundamental importance to the practical implementation
of optical microresonators.

2.1 Total Internal Reflection and Waveguide Confinement

Conventional dielectric waveguides channel light through transverse con-
finement in a dielectric “core,” of refractive index n1 surrounded by a
“cladding” often of lower refractive-index n2 . The cladding may consist
of a different material, a differently doped region, or in some cases the
surrounding air. Guidance in dielectric waveguides results from the phe-
nomenon of total internal reflection (TIR).
Light inside the core experiences total internal reflection when strik-
ing the core-cladding interface at angles greater (with respect interface
normal) than the critical angle. For most conventional waveguides, this
condition is satisfied when light enters the core at angles smaller than the
guide’s acceptance angle (with respect to the guide’s propagation axis) or
numerical aperture NA. Provided that the waveguide cross section does
not vary significantly or rapidly along the direction of propagation, the
light will be localized over long lengths with low loss. Figure 2.1 depicts
some of the more common waveguide cross sections. The critical an-
gle arises from the Fresnel equations that derive from applying bound-
ary conditions to the electric and magnetic fields in conjunction with
Maxwell’s equations. The Fresnel reflectivity expressions applicable to
angles of incidence within the critical angle for TIR are given as

n1 cos θ1 − i n21 sin2 θ1 − n22
rTE =  ≡ e−iφTE (2.1)
2 2 2
n1 cos θ1 + i n1 sin θ1 − n2
10 2. Optical Dielectric Waveguides

Fig. 2.1. Common waveguide geometries. Darker regions represent higher

refractive index.

i n12 n21 sin2 θ1 − n22 − n2 cos θ1
rTM =  ≡ e−iφTM . (2.2)
n1 2 2 2
i n2 n1 sin θ1 − n2 + n2 cos θ1

Here, transverse electric (TE) refers to linearly polarized light where the
electric field is perpendicular to the plane of incidence and transverse
magnetic (TM) refers to the case where the magnetic field is perpendicular
to the plane of incidence. It follows from these equations that for angles
of incidence satisfying (sin θ1 > n2 /n1 ), the modulus of the reflectivity is
unity. Thus, the critical angle for total internal reflection is defined solely
by the refractive index ratio as θc = arcsin(n2 /n1 ). For angles above θc ,
the complex reflectivities can be expressed in phasor representation with
unit amplitudes and phase shifts acquired upon reflection1 :
⎛ ⎞
n21 sin2 θ1 − n22  
φTE = 2 arctan ⎝ ⎠ = 2 arctan γx (2.3)
n1 cos θ1 kx
⎛  ⎞ 
n21 sin2 θ1 − n22 2
φTM = 2 arctan ⎝
n 1 ⎠ = 2 arctan n1 γx . (2.4)
n2 n2 cos θ1 n22 kx

In the final forms of the expressions, kx = n1 k0 sin θ1 refers to the perpendic-
ular component of the propagation vector k1 = n1 k0 incident on the interface
and γx refers to the decay constant associated with the evanescent tail of the
field that extends beyond the interface.
2.1 Total Internal Reflection and Waveguide Confinement 11

Fig. 2.2. Light confinement and guiding by total internal reflection. The
zigzag path can be decomposed into a transverse wavevector, kx describ-
ing purely oscillatory behavior and a longitudinal propagation vector,
β along the guide axis.

As we will see later, the angular dependence of the acquired phase shift
in reflection plays an important role in dictating the dispersion relation
for guided modes. In Fig. 2.2, light guided by total internal reflection is
depicted as resulting from the interference of plane waves alternately re-
flecting from both core-cladding interfaces of a planar slab waveguide of
thickness d. The dotted line represents the direction of the total wave-
vector k1 . The wave-vector can be decomposed into a component along
the guide propagation direction termed the effective propagation con-
stant β and another in the transverse direction kx . For an eigenmode
solution of a slab waveguide, the transverse field profile is oscillatory
and does not vary with propagation. Hence, the wave must experience a
phase shift of 2π m, where m is an integer, in one round-trip between the
two core-cladding interfaces. The phase shift consists of a contribution
from the transverse component of the propagation constant along with
the Fresnel phase shift acquired at each interface. The round-trip phase
requirement implies that for a given frequency ω, only certain discrete
incidence angles θm are allowable for the wavevector; each angle solu-
tion is associated with a transverse mode of the guide. Thus we arrive at
a very simple geometric derivation for the dispersion relation of modes
in a planar or slab waveguide for the TE case:
n1 k0 2d cos θm − 2φTE [θm ] = m2π (2.5)
⎛ ⎞
 β 2
m − n 2 2
2d n21 k20 − β2m − 4 arctan ⎝  ⎠ = m2π
2 0
n21 k20 − β2m
and for the TM case:
n1 k0 2d cos θm − 2φTM [θm ] = m2π (2.7)
⎛  ⎞
 2 β 2 − n2 k 2
n m
2d n1 k0 − βm − 4 arctan ⎝ 2  ⎠ = m2π .
2 2 2 1 2 0
n2 n2 k2 − β2m
1 0
12 2. Optical Dielectric Waveguides

This derivation has intuitive appeal in describing the allowable propa-

gation constants for a given frequency (ω = ck0 ) in terms of reflection
angles for geometric ray paths and Fresnel reflection phase shifts. From
these discrete propagation constants, the mode profile in the core and
cladding can be obtained. Unfortunately, this technique is of limited use
and is generally not directly applicable to the derivation of dispersion
relations associated with waveguides of a two-dimensional cross section.
The source of the limitation is that the Fresnel equations are directly ap-
plicable only for plane waves incident on planar interfaces in which the
angle of incidence is discrete. In the case of a planar waveguide, fields
confined inside the core can be described by two interfering plane waves
and the Fresnel equations are applicable. The intuition acquired through
this description is sometimes useful for quickly analyzing other guiding
structures, although with some degree of caution. In order to analyze
more generalized waveguide cross sections, we turn to a physical optics
description of wave propagation.

2.2 The Paraxial Waveguiding Equation

A physical optics model for propagation in a waveguide begins with

Maxwell’s two first-order vector curl equations for the electric and mag-
netic fields in dielectric media:
∇ × E = −µ0 (2.9)
∇×H=ε (2.10)
These equations are combined into a single second-order wave equation,

n2 ∂ 2 U
∇2 U − =0 (2.11)
c 2 ∂t 2
where U represents a component of the electric E or magnetic H field.
We assume that the fields are harmonic in time t and that propagation
takes place along the z axis. A temporal Fourier transform of the wave
equation results in the time-independent Helmholtz equation

+ ∇2T U + k2 U = 0, (2.12)

where the Laplacian ∇2 operator has been decomposed into longitudinal

∂ 2 /∂z2 and transverse ∇2T = ∂ 2 /∂x 2 + ∂ 2 /∂y 2 components. Next, we as-
sume a field solution such that for all points transverse to the direction
of propagation, the field accumulates phase uniformly with propagation
2.3 The Planar Slab Waveguide 13

constant β according to U (x, y, z) = A(x, y)eiβz , where A(x, y) repre-

sents the stationary transverse mode profile. Substituting this expression
into Eq. 2.12 results in [97]

∂2A ∂A
+ i2β + ∇2T A + (n2 k20 − β2 )A = 0. (2.13)
∂z2 ∂z
Since we wish to find the stationary transverse modes of the structure,
we set both longitudinal derivatives to zero and are left with the equation
to be solved:
∇2T A + (n2 k20 − β2 )A = 0. (2.14)
This equation is equivalent to the time-independent Schrödinger equa-
tion and is solved in both the core and cladding with the appropriate
boundary conditions to obtain β. For guided modes, the value of β lies
between the propagation constants of the core and cladding. This ensures
that light is confined to the core where the solutions to Eq. 2.14 (for which
β < nk0 ) are oscillatory. The evanescent tail of the mode in the cladding
(for which β > nk0 ) is described by solutions exhibiting exponential de-
cay away from the core. For a guided wave, the quantity neff = β/k0
is termed the “effective index” of the waveguide, since it represents the
ratio of the speed of light in vacuum to that of the propagating mode.
In certain special cases, analytic solutions for Eq. 2.14 are possible. In
general, however, approximate analytic expressions deliver quick and in-
tuitive understanding, whereas numerical methods are indispensable in
achieving sufficient accuracy for most modern waveguiding structures
of interest. In the following section we revisit the planar slab waveguide
within the physical optics formalism.

2.3 The Planar Slab Waveguide

For a planar slab waveguide, confinement exists only in one direction

(here along the x axis). Consequently, there is in infinite quantity of opti-
cal power in the mode, although a power per unit length along the trans-
lation invariant direction (y axis) can be ascribed. Localized solutions
require that the field decays to 0 at x → ±∞. We examine separately
the derivation [98] of the field profiles for TE and TM guided modes (for
which k2 < β < k1 ).

2.3.1 TE Mode Profiles

For TE polarized modes, Maxwell’s vector curl equations reduce from six
to three equations involving three field components (Ey , Hx , Hz ):

−iβEy = iωµ0 Hx (2.15)

14 2. Optical Dielectric Waveguides

= iωµ0 Hz (2.16)
∂Hx ∂Hz
− = −iωεEy (2.17)
∂z ∂x
that reduces to a single wave equation for the transverse electric field
∂ 2 Ey ∂Hx ∂Hx
= iωµ 0 + iωεE y = iωµ0 − ω2 µ0 εEy
∂x 2 ∂z ∂z

= β2 − k2 Ey (2.18)

that can be separated into transverse Helmholtz equations for each of the
core and the two cladding regions. For simplicity, we assume a symmetric
cladding configuration.

∂2 2
+ kx Ey = 0 (x < −d/2, x > d/2) (2.19)
∂x 2

− γx2 Ey = 0 (−d/2 < x < d/2) (2.20)
∂x 2

resulting in field profiles for the three regions here for even-order modes.2

n1 − n2
Ey = Ey0  2 eff −γx (x−d/2)
e (x > d/2) (2.21)
n1 − n22
= Ey0 cos (kx x) (x > −d/2, x < d/2) (2.22)

n1 − n2
= Ey0  2 eff γx (x+d/2)
e (x < −d/2). (2.23)
n1 − n22

The magnetic field components are then easily derived from this electric
field solution through the following relationships:
Hx = − Ey (2.24)
i ∂Ey
Hz = − . (2.25)
k0 Z0 ∂x

The power per unit length carried by the mode can be obtained by inte-
grating the time-averaged Poynting vector (S = 21 {E × H∗ }) across the
waveguide dimension.

For odd-order modes, the cosine is replaced with a sine and the two cladding
amplitudes are oppositely signed.
2.3 The Planar Slab Waveguide 15

P/L =− dxEy Hx∗ (2.26)

= dx Ey  (2.27)
neff Ey0  2
= d+ . (2.28)
4Z0 γx

The power per unit length thus reduces to an intuitive expression equal-
ing half the time-averaged peak intensity (I0 = neff Ey0  /2Z0 ) times
the effective width of the mode (deff = d + 2/γx ). In terms of the power
per unit length, the field amplitudes are given by

4Z0 P/L
Ey0 = (2.29)
neff deff

4neff P/L
Hx0 = (2.30)
Z0 deff
with the relationship: Ey0 Hx0 = 4P/L /deff .
As an example, the mode profile for an even mode is shown in Fig. 2.3a,
and that for an odd mode is shown in Fig. 2.3b. Shown are the lowest-
order modes, i.e., the modes with the minimum number of transverse
oscillations, that also have the highest effective indices. The thickness d
of the waveguide in this example is 1000 nm, and n2 = 3.17, n1 = 3.35.

2.3.2 TM Mode Profiles

The expressions for TM polarized modes have a similar derivation. The

Maxwell vector curl equations again reduce from six to three equations
involving three field components (Hy , Ex , Ez ):

−iβHy = −iωεEx (2.31)

= −iωεEz (2.32)
∂Ex ∂Ez
− = iωµ0 Hy (2.33)
∂z ∂x
that reduces to a single wave equation for the transverse magnetic field
∂ 2 Hy ∂Ex ∂Ex
= −iωε − iωµ 0 y = −iωε
H − ω2 µ0 εHy
∂x ∂z ∂z

= β2 − k 2 Hy (2.34)
16 2. Optical Dielectric Waveguides


Fig. 2.3. (a) An even and (b) an odd mode of a slab waveguide. The dotted
lines represent the waveguide boundaries.

that can be separated into transverse Helmholtz equations for each of

the core and the two (here symmetric) cladding regions:

∂2 2
+ kx Hy = 0 (x < −d/2, x > d/2) (2.35)
∂x 2

− γx2 Hy = 0 (−d/2 < x < d/2), (2.36)
∂x 2

resulting in the following transverse electric field profiles for the three
regions for even-order modes.3

For odd-order modes, the cosine is replaced with a sine and the two cladding
amplitudes are oppositely signed.
2.3 The Planar Slab Waveguide 17

Hy = Hy0
 e−γx (x−d/2) (x > d/2) (2.37)
n41 n2eff −n22
1+ n42 n21 −n2eff

= Hy0 cos (kx x) (x > −d/2, x < d/2) (2.38)

= Hy0
 eγx (x+d/2) (x < −d/2), (2.39)
n41 n2eff −n22
1 + n4 n2 −n2
2 1 eff

where field components are related through the following equations:

neff Z0
Ex = Hy (2.40)
iZ0 ∂Hy i ∂Ex
Ez = 2
= . (2.41)
k0 n ∂x k0 neff ∂x

Note that unlike the TE case, the transverse component of the electric
field Ex in the TM case displays a discontinuous increase of n22
from the
core to the cladding. This asymmetry results because we are only con-
sidering dielectric waveguides where only the permittivity varies while
the permeability remains constant. There is also a discontinuity in the
magnetic field gradient for the TM mode that forces it to have a lower
effective index and a longer evanescent tail. The power per unit length
carried by the mode is again obtained by integrating the time-averaged
Poynting vector across the waveguide dimension.

P/L = dxEx Hy∗ (2.42)

neff n2
= dx |Ex |2 (2.43)
2Z0 n2eff
⎡ ⎤
neff |Ex0 | 2 ⎢ n21 n22 2 n42
−1 n22 2 ⎥
= ⎢ + + ⎥.
4Z0 ⎣ n2 d n2eff γx n41 n2eff −n22 n2eff γx ⎦
eff 1+ n42 n21 −n2eff

The power per unit length again reduces to an expression equaling half
the peak intensity times the effective width (contained in square brack-
ets). The effective width has a more complicated expression than in the
TE case but reduces to the same value in the limit of low index contrast
(n1 ≈ n2 ≈ neff ). In terms of the power per unit length, the field ampli-
tudes are given by 
4Z0 P/L
Ex0 = (2.45)
neff deff
18 2. Optical Dielectric Waveguides

n21 4neff P/L
Hy0 = . (2.46)
n2eff Z0 deff

2.3.3 Planar Waveguide Dispersion Relations

The previous sections used a physical optics description to determine

the mode profiles for a planar waveguide assuming a known propagation
constant. We return to solve for the value and frequency dependence of
the propagation constant βm (ω) that will differ for each nondegenerate
mode m. For TE polarization, four equations and four unknowns result
from ensuring the continuity of the transverse electric field Ey and the
(boundary) transverse magnetic field Hz (related to ∂Ey /∂x) at each in-
terface. For clarity, we depart from the previous notation and label the
four field amplitudes as A1 , A2 , A3 , and A4 corresponding respectively
to the +x, −x propagating (oscillatory) fields in the core and the +x, −x
evanescently decaying fields in the claddings. The resulting four equa-
tions are, thus,

A1 e−ikx d/2 + A2 e+ikx d/2

A4 = (2.47)
−ikx d/2
A1 e − A2 e+ikx d/2
γx A4 = ikx (2.48)
A1 e+ikx d/2 + A2 e−ikx d/2
A3 = (2.49)
+ikx d/2
A1 e − A2 e−ikx d/2
−γx A3 = ikx . (2.50)
These equations can be solved simultaneously resulting in the equation:
(kx − iγx ) +ikx d
e = 1 = eim2π . (2.51)
(kx + iγx )

that can be rewritten in the form of a transcendental expression relating

the propagation constant βm to the radian frequency ω for TE modes:

βm − n2 ω2 /c 2
d n21 ω2 /c 2 − β2m − 2 arctan  2 2
= mπ . (2.52)
n1 ω2 /c 2 − β2m

For completeness, we repeat the derivation of the slab waveguide disper-

sion relation this time for the TM modes.
A1 e−ikx d/2 + A2 e+ikx d/2
A4 = (2.53)
−ikx d/2
1 1 A1 e − A2 e+ikx d/2
γ A = 2 ikx
2 x 4
n2 n1 2
2.3 The Planar Slab Waveguide 19

A1 e+ikx d/2 + A2 e−ikx d/2

A3 = (2.55)
1 1 A1 e+ikx d/2 − A2 e−ikx d/2
− 2 γx A3 = 2 ikx . (2.56)
n2 n1 2

These equations can be solved simultaneously resulting in the equation:

⎡  ⎤2
⎢ kx − i n12 γx ⎥
⎢ 2
 e+ikx d ⎥ im2π
⎣ n21 ⎦ =1=e . (2.57)
kx + i n2 γx

that is again rewritten in the form of a transcendental expression relating

the propagation constant βm to the radian frequency ω for TM modes:

n β2m − n22 ω2 /c 2
d n21 ω2 /c 2 − β2m − 2 arctan 12  2 = mπ (2.58)
n2 n1 ω2 /c 2 − β2m

Note that these dispersion relations Eqs. 2.52 and 2.58 are equivalent to
Eqs. 2.6 and 2.8 derived earlier using the Fresnel equations in conjunction
with a geometrical optics description.

2.3.4 Normalized Planar Waveguide Dispersion Relations

Kogelnik and Ramaswamy presented a universal set of normalized curves

from which the dispersion relation of an arbitrary one-dimensional (1D)
slab waveguide can be mapped [99]. Under this formalism, a normalized
frequency V incorporates
 the wave frequency ω, core dimension d, and
numerical aperture n21 − n22 , whereas a normalized propagation con-
stant b is a shifted and scaled form of the propagation constant:

V = d n21 − n22 (2.59)
β2 − k22
b= 2 . (2.60)
k1 − k22

Useful relations between the real and normalized parameters are given

neff = n22 + b n21 − n22 (2.61)

V 1−b
kx = (2.62)
20 2. Optical Dielectric Waveguides

V b
γx = (2.63)
deff = d 1 + √ (2.64)
V b

V b + 2b
Γ = √ , (2.65)
V b+2
where Γ represents the confinement factor or proportion of optical power
contained in the waveguide core. Implementing these relations with
Eq. 2.52 results in the normalized dispersion relation for a symmetric
TE planar waveguide,

V 1 − b − 2 arctan = mπ . (2.66)
This transcendental relation between V and b is plotted in Fig. 2.4. Note
that single-mode operation is guaranteed for V < π .

2.4 Analysis Methods for Rectangular Dielectric


The previous sections were intended to refresh the reader with the prop-
erties of one-dimensional planar waveguides. However, the guiding struc-
tures associated with ring resonators are generally two-dimensional (2D).
The most common 2D dielectric waveguide for which analytic solutions
exist is the circular waveguide or optical fiber. This results from the sym-
metry of the waveguide cross section. The geometry can be defined inde-
pendently along radial and azimuthal dimensions, and this fact is in turn
reflected in the separability of field solutions along each dimension. It
would seem to follow that other cross sections that possess high degrees
of symmetry might also support analytic solutions. Waveguides of rec-
tangular cross section are typically employed in planar geometries and
in particular for the construction of microring resonators. A rectangular
cross section, of course, possesses Cartesian symmetry. A rectangular
metallic waveguide possesses separable sinusoidal field variation along
two orthogonal dimensions. The same, however, is not true of a rectangu-
lar dielectric waveguide. The reason is that the fields extend beyond the
core into a region where the geometry cannot be defined independently
along orthogonal directions. Thus, the problem of the rectangular dielec-
tric waveguide has been addressed via a large number of approximate
analytic techniques and numerical methods. Although somewhat dated,
Saad presents an excellent review [100]. In the following discussion, we
describe two approximate analytical techniques: Marcatili’s method and
the effective index method, which through approximation, reduce 2D
2.4 Analysis Methods for Rectangular Dielectric Waveguides 21

Fig. 2.4. The dispersion relation for a planar slab waveguide with n1 =
3.35, n1 = 1.0, and d = 240 nm. For these parameters, light at λ =
1.55 µm, ω/c = 4.05 µm−1 , is just barely single-moded. The TE solutions
are in bold linetype, and the TM solutions are dashed. The gray region in
the top plot represents the guided wave solution space bounded by core
and cladding light lines. The lower plot is a normalized version of the
dispersion relation.

structures with rectangular symmetry to equivalent 1D structures sep-

arable along each dimension. Guidance along each dimension can then
be analyzed using the planar slab waveguide formalism derived earlier.
These techniques have their limitations, and so we follow up with a dis-
cussion of numerical techniques that have been implemented to solve
the waveguide equation.
22 2. Optical Dielectric Waveguides

2.4.1 Marcatili’s Method

Marcatili introduced a method whereby rectangular waveguides can be
analyzed through a slight alteration of the refractive index geometry
[101]. This modification forces the field solutions along Cartesian axes
to be separable. The technique works well for 2D rectangular waveguides
where the index contrast along each dimension is of similar magnitude.
Such is the case for a buried-channel waveguide as depicted in Fig. 2.5.
The corresponding Marcatili approximation and equations are shown in
the lower figure. Note that the relative permittivities (square of refrac-
tive indices) of the guiding channel and four cladding regions at its edge
boundaries remain unchanged, but the four corner regions are artificially
reduced by the permittivity difference n21 − n22 . For low index contrast
guidance of moderate to strongly confined modes, the field is expected
to be small in these corner regions and the approximation works rather
well. Notice that the same permittivity difference exists across each of
the three regions of each of the horizontal and vertical slab waveguide
components. Thus, this trick successfully separates the geometry into
two orthogonal slab waveguides, each of which can be solved indepen-
dently. Because the index in the corner regions is artificially lowered, the

Fig. 2.5. Marcatili approximation for a rectangular buried channel

2.4 Analysis Methods for Rectangular Dielectric Waveguides 23

result is an underestimated propagation constant. The severity of the

error is reduced proportional to the index contrast of the guide and the
confinement of the mode. Kumar et al. [102] extended Marcatili’s method
by taking the resultant field of the Marcatili solution and reintroducing
the corner permittivity difference n21 − n22 as a perturbation. Applying
this perturbative method results in greater accuracy albeit with increased

2.4.2 Effective Index Method

The effective index method (EIM) was introduced by Knox and Toulios
[103] as an alternative analytic method to that of Marcatili’s for the mod-
eling of rectangular dielectric waveguides. The method is extremely sim-
ple to implement and to this day retains popular appeal as a method for
quick estimation. Unlike the Marcatili method that solves each dimension
in parallel, the EIM proceeds in series solving one dimension first and
analyzing the second as a perturbation of the first. Kumar et al. [102]
pointed out that ambiguous results are obtained when trying to imple-
ment this method for waveguides of square cross section. The method is
better suited for modeling rib or ridge waveguides. In these structures,
the guidance is nearly a slab waveguide in one dimension with a small
perturbation to index or core thickness that achieves guidance in the
other dimension.
The EIM begins by analyzing the structure for each segmented slab
waveguide section and assigning effective indices for each section. Next,
the effective indices for each slab segment are used to solve for the re-
sulting slab waveguide geometry in the slightly perturbed dimension. As
an example, we apply the method to solve for the dispersion relation and
mode profiles of a rib waveguide. Figure 2.6 shows the structure along
with the first and second approximation steps. The first step results in
three normalized frequencies (V2 , V1 , V2 ) where here, due to the symme-
try, the first and third are equal. Using Kogelnik’s generalized dispersion
relation allows us to solve for (b2 , b1 , b2 ) and then the effective refractive
indices (neff,2 , neff,1 , neff,2 ) of each segment. Next, these auxilliary effec-
tive indices define a net normalized propagation constant (V ) that leads
to a unique effective index for the mode. In contrast to Marcatili’s method,
the EIM overestimates the coupling constant [104] but is more accurate
for analyzing the fundamental mode solution near cutoff. Lee et al. [105]
pointed out that this is because the effective index method makes the
one-mode approximation. Numerous extensions of the EIM have been
developed [106–108]. Chiang [108], for example, introduced a correction
factor that greatly improves the accuracy of the method. Because mod-
ern computers and commercial software can readily solve for modes of
waveguides of arbitrary cross section, overrefinement of this method is
usually unnecessary. The usefulness of the original method lies in its
24 2. Optical Dielectric Waveguides

waveguide n1 a2 a 1


n3 n3 n3
Effective index
approximation n1 n1 n1 a2 a1
(step 1)
n2 n2 n2

Effective index
approximation neff,2 neff,1 neff,2
(step 2)

Fig. 2.6. Effective index approximation for a rib waveguide.

2.4.3 Goell’s Circular Harmonic Method

Goell [109] introduced the first accurate method for computing the prop-
agation constants and mode profiles of rectangular dielectric waveguides.
The method is based on the expansion of the mode fields in circular har-
monics (Bessel functions in radius and sinusoids in azimuth). For a time,
it was the gold standard method against which all other approximate
methods were compared. Today if one requires a high degree of numeri-
cal accuracy, finite element methods can generally provide high accuracy
with better flexibility.
2.4 Analysis Methods for Rectangular Dielectric Waveguides 25

2.4.4 Finite Element Method

The finite element method (FEM) applied to waveguide analysis is a vari-

ational method for solving the Helmholtz equation along faceted ele-
ments of a 2D mesh filling a waveguide cross section and surrounding
cladding. The accuracy depends on the number of elements used to ap-
proximate the structural geometry. It has the advantage of being able to
discretize the region of an arbitrarily shaped waveguide cross section in
an adaptive manner. This allows increased numerical accuracy by adding
more elements in regions where the fields are expected to be rapidly vary-
ing (such as near dielectric discontinuities). Although somewhat difficult
to implement, it is both flexible and highly accurate particularly for com-
puting the full vectorial nature of guided mode fields. Many commercial
mode-solving packages implement this technique.

2.4.5 Beam Propagation Method

The beam propagation method is the name often given to fast Fourier
transform (FFT) methods of solving the paraxial diffraction equation [97,
110]. As the name implies, this is a method that can be used to predict
the evolution of an arbitrary field distribution injected into one end of a
waveguiding structure.
The scalar Fresnel paraxial diffraction equation applied to waveguides
results from allowing the transverse mode profile A to evolve slowly in z
by dropping the second longitudinal derivative in Eq. 2.13
∂A(x, y, z) i 2 i
= ∇ A+ (n(x, y)2 k20 − β2 )A.
∂z 2β T 2β
∂A(kx , ky , z) i i
=− (k2 + k2y )A + (n(kx , ky )2 k20 − β2 ) A.
∂z 2β x 2β
Here, we present the form of the equation in both the spatial (x, y) and
the wavevector (kx , ky ) or Fourier domains. The wavevector spectrum as-
sociated with a field is directly related to the angular spectrum through
the transformation n sin(θx,y )/λ = kx,y . The first term on the right side
of each equation corresponds to Fresnel diffraction. In the spatial domain,
the second-order derivatives associated with the transverse Laplacian op-
erator are cumbersome to compute. In the wavevector domain, however,
the plane wave or angular components of the field advance with a propa-
gation constant that exhibits a simple quadratic dependence with spatial
frequency. Hence it is both natural and simple to numerically compute
Fresnel diffracted fields in the Fourier domain. The second term on the
right side of each equation corresponds to the offset in propagation con-
stant from the assumed β value in each region of the waveguide struc-
ture. The index variation associated with a waveguide is represented in
26 2. Optical Dielectric Waveguides

this term and is of course defined locally in space. It follows that the
spatial domain is the natural domain in which to apply this perturbation
(the numerically taxing convolution operation denoted by is avoided).
The BPM takes advantage of the natural solution domains and solves the
equation alternately in the Fourier domain for Fresnel diffraction and in
the spatial domain for refractive index waveguiding. The main advantage
of this method is its speed. Limitations of the method are restriction to
paraxiality and low index contrasts as well as an inability in its basic form
to handle retro-reflections (which of course are highly non-paraxial).
The modes of a waveguide may be determined through use of the BPM
via impulse response methods. A spatially and temporally localized field
(such as a short-pulsed point source) is introduced into the waveguide
and is allowed to propagate. The spatial spectral content that does not lie
within one of the modes of the structure radiates into continuum modes
and what is left behind after a suitable propagation length is a superposi-
tion of all the exited modes of the structure. By Fourier transforming the
resulting field distribution both along the propagation axis, and in time,
the modes are clearly defined as sharp peaks in the propagation constant
spectrum for each spectral frequency.

2.4.6 Finite-Difference Time-Domain Method

The finite-difference time domain (FDTD) method typically refers to the

Yee method of solving Maxwell’s equations [111, 112]. As a result, it can
be very accurate for the solution of compact photonic structures, par-
ticularly those involving fully vectorial, highly nonparaxial propagation
(associated with high index contrasts) and rapidly time varying (ultra-
fast) pulse envelopes. The primary drawback associated with this method
is its computational intensiveness. For example, 1/20th wave resolution
at λ = 1 µm, over a 10-µm2 window results in a 40,000 count pixel
matrix for 2D (and 8,000,000 for 3D) in which the discretized form of
Maxwell’s equations must be solved every 1/20th of an optical cycle to
satisfy numerical dispersion conditions (Courant). Another drawback is a
somewhat inflexible grid. Despite these limitations, FDTD methods have
become very common for simulating the transient evolution of fields in
ring resonators and photonic crystals. The modes of a photonic structure
can be obtained much in the same way as with the BPM by introducing
temporal impulses and by spectrally resolving the fields after some suit-
able propagation distance or time to determine the dispersion relations.

2.5 Coupling

When two waveguiding cores are situated in close proximity to each other,
optical power can be exchanged between their supported modes. The
analysis of this problem can proceed by considering the collective normal
2.5 Coupling 27

Fig. 2.7. Coupling of two slab waveguides. The horizontal lines repre-
sent the interfaces between the core and the cladding. The direction of
propagation is left-to-right (or vice versa).

modes of the two waveguides to be superpositions of the modes of the in-

dividual guides. In the case of single-moded waveguides, both symmetric
and antisymmetric solutions exist for the coupled structure. In general,
the symmetric and antisymmetric modes possess different propagation
constants. If all the power is initially in one waveguide, the field distribu-
tion can be represented as one particular superposition of the symmetric
and antisymmetric modes adding coherently in that waveguide’s core.
Since the two modes have different phase velocities, over some length
(the “beat” length), optical power is transferred to the other waveguide
and continues to slosh back and forth while the two waveguide cores
maintain their close proximity (Fig. 2.7). The beat length depends on the
separation of the two waveguide cores. If the cores are far apart, the beat
length is infinitely long, and for all practical purposes, the modes will not
couple. Coupling coefficients can be calculated from the effective indices
of the symmetric and antisymmetric modes or by using the modes of
the individual waveguides and applying a perturbation-based approach
described in what follows.

2.5.1 Perturbation Method for Deriving Coupling Coefficients

There are a variety of methods for computing distributed coupling coef-
ficients κj,k . Finite-difference time-domain methods can give this result
directly but are computationally intensive. Finite element methods are of-
ten better suited to waveguides with non-Cartesian or noncircular cross
sections. By solving for the normal modes of the composite structure, one
can obtain the propagation constants for symmetric and antisymmetric
modes that are related to the coupling coefficient via κ = |∆β|/2. We next
28 2. Optical Dielectric Waveguides

present an analytic method that can be employed if the fields associated

with the coupled modes can be defined exactly or approximately with
analytic expressions.
The derivation of the coupling coefficients between two modes pro-
ceeds assuming that the field associated with mode j induces a polariza-
tion within the core of mode k. We begin by writing Maxwell’s vector curl
equations for the two modes.

∇ × Ek = +iωµ0 Hk (2.68)
∇ × Ej = +iωµ0 Hj (2.69)
∇ × Hk = −iωεEk (2.70)
∇ × Hj = −iωεEj − iωPper,j , (2.71)

where the asymmetry results from the fact that we are allowing mode k
to be perturbed by mode j inducing polarization Pper,j . Combining these
equations results in:

∇ · Ej × H∗ ∗
k = +iω µ0 Hj · Hk − ε Ej · Ek


∇ · E∗ ∗ ∗
k × Hj = −iω µ0 Hj · Hk − ε Ej · Ek − Pper,j · Ek


Adding these equations and integrating both sides across the transverse
dimension results in:

∇ · E j × H∗
k + E ∗
k × H j dxdy = iω Pper,j · E∗
k dxdy. (2.74)

Gauss’s theorem eliminates the transverse components of the divergence

operator yielding:

∂ ∗ ∗
EjT × HkT + EkT × HjT dxdy = iω Pper,j · E∗
k dxdy. (2.75)
where the “T” subscripts refer to the transverse components of the fields.
The orthonormality of modes can be written in the following manner:

z · ejT x, y × hkT x, y dxdy = δj,k , (2.76)
where the variables e and h represent the field strengths for modes nor-
malized to possess unit power per length. The Kronecker delta δj,k is zero
for j ≠ k and unity for j = k. Implementing these orthonormal fields and
extracting the longitudinal field dependence for the perturbed mode as
Ak (z), the expression for the longitudinal variation in field strength of
the perturbed field begins to take form:

∂ iω
Ak = Pper,j · Ek ∗ dxdy. (2.77)
∂z 4
2.5 Coupling 29

Finally, implementing orthonormal fields for the induced polarization

and extracting the longitudinal field dependence for the perturbing mode
Aj (z):
 !  "
∂ iω ε∆ε
Ak = ∆εejT Aj eiβj z + ejz Aj eiβj z · e∗
k e −iβk z
∂z 4 ε + ∆ε
⎧ ⎫
⎨ ik  ∗ ε ∆ε ∗
∆εr ejT · ekT + εr +∆εr ejz · ekz
r r ⎬
∂ 0
Ak = dxdy Aj ei(βj −βk )z .
∂z ⎩ 2 2Z0 ⎭

From this expression we see that a coupling coefficient can be extracted

as: ⎧ ⎫

k0 ⎨ ∆εr ejT · ekT + εr +∆εr ejz · ekz ⎬
∗ εr ∆εr ∗
κjk = dxdy . (2.80)
2 ⎩ 2Z0 ⎭

This overlap integral must be solved to determine the coupling per unit
length. In the following section, we apply this integral to the coupling
between the lowest order modes of two symmetric TE planar waveguides.

2.5.2 Coupling Between Symmetric TE Planar Waveguides

The expression for the coupling coefficient between symmetric TE planar

waveguides results from the following overlap integral:

d ∗
n21 − n22 Ey1 Ey2
κ21 = dx . (2.81)
2 2Z0 P/L

Substituting the field profile expressions derived earlier for a core sepa-
ration of s yields:

|E |2 

n21 − n2eff −γ (s+x−d/2)
κ21 = k0 n21 − n22 dx cos (kx x)  e x
4Z0 P/L n 2
− n 2
0 1 2
that can be shown to reduce to the following expression:

2 n2eff − n22 n21 − n2eff e−γx s 2b (1 − b) n21 − n22
κ21 =
= k0
√  e−γx s .
neff n21 − n22 deff 2+V b 1
b + n2 /n2 −1
1 2
30 2. Optical Dielectric Waveguides

Making the following substitutions:

n ≡ n1 /n2 (2.84)
V1 ≡ n1 k0 d = √ (2.85)
1 − n−2
G≡ , (2.86)
the coupling coefficient can be written in a slightly different normalized
√ √
√ √ √
κ21 2b (1 − b) 1 − n−2 e−V1 1−n bG 2 b (1 − b) e−V bG
√ √  →
√ ,
n1 k0 2 + V1 1 − n−2 b b + 2
1 n→∞ 2+V b
n −1
where the last form of the expression refers to the high index contrast
limit applicable to pedestal waveguides.

2.5.3 Coupled Wave Formalism

We next introduce a set of first-order coupled-wave equations that model
the exchange of power between two waves of arbitrary propagation con-
stants β1 , β2 and coupling coefficients κi,j .
A1 = iκ11 A1 + iκ21 A2 ei(β2 −β1 )z (2.88)
A2 = iκ12 A1 ei(β1 −β2 )z + iκ22 A2 . (2.89)
These equations can be combined into a single second-order differential
)1 = A1 e−i(κ11 −δ)z or A
equation for either A )2 = A2 e−i(κ22 +δ)z

+ δ 2
+ κ κ
12 21 )j = 0,
A (2.90)
where the detuning parameter is defined as 2δ = (β1 + κ11 ) − (β2 + κ22 ).
Integration of this equation along the mutual propagation direction z
yields lumped self- and cross-coupling coefficients:

r 2 = cos2 δ2 + κ 2 z + 2 sin δ 2 + κ2 z (2.91)
δ + κ2
t2 = 2 sin2 δ2 + κ 2 z . (2.92)
δ +κ 2

Here, it is assumed that κ12 = κ21 ≡ κ. Note that the lumped cross-
coupling efficiency can be limited to a value below 100% for fields that
are phase mismatched (δ ≠ 0). The maximum achievable coupling effi-
ciency (tmax = κ√2 /(δ2 + κ 2 )) is achieved at a minimum interaction length
of zmax = π /(2 δ2 + κ 2 ). Extensive treatment of this coupled mode for-
malism can be found in the literature [113–116].
2.5 Coupling 31

a1 b1

a2 b2
Fig. 2.8. Coupling of fields in a 2 × 2 directional coupler.

2.5.4 The Scattering Matrix

The lumped coupling coefficients are typically implemented in a scatter-

ing matrix formalism. Consider a set of coupled waveguides (Fig. 2.8).
The output fields are b = [b1 , b2 ]t , and the input fields are a = [a1 , a2 ]t .
The scattering matrix S = sij , that relates the output fields to the input
fields, is
b = Sa. (2.93)
Then, using power conservation and time-reversal, we can show that the
scattering matrix is symmetric and has the form [117]
rc tc
S= t , (2.94)
tc −rc∗ tc∗c

tc∗ rc + rc∗ tc = 1. (2.95)
We can choose our reference planes such that rc = −rc∗ tc /tc∗ = r ∈ R
(where R is the set of real numbers). Then, we can define t ∈ R such that
tc = ±it. Then r 2 + t 2 = 1 for lossless coupling, and eq. 2.94 becomes
r it
S= . (2.96)
it r

2.5.5 Optimized Coupling for Waveguides and Resonators

In practice, the coupling strength may need to be optimized so as to

to achieve the greatest possible coupling coefficient to keep device di-
mensions small. An accurate modeling tool (such as FDTD) should be
performed as a final check, but there are fundamental limitations on
the maximum achievable coupling per unit length. At best, a distributed
coupling of no more than κ ≈ 2n1 /λ is achievable. The required dis-
tributed coupling is obtained from the lumped cross-coupling coefficient
t from t = sin(κzint ), where zint is the effective interaction length. An
32 2. Optical Dielectric Waveguides

analytic expression for the TM4 distributed coupling in terms of normal-

ized waveguide parameters (derived earlier) is given by
√ √
2 b (1 − b) e−V b(s/d)
κ → n1 k0 √ . (2.97)
n1 >>n2 2+V b
Two crucial implications result from this high index contrast limit [118].
The first is that in the high contrast approximation, the coupling coef-
ficient is independent of index contrast (although it implicitly appears
in the normalized frequency V -parameter). The second is that for high
contrast laterally coupled TM modes,5 the normalized frequency V and
normalized gap width G completely dictate the coupling. Thus, practi-
cally, there are only two parameters that are critical and that can be cho-
sen to determine the coupling. Provided that the material parameters
and wavelength are fixed, both the guide width and the gap width must
be chosen appropriately. Even if the guides are touching, their coupling
may still be weaker than needed over a short distance because the guide
dimension may be too wide. The TE distributed coupling is more compli-
cated and cannot be written in such a compact form. Figure 2.9 shows the
variation of coupling strength versus the normalized frequency and gap
width in the high index contrast planar waveguide approximation for TM
modes. Note that it predicts a maximum value of the coupling strength
lying between V = 0 and V = π in the region of single-moded behavior.
Recall that for evanescent coupling, the evanescent tail of a mode extend-
ing beyond one guide excites the material polarization in a second guide
that in turn reradiates into its mode. For large enough guides, the por-
tion of confined optical power present in the evanescent tail is negligibly
small leading to low overlap. In the other limit, for small enough guides,
the evanescent tail becomes the dominant portion of the confined opti-
cal power and is spread widely throughout space such that only a small
fraction of it overlaps the other waveguide.
In a racetrack geometry, the coupling region is extended with a straight
waveguide section to greatly increase the coupling, at a modest reduction
of the FSR and introduction of some junction loss [119]. This workaround
can achieve a low loading finesse in small resonators where the effective
interaction length is limited. Figure 2.10 displays the required interaction
length to achieve a given loading finesse associated with a single coupler.
Other considerations for tweaking the coupling strength lie in the choice
of air-cladding, filled cladding, or rib waveguiding. In order to confine
light in a microresonator, it is necessary that the lateral index contrast
TM with respect to the substrate (vertical electric field) is really TE with respect
to the coupling interfaces for laterally coupled waveguides.
Again, to avoid confusion, the convention for distinguishing between TE and
TM modes is with reference to the substrate. A mode with a dominant electric
field perpendicular to the substrate surface is denoted TM. Such a mode field
is, however, parallel with respect to the lateral interface, and thus, the TE
Fresnel reflection and planar waveguide laws apply.
2.5 Coupling 33

Normalized coupling, κ / k1
s/d=0 s


gap/width ratio
0 π/2 π 3π/2
Normalized frequency, V=k0d NA
Fig. 2.9. Normalized coupling strength (κ/k1 ) versus normalized fre-
quency (V = k0 dNA), and normalized gap (s/d) in the limit of high index
contrast for the planar waveguide approximation to TM guide-to-disk

Fig. 2.10. The required interaction lengths for TM waveguides in the

high index contrast limit. Shown are the lengths required to achieve
a resonator loading finesse of 10, 100, and 1000. Note that in many
cases, the required interaction length varies greatly with differing gap
and guide widths indicating a need for extremely fine tolerances on de-
sign geometry.
34 2. Optical Dielectric Waveguides

be high. This implies that, in order to make the guides single-moded lat-
erally and to maintain high lateral coupling by exposing a higher fraction
of the evanescent tail of the modes, the guides would have to be very
thin in the lateral dimension. This still makes it difficult to couple a sig-
nificant fraction of light from free space or low NA single-mode fiber.
However, although it is difficult to taper the guides out in the vertical
dimension, it is generally a straightforward task in the lateral dimension.
Thus, a lateral-tapering guide with high lateral index contrast that is ini-
tially highly multi-moded can be used. In the vertical dimension, how-
ever, there is no requirement for high index contrast. As such, the input
coupling efficiency can be improved by making the guides possess a rea-
sonably large vertical core dimension that is maintained single-moded by
employing low vertical index contrast.
A finite-element solver can be used to solve for the 2D modes and cou-
pling coefficients associated with the cross section of a pair of waveguides
[Fig. 2.11]. This can be most useful for determining the coupling coeffi-
cient (per unit length) that can then be integrated (along z) to find the
required coupling length for a desired coupling coefficient. This method is
appropriate for modeling the couplers in a Mach–Zehnder or the coupling
to and from racetrack resonators in which the gap per unit length does not
vary longitudinally over most of the coupling region. For more complicated
geometries, such as that of bent resonator-to-straight waveguide coupling,
numerical techniques are generally required. The three-dimensional na-
ture of the problem can sometimes be reduced to two dimensions by
employing the effective index method. This method is implemented by
first solving for the propagation constants of equivalent planar waveguides
of regions that possess similar vertical guidance. Once the propagation
constants for differing regions is known, an effective refractive index is
assigned to them and the problem eliminates any further need to calcu-
late along the vertical dimension. These effective indices are then used to
model the photonic structure in only two dimensions by using an FDTD
solver. By measuring the power associated with fields coupled into a ring
or disk resonator after a single round-trip, the FDTD method readily gives
estimates of the coupling coefficient. Phase matching the waveguide mode
to the mode of the ring or disk may be an important consideration when
requiring a high lumped coupling strength. Specifically, ensuring that the
modes are phase matched is important when the difference in propagation
constants ∆β is greater than a critical value determined by the coupling
per unit length,
 κ and the desired lumped coupling coefficient t 2 such
that ∆β < κ (1 − t 2 )/t 2 . Fortunately, for small enough resonators, phase
matching is not a significant issue. The tolerance on maximum allowable
beta-mismatch increases in inverse proportion to the interaction length
of guide and resonator. The tolerance on the gap dimension and fidelity,
however, remain quite strict.
2.5 Coupling 35

Fig. 2.11. Finite element simulation (in the commercial software pack-
age Comsol) of the normal modes of a coupled pedestal waveguiding
structure. The n = 3.4 cores are 1-µm tall and 0.5-µm wide supported
and capped by n = 3.2 cladding layers. For a gap width of 100 nm,
the symmetric and antisymmetric modes at λ = 1.55 µm propagate
with different propagation constants (βsym = 12.85 µm−1 , βantisym =
12.62 µm−1 ). The spatial beat length is directly related to the coupling
strength and this configuration exchanges 100% power over 13.7 µm. In
these plots, the contours represent the electric field magnitude, arrows
represent the electric field direction (TM), and the shading represents the
36 2. Optical Dielectric Waveguides

2.6 Bending Loss

A critical factor in the fabrication of small ring resonators is control-
ling the loss in the ring. Loss in microrings and pedestal waveguides
is typically dominated by leakage to the substrate, scattering or contra-
directional coupling due to edge roughness [120–122] and bending
[2, 66–69].
Loss due to substrate leakage can be solved by waveguide design,
and that from edge surface roughness can be solved with process im-
provements, such as high-resolution lithography, vertical photoresist
sidewalls, and hard masks. However, bending losses are dictated by the
refractive index contrast associated with the choice of materials. The in-
dex contrast prescribes a minimum size for ring resonators in a material
system. The index contrast that determines the minimum bending ra-
dius is that in the lateral direction (the plane of the ring). The cladding
in the plane of the ring, for a pedestal waveguide, is typically air, a low-
index polymer, like amorphous polytetrafluoroethane (better known by
the DuPont brand Teflon, n = 1.33) or benzocyclobutene (n = 1.51), SiO2
(n ∼ 1.5), spin-on glasses (n ∼ 1.5), SiON (n ∼ 1.5–2.1), or polyimides
(n ∼ 1.7). Since the core in the case of III–V waveguides typically has a
refractive index >3, the minimum bend radius possible is <1 µm, and in
practice bending losses are rarely a concern with proper design.
Before delving into the origin of bending loss [2, 66–69], let us ex-
amine the effect of index contrast and waveguide type (strip-loaded
versus pedestal) on the mode profile (intensity distribution across the
waveguide); the next section establishes the connection between the in-
dex contrast and the bending loss. For optical fibers that have a (doped)
silica core, the index contrast is very small — around a few tenths of
a percent. For rib or strip-loaded III–V semiconductor waveguides, the
index contrast is generally higher around 5 − 10% in the vertical dimen-
sion. Along the lateral dimension, the index contrast is provided by the
presence of three slab-like regions with small effective-index difference,
with the central slab having slightly higher effective index than the outer
slabs. The confinement in the lateral dimension is therefore weaker than
in the vertical [Fig. 2.12(a)]. The waveguide mode is very asymmetric and
elliptical in shape. By contrast in the case of pedestal waveguides, the
index contrast is high in both dimensions but particularly in the lateral
dimension. The mode is well-confined in the waveguide core [Fig. 2.12(b)].
Bending losses result from the inability of light to remain confined
inside a bending open waveguide. Radiation takes the form of an out-
going cylindrical accompanied by an azimuthal component resulting in
spiral phase contours. In the low Q limit, the bending loss associated
with a cylindrical geometry can be computed exactly. In the high Q limit,
it makes sense to employ other methods. In a finite 2D geometry, there
is some out-of-plane radiation loss.
2.6 Bending Loss 37

Strong confinement Air


Core, n2
GaInAsP-InP: n1
n1 ∼ n3 ≈ 3.17 (InP)
n2 ∼ 3.3 − 3.5
Weak confinement

Strong confinement

Air n3 Air

Core, n2
Strong confinement
n1 ∼ n3 ≈ 3.17 (InP)
n2 ∼ 3.3 − 3.5
Fig. 2.12. Schematic of (a) a strip loaded waveguide and corresponding
mode profile and (b) a pedestal waveguide and corresponding mode pro-
file. Shown in each case are nine contours, with the innermost represent-
ing the locus of points where the intensity falls to 0.9 times the maximum,
and the outermost representing the locus of points where the intensity
falls to 0.1 times the maximum.

Analytical solutions to the problem of bending loss can be obtained

by using a conformal transformation to convert the problem to an equiv-
alent slab waveguide with a graded index [67]. Marcuse used a different
approach to provide an analytical form; he calculated the loss from the
38 2. Optical Dielectric Waveguides

10 % loss in one wavelength, µm 100

ncore = 1.5
Minimum radius for


ncore = 2.0

1 ncore = 3.5

1 1.5 2 2.5 3 3.5

Cladding refractive index
Fig. 2.13. Minimum radius for 10% loss in one wavelength versus cladding
index for various core indices.

radial component of the field at large distances from the waveguide [69].
Marcuse’s analytical form is used to plot Fig. 2.13.
Regardless of the approach used, we can estimate trends in waveguide
behavior from index contrast (the difference in index between core and
cladding). Since pedestal waveguides provide the strong index contrast
necessary for minimizing bending loss, they are are preferred for photonic
circuits with very small bending radii. Pedestal waveguides are referred to
as “tight-confining” or “strong-confining” because of the corresponding
mode shape. Although it may seem like a good idea to use ever-higher
index contrasts, in practice, high index contrast waveguides pose seri-
ous problems. Fabrication tolerances are reduced because the guide di-
mensions must be much smaller than low-index-contrast waveguides to
maintain single-mode behavior. Finally, since the modes are much smaller
than in conventional fibers, coupling inefficiencies can be quite high. In
the next section, we reexamine the phenomena of bending loss from the
formalism of whispering gallery modes.

2.7 Whispering Gallery Modes

In 1912, Lord Raleigh analyzed the curiosity of the “whispering gallery”

that involved the propagation of acoustic waves skimming along the in-
terior of the dome of St. Paul’s cathedral [1]. The term has since come to
be applied to a family of modes of a cyclical curved interface such as a
2.7 Whispering Gallery Modes 39

Fig. 2.14. Geometry for a TM whispering gallery mode.

cylindrical or spherical surface, where wave propagation is confined pri-

marily to the inside surface of an interface and guided by it by repeated
Light propagating in a conventional optical waveguide is confined via
total internal reflection by two dielectric interfaces (along each dimen-
sion) to a region of high refractive index. The guiding region of high
refractive index is formally equivalent to a potential well wherein the
electric fields may be decomposed into eigenmodes that are solutions
of the Schrödinger equation. Light propagating in a curved waveguide
is still guided via TIR at the outer interface, but it no longer demands
an inner interface to complete the confinement. Elimination of the in-
ner boundary leaves a dielectric disk that supports whispering gallery
modes. These modes consist of azimuthally propagating fields guided
by TIR at the dielectric interface and “optical inertia” that prevents the
field from penetrating inward beyond a fixed radius termed the inner
caustic. Mathematically, a whispering gallery mode (WGM) is a solution
of the Helmholtz equation in a curved coordinate geometry as in Fig. 2.14.
Attention is restricted to a cylindrical geometry appropriate for the analy-
sis of planar disk and ring resonators.

2.7.1 TM Whispering Gallery Modes

The Helmholtz equation ∇2 + k2 Ez = 0, written in cylindrical coordi-
nates for the axial field of a TM6 whispering gallery mode is

∂2 1 ∂ 1 ∂2 2
+ + 2 + k Ez (r , ϕ) = 0. (2.98)
∂r 2 r ∂r r ∂ϕ2
The method of separation of variables can be employed to simplify
this equation by splitting it into two separate equations for radial and
azimuthal dependance. An integer7 parameter m is introduced to con-
nect the two equations. This parameter physically corresponds to the
Once again, the accepted nomenclature for assigning TE or TM designations
for cylindrical whispering gallery modes, is to follow the conventions for that
of planar waveguides. For example, in the limit of an infinitely long cylinder,
only one electric field component exists in the TM mode solution - the one
directed axially. As a result, the TE and TM designations with respect to the
sidewall seem to be reversed.
The restriction that m be an integer assumes a resonance.
40 2. Optical Dielectric Waveguides

number of optical cycles (effective wavelengths) the field undergoes in

space when completing one revolution around the disk. The azimuthal
equation takes the form,

∂2 2
+ m Ez (ϕ) = 0, (2.99)

and has solutions that are simply complex exponentials:

Ez (ϕ) = e±imϕ . (2.100)

The resulting (Bessel) equation for the radial field dependence becomes:

∂2 1 ∂ 2 m2
+ + k − 2 Ez (r ) = 0. (2.101)
∂r 2 r ∂r r

The solutions of Bessel’s equation are the Bessel functions of the first
Jm and second Ym kind. Because the second kind function is singular at
the origin, only the first kind function is retained inside the disk. Outside
the disk, both functions are well behaved8 and must be retained. The
Hankel functions are linear superpositions of the two Bessel function
(1) (2)
solutions corresponding to outward Hm = Jm + iYm and inward Hm =
Jm − iYm propagating cylindrical waves. The analysis of waves arriving
back at the resonator from the radial horizon is not considered here,
and thus, only the Hankel function of the first kind, Hm is retained.
The appropriate solutions for the radial field dependence both interior
(r < R) and exterior (r > R) to the dielectric disk boundary [123] are

)1 r )
Ez (r < R) = Am Jm (k (2.102)
(1) )
Ez (r > R) = B m Hm (k2 r ), (2.103)

where a complex propagation constant and frequency k ) j = nj ω

*/c are
introduced for reasons that will become apparent later. The complete
axial electric field interior and exterior to the disk is constructed from the
azimuthal and radial solutions, including the boundary condition at the
interface (r = R) that forces the tangential electric field to be continuous:

Ez (r , ϕ) = Am Jm k )1 r ei(±mϕ−* ωt )

Jm k )1 R

(1) )
Ez (r , ϕ) = Am (1)
Hm k2 r ei(±mϕ−*ωt )
. (2.105)
Hm k )2 R

The fields outside the disk are not modified Bessel functions of the first kind,
Km as in the case of bound modes of a circular dielectric waveguide or optical
fiber because the absence of an axial propagation constant eliminates the
possibility of modified Bessel function solutions.
2.7 Whispering Gallery Modes 41

Finally, the radial and azimuthal magnetic field components are easily
derived from the axial electric field by use of Maxwell’s equations

−i 1 ∂ m
Hr = Ez = Ez (2.106)
)0 r ∂ϕ
Z0 k )0 r
Z0 k
i ∂
Hϕ = Ez . (2.107)
)0 ∂r
Z0 k

2.7.2 TE Whispering Gallery Modes

The Helmholtz equation for the axial field of a TE whispering gallery

mode is 
∂2 1 ∂ 1 ∂2 2
+ + 2 + k Hz (r , ϕ) = 0. (2.108)
∂r 2 r ∂r r ∂ϕ2
The equation is again simplified via the method of separation of variables,
and the azimuthal equation takes the form,

∂2 2
+ m Hz (ϕ) = 0, (2.109)

possessing complex exponential solutions, Hz (ϕ) = e+imϕ , e−imϕ . The

radial equation:

∂2 1 ∂ 2 m2
+ + k − 2 Hz (r ) = 0, (2.110)
∂r 2 r ∂r r

is Bessel’s equation with equivalent solutions as in the TM case. The ap-

propriate solutions for the radial dependence of the magnetic field are

)1 r )
Hz (r < R) = Am Jm (k (2.111)
(1) )
Hz (r > R) = Bm H m (k2 r ). (2.112)

The complete axial magnetic field interior and exterior to the disk is con-
structed from the azimuthal and radial solutions, including the boundary
condition at the interface (r = R) that forces the tangential magnetic field
to be continuous:

Hz (r , ϕ) = Am Jm k )1 r ei(±mϕ−* ωt )

Jm k )1 R

(1) )
Hz (r , ϕ) = Am (1)
Hm k2 r ei(±mϕ−*
ωt )
. (2.114)
Hm k 2 R

Finally, the radial and azimuthal electric fields are easily derived from
the axial magnetic field by use of Maxwell’s equations,
42 2. Optical Dielectric Waveguides

iZ0 1 ∂ −mZ0
Er = Hz = Hz (2.115)
n k0) r ∂ϕ )0 r
n2 k
−iZ0 ∂
Eϕ = Hz . (2.116)
n2 k)0 ∂r

2.7.3 Radiation Loss

For an open-boundary structure such as a dielectric disk,9 whispering

gallery modes are inherently “leaky” [124]. The mechanism for loss is
a tunneling or coupling of the azimuthally guided mode into radially
outward-going radiation modes. This phenomenon is directly related to
bending loss and is sometimes referred to as whispering gallery or radi-
ation loss. It is a property of any curved open boundary waveguide con-
figuration. A convenient measure parameterizing this loss is the intrinsic
quality factor that quantifies the number of optical cycles a mode will
last confined within a resonator. Theoretical predictions show that the
intrinsic quality factor (Q) can be as high as 1011 , and experiments have
confirmed this fact [125].
Intuitively, the WGM is confined within a radial potential well (m/nr )2
that can also be expressed as a effective radial index nr :

nr = n2 − 2 . (2.117)
k0 r 2

The inner caustic boundary, below which the “optical inertia” is too great,
is defined at R1 = nm 1 k0
. There is an outer radiation boundary, beyond
which the azimuthal phase velocity exceeds the speed of light in vacuum
is defined at R2 = nm2 k0
. Figure 2.15 depicts an optical whispering gallery
mode with the associated radial potential well and illustrates the regions
where the fields in the mode are bounded, evanescent, and radiative
(outwardly propagating). The radiation loss associated with whispering
gallery resonators may be viewed physically as a tunneling of the confined
field through a potential barrier defined by the disk edge and radiation
boundary into a region of lower potential. Beyond the radiation boundary,
the radially evanescent tail of the field becomes propagating again. For a
typical high Q disk resonator, the field has decayed to such a low value
that its leakage into cylindrical radiating waves is very small. The pre-
dominant propagation direction for a WGM is, of course, primarily in the
azimuthal direction such that the phase contours behave like revolving
spokes of a wheel. The pattern revolves about the disk center with a angu-
lar frequency of ω/m. Because the azimuthal phase contours increase in
separation with radius, the azimuthal phase velocity likewise increases
As opposed to a closed-boundary structure, i.e., dielectric guiding region with
perfectly conducting walls.
2.7 Whispering Gallery Modes 43

Fig. 2.15. Propagation constant kφ as a function of radial distance from

disk axis for a whispering gallery mode. The circulating power is confined
between an inner caustic and the disk edge. Beyond the outer radiation
boundary, the radially evanescent field becomes propagating and acts as
a loss mechanism.

without bound. At the radiation boundary, the azimuthal phase veloc-

ity is equal to the phase velocity in the surrounding medium. Beyond
this radius, the fields cannot keep up and thus spiral away. Figure 2.16
graphically illustrates this fact for a very low Q WGM.

2.7.4 WGM Dispersion Relations (Resonance Maps)

In order to calculate the mode propagation constants and quality fac-
tors for particular WGMs, one must solve the complex WGM dispersion
relation. The dispersion relation for whispering gallery modes is similar
to the relation for fiber modes with the exception that the axial propaga-
tion constant kz is much smaller and perhaps negligible. The dominant
propagation constant is of course directed in the azimuthal direction.
For an infinite cylinder and a longitudinal propagation constant of zero,
the dispersion relation has eigensolutions with complex propagation con-
stants, which implies that either the refractive index and/or the frequency
44 2. Optical Dielectric Waveguides

Fig. 2.16. Plot of the electric field associated with the sixth order TM (axial
E-field) whispering gallery mode. Here n1 = 2, n2 = 1, and the resonant
radius (solid line) is 1.04 µm at λ = 1.55 µm. This configuration was
chosen because it is poorly confined with a low Q of 64 and thus allows
easy visualization of the super-evanescent component of the field below
the caustic radius (inner dotted line) and the radiating component of the
field past the radiation boundary (outer dotted line).

must be complex — a consequence of the radiation losses present in the

system. Should the disk possess a negative imaginary component of the
refractive index (representing gain), solutions exist that maintain a con-
stant field energy in the resonator while power is steadily radiated. In
the case of a lossless/gainless dielectric resonator, however, solutions of
the equation necessarily involve a complex frequency, implying a decay
rate of energy confined within the resonator. The complex frequency is
related to the temporal decay rate γ according to ω* = ω − i 2 , and a qual-
ity factor may be ascribed Q = γ defined as the characteristic number
2.7 Whispering Gallery Modes 45

of optical cycles before confined energy is lost to the radiation contin-

uum. The Pythagorean theorem does not apply for the variables ω, γ as
it does for the traditional dispersion relation variables for a waveguide
β, kx . This fact along with the presence of discrete resonances give
the dispersion relation a different character — thus, perhaps “complex
resonance map” is a more appropriate description. Obtaining solutions
involves solving for the complex roots of a complex equation. Many ap-
proximations have been employed to simplify this equation, for example,
conformal transformation [67, 68, 126], WKB [127], and volume current
methods [120, 128]. Because the validity of these methods are in doubt
when the resonator circumference approaches a small number of optical
cycles, we choose to solve the equation numerically from the dispersion
For an infinite cylinder of dielectric material with negligible absorp-
tion and no axial component of the propagation vector, we proceed to
derive the TM WGM dispersion relation. The field matching equations at
the disk boundary (r = R) are expressed as:

A1m Jm k )1 R = A2m H (1) k )2 R (2.118)

)1 A1m J ) ) (1) )2 R .
k m k1 R = k2 A2m Hm k (2.119)

The TM dispersion relation can be written as

)1 J k )1 R ) 2 Hm (1) )
k m k k2 R

. (2.120)
Jm k )1 R (1) )
Hm k 2R

Next, we proceed to derive the TE WGM dispersion relation. The field

matching equations at the disk boundary (r = R) are expressed as:

A1m Jm k )1 R = A2m H (1) k )2 R (2.121)

A1m Jm )1 R = 1 A2m H (1) k
k )2 R . (2.122)
k1 )2
The TE dispersion relation can be written as

)1 R (1) )
Jm k Hm k 2R

. (2.123)
)1 Jm k
k )1 R ) 2 Hm
(1) )
k2 R

The complete vector WGM dispersion relation is obtained by combining

the TE and TM equations

⎤ ⎡

)1 J k )1 R ) 2 Hm (1) ) )1 R (1) )
k m k k2 R Jm k Hm k 2R


⎦ = 0.
Jm k )1 R (1) )
Hm k 2R
) 1 Jm k
k )1 R ) 2 Hm
(1) )
k2 R
46 2. Optical Dielectric Waveguides

Note the equivalence with the complete vector dispersion relation for
circular step index fibers (for β = 0):


k)1 J k )1 R )2 K −ik
k )2 R Jm )1 R
k Km
−ik)2 R
m m

⎦ = 0,
Jm k )1 R −iKm −ik )2 R ) 1 Jm k
k )1 R )2 Km −ik
−ik )2 R
where the following relations hold:

iπ i mπ (1)
Km (−iz) = e 2 Hm (z) (2.126)

iπ i mπ (1)
−iKm (−iz) = e 2 Hm (z) (2.127)
Km (−iz) Hm (z)
= (1)
. (2.128)
Km (−iz) −iHm (z)

2.7.5 Normalized WGM Dispersion Relations (Resonance Maps)

In terms of the quality factor, the TM and TE dispersion relations can be

written, respectively, as


2π R 1 (1)

n1 Jm n1 λ 1 − i 2Q n2 Hm n2 2πλ R 1 − i 2Q1

2π R 1 (1) 2π R 1
Jm n1 λ 1 − i 2Q Hm n2 λ 1 − i 2Q


Jm 1
n1 2πλ R 1 − i 2Q 1
Hm n2 2πλ R 1 − i 2Q

 . (2.130)
2π R 1 (1) 2π R 1
n1 Jm n1 λ 1 − i 2Q n2 Hm n2 λ 1 − i 2Q

For a given index ratio, n = n1 /n2 and azimuthal mode number m, a

normalized radius X = n1 2π R/λ, and intrinsic quality factor Q may be
obtained for a particular WGM solution. Equations 2.130 can be rewritten
in terms of the normalized radius and the intrinsic, radiation-limited Q
or finesse (F = Q/m).


1 (1) 1
Jm X 1 − i 2mF Hm X/n 1 − i 2mF
1 (1) 1
Jm X 1 − i 2mF Hm X/n 1 − i 2mF


1 (1) 1
Jm X 1 − i 2mF Hm X/n 1 − i 2mF

 = n 
 . (2.132)
1 (1) 1
Jm X 1 − i 2mF Hm X/n 1 − i 2mF

To solve for the roots of these equations, a global optimization scheme

can be used to minimize the absolute value of each equation over two
2.7 Whispering Gallery Modes 47

variables: the real and imaginary parts of the propagation constant or

X and Q in the normalized formulation. Using this method, generalized
plots of limiting finesse against normalized radius may be obtained for
the whispering gallery modes of a dielectric cylinder. Figure 2.17 displays
the Radiation-loss-limited finesse vs. normalized radius for a variety of
azimuthal mode numbers and index ratios.

8 n1/n2 = 3.5
7 Eφ 3.0
6 Er 2.5
Log10 F

5 2.0

4 1.7 100
3 90
2 1.5 70
1 1.35 60
0 1.2530 40
m = 2 5 10 20
0 20 40 60 80 100
Normalized radius, (n1ω/c)R

8 n1/n2 = 3.5
Ez 3.0
7 Hφ
6 Hr
Log10 F

4 1.7
3 90
2 1.5 70
1 1.35 60
1.25 40
0 30
m = 2 5 10 20
0 20 40 60 80 100
Normalized radius, (n1ω/c)R

Fig. 2.17. Radiation-loss-limited finesse of the lowest order radial TE

and TM whispering gallery modes of a dielectric cylinder of index n1
in a medium of index n2 plotted against normalized radius (n1 ω/c) R.
The family of diagonal lines represents varying refractive index contrast
(n1 /n2 ). The family of nearly vertical lines corresponds to whispering
gallery mode resonances, each characterized by an azimuthal mode num-
ber m. The plots were obtained by numerically solving the dispersion re-
lation for whispering-gallery modes. (After [129], ©2002, Optical Society
of America.)
48 2. Optical Dielectric Waveguides

2.7.6 Spheres, Rings, and Disks

The quality factors associated with silica microspheres [86–88, 94, 125,
130–138] have attracted much attention for their ultra-high Q factors.
Their traditional method of production relies on melting silica (usually a
fiber tip) and allowing surface tension forces to reshape the liquid glass
into a sphere that is then allowed to cool. Unfortunately, this method
does not allow for the fabrication of spheres of reproducible diameters,
nor does it allow for spheres much smaller than 20 µm. Recent advances,
however, have shown that selective reflow of patterned silica disk edges
can result in resonators possessing ultra-high Q factors typically found in
microspheres [139]. Propagating whispering gallery modes have been im-
aged by scanning a near-field probe across the surface of fused-silica mi-
crospheres [140, 141], and silicon nitride cylindrical disks [142] whereby
light is collected (as it is injected [143]) via frustrated total internal re-
flection. In the past decade WGMs have attracted increasing attention
for their ability to sustain very high Q-factors and low mode volumes.
Microdisks and microrings constructed using the techniques of micro-
fabrication and nanofabrication, however, can be readily constructed in
a reproducible manner to designed dimensions down to dimensions less
than 1 µm. The main drawback of planar fabricated microresonators is
the surface roughness left on the edges due to etching processes that re-
sult in resonators of much lower quality factors. A ring has two bounding
edges that can be used to properly design a single-moded guide. A disk
only has one bounding edge; the other boundary is an effective one arising
from an inner caustic. As a result, a disk can possess less scattering loss
than that of a corresponding ring geometry but is multi moded. If the FSR
is high, and/or the side coupling guides preferentially couple to one of
the radial modes of the disk, then disks are preferable. If not, then a ring
geometry is better and the extra loss must be taken into account [144].
A mode may be considered a “whispering galley mode” if the confine-
ment along some dimension is provided by only a single reflective in-
terface. For a given core-to-cladding refractive-index difference, the loss
at a given bend radius decreases with increasing waveguide width until
a limit is reached where only the outer core interface is important for
guiding. A mode of a curved waveguide (forming a ring) defined by two
interfaces would be considered a whispering gallery mode if the inner
caustic radius (defined by the azimuthal index m and the wavelength)
lies between the inner and outer interfaces. In this regime, light cannot
penetrate (towards the origin) beyond the inner caustic, and thus, the
interior interface plays a negligible role in the guidance. Thus, a curved
waveguide will have similar bending loss per unit radian as a disk with
the same exterior radius. It is worthwhile pointing out the advantages
and disadvantages of disks and rings. A microdisk may possess higher
order radial modes primarily depending on the location of the inner caus-
tic. These radial modes possess differing resonant wavelengths and thus
2.7 Whispering Gallery Modes 49

Disk Single- Azimuthal #
moded m=8
100 ring
Radial potential, µm-2

Radial Index contrast

potential n = 2.5 : 1.0

60 3rd mode λ=874nm, Q=72

40 2nd mode λ=1079nm, Q=224

1st mode
λ=1438nm, Q=3247

0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Radius, µm
Fig. 2.18. A plot of the three supported TM radial modes of an m = 8,
1-µm radius microdisk resonator with index contrast n = 2.5 : 1. The res-
onance wavelengths for the first, second, and third modes are 1438, 1079,
and 874 nm, respectively. Mode quality factors are 3247, 224, and 72, re-
spectively. The suppression of the higher order modes can be achieved
by eliminating the interior of the disk to form a ring, here illustrated by
a dashed line.

might be discriminated against by properly choosing the excitation wave-

length. Depending on the quality factors associated with the modes, the
resonant wavelength of a particular mode of a microdisk (defined by a
radial and azimuthal number) might still overlap another. This may be
a problem if only a single-mode is desired.10 A properly designed ring
can be used instead to force single-radial-mode operation. However, as
will be investigated in more detail later, the presence of an extra sidewall
contributes to additional scattering losses. Figure 2.18 shows the radial
field distribution associated with the TM modes of a 1-µm-radius disk res-
onator supporting three radial modes. Each mode possesses an “energy,”
Although the modes may overlap, the higher order modes will typically posses
a lower coupling coefficient and higher loss.
50 2. Optical Dielectric Waveguides

k20 that lies within the potential energy well (m/nr )2 , which is dictated
by the azimuthal number and structure.11 A ring resonator formed by
removing the material within a 0.65-µm radius would cut off the second-
and third-order radial modes. Its potential energy distribution, shown as
a dashed line, only supports the lowest order mode.

2.8 Scattering Losses Resulting From Edge Roughness

For microresonators constructed using planar fabrication technology, the

dominant loss mechanism is scattering due to edge roughness. In the
following discussion we develop further the volume current-based analy-
sis of Kuznetsov, Haus, and Little [120, 128], Borselli et al. [145], and
Rabiei [146]. For the design of other waveguide geometries, generalized
formulations such as that of Kogelnik and Ramaswamy [99] have been
instrumental both in fostering intuition and in making early, high-level
design choices. The following derivation proceeds in the same spirit.

2.8.1 Volume Current Method Formulation for Scattering Losses

The volume current method is used to determine edge scattering losses

associated with microdisk resonators in the whispering gallery mode
regime. Only the fundamental whispering gallery mode characterized by
a single radial lobe is considered. The time-independent modal field am-
plitude for a WGM can be decomposed into a transverse mode profile and
an azimuthal phase dependence:

E(r , z )eimϕ (2.133)

where E is the cross section of the modal field amplitude and m is

the azimuthal quantization number. A dielectric perturbation on the
disk edge may be written as a spatially dependent permittivity distribu-
tion ∆ε(r , z , ϕ ). The dielectric perturbation introduces dipole currents
that contribute to outwardly radiated (scattered) fields. The expressions
for the perturbed current densities manifest themselves in the form of
surface-parallel and surface-perpendicular contributions.

2.8.2 Current Density Contributions

The surface-parallel current contribution is readily derived starting from

the curl of the Maxwell equation:
Strictly speaking, these “energies” do not have the correct units of energy.
However, they are formally mathematically equivalent to the energies involved
in the solution of the radial Helmholtz equation.
2.8 Scattering Losses Resulting From Edge Roughness 51

∇×J = − ∇×D+∇×∇×H (2.134)
= +iω∇ × (εE) + ∇ × ∇ × H (2.135)
= +iω∇ε × E + iωε∇ × E + ∇ × ∇ × H (2.136)
= +iω∇ε × E − ω µ0 εH + ∇ × ∇ × H (2.137)
= +iω∇ε × E. (2.138)

By definition, the permittivity gradient is oriented normal to the inter-

face; taking the line integral from just below to just above the interface
eliminates the curl:
 +  +
J = lim du û × (∇ × J) = lim du iωû × (∇ε × E ) = −iω∆ε E .
→0 − →0 −
The surface-perpendicular current contribution is readily derived start-
ing from the continuity relation in the absence of free charges. In com-
bination with the expanded Maxwell equation for the divergence of the
displacement vector:
∇·J = − = iωε0 ∇ · E (2.140)
∇·D = ε∇ · E + ∇ε · E = 0 (2.141)
∇·J = −iωε0 2 · D. (2.142)
Again, the permittivity gradient is oriented normal to the interface; taking
the integral eliminates the divergence:

J⊥ = û du∇ · J = iωε0 du∇ D⊥ = iωε0 ∆ ε−1 D⊥ . (2.143)
The current density associated with boundary-continuous parallel elec-
tric and perpendicular displacement fields in the presence of the dielec-
tric perturbation is thus given as

J(r , z , ϕ ) = −iω ∆εE (r , z ) − ε0 ∆ ε−1 D⊥ (r , z ) eimϕ . (2.144)

The dielectric perturbation at the disk boundary can be written as a ra-

dial step variation at the interface between the two dissimilar refractive
indices (n1 for the core, and n2 for the cladding),
+ ,
∆εin = ε0 (n22 − n21 ) step −∆R(z , ϕ ) (2.145)
2 2
∆εout = ε0 (n1 − n2 ) step ∆R(z , ϕ ) (2.146)

−1 1 1 1 + ,
∆(εin ) = 2
− 2 step −∆R(z , ϕ ) (2.147)
ε0 n2 n1

−1 1 1 1 + ,
∆(εout ) = 2
− 2 step ∆R(z , ϕ ) . (2.148)
ε0 n1 n2
52 2. Optical Dielectric Waveguides

The interpretation is that the field just inside the interface is perturbed
by a lower permittivity when the radial variation is negative and that the
field just outside the interface is perturbed by a higher permittivity when
the radial variation is positive.

2.8.3 Spectral Density Formulation for Edge Roughness

The mode is primarily affected by perturbations along the direction in

which the propagation vector is dominant — here azimuthal. Moreover,
etch processes tend to deliver uniform corrugations along height, z. This
justifies a decomposition of the radial variation into a Fourier series ex-
pansion of corrugation harmonics of azimuthal quantization number M

∆R(z , ϕ ) = ∆RM (z )e−iMϕ . (2.149)

The amplitude of each corrugation harmonic may be determined from an

experimental measurement of the corrugation. Typically, Gaussian statis-
tics apply as in Borselli [145] where the roughness can be characterized
by two parameters: the rms value of the roughness σ and its correlation
length Sc . Here, the correlation function for edge roughness, C(s) is re-
lated to the measured variation in radius ∆R(s) with respect to arc length
s at the disk edge over a suitable measured total arc length Smeas .

C(s) = ds ∆R(s )∆R(s − s). (2.150)

The value of the correlation function at zero is equal to the mean squared
roughness C(0) = σ 2 . A Gaussian correlation function can be defined as:

C(s) = σ 2 e Sc . (2.151)

 written in this form, the correlation length is within a factor of
π /4 ln 2 = 1.064 of a full width at half maximum (FWHM) definition.
The result of further manipulations is also cleaner, hence, the motivation.
The spectral density is equal to the Fourier transform of the correlation
function, which is a Gaussian function of spatial frequency variable fs ,
C(fs ) = σ 2 Sc e−π (Sc fs ) . (2.152)

If the entire circumference of a disk were to be mapped, a Fourier se-

ries representation with harmonics of integer azimuthal quantization
numbers would emerge naturally. In practice, it is not often feasible to
measure the entire circumference; thus, the amplitude coefficients of the
2.8 Scattering Losses Resulting From Edge Roughness 53

Fourier series expansion of the corrugation must be extrapolated from

the limited data. To obtain those amplitudes, the spatial frequency vari-
able is expressed as fs = M/2π R and the spectral density is integrated
around each integer M.
M+ 12
2 1 M
∆RM = dM C (2.153)
2π R 2π R
M− 2

M+ 12

Sc −π
2π R M
= σ2 dM e (2.154)
2π R
M− 12

Sc −π
2π R M
≈ σ2 e . (2.155)
2π R
For most cases of interest, in comparison with the disk circumference, the
correlation length is very small Sc /2π R  1 allowing the final trapezoidal
approximation to the integral to hold.

2.8.4 Far-Field Scattered Power

Returning now to the electrodynamics of scattering, the vector potential

in the far field consists of the volume-integrated current density vector
with a retardation phasor term to account for coherent interaction among
the current density elements,

A= dV J(r , z , ϕ )eikr cos ψ . (2.156)
4π r
Here, the volume integral is represented in cylindrical coordinates appro-
priate to the geometry of the circulating mode while the far-field scatter-
ing direction is represented in spherical coordinates (see Fig. 2.19). The

Fig. 2.19. The geometry used in the volume current method formula-
tion for edge scattering losses in microresonators, here shown for a mi-
crodisk. The roughness perturbations on the disk edge are parameterized
in cylindrical coordinates (r , z , ϕ ) whereas the scattered radiation is
parameterized in spherical coordinates (r , θ, ϕ).
54 2. Optical Dielectric Waveguides

angle cosine between the current density element and the observation
point is expanded in spherical coordinates as
cos ψ ≈ cos(θ) cos(θ ) + sin(θ) sin(θ ) cos(ϕ − ϕ ). (2.157)
For most geometries in which the disk height is smaller than the radius
d  R, the small polar angle approximations about θ = 90◦ can be made,
sin(θ ) ≈ 1 (2.158)
cos(θ ) ≈ . (2.159)
Incorporating these approximations results in a volume integral written
completely in cylindrical coordinates:


µ0 cos θ ikr sin(θ) cos(ϕ−ϕ )
A= dz dr r dϕ J(r , z , ϕ )eikz e .
4π r
−d/2 0 0
Note that the integrated vector potential will be azimuthally independent
due to the inherent symmetry of the geometry yet retain a polar depen-
dence. In the far field, the electric and magnetic fields and Poynting vector
are expressed in terms of the vector potential as
EFF = iωr̂ × (A × r̂) (2.161)

HFF = −iω ε2 /µ0 (A × r̂) (2.162)
ω 2
SFF = EFF × HFF = |r̂ × A| r̂. (2.163)
2µ0 c
The power radiates as transverse electromagnetic waves into all angles
of the far field. The scattered power per unit solid angle thus consists of
only polar and azimuthal contributions,
dPs Z0  2
Nϕ 2 ,
= r 2 SFF · r̂ = |N θ | + (2.164)
dΩ 8λ2
where the radiation vector N = 4π r A/µ0 is introduced for convenience
[146]. The total scattered power results from the solid angle integral:
dPs dPs
Ps = dϕ sin θ dθ = 2π sin θ dθ . (2.165)
dΩ dΩ

Finally, the loss per unit length attributed to scattering is directly related
to the scattered power according to
1 Ps
αs = , (2.166)
2π R Pg
where Pg is the power in the guided mode.
2.8 Scattering Losses Resulting From Edge Roughness 55

2.8.5 TM Scattering Losses

For TM WGM modes, the modal electric field is perpendicular to the plane
of the disk (directed along the z-axis) and thus does not give rise to radi-
ated fields polarized in the azimuthal direction. Thus, the radiation vector
only possesses a polar component (the projection of the z-component),

Nθ = − sin θ Nz , (2.167)

resulting in


Nθ = iω dz dϕ r dr
−d/2 0 0
cos θ ikr sin(θ) cos(ϕ−ϕ )
∆ε sin θ Ez (r , z )eimϕ eikz e . (2.168)

The resulting scattered field retains the orthogonality of the corrugation

harmonics; thus, the integral for the radiation vector can be treated sepa-
rately for each harmonic and later summed incoherently. The modal field
is continuous across the interface simplifying greatly the result. Incorpo-
rating the unit step perturbation as a limit in the integral results in

2π  )e−iMϕ
R+∆RM (z

NθM = iωε0 sin θ dz dϕ r dr

−d/2 0 R

(n21 − n22 )Ez (r , z )eikz cos θ eimϕ eikr sin(θ) cos(ϕ−ϕ ) . (2.169)

Because the perturbation is small ∆RM  R and localized to the disk edge
surface all radial variables are replaced with the nominal disk radius and
the integral is collapsed:


NθM = iωε0 R(n21 − n22 ) sin θ dz ∆RM (z )Ez (R, z )eikz cos θ


dϕ ei(m−M)ϕ eikR sin(θ) cos(ϕ−ϕ ) . (2.170)

Implementing the identity:

dϕ ei(m−M)ϕ eikR sin(θ) cos(ϕ−ϕ ) = 2π im−M Jm−M (kR sin θ) ei(m−M)ϕ
56 2. Optical Dielectric Waveguides

and retaining only the square modulus of the polar radiation vector com-
ponent yields
 M 2
Nθ  = 2π ωε0 R(n21 − n22 ) sin2 θ |Jm−M (kR sin θ)|2
 ikz cos θ 

 dz ∆RM (z )Ez (R, z )e  . (2.172)

Next the scattered power is calculated assuming that the field and corru-
gation are z-independent:

- k40 (n21 − n22 )2
Ps = 2π R 2 2
∆RM |Ez (R)|2
3 2 ikz cos θ 

dθ sin θ |Jm−M (kR sin θ)|  dz e  . (2.173)

The calculation of the z integral is straightforward and results in a polar

sinc pattern. Incorporating both this and the Gaussian correlation func-
tion, the expression for the scattering loss becomes

k40 (n21 − n22 )2 σ 2 Sc 1 λd |Ez (R)|2 - Sc
−π 2π
αs = R e RM
4 2π R Pg 2Z0 M=−∞

d d cos θ
dθ sin3 θ |Jm−M (kR sin θ)|2 sinc2 . (2.174)
λ λ

2.8.6 TE Scattering Losses

The edge scattering loss derivation for TE WGM modes is considerably

more complicated. First, the modal electric field lies in the plane of the
disk with both radial Er and azimuthal Eϕ components. Although the
azimuthal component may be negligible for very low index contrast mi-
croresonators, in general it can be quite strong and cannot be neglected.
Second, each of these components couples to both polar and azimuthal
components of the radiated fields. Third, discontinuities in the planar
electric field components exist unless the roughness is locally flat [147].
Fortunately, this approximation is valid for typical (shallow) roughness
distributions where the corrugation depth is much smaller than the cor-
relation length (σ  Sc ). For a treatment of deep perturbations, see
Johnson et al. [148].
2.8 Scattering Losses Resulting From Edge Roughness 57

The modal field amplitudes projected in Cartesian coordinates are

Ex = cos ϕ Er − sin ϕ Eϕ (2.175)

Ey = sin ϕ Er + cos ϕ Eϕ . (2.176)

The radiation vector will consist of both polar θ and azimuthal ϕ com-
ponents each arising from these modal field components.

Nθ = cos θ cos ϕ Nx + sin ϕ Ny (2.177)
Nϕ = − sin ϕ Nx + cos ϕ Ny . (2.178)

For convenience, field projection variables are defined:

Kθ (θ, ϕ, r , z , ϕ ) =
.   /
cos θ cos ϕ − ϕ ε0 ∆ε−1 Dr + sin ϕ − ϕ ∆εEϕ (2.179)

Kϕ (θ, ϕ, r , z , ϕ ) =
.   /
− sin ϕ − ϕ ε0 ∆ε−1 Dr + cos ϕ − ϕ ∆εEϕ , (2.180)

resulting in a compact expression for the radiation vector components


Nθ = −iω dz dϕ r dr
−d/2 0 0
cos θ ikr sin(θ) cos(ϕ−ϕ )
Kθ (θ, ϕ, r , z , ϕ )eimϕ eikz

e (2.181)


Nϕ = −iω dz dϕ r dr
−d/2 0 0
cos θ ikr sin(θ) cos(ϕ−ϕ )
Kϕ (θ, ϕ, r , z , ϕ )eimϕ eikz

e . (2.182)

The integral for the radiation vector is treated separately for each har-
monic. Furthermore, for convenience, the radial and azimuthal field con-
tributions can be treated separately and summed later. Incorporating the
unit step perturbation as a limit in the integral results in

R+∆RM(z )e−iMϕ

M 1 1
Nθ,r = −iω cos θ dz dϕ r dr − 2
n21 n2
−d/2 0 R
+  ,
cos ϕ − ϕ Dr (r , z ) eikz cos θ eimϕ eikr sin(θ) cos(ϕ−ϕ ) (2.183)
58 2. Optical Dielectric Waveguides

2π  R+∆RM (z )e−iMϕ
Nθ,ϕ = −iωε0 cos θ dz dϕ r dr (n21 − n22 )
−d/2 0 R
+  ,
sin ϕ − ϕ Eϕ (r , z ) eikz cos θ eimϕ eikr sin(θ) cos(ϕ−ϕ ) (2.184)

R+∆RM(z )e−iMϕ

1 1
Nϕ,r = −iω dz dϕ r dr − 2
n21 n2
−d/2 0 R
− sin ϕ − ϕ Dr (r , z ) eikz cos θ eimϕ eikr sin(θ) cos(ϕ−ϕ )


2π  R+∆RM (z )e−iMϕ
Nϕ,ϕ = −iωε0 dz dϕ r dr (n21 − n22 )
−d/2 0 R

cos ϕ − ϕ Eϕ (r , z ) eikz cos θ eimϕ eikr sin(θ) cos(ϕ−ϕ ) . (2.186)
Because the perturbation is small ∆RM  R and localized to the disk
edge surface, all radial variables can be replaced with the disk radius to
collapse the integral.

1 1
Nθ,r = −iωR 2
− 2 cos θ dz ∆RM (z )Dr (R, z )eikz cos θ
n1 n2

+  ,
dϕ cos ϕ − ϕ ei(m−M)ϕ eikR sin(θ) cos(ϕ−ϕ ) (2.187)


Nθ,ϕ = −iωε0 R(n21 − n22 ) cos θ dz ∆RM (z )Eϕ (R, z )eikz cos θ


+  ,
dϕ sin ϕ − ϕ ei(m−M)ϕ eikR sin(θ) cos(ϕ−ϕ ) (2.188)


1 1
Nϕ,r = −iωR − 2 dz ∆RM (z )Dr (R, z )eikz cos θ
n21 n2

+  ,
dϕ − sin ϕ − ϕ ei(m−M)ϕ eikR sin(θ) cos(ϕ−ϕ ) (2.189)
2.8 Scattering Losses Resulting From Edge Roughness 59


Nϕ,ϕ = −iωε0 R(n21 − n22 ) dz ∆RM (z )Eϕ (R, z )eikz cos θ


+  ,
dϕ cos ϕ − ϕ ei(m−M)ϕ eikR sin(θ) cos(ϕ−ϕ ) . (2.190)

Implementing the identity

dϕ e±i(ϕ−ϕ ) ei(m−M)ϕ eikR sin(θ) cos(ϕ−ϕ ) =

2π im−M±1 Jm−M±1 (kR sin θ) ei(m−M)ϕ (2.191)

and retaining only the square modulus of the radiation vector compo-
nents yields
 M 2 1 1
Nθ,r  = 2π ωR 2
− 2 cos2 θ
n1 n2
|Jm−M+1 (kR sin θ) − Jm−M−1 (kR sin θ)|2
 ikz cos θ 
 dz ∆RM (z )Dr (R, z )e  (2.192)

 M 2
Nθ,ϕ  = 2π ωε0 R(n21 − n22 ) cos2 θ

|Jm−M+1 (kR sin θ) + Jm−M−1 (kR sin θ)|2

 ikz cos θ 
 dz ∆RM (z )Eϕ (R, z )e  (2.193)

 M 2 1 1
Nϕ,r  = 2π ωR − 2
n21 n2
|Jm−M+1 (kR sin θ) + Jm−M−1 (kR sin θ)|2
 ikz cos θ 
 dz ∆RM (z )Dr (R, z )e  (2.194)
60 2. Optical Dielectric Waveguides

 M 2
Nϕ,ϕ  = 2π ωε0 R(n21 − n22 )

|Jm−M+1 (kR sin θ) − Jm−M−1 (kR sin θ)|2

 ikz cos θ 
 dz ∆RM (z )Eϕ (R, z )e  . (2.195)

Assuming that the field and corrugation are z-independent, the scattered
power is given as

- k40 (n21 − n22 )2 2
Ps = 2π R 2 ∆RM

π 0
|Jm−M+1 (kR sin θ) − Jm−M−1 (kR sin θ)|2
dθ sin θ cos2 θ

|Jm−M+1 (kR sin θ) + Jm−M−1 (kR sin θ)|2 |Dr (R)|2
+ sin θ
4 n2 n2 ε 1 2 0

|Jm−M+1 (kR sin θ) + Jm−M−1 (kR sin θ)|2
+ sin θ cos2 θ
|Jm−M+1 (kR sin θ) − Jm−M−1 (kR sin θ)|2  

+ sin θ Eϕ (R)
 ikz cos θ 
 dz e  . (2.196)

The calculation of the z integral is straightforward and, again, results in a

polar sinc pattern. Incorporating both this and the Gaussian correlation
function, the expression for the scattering loss results, here split into
azimuthal and radial field contributions:

k40 (n21 − n22 )2 σ 2 Sc

αs = R
4⎧ 2π R
⎨ 1 λd |D (R)|2 - ∞

−π 2π

r e RM
⎩ Pg 2Z n4 n4 ε2
0 1 2 0 M=−∞

dθ sin θ cos2 θ Jm−M (kR sin θ)
2.8 Scattering Losses Resulting From Edge Roughness 61

(m − M)2 |Jm−M (kR sin θ)|2 d d cos θ
+ sin θ 2
(kR sin θ) λ λ
 2 ∞
1 λd Eϕ (R) -

−π 2π RM
+ e
Pg 2Z0 M=−∞

(m − M)2 |Jm−M (kR sin θ)|2
dθ sin θ cos2 θ 2
(kR sin θ)
 2  d d cos θ
+ sin θ Jm−M (kR sin θ) sinc2 . (2.197)
λ λ

2.8.7 Normalized Formulation for Edge Scattering Losses

The expressions for the edge scattering loss scale with the inverse fourth
power of the wavelength, typical of scattering processes. They also pre-
dict that the loss is strongly dependent on index contrast in proportion to
at least the square of the permittivity difference. Finally, the expression is
insensitive to disk height d when the height exceeds the wavelength as we
will show later. The quality factor can be expressed in normalized units,
from which useful limiting approximate forms can be derived. The quality
factor is given by the radians per cycle divided by the fractional loss per
cycle. Its association with the scattering loss is given by Qs−1 = αs R/m,

 2  2
1 3 n1 2π R n22 σ Sc 1 d |Ez (R)|2
TM = 4π 1−
Qs mλ n21 λ/n1 λ/n1 Pg 4kZ0

- −π Sc λ
λ/n1 n1 2π R M

3 d 2 d cos θ 2
dθ sin θ |Jm−M (kR sin θ)| sinc (2.198)
λ λ

 2  2
1 n1 2π R n22 σ Sc
= 4π 3 1−
QsTEr mλ n21 λ/n1 λ/n1

1 d |Dr (R)|2 - Sc
−π λ/n λ

n1 2π R M
e 1
Pg 4kZ0 n n ε M=−∞
4 4 2
1 2 0

dθ sin θ cos2 θ Jm−M (kR sin θ)
(m − M)2 |Jm−M (kR sin θ)|2 d d cos θ
+ sin θ 2
sinc2 (2.199)
(kR sin θ) λ λ
62 2. Optical Dielectric Waveguides

 2  2
1 3 n1 2π R n22 σ
TEϕ = 4π 1−
Qs mλ n21 λ/n1
 2 ∞
1 d Eϕ (R) - Sc
−π λ/n λ
n1 2π R M
e 1
Pg 4kZ0 M=−∞

(m − M)2 |Jm−M (kR sin θ)|2
dθ sin θ cos2 θ 2
(kR sin θ)
 2  d d cos θ
+ sin θ Jm−M (kR sin θ) sinc2 . (2.200)
λ λ

These expressions can be written in a compact form by defining normal-

ized units:  
1 3 X 1 2 2 p p p
p = 4π 1 − 2 ξ c Γz Γr Gm , (2.201)
Qs m n
where X, ξ, c , respectively, are normalized quantities representing the
radius X = n1 2π R/λ, roughness ξ = n1 σ /λ, and correlation length
c = n1 Sc /λ. The superscript symbol p refers to the polarization state
(TM, TEr , or TEϕ ). The quantity Γz = Pc /Pg is the ratio of power vertically
confined to total power guided or simply the vertical confinement factor.
The usual confinement factor for a planar waveguide can be applied here
when using the effective index method. This requires solution of the sim-
ple planar slab waveguide dispersion relation (kz vs. ω) as in the Kogelnik
formulation [99]. The quantity Γr is the edge confinement factor and is
defined as the ratio of the intensity at the disk edge (|E(R)|2 /2Z0 ) to the
characteristic intensity of the mode (2kPc /d). This requires solution of
the simple infinite cylinder whispering gallery mode dispersion relation
(X, Qi ) vs. (m, n) vs. (m, n). The radial mode profile is normalized to a
power per unit disk height of Pc /d by integrating the azimuthal compo-
nent of the Poynting vector along the radial dimension from the disk cen-
ter out to the radiation boundary (Rr = mλ/2π n2 ). Finally, a geometric
factor resulting from the sum/integral of the far field scattering pattern

p 3

−π Xc M p
into azimuthal and polar angles is defined, Gm = e Pm−M ,
where Pm−M refers to the associated polar integral.
There are useful limits to consider: thick and thin cylinder. In the
limit of a thick cylinder (d  λ, but where d  R such that the small
polar angle approximation, Eq. 2.159 still holds), the scattered radiation
is directed outward at θ = 90◦ . The following expressions result for the
polar integrals where the derivation is assisted by making the change of
variables d cos θ/λ ≡ τ ):
2.8 Scattering Losses Resulting From Edge Roughness 63

d d cos θ
Pm−M = dθ sin3 θ |Jm−M (kR sin θ)|2 sinc2
  ⎛ λ λ ⎞
d/λ  2     2 2
λ  λ 
= dτ
1− τ Jm−M ⎝kR 1 − τ ⎠ 2
  sinc τ
d  d 

→ dτ |Jm−M (kR)|2 sinc2 τ = |Jm−M (kR)|2 (2.202)


Pm−M = dθ sin θ cos2 θ Jm−M (kR sin θ)
(m − M)2 |Jm−M (kR sin θ)|2 d 2 d cos θ
+ sin θ 2
(kR sin θ) λ λ
⎡  ⎛  ⎞ 

d/λ  2   2 2
⎢ λ  λ 
= dτ ⎣ τ  ⎝
Jm−M kR 1 − d τ


2  2
(m − M)2 Jm−M kR 1 − dλ τ  ⎥
+   2 ⎥ sinc2 τ

2 ⎥
λ ⎦
kR 1 − d τ

(m − M)2 |Jm−M (kR)|2 (m − M)2 |Jm−M (kR)|2
→ dτ 2
sinc2 τ = 2
dλ (kR) (kR)

TEϕ (m − M)2 |Jm−M (kR sin θ)|2
Pm−M = dθ sin θ cos2 θ 2
(kR sin θ)
 2  d 2 d cos θ

+ sin θ Jm−M (kR sin θ)  sinc
⎡  λ   λ 

⎢ 2
 λ 2 

⎢ 2 (m − M) Jm−M kR 1 − d τ 
⎢ λ
= dτ ⎢ τ   2
⎢ d
⎣ λ
−d/λ kR 1 − d τ
 ⎛  ⎞ ⎤
  2 2
+ ⎝ λ ⎠ ⎥ 2
Jm−M kR 1 − d τ  ⎦ sinc τ

 2  2
→ dτ Jm−M (kR) sinc2 τ = Jm−M (kR) . (2.204)
64 2. Optical Dielectric Waveguides

In the limit of a thin cylinder (d  λ), the following expressions result

for the polar integrals:

Pm−M → dθ sin3 θ |Jm−M (kR sin θ)|2 (2.205)
dλ λ


d  2
Pm−M → dθ sin θ cos2 θ Jm−M (kR sin θ)
dλ λ

(m − M)2 |Jm−M (kR sin θ)|2
+ sin θ 2
(kR sin θ)

TEϕ d (m − M)2 |Jm−M (kR sin θ)|2
Pm−M → dθ sin θ cos2 θ 2
dλ λ (kR sin θ)
+ sin θ Jm−M (kR sin θ) . (2.207)

The polar integrals for scattering as a function of corrugation order are

plotted for comparison in Fig. 2.20. The scattering as a function of cor-
rugation order can be expressed as scattering as a function of local az-
imuthal coordinate ∆ϕ through a modified grating equation,
!  "
∆ϕ = arccos n 1 − . (2.208)
If the correlation length is small such that the width of the spec-
tral density of the roughness distribution (Xλ/n1 Sc ) is much wider than
that of the polar integral (2X/n), then the summation term is simplified
greatly. For typical fabrication processes, the relation (Sc  λ/2n2 ) is
generally the case.

p p
Gm → Pm−M . (2.209)
Sc λ/2n2
This allows the further reduction of the Gm parameter to simple limiting
Gm → 1, , (2.210)
dλ 2 2
TM,TEr ,TEϕ THIN 4 4 4
Gm → δ, δ, δ. (2.211)
dλ 3 3 3
Here, a normalized thickness has been defined as δ = d/λ. As a final use-
ful approximation, the edge confinement factor may be approximated in
the high index contrast limit as Γr ≈ 1/X. This simple inverse radius
2.8 Scattering Losses Resulting From Edge Roughness 65

Fig. 2.20. Distribution of the polar integral terms Pm−M versus corru-
gation order M for (a) TM, (b) TE radial, and (c) TE azimuthal. Here, the
azimuthal order for the mode, and hence the center for the distribution
is m = 50. The index ratio (for this example n = 3) restricts the partic-
ipating corrugation orders from the full 0 < M < 2m because of Snell’s
law or phase matching conditions. For λ = 1.55 µm, the resonant radii
for TM and TE at m = 50 are 4.605 and 4.686 µm respectively. The total
sums Gm are shown for each component in the thick and thin limits. The
thick/thin cylinder limits are denoted by thick/thin linewidths, respec-
tively. The thin curves have been normalized by factoring out the nor-
malized thickness δ = d/λ parameter. The scattering distributions are
also plotted as an angular distribution on the right. (After [296], ©2002,
Optical Society of America.)
66 2. Optical Dielectric Waveguides

Fig. 2.21. Variation in the edge confinement factor associated with the
electric field amplidudes of (a) TM, (b) TE radial, (c) TE azimuthal, and
(d) TE net as a function of normalized radius for varying index contrasts
(n = 1.25, 1.35, 1.5, 1.7, 2.0, 2.5, 3.0, 3.5). Note that (a) and (d) approach
the approximate form 1/X for high index contrasts. (After [296], ©2002,
Optical Society of America.)




0 20 40 60 80 100
Normalized radius, X = n12πR/λ

Fig. 2.22. Finesse limited by bending and edge scattering losses, for TM
polarization, λ = 1.55 µm, d = 300 nm, n1 = 1.5, 3.0, n2 = 1, σ = 1 nm,
Sc = 75 nm. Note in the asymptotic limit, the validity of the edge scatter-
ing limited finesse approximation (dashed line) for thin microresonator
disks. (After [296], ©2002, Optical Society of America.)
2.8 Scattering Losses Resulting From Edge Roughness 67

dependence is found for both TM and TE when the radial and azimuthal
contributions are summed, although it is dominated by the azimuthal
component in the high index contrast limit. For a breakdown of these
contributions, see Fig. 2.21. Because the edge confinement is inversely
related to the normalized disk radius, it directly cancels the increased
circumferential path length per round-trip. Ultimately this leads to an
edge scattering limited finesse value that is independent of radius. Incor-
porating all these approximations results in simple forms for the edge
scattering limited finesse. For thick, vertically extended (d > λ) microres-
onators, the expressions for TM and TE differ slightly:

Fig. 2.23. Finesse limited by bending and edge scattering losses, for both
TM and TE polarization, λ = 1.55 µm, d = 300 nm, n2 = 1, σ = 1, 10 nm,
Sc = 75 nm. The index ratios are n = 1.25, 1.35, 1.5, 1.7, 2.0, 2.5, 3.0, 3.5.
Note the clamping of finesse with increasing normalized radius in the
edge scattering limited regime. (After [296], ©2002, Optical Society of
68 2. Optical Dielectric Waveguides

1 3 1 2 2
= 4π 1 − ξ c (2.212)
FsTM n2
1 3 1 2 2
= 2π 1 − ξ c . (2.213)
FsTE n2
However, for thin, thumbtack-like (d < λ) microresonators, they are
1 16π 3 1 2 2
= 1 − ξ c Γz δ. (2.214)
Fs 3 n2
In practice it is often found that TE operation results in higher losses.
This has been attributed to a variety of interpretations such as a higher
field strength at the sidewall and arguments involving erroneous deriva-
tions of geometric factors. The derivation shows that neither is the case.

Fig. 2.24. Quality factor limited by bending and edge scattering losses,
for both TM and TE polarization, λ = 1.55 µm, d = 300 nm,
n2 = 1, σ = 1, 10 nm, Sc = 75 nm. The index ratios are n =
1.25, 1.35, 1.5, 1.7, 2.0, 2.5, 3.0, 3.5.
2.8 Scattering Losses Resulting From Edge Roughness 69

Rather, it is likely that other mechanisms are responsible for the discrep-
ancy such as the lower vertical confinement factors associated with TM
operation. The validity of this approximation for a thin disk is shown
by plotting the bending and edge scattering limited finesse as a func-
tion of normalized radius in Fig. 2.22. Figures 2.23 and 2.24 show the
TM and TE finesse and Q curves for two different values of roughness
across a variety of refractive index ratios. Note that at small radii, the
advantage of high indices to combat bending loss is apparent as higher
indices lead to higher finesse. For larger radii, however, where bending
loss is insignificant, the ordering of the curves interchanges where edge
scattering limited operation becomes prevalent.
Figure 2.25 displays the tradeoff from a different perspective, plotting
the finesse versus refractive index ratio for a few radii (parameterized by
m). With increasing refractive index, bending losses decrease whereas
edge scattering losses increase. The tradeoff is thus clear: if an index
ratio is a design variable, one desires enough to negate the effects of
bending loss, but only just enough, as any more leads to increased edge
scattering loss in a practical device.
To summarize, this generalized formulation predicts an edge scatter-
ing limited finesse that is independent of disk radius and polarization
in the thin disk regime. The derived simplified expressions should be

100 Scatterin
50 g loss li
25 mited
10 15



1.0 1.5 2.0 2.5 3.0 3.5

Index ratio, n1/n2

Fig. 2.25. The tradeoff between edge scattering and bending loss as a
function of index contrast. An optimum index contrast exists whose value
increases as the resonator is made smaller (lower azimuthal number m).
Specific choices made for this plot are: TM polarization, λ = 1.55 µm,
d = 300 nm, n2 = 1, σ = 3 nm, Sc = 75 nm. (After [296], ©2002, Optical
Society of America.)
70 2. Optical Dielectric Waveguides

widely applicable both in predicting the impact of fabrication imperfec-

tions and/or in selecting the lateral index contrast for optimizing the
performance of microresonators.

2.9 Summary

In this chapter we introduced the basic formalisms for analyzing propa-

gation, coupling, and loss mechanisms in optical waveguides and whis-
pering gallerys. In the next chapter, we examine a generic microresonator
from a slightly higher vantage point where these distributed parameters
are condensed into the single-pass phase shift, coupling constant, and
loss parameter.
3. Optical Microresonator Theory

Optical resonators manifest themselves everyday in our world: from the

lasers in supermarket scanners to the colorful patterns reflected from
oil slicks. They all share the ability to selectively modify certain optical
properties within narrow spectral ranges. In this chapter, we introduce
fundamental concepts associated with optical microresonators. Although
introduced in the context of optical microring resonators, the concepts
may be applied to a variety of resonant physical systems independent of
application or scale. Many of the ideas presented in the chapter are thus
equally applicable to analogous microwave and acoustic resonators.

3.1 Resonator Fundamentals

3.1.1 Fabry–Perot Resonators

Optical resonators were employed as useful devices as early as 1899,

when Fabry and Perot described the use of a parallel-plate resonator
as a multipass interferometer [149]. Light incident on this Fabry–Perot
resonator is split into transmitted and reflected components with power
fractions that depend on many variables. If temporally incoherent
(“white”) light is incident on the resonator, then the transmission and
reflection coefficients depend only on the mirror reflectivities. The total
reflected power consists of the power reflected from the first mirror plus
all the multiple reflections between the mirrors that contribute to overall
reflection. The sum of these contributions takes the form:

- R1 − 2R1 R2 + R2 2R
R = R1 + T12 R2 (R1 R2 )m−1 = → . (3.1)
1 − R1 R 2 R1 =R2 ≡R 1+R

Likewise, the transmitted power fraction takes the form:

- T1 T2 T2 1−R
T = T1 T2 (R1 R2 )m−1 = → = . (3.2)
1 − R1 R 2 R1 =R2 ≡R 1 − R2 1+R

If, however, the incident light consists of a temporally coherent (mono-

chromatic) plane wave, then the reflected power will be proportional to
72 3. Optical Microresonator Theory

the square of the coherent sum of all reflected fields. Because the fields
carry phase information in addition to their amplitude, the fraction of re-
flected and transmitted light depends not only on the mirror reflectivities,
but also on the mirror spacing and excitation wavelength. The coherent
sum of fields is maximized when all the fields interfere constructively (in
phase) and minimized when they interfere destructively (out of phase).
Phase accumulates with propagation distance as φ(z) = βz and may also
be acquired upon interaction with the mirrors. The coherent versions of
Eqs. 3.1 and 3.2 include an accumulated phase factor per round-trip that
can be interpreted as a normalized detuning φ = TR ω, where TR is the
cavity transit time, TR = neff L/c for the circumference, L and effective
index neff . Now, r̃ represents the complex reflectivity:

m−1 r 1 − e+iφ
2 imφ imφ r1 − r2 eiφ
r̃ = r1 −t1 r2 e r1 r2 e = → ,
1 − r1 r2 eiφ r1 =r2 ≡r 1 − r 2 e+iφ
and t̃ represents the complex transmittivity:

φ φ
im 2 imφ −t1 t2 eim 2 − 1 − r 2 eim 2
t̃ = −t1 t2 e r1 r2 e = → .
1 − r1 r2
r1 =r2 ≡r 1 − r2
The square modulus of these complex quantities gives the reflection
R and transmission T coefficients. Antiresonant wavelengths are more
strongly reflected than in the incoherent case, whereas resonant wave-
lengths are transmitted 100% for balanced reflectors (r1 = r2 ). For a fixed
mirror spacing, the transmission and reflection spectra thus exhibit peri-
odic peaks and valleys. Figure 3.1 displays the transmission and reflection
spectra for a lossless, balanced Fabry–Perot resonator. The fraction of re-
flected and transmitted power for incoherent excitation is equivalent to
the respective spectrally averaged reflection and transmission across a
period of the spectrum.

Fig. 3.1. The transmission and reflection spectra for a Fabry–Perot

3.1 Resonator Fundamentals 73

3.1.2 Gires–Tournois Resonators

A lossless Fabry–Perot resonator with a 100% reflecting rear mirror con-

stitutes a device that is 100% reflecting at all frequencies. Nevertheless,
resonant frequencies spend more time circulating in the resonator and
experience longer group delays than do non-resonant frequencies. This
deceptively simple device, termed a Gires–Tournois resonator, provides
a means for preserving the spectral power of light reflected from it while
modifying its phase. Consequently it is often referred to as a “phase-only”
filter. The complex reflectivity of this device may be taken as a limiting
case of Eq. 3.3 with r1 ≡ r and r2 = 1:

- m−1 r − eiφ
r̃ = r − t 2 eimφ r eimφ = . (3.5)
1 − r eiφ

The square modulus of this expression is unity for all values of the de-
tuning parameter, φ. The phase argument of this expression, however,
varies with detuning according to
r sin(φ)
Φ = π + φ + 2 arctan . (3.6)
1 − r cos(φ)
For a fixed mirror spacing, the spectral phase exhibits a staircase-like as-
cension with frequency. Figure 3.2 displays the spectral phase for a loss-
less Gires–Tournois resonator. The increase in phase sensitivity (slope)
near each resonance is related to an increased group delay (to be inves-
tigated in a later section).

3.1.3 Ring Resonators

Fabry–Perot and Gires–Tournois resonators are extremely versatile de-

vices finding application as spectroscopic tools, add–drop filters, dis-
persion compensators, and laser cavities. Unfortunately, their free-space

Fig. 3.2. The spectral phase versus detuning for a Gires–Tournois

74 3. Optical Microresonator Theory

Fig. 3.3. Two ring resonator devices and their free-space embodiments.

mirror embodiment is incompatible with planar integrated technology.

Devices nearly equivalent to these resonators can be constructed in a
planar waveguiding geometry through the use of a ring waveguide cou-
pled to one or two ordinary waveguides. For these ring resonators, the
coupling strengths play the analogous role of mirror transmission coef-
ficients. Figure 3.3 shows schematics of these two ring resonator devices
and their free-space embodiments.

3.2 All-Pass Ring Resonators

A simple ring resonator is created by taking one output of a generic di-

rectional coupler and feeding it back into one input (see Fig. 3.4). Such a
device exhibits a periodic cavity resonance when light traversing the ring
acquires a phase shift corresponding to an integer multiple of 2π radi-
ans. The resonator is mathematically formulated from two components:
a coupling strength and a feedback path. Contrary to the infinite sum
derivations performed earlier for the Fabry–Perot and Gires–Tournois, in
what follows we derive the basic spectral properties by assuming steady-
state operation and matching fields. Although both methods are equally
valid, the field-matching method has the advantage of simplicity.
3.2 All-Pass Ring Resonators 75

Fig. 3.4. Fields associated with an all-pass ring resonator.

The basic relations amongst the incident E1 , transmitted E2 , and cir-

culating E3 , E4 fields of a single resonator are derived by combining the
relations for the coupler with that of the feedback path. In the spectral
domain, the fields exiting the coupling region1 are related to the input
fields via the following unitary matrix (section 2.5.4):
E4 (ω) r it E3 (ω)
= , (3.7)
E2 (ω) it r E1 (ω)

where the lumped self- and cross-coupling coefficients r and t are as-
sumed to be independent of frequency 2 and satisfy the relation r 2 +t 2 =
1. The feedback path (of length 2π R) connects the output from port 4
back into input port 3 where the field is expressed as
E3 = e − 2 2π R ik2π R
e E4 ≡ aeiφ E4 . (3.8)

Here, a represents the single-pass amplitude transmission and φ repre-

sents the single-pass phase shift. Because adding or subtracting an inte-
ger number m of 2π radians from the single-pass phase shift does not
change the value of the function, the single-pass phase shift for all res-
onances is defined such that its value is zero for a local resonance of
interest. Furthermore, because the single-pass phase shift is directly re-
lated to the radian frequency as φ = ωTR , where TR is the transit time of
the resonator and φ is clearly representative of a normalized frequency

3.2.1 Intensity Buildup

As we will see in the next section, all-pass resonators delay incoming

signals via the temporary storage of optical energy within the resonator.
Constructive interference at the coupler port entering the ring ensures
that circulating optical intensity is built up to a higher value than that
In the case of a microresonator, although the coupling is distributed over a
significant angular portion of the disk, the coupling can be treated as being
lumped and localized to a single point without loss of generality.
In most cases, this is valid for microresonators since the coupling interaction
length is much smaller than the cavity circumference, yielding a much broader
bandwidth for the coupling interaction than for the cavity resonance.
76 3. Optical Microresonator Theory

initially injected. A coherent source is essential to achieving a buildup of

intensity. In the case of perfectly incoherent excitation of an all-pass res-
onator (and no attenuation) the intensity in the cavity equalizes with the
incident intensity. This result is analogous to the equalization of pressure
between coupled pipelines in hydrostatics. Implementation of a coherent
source does not change the average intensity buildup across a free spec-
tral range. Rather, with coherent excitation, the buildup can greatly ex-
ceed unity near resonances at the expense of being reduced below unity
away from them such that the average never deviates from the incoherent
case. A coherent buildup of intensity can lead to a dramatically enhanced
nonlinear response. We will return to this point later in this volume.
Equations 3.7 and 3.8 are solved to obtain an expression for the ratio
of the circulating field to the incident field:
E3 itaeiφ
= . (3.9)
E1 1 − r aeiφ
The ratio of circulating intensity to incident intensity, or the buildup fac-
tor B, is given by the squared modulus of this result,
I3  E3 
B= =E 
I1  1 2 2
1−r a
= (3.11)
1 − 2r a cos φ + r 2 a2
1+r 4
→ ≈ 2, (3.12)
φ=m2π ,a=1 1 − r t

where the last result refers to the situation in which the incident light is
resonant with the ring (φ = m2π ) and attenuation is negligible (a = 1).
A passive ring resonator under these conditions attains the maximum
ratio of circulating power to incident power that can be achieved, which
is illustrated in Fig. 3.5 for two successive resonances. For cross-coupling

Fig. 3.5. A plot of the buildup factor versus detuning for an all-pass ring
3.2 All-Pass Ring Resonators 77

Fig. 3.6. Finite-difference time-domain simulation demonstrating coher-

ent buildup of intensity at 1.5749 µm. Guide and ring widths are 0.4 µm.
Resonator radii are 2.5 µm (outer) and 2.1 µm (inner). All guiding struc-
tures have a refractive index of 2.5 that is cladded by air.

values of 10% (t 2 = 0.1), the intensity in the ring can be 40 times higher
than the intensity incident on the resonator in the input waveguide. Since
the intensity in the ring can be much higher than in the bus, ring res-
onators can be used for nonlinear optics applications with moderate in-
put intensities. Figure 3.6 displays the coherent buildup of intensity in
an all-pass microring resonator.

3.2.2 Finesse F
The spectral shape of the buildup factor displays sharply peak reso-
nances that are characterized by a finesse parameter. The finesse is de-
fined as the free spectral range (FSR) between resonance peaks divided
78 3. Optical Microresonator Theory

by the full width at half depth (FWHD) of a resonance. Implementing this

definition in the absence of internal losses results in an expression for
the finesse of an all-pass resonator:

2π π 2π
F=   → ≈ 2 . (3.13)
2r r ≈1 1−r t
2 arccos
1+(r )2

It is through the coherent constructive interference of recirculating feed-

back that resonators are able to increase the effective path length (and
interaction time) of light traversing them by a factor equivalent to the

3.2.3 Effective Phase Shift

An examination of the transfer characteristics of the resonator reveals

another periodically resonant feature. Equations 3.7 and 3.8 are solved
to obtain the ratio of the transmitted field to the incident field:

E2 a − r e−iφ
= ei(π +φ) . (3.14)
E1 1 − r ae+iφ

The intensity transmission is given by the squared modulus of this quan-

tity. For negligible attenuation, i.e., a = 1, the equation predicts a unit
intensity transmission for all values of detuning φ. Such a device is di-
rectly useless as an amplitude filter, allowing 100% transmission for all
frequencies and is aptly termed an “all-pass” filter. This result is satisfy-
ing from an intuitive standpoint because light is offered only two choices:
leave the device at port 2 or (re)enter the resonator at port 4. No mecha-
nism exists for light to exit via port 1, and thus, in steady state, the optical
powers entering and exiting the resonator are equal. The device, however,
does not impose a uniform requirement of constant phase across all fre-
quencies. The phase of the transmitted light, as in the Gires–Tournois
resonator, can be dramatically different for different frequencies, espe-
cially those near resonance. The effective phase shift is defined as the
phase argument of the field transmission factor and is the phase shift
acquired by light in crossing the coupler from port 1 to port 2:

r sin(φ) r a sin(φ)
Φ = π + φ + arctan + arctan
a − r cos(φ) 1 − r a cos(φ)
r sin(φ)
→ π + φ + 2 arctan . (3.15)
a=1 1 − r cos(φ)

A plot of the effective phase shift versus the single-pass phase shift φ for
different values of r 2 is shown in Fig. 3.7. Near resonances (φ ≈ m2π ) the
slope of the curve becomes very steep indicating that the phase that the
3.2 All-Pass Ring Resonators 79

Fig. 3.7. A plot of the effective phase shift versus the single-pass phase
shift or normalized detuning for an all-pass ring resonator. Note the in-
creasing sensitivity near resonance for increasing self-coupling parame-
ter r 2 .

device imparts is sensitively dependent on the normalized detuning [150,

151]. The phase sensitivity is obtained by differentiating the effective
phase shift with respect to the detuning to obtain

Φ =

1 − r 2 a2

1+a2 ) 2
1 − 2r a cos(φ) 2 + r 2 a2 + sin2 (φ) (1 − a2 ) r 2 − (1 − a2 )
→ . (3.16)
φ=m2π ,a=1 1−r

The last form of this result refers to the situation in which the incident
light is resonant and attenuation is negligible (a = 1). A comparison of
Eqs. 3.16 and 3.12 reveals that under these conditions, the level of phase
sensitivity is exactly equal to the level of intensity buildup across the
entire spectrum.

3.2.4 Group Delay and Group Delay Dispersion

The increased phase sensitivity is directly related to the increase in effec-

tive path length. This correspondence can be observed by examining the
delay imposed by the resonator on a resonant pulse. The group delay for
a linear device is given by the radian frequency derivative of the phase of
the transfer function,

TD = − = Φ TR . (3.17)

Because the detuning is related to the radian frequency as φ = ωTR , the
group delay can be expressed as the cavity transit time enhanced by the
phase sensitivity. Although a pulse is being delayed in a resonator, it’s
80 3. Optical Microresonator Theory

energy is stored inside the cavity; hence, the group delay is also equal to
the cavity lifetime. Conversely, the phase sensitivity is interpreted as the
effective number of round-trips light traverses in the resonator.
Because the group delay associated with an all-pass filter is a frequency-
dependent function, its transmission characteristics are inherently dis-
persive. The group delay dispersion (GDD) for a linear device is defined
as the radian frequency derivative of the group delay,

d2 Φ
GDD = = Φ TR2 . (3.18)
The GDD can be strong enough to significantly disperse a pulse. On res-
onance, the GDD (and all even dispersive orders) is zero although
√ higher
order dispersion exists.
√ The GDD has extrema at φ = ±π /F 3 where it
attains the value ∓3 3F 2 TR2 /4π 2 . Assuming a Gaussian pulse of FWHM
equal to the cavity lifetime of a resonator, the quadratic depth of phase

1 2 ln 2 2
imparted across the FWHM of the spectrum is equal to 2 GDDmax F TR
or approximately 0.1265 radians. A convenient parameter characterizing
the depth of the spectral quadratic phase resulting from GDD across a
pulse spectrum is the chirp parameter. The chirp parameter (C) is de-
fined by the following expression for the complex spectral amplitude of
a Gaussian pulse:
− 1+iC ω
E(ω) = E0 e 2 ωp
. (3.19)
At a spectral
√ chirp of unity, the pulse spreads such that its peak intensity
falls to 1/ 2 of its minimum
√ value . The maximum chirp per resonator
then is of the order of 3 3 ln 2/π 2 or approximately 0.365. Thus, approx-
imately three resonators are required to impart a chirp of unity.
Because the properties of resonators are periodic in frequency, a
possibility exists for imparting equivalent phase profiles across mul-
tiple spectral bands. In this manner, ring resonators may be used to
perform dispersion compensation across multiple wavelength division
multiplexed channels simultaneously [152]. Similarly, a delay line may
be built that operates for multiple wavelength multiplexed channels si-
multaneously. Furthermore, carrying this concept over for time-division
multiplexed signals, it also seems that it is possible to operate a resonator
with extremely short pulses (much sorter than the usual cavity lifetime
imposed restriction) as long as the pulses are spaced by the round-trip
time. This method of operation is termed synchronous pumping. Such a
pulse train possesses a wideband spectrum but only in the discrete sense.
That is, the pulse-train spectrum may completely lie within multiple res-
onance bandwidths and can thus take advantage of the phase-enhancing
properties of the resonator. However, unless changes (on–off switches)
This is analogous to the Raleigh range for a Gaussian beam.
3.2 All-Pass Ring Resonators 81

in the pulse train take place at a time scale that is longer than the cav-
ity lifetime, spectral components will be present in the signal that will
fall outside the resonance bandwidths and the synchronous operation
will fail. Thus, it is fallacious to think that employing a synchronous op-
eration can circumvent the bandwidth limitation imposed by a periodic
resonance on an information-carrying pulse train.
For excitation pulse widths less than or equal to the cavity lifetime, all
orders of dispersion imparted by the resonator are important. Expand-
ing the phase response as a Taylor series does not make much sense in
this regime. The high orders of dispersion are responsible for introduc-
ing ringing and stepwise behavior that is intuitively understood in the
time-domain. Figure 3.8 demonstrates the rect and Gaussian responses
for comparable pulse widths slightly larger than the resonator cavity

10 Rect response

8 Circulating pulse
Power, arb. units

Output pulse
Input pulse
0 2 4 6 8 10 12 14
Time, ps

10 Gaussian response

Power, arb. units

6 Circulating pulse

2 Input pulse Output pulse

0 2 4 6 8 10 12 14
Time, ps

Fig. 3.8. Rect response and Gaussian response of a ring resonator with
r = 0.8, TR = 0.131 ps, TC = 1.9 ps, and TP = 4 ps.
82 3. Optical Microresonator Theory

(a) Amplitude, arb. Anti-resonant (d) Resonant

Amplitude, arb.
.035 ps (0.01 TC) .035 ps (0.01 TC)
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time, ps Time, ps
(b) (e)
Amplitude, arb.

Amplitude, arb.
0.35 ps (0.10 TC) 0.35 ps (0.10 TC)
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time, ps Time, ps
(c) (f)
Amplitude, arb.

Amplitude, arb.

3.50 ps (1.00 TC) 3.50 ps (1.00 TC)

0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time, ps Time, ps

(g) (h)
Spectrum, arb.

Spectrum, arb.

effective effective
phase phase

a b c d e f

-20 -15 -10 -5 0 5 10 15 20 -20 -15 -10 -5 0 5 10 15 20

Frequency, THz Frequency, THz

Fig. 3.9. Simulations of interfering output field amplitudes for 6 input

pulse cases. Pulse widths of 3.5, 0.35, and 0.035 ps are injected into a
10 µm diameter resonator (n = 3, r = 0.75). In (a), (b), and (c), the carrier
frequency is completely detuned from resonance. In (d), (e), and (f), the
carrier frequency is tuned directly on resonance. Note that for ultrashort
pulse excitation, the output pulses are representative of the impulse re-
sponse of an all-pass resonator. For pulse widths of the order of the cavity
lifetime (TC ), the pulse remains mostly undistorted. Plots g and h show
the corresponding pulse spectra superimposed on the effective phase of
the transfer function.

lifetime. The hard edges associated with the rect pulse lead to ringing
behavior in the transmitted pulse. The adiabatically changing Gaussian
pulse, however, passes delayed with minor distortion. Figure 3.9 shows
the transmitted field amplitudes for resonant and antiresonant cases for
pulse widths that are 0.01, 0.1, and 1.0 times the cavity lifetime. Note how
the subsequent impulses interfere destructively, forming an undisturbed
3.2 All-Pass Ring Resonators 83

pulse in the antiresonant case and interfere constructively, forming a

delayed (and inverted) pulse in the resonant case.

3.2.5 Attenuation
In reality, internal attenuation mechanisms are always present and thus
render limitations as to when a ring resonator may closely approximate
a true all-pass, phase-only filter. In particular, near resonance, the inter-
nal attenuation is increased such that dips appear in the transmission
a2 − 2r a cos φ + r 2
T = . (3.20)
1 − 2r a cos φ + (r a)2
The attenuation at the dips is equal to the single-pass attenuation magni-
fied by the phase sensitivity (for r < a). The width of the resonance also
broadens, lowering the finesse:
2π π 2π
F=   → ≈ 2 , (3.21)
2r a r a≈1 1 − ra t + αL
2 arccos
1+(r a)2

where for small losses, αL is the fraction of power lost per round-trip. If
the attenuation is comparable to the cross-coupling, light is resonantly
attenuated strongly. Under critical coupling, (r = a) or (t 2 = αL) the fi-
nesse drops by a factor of 2 and more importantly the transmission at res-
onances drops to zero. The circulating intensity peaks are diminished and
the phase sensitivity is paradoxically increased (Eq. 3.16). At resonances,
the phase sensitivity increases without bound at the expense of a de-
creasing transmitted signal until the transmission is zero and the phase
sensitivity is infinite. Of course the phase only undergoes a finite and dis-
crete phase jump at this point. If the resonator is dominated by bending
or scattering loss, then the waveguide mode is coupled perfectly to the
continuum of outward propagating waves outside the resonator. Under-
coupling occurs when the loss exceeds the coupling strength (r > a) or
(t 2 < αL). Many counter-intuitive effects may take place in this regime
such as the inversion of the phase sensitivity. Overcoupling occurs when
the round-trip loss does not exceed the coupling strength (r < a) or
(t 2 > αL) and is the conventional mode of operation for an all-pass res-
onator. Figure 3.10 displays the effect of attenuation on the transmission
and build-up for the overcoupled, critically-coupled, and undercoupled
regimes. An experimental demonstration of the transmission character-
istics in each of these regimes for a fiber ring resonator can be found
in [153]. Finally it is worth examining the introduction of gain. Gain may
be implemented, if possible, to offset loss mechanisms and to restore the
all-pass nature of a normally lossy ring resonator. Of course if the round
trip gain exceeds the net round-trip loss due to attenuation and coupling,
the resonator can exceed the threshold for lasing.
84 3. Optical Microresonator Theory

1.0 7
a = 0.99
a = 0.90
a = 0.99
a = 0.75 6
a = 0.96
a = 0.50 a = 0.96

0.7 5
a = 0.20

Build-up factor

a = 0.00 a = 0.90

0.3 2
a = 0.75
0.1 a = 0.50
a = 0.20

0.0 0
-2 0 2 -2 0 2
Normalized detuning, rad Normalized detuning, rad

Fig. 3.10. A plot of the (a) net transmission and (b) buildup versus nor-
malized detuning for an all-pass resonator with r = 0.75 and varying
loss. The single-pass field transmission a is displayed for each curve in
the figure.

3.3 Add–Drop Ring Resonators

The direct waveguide analogy of a free-space Fabry–Perot is obtained by

adding a second guide that side-couples to the resonator as in Fig. 3.11.
Because this configuration behaves as a narrow-band amplitude filter
that can add or drop a frequency band from an incoming signal, it
is commonly termed an add–drop filter. Because this configuration is
mathematically equivalent to the extensively studied classic Fabry–Perot
interferometer, the equations for transmission coefficients are simply
stated: The intensity throughput coefficient corresponding to light by-
passing the lower excitation waveguide is

4r 2 φ
I2 r22 a2
− 2r1 r2 a cos φ + r12 (
1−r 2
) 2
T1 = = →
I1 1 − 2r1 r2 a cos φ + (r1 r2 a)2 a=1,r1 =r2 ≡r 1+
4r 2
2 sin2
( 1−r 2 ) 2

3.3 Add–Drop Ring Resonators 85

Fig. 3.11. Fields associated with an add–drop ring resonator.

This corresponds to a transmitted signal modified such that a narrow

frequency band4 has been extracted; see Fig. 3.1. The extracted band exits
at the drop port with transmission coefficient:

I5 1 − r12 1 − r22 a 1
T2 = = →
2 a=1,r =r ≡r 4r 2
2 φ
I1 1 − 2r1 r2 a cos φ + (r1 r2 a) 1 2 1+ 2 sin 2
(1−r 2 )
Although there are similar enhancements (as in the all-pass resonator) in
the effective phase shifts at the two output ports, they are intermingled
with the dominant amplitude effects. Hence, although the phase response
should not be ignored (particularly for multiple interacting resonators)
it is rarely implemented for its phase properties.
It is worth stating an interpretation of the add–drop configuration.
Whereas in the lower coupler, interference exists between the input and
the circulating field, the upper coupler does not display any interference
(provided that excitation is from the lower guide only). The upper coupler
may thus be viewed simply as a “tap” waveguide that leaks power out of
the cavity into the drop port. This tap is formally equivalent to a lumped
loss. For an add–drop filter, it is desirable to operate at critical coupling
for complete extinguishment of a band in the through guide. Here the
sum of all losses incurred in the resonator including at the out-coupled
drop port must be taken into account.

3.3.1 Intensity Buildup

The buildup factor for an add–drop resonator is given by

I3  E3 
B= =E 
I1 1

1 − r12 r22 a2
= (3.25)
1 − 2r1 r2 a cos φ + (r1 r2 a)2
The shape of the transmission curve, as with the buildup, is approximately
Lorentzian centered on a resonance.
86 3. Optical Microresonator Theory

1 − r2 r2
→ (3.26)
a=1,r1 =r2 ≡r 1 − 2r 2 cos φ + r 4

r2 1
→ ≈ 2, (3.27)
φ=m2π 1−r 2 t

where the second-to-last result refers to the situation where attenuation

is negligible and the resonator is coupled critically through balanced cou-
plers (a = 1,r1 = r2 ). The last result refers to the situation in which addi-
tionally, the incident light is resonant with the ring (φ = m2π ). A com-
parison with the buildup expression for an all-pass resonator (Eq. 3.12)
reveals that the intensity buildup in the add–drop configuration is only
1/4-th that in the all-pass configuration for the same coupling strength.
Figure 3.12 displays the coherent buildup of intensity in an add–drop
microring resonator. Figure 3.13 compares the transmission, phase, in-
tensity buildup, and group delay for all-pass and add–drop resonators
with common coupling strengths.

Fig. 3.12. Finite-difference time-domain simulation demonstrating res-

onator buildup and rerouting (channel dropping) at 1.5736 µm. Parame-
ters are the same as in Fig. 3.6.
3.3 Add–Drop Ring Resonators 87

All-pass resonator Add-drop resonator

T2 , Φ 2

φ, B φ, B
Input T,Φ Input T1 , Φ1

(a) 1.0 (c) (e) 5π

Effective phase shift

T T1 T2 4π


.50 Φ2 2π
Φ Φ1
0 -π
(b) 20 (d) (f) 20

Norm. group delay

15 15

10 B dΦ/dφ 10

5 5
B B dΦ2/dφ
0 0
-π 0 π 2π 3π -π 0 π 2π 3π -π 0 π 2π 3π
Norm. detuning, φ =TRω Norm. detuning, φ =TRω Norm. detuning, φ =TRω

Fig. 3.13. Amplitude transmission (solid lines) and effective phase shift
(dashed lines) for (a) an all-pass resonator, (c) through port of an add–
drop resonator, and (e) drop port of an add–drop resonator. Plots (b), (d),
and (f) display the coherent intensity buildup (solid lines) and group delay
normalized with respect to the cavity transit time (dashed lines) for the
same ports as in (a), (c), and (e), respectively. The independent variable
on all plots is the normalized detuning, and all coupling coefficients are
t 2 = 0.1814.

3.3.2 Add–Drop Resonance Width ∆ω or ∆λ

The resonance width is defined as the FWHD of the resonance lineshape.

Using the expression for the drop-port output of an add–drop resonator,
we can write the transmission as

t12 t22 a 1 t12 t22 a

= . (3.28)
1 − 2r1 r2 a cos φ + (r1 r2 a)2 2 1 − 2r1 r2 a + (r1 r2 a)2

Solving this equation for the FWHD 2φ = ∆ωTR results in

1−(1−r1 r2 a)2
2 arccos 2r1 r2 a
∆ω = . (3.29)
88 3. Optical Microresonator Theory

Per the Euler formula, for small φ, cos φ = 1 − φ2 /2, so

(1 − r1 r2 a)2
φ2 = . (3.30)
r1 r2 a

In the case where the loss is negligible, i.e., a = 1, and the coupling is
symmetric, i.e., r1 = r2 ≡ r , the RHS of Eq. 3.30 is (1 − r 2 )2 /r 2 . Then,
(1 − r 2 )2
φ = (3.31)
2(1 − r 2 )
⇒ ∆ω = , (3.32)
r TR

and for weak coupling,

2t 2 c
∆ω = . (3.33)
Translating to wavelength,

∆λ ≈ ∆ω, (3.34)
2π c
where λ0 is the free-space wavelength, and we have assumed λ0  ∆λ.
t 2 λ20
∆λ ≈ . (3.35)
π Lneff
A more elegant expression for ∆ω may be obtained by treating the
coupling to the bus waveguides as a distributed loss. We define αdis
as αdis = αring + αthrough + αdrop , exp(−αthrough L) = r12 = 1 − t12 , and
exp(−αdrop L) = r22 = 1 − t22 . Then, eq. 3.30 gives

∆ωTR 1 − exp(−αdis L/2)

= (3.36)
2 exp(−αdis L/4)

If αdis L  1, we get

2 αdis L
∆ω =
TR 2
αdis L cαdis
= = . (3.37)
TR neff

The above equation reduces to Eq. 3.33 for t1 = t2 ≡ t  1 since

exp(αdis L) = 1 − αdis L ⇒ αdis L = 2t 2 . (3.38)

3.4 More on Concepts Associated with Resonators 89

3.3.3 Free Spectral Range (FSR)

The separation of successive resonances is termed the FSR. For negligible

internal GDD, ring resonators possess a discrete impulse response and
hence a periodic spectral response. At resonance, ωTR = m2π , where TR
is the round-trip time and m is an integer. Two successive resonances,
ω1 and ω2 , are then related as

FSRfrequency = ω2 − ω1 =
2π c
= . (3.39)
Translating to wavelength, we get

FSRwavelength = . (3.40)

3.3.4 Finesse F

The finesse F is defined as the ratio of FSR and resonance width. It is

thus a convenient measure of the sharpness of resonances relative to
their spacings. Using Eqs. 3.37 and 3.39,
1 2π c
F = cαdis
neff Lneff

= . (3.41)
αdis L
In the case where internal loss is negligible and coupling to the bus
waveguides is symmetric and weak (t1 = t2 ≡ t  1), using Eq. 3.38,
we get
F = 2. (3.42)
A particularly compact approximate expression for the drop-port trans-
mission incorporating the finesse results from making the Lorentzian
T2 →
2 . (3.43)
a=1,r1 =r2 ≡r ≈1 φ
1 + π /F

3.4 More on Concepts Associated with Resonators

3.4.1 Quality Factor Q

The quality factor of a resonator is a measure of the sharpness of the

resonance relative to its central frequency. The Q is formally defined
90 3. Optical Microresonator Theory

as the ratio of the stored energy circulating inside the resonator to the
energy lost per optical cycle:

Stored energy
Q = ω0 . (3.44)
Power loss
Since power loss is a temporal phenomenon, here, we must examine
the transient response. Let us consider the behavior of a ring that has
been charged to an intensity |E0 |2 , after which time the input is abruptly
switched off. For a circularly symmetric microring resonator, the loca-
tion inside the resonator where we measure the circulating intensity is
arbitrary. The intensity after the nth round-trip is:

|En |2 = exp(−αdis L)|En−1 |2 (3.45)

= exp (−nαdis L) |E0 | . (3.46)

If n is large, we can treat it as a continuous variable and get

d|En |2
= −αdis L|En |2 . (3.47)
We can now relate this to the power loss, as each round-trip takes
time TR . Since the power loss is energy lost per unit time, d|En |2 /dt =
(1/TR )d|En |2 /dn. Therefore,

|En |2
Q = ω0
−d|En |2 /dt
ω0 T R
= . (3.48)
αdis L
From Eqs. 3.37 and 3.48,

ω λ
Q= ≈ . (3.49)
∆ω ∆λ
Implementing Eq. 3.41, we arrive at the relationship between the quality
factor and the finesse:
ω0 T R
Q= F (3.50)

ω0 Lneff
Q = F (3.51)
2π c
neff L
≈ F (3.52)
= mF . (3.53)

For most optical resonators of interest, the optical path length within
a cavity cycle or ring circumference is typically many wavelengths long.
3.4 More on Concepts Associated with Resonators 91

The order (azimuthal number) of a particular resonance is a measure

of the number of wavelengths within the circumference, m = neff L/λ0 .
The order is also indicative of the mth peak in the spectrum and directly
relates the quality factor to the finesse.
A more rigorous approach in deriving the quality factor uses the
Laplace transform to go from the steady-state frequency response to the
transient response. We consider the circulating field E4 in an add–drop
E4 it1
= . (3.54)
E1 1 − r1 r2 aei(δω)TR
Although the choice of E4 is arbitrary, for a low-loss ring, E4 is a reason-
able approximation to the field everywhere in the ring. We can identify
iδω as s and use the Euler formula to write exp[i(δω)TR ] = exp(sTR ) =
1 + sTR . Then,
E4 it1 /TR
= , (3.55)
E1 (1 − A )/TR − A s
where A = r1 r2 A. If we switch off the input, the transient response is
given by L−1 1/[(1 − A )/TR − A s], so that

E4 (t) ∼ L−1 (3.56)
(1 − A )/TR − A s
1 − A
∼ exp − t . (3.57)

Identifying the coupling to the bus waveguides as a distributed loss, we

get A = exp(−αdis L/2) = 1 − αdis L/2. The Q is then

|E4 |2
Q = ω0
−d|E4 |2 /dt
= ω0
2 (1 − A )
ω0 TR
= , (3.58)
αdis L

which is the same result as obtained in Eq. 3.48.

3.4.2 Physical Significance of F and Q

To find the physical meaning of the finesse and Q, we consider the num-
ber of round-trips made by the energy in the resonator before being lost
to internal loss and the bus waveguides. If we define N as the number of
round-trips required to reduce the energy to 1/e of its initial value, we
92 3. Optical Microresonator Theory

exp (−αdis NL) = (3.59)
⇒N = (3.60)
αdis L
⇒F = 2π N. (3.61)
Equation 3.61 tells us that the finesse represents, within a factor of 2π ,
the number of round-trips made by light in the ring. Similarly,
Q = ω0 TR N (3.62)
tells us that Q represents the number of oscillations of the field before
the circulating energy is depleted to 1/e of the initial energy. In summary,
the expressions for buildup, finesse, and Q for all-pass and add–drop
resonators are related in the following manner:
π Q
Ball-pass = F = m (3.63)
π Badd–drop = F = m . (3.64)
Thus, the finesse and Q represent metrics for the intensity buildup and
effective interaction time, respectively, in a microresonator. Light inter-
acts with the coupling interface for a finesse number of times while in-
teracting with the cavity interior for a Q number of cycles. This insight
has implications for the design of applications relying on the change in
the transfer characteristics of a microresonator brought about by vari-
ations to the microresonator constituents. For instance, to construct an
enhanced microresonator-based sensor or switch operating on the vari-
ation of a distributed optical property such as the refractive index, it is
beneficial to make the cavity both large and with a high finesse. For this
application, the figure of merit characterizing the potential enhancement
is the quality factor. Alternatively, to construct a sensor or switch oper-
ating on the variation of a localized optical property such as the displace-
ment of a coupler or end mirror, it is no longer beneficial to make the
cavity large. Rather, it is often detrimental due to the increased sensi-
tivity to thermal and vibrational noise. For such applications, the finesse
completely characterizes the enhancement.

3.4.3 Phasor Representation

Phasor diagrams are highly intuitive representations of the interfer-
ence of successively delayed complex field amplitudes in a resonator.
Figure 3.14 graphically depicts the sum of phasors contributing to the
net complex output field for both an all-pass resonator and a critically
coupled or add–drop resonator. Note that for small detunings near reso-
nance (φ = m2π ), the net phasor in each case sweeps through π radians
very rapidly. In the critically coupled case, the output amplitude is zero
on resonance and grows as the net phasor sweeps away from the origin.
3.4 More on Concepts Associated with Resonators 93

All-pass resonator, r=0.9, a=1


−π / F
Imaginary( E2 / E1 )


0.0 φ = 0



-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
Real( E2 / E1 )

Critically coupled resonator, r1=r2=0.9


Imaginary( E2 / E1 )

0.5 −π/2F −4π/F

0.0 φ=0

+π/2F +4π/F


-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
Real( E2 / E1 )

Fig. 3.14. Graphical sum of phasors contributing to the transmitted com-

plex field for a single all-pass (r = 0.9) and critically coupled (r1 = r2 =
0.9 or r = a = 0.9 ) resonator. Shown are the phasor sums for equal
increments of π /2F . Note that the effective (net) phase sweeps through
π radians over the ±π /F bandwidth.
94 3. Optical Microresonator Theory

3.4.4 Kramers–Kronig Relations

A true all-pass resonator is necessarily over-coupled to achieve lossless

operation. The spectral phase response of a lossless ring resonator cou-
pled to a single waveguide varies dramatically with the coupling coeffi-
cient. In the limit of 100% cross-coupling, the ring is traversed only once
and the resonator simply imparts a linear phase and group delay that cor-
responds to the single-pass transit time. Seemingly paradoxically, as the
cross-coupling coefficient t 2 is reduced from 100%, the phase sensitivity
and group delay are increased near resonance. Intuitively, this increase
occurs because although smaller fractions of power are initially injected
into the ring, larger fractions of power are subsequently coherently main-
tained within the ring after each pass. As the coupling coefficient ap-
proaches zero, the slope or group delay increases without bound for a
lossless resonator. The phase difference through a resonance, however,
remains clamped at 2π radians for a single resonator. In practice, the
group delay cannot be made arbitrarily small due to attenuation mech-
anisms that can dramatically reduce throughput to zero at critical cou-
Because a resonator introduces a frequency-dependent group delay
peaking at resonance, it necessarily introduces GDD. Depending on the
application, this may be seen as an intended feature or an undesirable
side effect. From a theoretical standpoint, since a phase-only filter mod-
ifies the spectral phase of a signal without modification to its spectral
amplitude, it seems at first glance that this device has violated the well
known Hibert and Kramers–Kronig relations that hold for a causal de-
vice. However, it is also clear from examination of the impulse response
(a comb of weighted delta functions appearing only for positive times)
that causality has not been not violated.
Let us examine this apparent inconsistency in more detail. If the re-
sponse of system is causal (i.e., causes precede effects), then the impulse
response function is zero for all time values prior to zero. This allows one
to separate any function into even and odd components that are exact
although of opposite sign for all t < 0 and are identical for t > 0. Thus,
for all t < 0, they add destructively to yield zero, and for all t > 0, they
add constructively in equal proportions to form the impulse response
h(t) = heven (t) + hodd (t). (3.65)
The even and odd components are simply related in the time-domain by
the signum function

heven (t) = sgn(t)hodd (t) (3.66)

hodd (t) = sgn(t)heven (t). (3.67)
3.4 More on Concepts Associated with Resonators 95

If, furthermore, the system is linear, and time-translation invariant, then

a frequency-dependent transfer function can be assigned to spectral com-
ponents traversing the system. The Fourier transform of the impulse re-
sponse equations may be employed to generate the Hilbert transform

2 Himag (ω )
Hreal (ω) = + Himag (ω) = +2 dω (3.68)
ω ω − ω

2 Hreal (ω )
Himag (ω) = − Hreal (ω) = −2 dω . (3.69)
ω ω − ω

These relations apply to the real and imaginary parts of the transfer
functions of any causal signal. They are also known as one form of the
Kramers–Kronig relations [154, 155].
The impulse response function of a ring resonator with no internal
dispersion (i.e., impulses propagate within the ring without dispersing)
is a weighted sum of equally spaced delta functions.

- m−1
h(t) = r δ(t) − 1 − r 2 e+iφ0 r e+iφ0 δ(t − mTR ). (3.70)

An impulse response function consisting of a weighted sum of delta im-

pulses spaced by the cavity transit time TR in the time domain can be
interpreted as the Fourier series for some periodic function in the fre-
quency domain. The periodic function is of course the transfer function
whose fundamental period or FSR is equal to the inverse of the transit
time. The functions form a Z-transform pair that is simply a time-domain
version of a Fourier series equation pair for discrete time signals. The
transfer function is given by the following expression that may be sim-
plified by taking the limiting value of the infinite series (provided it is
convergent, satisfied by r < 1):

m−1 r − e+iωTR
H(ω) = r − 1 − r 2 e+iωTR r e+iωTR = . (3.71)
1 − r e+iωTR

Because the system is linear, time-translation invariant, and causal, the

real and imaginary parts of its complex transfer function form a Hilbert
transform pair

− cos ωTR + 2r − r 2 cos ωTR

Hreal(ω) = (3.72)
1 − 2r cos ωTR + r 2

− 1 − r 2 sin ωTR
Himag(ω) = . (3.73)
1 − 2r cos ωTR + r 2
96 3. Optical Microresonator Theory

Fig. 3.15. (a) Plots of the real and imaginary components of the com-
plex field transfer function for an all-pass resonator with r = 0.9. (b)
Plots of the amplitude square modulus (transmission) and phase. Note
that although an all-pass resonator selectively modifies the phase spec-
trally preserving flat, unit transmission, there is no violation of Hilbert
or Kramers–Kronig relations. The real and imaginary components of the
transmissivity do satisfy these relations but in a manner that results in a
flat square modulus transmission response.

Thus, the real and imaginary parts of the transfer function do in fact
satisfy the Kramers–Kronig relations. But although the real and imaginary
parts of the transfer function vary in a complicated manner, they do so
in such a way that the amplitude is always preserved to be unity. Figure
3.15 illustrates this process graphically. In certain cases it is possible to
formulate Kramers–Kronig relations for the amplitude and phase of a
transfer function. It is accomplished by taking the natural logarithm of
the transfer function — a procedure that maps the amplitude and phase
into real and imaginary components:

|H(ω)| = 1 (3.74)
r sin ωTR
arg [H(ω)] = π + ωTR + 2 arctan . (3.75)
1 − r cos ωTR
3.5 Higher Order Filters 97

Thus, in certain systems, if the natural logarithm of the transfer func-

tion is analytic in the upper half complex frequency plane, the amplitude
and phase of a transfer function form a Hilbert transform pair as well.
For all values of gain and some values of attenuation, with the excep-
tion of the undercoupled regime, zeros are present in the upper half
complex frequency plane of the response of the all-pass resonator. The
natural logarithm of the response is nonanalytic at the zeros. As a result,
Kramers–Kronig relations cannot be applied to the amplitude and phase
response of an all-pass resonator [156–158]. Nevertheless, In the under-
coupled regime, Kramers–Kronig relations for amplitude and phase do in
fact exist. These concepts apply equally to add–drop resonators treating
the drop coupling as an internal (tap) loss. Useful phase-modifying de-
vices can also be constructed from under-coupled ring resonators. Such
devices possess an inverted phase response near resonance, meaning that
the phase decreases with increasing frequency. However, high losses nec-
essarily accompany operation in this regime.

3.5 Higher Order Filters

The design of an optical amplitude filter base on ring resonators typ-

ically involves fixed trade-offs among extinction, ripple, and complex-
ity. The filter-order describes how the filter response behaves away from
the resonant wavelength or frequency. The drop-port amplitude for a
single-ring OCDF behaves as 1/{1 − r1 r2 A[1 + i(ω − ω0 )TR ]}, so the
drop-port intensity behaves as 1/[(r1 r2 ATR )2 (ω − ω0 )2 + (1 − r1 r2 A)2 ]
(near ω0 ), a Lorentzian. By higher order response, we refer to the roll-off
compared to a single resonator device. A single-ring device has a roll-
off of 20 dB/decade and is referred to as a first-order filter, because the
drop-port amplitude behaves as 1/ω. In some applications, such as band-
pass filters, we require improved rejection out-of-band (away from res-
onance). We can draw an analogy between the ring as a reactive optical
element and an electronic filter with LC elements; cascading reactive el-
ements makes the filter response steeper away from resonance. Rings
can be cascaded serially (Fig. 3.16) or in parallel (Fig. 3.17), allowing the
fabrication of higher order filters with increased FSR and faster roll-off
compared with a single ring resonator device. In both cases, the Vernier
effect can enhance certain resonances and suppress others, resulting in
a wide free spectral range. In a serial cascade, the resonators are coupled
directly to each other. In a parallel cascade the resonators are coupled
via the bus, and the filter response depends on the length of the bus
waveguide between the rings. For example, Fig. 3.18 compares the roll-
off of a parallel-cascade of three rings and a single-ring filter. The x-axis
is normalized to the bandwidths of the two filters.
98 3. Optical Microresonator Theory

Fig. 3.16. Serially cascaded rings.

Fig. 3.17. Rings cascaded in parallel.

Fig. 3.18. A comparison of the roll-off for a first- and third-order optical
ring resonator filter.
3.5 Higher Order Filters 99

Mathematically, the reponse of some filters may be written as [152,

P1 (s)
H(s) = , (3.76)
P2 (s)
where P1 and P2 are polynomials, and the filter response is evaluated near
a resonance; sTR  1, s = −i(ω−ω0 ), where ω0 is a resonant frequency.
For a single-ring OCDF, the drop-port response is

Ed −t1 t2 A1/2 (1 + s)
= . (3.77)
Ei (1 − r1 r2 A) − sr1 r2 A
The zeros of P1 are called the zeros of the filter, and the zeros of P2
are called the poles of the filter. If we assume that P1 and P2 have no
zeros in common, the filter-order is determined from the order of P2 ; in
a first-order filter, the polynomial P2 is of the form (s − p), i.e., a first-
order polynomial; the single-ring OCDF response shown above, therefore,
qualifies as a first-order
4 response. For an nth order filter, the polynomial
P2 is of the form n (s −pn )n , where the poles pn may be degenerate. The
asymptotic behavior of the filter is s −n , so it has a sharper roll-off than a
first-order filter. At the same time, by using interference effects between
multiple filters, a flatter top compared with lower order filters can be
obtained. Sharper roll-off is desirable in applications such as band-pass
filters for better isolation between channels. A flat-top minimizes signal
distortion in similar applications.
We go over parallel-cascaded ring filters here. Serially coupled ring
filters have been analyzed elsewhere [15,152,160,161] and demonstrated
by several groups [15,18,55,162]. Parallel-cascaded ring filters have been
discussed by Little et al. [163], Griffel [164], Melloni [165], and Grover
et al. [63], and have been demonstrated by a few groups [22, 63]. The
analysis follows [63].
The single microring device with the transfer matrix formalism is rep-
resented as:     
Ei T11 T12 Et
= , where (3.78)
Ed T21 T22 Ea

1 − r1 r2 AΦ
T11 = , (3.79)
r1 − r2 AΦ
t1 t2 A1/2 Φ1/2
T12 = , (3.80)
r1 − r2 AΦ
T21 = −T12 , and (3.81)
r1 r2 − AΦ
T22 = . (3.82)
r1 − r2 AΦ
Now, consider a parallel cascade of microresonators separated by Λ
(Fig. 3.17). The transfer matrix of each bus section of length Λ, Tφ , is
100 3. Optical Microresonator Theory

∗ Λ
Φbus exp −αbus 2 0
Tbus = ⎣
⎦, (3.83)
0 Φbus exp −αbus Λ

where Φbus = exp(iβbus Λ), βbus is the propagation constant in the bus
waveguide, and αbus is the loss per unit length in the bus. Using the
transfer matrix defined in Eq. 3.78,
Ei1 EtN
= T1 · Tbus · T2 · Tbus . . . TN , (3.84)
Ed1 EaN

where Eim is the input-port field for the m-th resonator, Edm is the drop-
port field, Etm is the through-port field, and Eam is the add-port field.
To show that the response of a multiring cascade has higher order
behavior, we only need to show that a double-ring cascade has second-
order behavior; the behavior of cascades with more rings will follow by
induction. The response for a double ring cascade is:
Ei1 Et2
= T1 · Tbus · T2 . (3.85)
Ed1 Ea2

If we assume that the two rings are similar and assume that the bus
waveguide has no loss, the net transfer function T2RP is given as

Fig. 3.19. A triple ring parallel cascade (3R) has faster roll-off than a single
ring (1R). The simulation parameters are αring = 5 × 10−8 nm−1 , L =
√ ring
60004 nm, t1 = t2 = 0.1, neff = 3.1, αbus = 0, Λ = 39000 nm, nbus eff =
3.15. The x-axis has been normalized to the bandwidths of the two filters.
3.5 Higher Order Filters 101




Wavelength, a. u.
Fig. 3.20. Vernier effect with a multi-ring cascade. The resonances of the
rings that coincide with those of the bus are enhanced while the others
are suppressed.


T11 T12 Φbus 0 T11 T12
T2RP =
T21 T22 0 Φbus T21 T22

T11 T12 T11 Φbus T12 Φbus
= ∗
T21 T22 T21 Φbus T22 Φbus
2 2 ∗
(T11 + T21 )Φbus (T11 + T22 )T12 Φbus
= ∗ 2 2 . (3.86)
(T11 + T22 )T21 Φbus (T22 + T21 )Φbus
102 3. Optical Microresonator Theory

Normalized power



1550 1560 1570 1580 1590
Wavelength, nm
Fig. 3.21. Simulated drop port response of triple ring parallel cascade
OCDF (bottom) and single ring OCDF (top). The simulation parameters
are αring = 5 × 10−6 nm−1 , L = 60004 nm, t1 = t2 = 0.1, neff = 3.1,
αbus = 10−6 nm−1 , Λ = 39368 nm, nbus eff = 3.15. With appropriate choice
of bus length, it is possible to suppress some resonances, giving a wide

Ed2 T12
= 2RP
Ei2 T11
(T11 + T22 )T12 Φbus
= 2 2 ∗
. (3.87)
(T11 + T21 )Φbus
We can now substitute the expressions for the various terms on the RHS
from Eq. 3.78 and see that the numerator and denominator on the RHS
have no common factors. So the roll-off near a resonance is determined by
the denominator. Since both T11 and T21 are polynomials of first-order in
Φ = exp(−iωTR ) ≈ 1 − i(ω − ω0 )TR , or s = −i(ω − ω0 ), the denominator
3.6 Summary 103

is a polynomial of second order in s. Thus the double-ring parallel cas-

cade behaves as a second-order filter. Similarly, a single ring cascaded in
parallel with an n-ring cascade (with n-th order response) has a response
of order (n + 1), etc. In other words, a parallel-cascaded N-ring OCDF has
a roll-off N-times faster than a single-ring OCDF. The faster roll-off possi-
ble with a multi ring cascade is shown in Figs. 3.18 and 3.19 for the case
where there are three rings. If the spacing, Λ, between the resonators is
chosen such that
Nr · FSRr = Nb · FSRb = FSRa , (3.88)
where FSRr is the FSR of the resonator, FSRb is the FSR of the bus section
between the resonators, and Nr and Nb are integers, then the Vernier
effect causes the suppression of transmission peaks within the spectral
range FSRa , resulting in an Nr -fold increase of the effective FSR. Since
FSRr = 2π c/nr L, and FSRb = π c/nb Λ, Eq. 3.88 gives

Nb nr L
Λ= (3.89)
Nr n b 2

for an effective Nr -fold increase in the FSR of the filter. The Vernier effect
is illustrated in Fig. 3.20, and the simulated drop-port response of a triple-
ring parallel-cascade OCDF is shown in Fig. 3.21.
There are significant practical limitations to the number of resonators
that can be cascaded in parallel or series. Both resonator and bus loss
limit the number of resonators, as does the fabrication accuracy, in turn
affecting the ability to exactly tailor the resonator and bus response for
using the Vernier effect. In a parallel cascade, if the loss in the bus is
high, the amount of light reaching the rings down the cascade will not
be enough to enable them to affect the filter response; in that case, the
response will be of a lower order than designed. In a serial cascade, if the
loss in the ring is high, no power will reach the output bus.

3.6 Summary

In this chapter we introduced fundamental concepts associated with opti-

cal microresonators. We began with a treatment of the traditional Fabry–
Perot resonator, using it as a familiar starting point to present the key
concepts associated with all-pass and add–drop microresonators. We con-
cluded with a treatment of higher order filters - a topic we will return to
albeit with a different slant in the chapter 7.
4. Microring Filters: Experimental Results

This chapter presents the results of experiments performed on both

passive and actively tunable microring resonators functioning as optical

4.1 Passive Resonators

The results presented in this section represent a small subset of work in
the linear optics regime on microresonators from our group and other
groups. We refer the reader to [51, 55, 57, 58, 60, 62–64, 71, 76] for more.
Examples of linear performance are summarized in Table 4.1.
Optical microring resonators require a mechanism for coupling light
into and out of a mode of the ring. This coupling is best accomplished
through evanescent coupling of the resonator mode to that of the bus
waveguide(s) situated in close proximity. Microring resonators can be
made with laterally (in-plane) or vertically (out-of-plane) coupled bus
waveguides (Fig. 4.1). III–V semiconductor devices based on lateral coupl-
ing require the use of advanced fabrication technologies (high-resolution
e-beam lithography, facet-quality-dry-etching with etch rate independent
of trench width) to achieve reproducible filter bandwidths and high drop-
ping efficiencies. In the lateral coupling case, the separation between the
waveguides and the ring, typically less than 0.3 µm, is achieved by e-
beam lithography. E-beam lithography is affected by processing condi-
tions (humidity, temperature), the age of the cathode, and time taken by
the machine to be stable. These problems are exacerbated by the require-
ments of small gaps that can be difficult to etch, making reproducible
bandwidth and high dropping efficiency challenging to achieve.
Vertical coupling is a more robust and reproducible architecture as
it can be accomplished with standard optical lithography. In the verti-
cal coupling approach, the coupling gap is defined by material growth
or deposition. Furthermore, the inherently symmetric structure with re-
spect to resonator input and output coupling can result in better transfer
efficiency, as demonstrated in glass with a process that requires redepo-
sition and planarization [21]. Vertical coupling typically requires wafer
bonding and growth-substrate removal, which can be problematic and
106 4. Microring Filters: Experimental Results

(a) Lateral coupling

(b) Vertical coupling

Fig. 4.1. Schematic of a microring with (a) laterally and (b) vertically
coupled bus waveguides. The waveguide core is shown in red, and the
cladding is shown in blue.
4.2 Active Resonators 107

Table 4.1. Examples of device performance in the linear optical regime.

L: circumference, L-GaAs: laterally coupled GaAs-AlGaAs, V-GaAs: verti-
cally coupled GaAs-AlGaAs, L-InP: laterally coupled GaInAsP-InP, V-InP:
vertically coupled GaInAsP-InP.
Type L, µm ∆λ, nm FSR, nm Q F Reference
L-GaAs 51 1.3 18 1200 14 [173]
V-GaAs 63 0.22 10.4 7040 47 [173]
L-InP 82 0.25 8 6250 32 [174]
V-InP 29 0.24 24 6200 100 [174]

may limit yield in commercial applications by dictating a maximum size

for void-free bonding. Wafer bonding can be achieved with polymers like
benzocyclobutene or polyimides [166]; planarization followed by solders
like Pb-Sn [167] Pd [168], Au-Sn [169], or Pd-In [170, 171]; and wafer fu-
sion [172]. The use of solders and wafer fusion provides a conducting
interface, which is advantageous for active devices, as contacts can be
made to the (doped) substrate.
Laterally coupled microring resonators remain important for the even-
tual realization of large-scale integrated photonic circuits, where we must
integrate active and passive devices on the same chip. By combining the
vertical- and lateral-coupling approaches, all the passive elements can be
made with lateral coupling on the same layer as the (passive) bus, whereas
the active elements can be coupled to the bus vertically.

4.1.1 GaAs-AlGaAs

Fabricated GaAs-AlGaAs devices possess a GaAs core and an AlGaAs

cladding. The response of single- and double-ring GaAs-AlGaAs laterally
coupled resonators is shown in Fig. 4.2. Note the Lorentzian first-order
response, and the steeper roll-off for the double ring device. The response
of a vertically coupled device is shown in Fig. 4.3.

4.1.2 GaInAsP-InP

Fabricated GaInAsP-InP devices possess a GaInAsP core and InP cladding.

The response of a single-ring GaInAsP-InP laterally coupled resonator is
shown in Fig. 4.4. The response of a vertically coupled device is shown in
Fig. 4.5.

4.2 Active Resonators

Most devices demonstrated so far have been static passive. Static passive
devices can function only at certain wavelengths that cannot be changed
108 4. Microring Filters: Experimental Results

Fig. 4.2. Single and multiple ring GaAs-AlGaAs laterally coupled micror-
ing add/drop filters (left), and single and double-ring filter drop-port re-
sponses (right). Copyright Philippe P. Absil, 2000, used with permission.

Fig. 4.3. Optical micrograph of vertically coupled GaAs-AlGaAs micror-

ing add/drop filter (right), and 10 µm-radius ring filter response (left).
Copyright Philippe P. Absil, 2000, used with permission.

once the device has been fabricated. This limitation is serious, because it
is not always possible to fine-tune the fabrication process for the exact fil-
ter wavelength. Even small variations (tens of nanpmeters) in waveguide
width can affect the resonance wavelength.
However, if it were possible to modify the wavelength of operation
after fabrication — say, when the device is part of an optical network —
fabrication errors might be trimmed. Also through the use of gain and
loss modulation, resonators can be implemented as switches, amplifiers,
4.2 Active Resonators 109

Fig. 4.4. Scanning electron micrograph of a laterally coupled microring

add/drop filter (upper), and drop port response of an InP-based laterally
coupled add/drop filters (lower).

Fig. 4.5. Optical micrograph of vertically coupled GaInAsP-InP microring

add/drop filter (left), and 10 µm-radius ring filter response (right).

and routers. Thus, active microring resonator-based devices continue to

be an important topic of research. The first active ring resonator was
demonstrated by Tietgen in 1984 [13]. Since then, Rabiei et al. have
110 4. Microring Filters: Experimental Results

demonstrated active polymer microring resonators [40], Rabus et al. have

demonstrated active GaInAsP-InP microring resonators [73], and Grover
et al. have demonstrated active GaInAsP-InP microring resonators [77].
Microresonators can be tuned with temperature [52, 71]. Some of the
first devices on the market (from Little Optics, Inc.) also use thermal tun-
ing. Thermal tuning has the advantage that it has no polarization depen-
dence. However, it can be very slow (tuning speeds of milliseconds), and
can create polarization dependence in waveguide properties by causing
stress and consequent bireferingence.
The refractive index can also be altered via by carrier injection. Such
resonators have been demonstrated by the group at the University of
Southern California [65, 175].
The electro-optic effect has the highest potential for low-loss and high-
speed operation. The electro-optic effect is related to the nonlinear sus-
ceptibility of the material, so we can use the values of χ (2) or r (the lin-
ear EO coefficient) and χ (3) or s (the quadratic EO coefficient) to choose
the material system (GaAs or InP in the case of III–V semiconductors)
for operation at the signal wavelength. For GaInAsP, Adachi and Oe pro-
vide analytical expressions for the linear electro-optic effect in [176], and
the quadratic electro-optic in [177]. However, published data are of lim-
ited usefulness because of the wide range of parameters in the material
and waveguide design and the detuning of signal wavelength from the
bandgap. As detuning from the bandedge is reduced, the electro-optic
effect becomes stronger, albeit at the expense of increased insertion loss.
Nevertheless, the published data can prove useful as a sanity check on
experimental results.

4.2.1 Electro-Optic Tuning

The general form of the nonlinear polarization is [178]

(2) (3)
PiNL (t) = 4π χijk Ej Ek + χijkl Ej Ek El , (4.1)

where repeated indices imply summation and c.g.s. units are used. InP
and its quaternaries have a tetragonal structure, and belong to the crystal
class 422. The nonvanishing elements of the tensor χ (2) for this class are
xyz = −yxz, xzy = −yzx, zxy = −zyx. Of the 81 elements of the
tensor χ (3) , 21 are nonvanishing. They are

xxxx = yyyy, zzzz

yyzz = xxzz, yzzy = xzzx, xxyy = yyxx,
zzyy = zzxx, yzyz = xzxz, xyxy = yxyx,
zyyz = zxxz, zyzy = zxzx, xyyx = yxxy.

Consider the case where the direction of propagation is z, in the plane

of the wafer, and the x-axis is the direction of crystal growth, (100), i.e.,
4.2 Active Resonators 111

out-of-wafer-plane. In the presence of a constant electric field along e1 ,

and harmonic field Ei cos(ωt)e1 , the nonlinear polarization, P1NL (t), has
the form
. /
(2) (3)
P1NL (t) = 4π χ111 E1 (t)2 + χ1111 E1 (t)3
. /
(2) (3)
= 4π χ111 [Ei cos(ωt) + Edc ]2 + χ1111 [Ei cos(ωt) + Edc ]3
.  /
(2) (3)
P1NL (ω, t) = 4π χ111 [2Edc Ei cos(ωt)] + χ1111 3Edc 2
Ei cos(ωt) . (4.2)

In Eq. 4.2, the P1NL (ω, t) notation indicates that we are considering only
that part of the polarization that oscillates at the signal frequency; the
other components oscillate at 2ω or just add a constant background. The
nonlinear polarization has two terms: the first is linear in Edc and gives
rise to the linear electro-optic effect. The second term is quadratic in Edc
and causes the quadratic electro-optic effect. For III–V semiconductors,
where χ111 = 0, the relevant effect for the polarization and direction of
electric field chosen is the quadratic effect. Then, since n2 = 1 + 4π χeff ,
(3) 2
6π χ1111 Edc
n(ω) = n0 (ω) + , (4.3)
n0 (ω)

for the direction of the constant electric field and the polarization chosen.
This result is referred to as the quadratic electro-optic effect since the
refractive index can be varied as the square of the electric field.
Near the band-edge, the quadratic effect is caused primarily by the
Franz–Keldysh effect in bulk and the quantum-confined Stark effect
(QCSE) in multiple-quantum-well (MQW) waveguides. The two effects are
related; Miller, Chemla, and Schmitt-Ring have shown that the Franz-
Keldysh effect is a limiting case of the QCSE [179]. Typical values of
χ (3) for InP are ∼10−13 cm2 /W [180]. The QCSE develops from bandgap
shrinkage shifting the excitonic resonance closer to the signal wave-
length [181, 182]. The resultant change in the absorption coefficient
causes a change in refractive index via the Kramers–Kronig relations.
The Franz–Keldysh effect has a similar mechanism, except that exci-
tons are not involved; when a bias is applied, the carriers move apart in
bulk materials destroying the exciton resonance. The exciton resonance
is maintained in quantum wells due to carrier confinement [183]. The
quantum-confined effect is stronger than the bulk effect.
Although both the linear and the quadratic electro-optic effect can be
described by the nonlinear susceptibility, they have traditionally been de-
scribed in terms of the linear and quadratic electro-optic coefficients [178,
184] given as

1 1
∆n1 = − Γ n30,1 r11 E1 − Γ n30,1 s1111 E12 , (4.4)
2 2
112 4. Microring Filters: Experimental Results

where E1 e1 is the electric field, n0,1 is the refractive index in the absence
of field for the out-of-plane polarization, Γ is the confinement factor, r11
is the linear electro-optic coefficient (contracted notation), and s1111 is an
element of the quadratic electro-optic coefficient tensor.

4.2.2 Material Characteristics

To demonstrate electro-optic tuning, Grover et al. use a p-i-n diode struc-

ture, with the intrinsic region forming the core of the waveguide. Their
intrinsic region consists of a 25-quantum-well superlattice, to maximize
refractive index change via the QCSE.
To confirm that the wafer behaved as a diode, the electrical character-
istics of the wafer were tested by putting contacts on either side of the
wafer, and by using a semiconductor parameter analyzer (Agilent 4155B).
The characteristics are shown in Fig. 4.6 and validate the electrical de-
sign, but they introduce a wrinkle in the form of a high reverse leakage
current leading to significant thermal effects playing a role in devices em-
ploying reverse bias. To find the electro-optic coefficient of the material,
a free-space Mach–Zehnder interferometer was constructed (Fig. 4.7) to
test a slab waveguide made from the material. Composite images of the
interference pattern versus voltage for the S (out-of-wafer-plane) and P
(in-wafer-plane) polarizations are shown in Fig. 4.8.
As can be observed, the interference pattern shifts with voltage for
both polarizations. For the S-polarization, at 5 V, the phase change is
2π because the maxima and minima coincide with those at 0 V. If we
assume that the effect is purely quadratic in nature (such is the case for
the quadratic EO effect in InP for the S-polarization), we obtain

Fig. 4.6. Electrical characteristics of diode made with wafer used for
electro-optically tuned ring resonators. The diode area is 100×2000 µm2 .
4.2 Active Resonators 113

Fig. 4.7. Mach-Zehnder interferometer for EO coefficient measurement.

1 3 2
∆n = n sE , and (4.5)
2π L
2π = ∆n . (4.6)
Using λ = 1.55 µm, L = 0.2 cm, E = 10 V/µm (5 V), and n = 3.22 (effective
index of slab) gives

s = 2.3 × 10−15 cm2 /V2 . (4.7)

At this stage, it is tempting to identify s with the quadratic electro-optic

coefficient, but the high reverse leakage current means that the effect may
well be thermal — the voltage curve for the material is approximately
linear below 0 V to around 6 V, so both the thermal and the EO effect
are quadratic with voltage for the S-polarization. To distinguish between
the thermal and the electro-optic effects, we require high-speed (pulsed)
testing. If the absorption changes at high speeds, it implies that the QCSE
is changing absorption; hence, it is at least partly responsible for the
index change. Since the EO coefficient was measured under dc conditions,
the modulation depth should match low-frequency measurements.
First, the impedance of the device was measured up to 2 MHz by plac-
ing it in series with a 10 Ω resistor as shown in Fig. 4.9(a). The resistor had
an inductance of 4 µH (as measured by an inductance meter), so it is de-
picted as two elements, L2 and R2 . The device is represented as a resistor
114 4. Microring Filters: Experimental Results

(a) S-polarization (out-of-wafer-plane)

(b) P-polarization (in-wafer-plane)

Fig. 4.8. Interference pattern versus voltage. The empty rectangular box
represents the orientation of the slab, and the arrow indicates the E-field
4.2 Active Resonators 115


Fig. 4.9. (a) Setup for impedance measurement of slab waveguide, and
(b) voltage across R2 and L2 versus frequency.

R1 in parallel with a capacitance C. The capacitance is calculated from

the dimensions of the slab (100 µm × 2 mm plate, separated by 0.5 µm,
n2 = 3.3). R1 is calculated from the voltage drop across the resistor at low
frequencies. The measured data and the theoretical prediction are shown
in Fig. 4.9(b). Since the predicted curve provides a reasonable fit, it can
be concluded that the device impedance does not change appreciably up
to 2 MHz.
Next, the waveguide was modulated with sinusoidal signals from 1 Hz
to 2 MHz. The response was measured below 2 MHz with a 2 V peak-to-
peak signal offset below 0 by 1 V, a large-area photodetector (ThorLabs
PDA400) and a lock-in amplifier (Perkin-Elmer 7081 DSP Lock-in ampli-
fier); the frequency response is shown in Fig. 4.10. The thermal roll-off
frequency for the slab structure can be estimated to be 14 kHz or less by
solving the heat diffusion equation.
Since the frequency response is flat to 2 MHz, which is considerably
higher than the maximum predicted thermal roll-off frequency of 14 kHz,
116 4. Microring Filters: Experimental Results

Fig. 4.10. Frequency response of slab waveguide.

it is a reasonable conclusion that the absorption modulation is not ther-

mal in nature. It can be concluded that the absorption modulation is from
the quantum-confined Stark effect and that the index change observed in
the waveguide is due to the electro-optic effect. Since there is a large leak-
age current, the electro-optic coefficient measured (s) represents a lower
bound for the quadratic electro-optic coefficient of the waveguide.

4.2.3 Tuning Measurements

For the tuning measurements, the device was tested without antireflec-
tion coatings. Amplified spontaneous emission (ASE) from two cascaded
erbium-doped fiber amplifiers (EDFAs) was used as a source in obtaining
the spectrum of the device at various voltages.
The spectrum was normalized to account for the input EDFA spec-
trum, and increased absorption with bias in the bus waveguide due to
QCSE or increased temperature. The latter effect can be mitigated by pat-
terning the contact so that it is only on the ring, and by using a waveguide
core with larger bandgap. The data prior to normalization for two volt-
ages is shown in Fig. 4.11. To normalize the data to the EDFA spectrum,
the measured output power from the device was divded into a polynomial
fit to the EDFA power reading at each measurement wavelength (Fig. 4.12).
Finally, to account for the changing loss in the bus, the power was normal-
ized to the maximum power over the wavelength range at each voltage.
The device spectrum at two voltages is shown in Fig. 4.13. Although the
input has a mixed polarization, the device has very high loss for the in-
plane polarization, effectively filtering the in-plane polarization. So the
response shown in Fig. 4.13 is for the out-of-plane polarization.
The free spectral range of the device is >30 nm, so other peaks lie
outside the EDFA band. Due to the fused bus and ring, the resonator
possesses high loss, increasing the bandwidth to a few nanometers; as a
4.2 Active Resonators 117

Fig. 4.11. Raw data for active resonator at two voltages.

Fig. 4.12. EDFA spectrum and polynomial fit used to normalize the raw
spectral data for the active resonator.

1 ♦

+ ♦ ♦♦






+ ♦






+ ♦

+ ♦




+ ♦
♦ ♦


♦+ ♦
+ +

+ ♦



+ ♦



++ ♦
Normalized power

♦♦+ ♦♦


♦ + ♦
♦ ++

+ ♦♦+
♦ +
+ ♦♦

♦+ ♦

♦ +
+ ♦♦

+ ♦
♦ +
0.1 ♦ + ♦ + 0V ♦
♦♦++ ♦ ♦+

6V +

♦ + ♦ +
♦ +♦ ♦+ +

♦ +♦ + +



+ +

1548 1550 1552 1554 1556 1558 1560 1562
Wavelength, nm
Fig. 4.13. Change in resonance wavelength with voltage.
118 4. Microring Filters: Experimental Results

2.8882 ♦

Resonance wavelength, nm
♦ 1553.5
2.888 Experiment ♦
Quadratic fit ♦
Refractive index

2.8878 ♦
♦ 1553.3
♦ 1553.2
2.8874 ♦ ♦
2.8872 ♦ ♦
♦ ♦ 1553
♦ ♦ ♦ ♦
0 1 2 3 4 5 6 7 8
Reverse bias, V
Fig. 4.14. Effective refractive index and resonance wavelength versus volt-
age. The step-like behavior is from the resolution of the optical spectrum
analyzer used to acquire the device spectrum.

result, the Q of the resonator is low, about a few hundred. The resonance
wavelength of the resonator is red-shifted with reverse bias. For clarity,
the spectra for only two voltages are shown.
Figure 4.14 shows the change in effective refractive index with volt-
age. The resonance locations were determined by a minimum search rou-
tine employing polynomial fits on each resonance shape to overcome
noise, and then they were searched for the local minimum of the poly-
nomial. The resonance locations are stepped because of the resolution
of the optical spectrum analyzer (0.08 nm) used to collect the spectra.
The corresponding value of the resonance wavelength is shown on the
right side. The refractive index in the absence of a reverse bias is ob-
tained from simulations using a commercial waveguide mode solver (Op-
tical Waveguide Mode Solver from Apollo Photonics). Using Eq. 4.4 and
a quadratic fit to the data in Fig. 4.14, the quadratic electro-optic coef-
ficient for the waveguide can be obtained. The value of the coefficient
for the waveguide is a lower bound for the quadratic electro-optic coeffi-
cient of the core because of the high leakage current in the material when
under reverse bias. The tuning range of 0.8 nm, or 100 GHz, means that
the device is suitable for dense wavelength division multiplexing applica-
tions. The curve-fit in Fig. 4.14 allows us to calculate the quadratic electro-
optic coefficient for the waveguide as 4.9 × 10−15 cm2 /V2 at ∼ 65 meV
from the band edge for the waveguide. Since the quantum wells oc-
cupy roughly half the core of the waveguide, the quadratic electro-optic
4.2 Active Resonators 119

coefficient for the wells is double that value; i.e., sqw = 9.8×10−15 cm2 /V2 .
The obtained quadratic electro-optic coefficient value compares favorably
with the value of 5 × 10−15 cm2 /V2 reported by Fetterman et al. for a sim-
ilar MQW structure, at 113 meV from the band edge [184]. Higher values
of the quadratic coefficient may be obtained by operating closer to the
band edge; however, that has the deleterious effect of decreasing the sig-
nal transmission substantially. Zucker et al. report a much higher value
for their structure by operating 32 meV from the band edge [185].
The change in throughput with voltage at off-resonance wavelengths
allows us to estimate the increase in loss with electric field (Fig. 4.15). The
increased loss is due to bandgap shrinkage in the quantum wells [186].
The loss in the absence of any field is due to scattering by rough side-
walls and mode-mismatch between the modes of the straight and curved
sections of the resonator [122].
In earlier work, Grover et al. demonstrated resonance bandwidths as
narrow as 0.25 nm in the same material system, so it is possible to make
substantial improvements in the filter characteristics. Imperfect mask
definition in the coupling region during fabrication of the current series
of devices resulted in fused bus and resonators. The effect was a high
coupling between the bus and the resonator, giving us the broad (a few
nanometers) resonance of the device in this report.
As an example, let us consider the behavior of a high-finesse microring
resonator add–drop filter made from the same layer structure described
in this chapter. We can use the index change from Fig. 4.14 and the loss
from Fig. 4.15, τ1 = τ2 = 0.989, and A = 0.99 to predict the behav-
ior of a ring resonator with achievable behavior — the corresponding


Loss, cm−1



0 2 4 6 8 10 12 14 16
Electric field, V/µm

Fig. 4.15. Measured loss versus electric field in straight waveguides. The
loss in the absence of electric field is estimated as 15 cm−1 based on
results from similar passive waveguides. The change in loss will be the
same in the ring as in the bus, as the electric field affects only the internal
120 4. Microring Filters: Experimental Results

bandwidth is a little over 0.5 nm. The simulated drop-port response for
an add/drop filter at two voltages is shown in Fig. 4.16. The increased
loss affects the bandwidth and decreases the drop-port amplitude, and
it changes the extinction on the through port (not shown in the graph).
The simulated bandwidth versus electric field is shown in Fig. 4.17. Since
the bandwidth increases drastically over the tuning range, the material
bandgap must be modified to decrease the change in loss. Ultra-compact
tunable microring notch filters have been demonstrated in a p-i-n diode
geometry, with the intrinsic region forming the waveguide core. The in-
trinsic region is composed of a superlattice of quantum wells. By applying
a reverse bias, the resonance wavelength was made tunable over 0.8 nm
(100 GHz) by the application of over 8 V. Direct material property mea-
surements indicate that the effect is electro-optic, and that the quadratic
electro-optic coefficient is at least 2.3×10−15 cm2 /V2 , which is consistent
with values reported in literature. The quadratic electro-optic coefficient
for the waveguide is estimated as 4.9 × 10−15 cm2 /V2 ; the corresponding
value for the wells that comprise the waveguide compares favorably with
previous reports. The tuning range is suitable for wavelength-division
multiplexing applications.

Normalized power

0V 7.5 V

-3 -2 -1 0 1 2 3
Detuning, nm
Fig. 4.16. Simulated change in spectral response with reverse bias for the
layer structure used in this chapter.
4.3 Summary 121



Bandwidth, nm


0.65 ♦


0.55 ♦
♦ ♦
♦ ♦ ♦ ♦ ♦ ♦
0 1 2 3 4 5 6 7 8
Voltage, V
Fig. 4.17. Simulated change in bandwidth with electric field across the
core for the layer structure used in this chapter.

4.3 Summary

In this chapter we presented experimental results of microrings acting

as both passive and active tunable filters. The performance associated
with passive microring resonators is limited by fabrication imperfections.
Actively tunable (trimmable) microrings not only forgive fabrication er-
rors but also allow the possibility of dynamic switching and routing.
5. Nonlinear Optics with Microresonators

The previous chapters were concerned with the fundamental linear

optical properties of microresonators. In this section, we examine mech-
anisms whereby the coherent buildup of intensity and increased inter-
action length can enhance the nonlinear or intensity-dependent optical
properties of microresonators. As we shall see, such enhancements can
lead to all-optical functionalities in an ultra-compact geometry.

5.1 Nonlinear Susceptibility

We proceed to describe the formalism of nonlinear optics through the
perturbation approach. The material polarization P may be expanded in
a Taylor expansion of electric field strength E as follows [178]:

P = χ (0) + χ (1) E + χ (2) E 2 + χ (3) E 3 . . . (5.1)

where χ (N) refers to the N th -order susceptibility. The first-order suscep-

tibility gives rise to the linear refractive index
! "
n =  1 + χ (1) (5.2)

and linear attenuation coefficient,

! "
α =  1 + χ (1) 4π /λ. (5.3)

The second-order polarization term cannot have a frequency component

that is composed of field components at that same frequency. This term
is responsible for describing second-harmonic generation, the more gen-
eralized sum-frequency mixing, degenerate one-half subharmonic gener-
ation, the more general difference frequency mixing, optical rectification,
and the electro-optic (Pockels) effect.
In the case of a centrosymmetric material, the second-order polariza-
tion and all subsequent even orders vanish. The third-order polarization
term describes third harmonic generation, four-wave mixing, intensity-
dependent refractive index, saturable absorption, and two-photon
124 5. Nonlinear Optics with Microresonators

P (3) (ω = ω1 + ω2 + ω3 ) = χ (3) (ω1 , ω2 , ω3 )E(ω1 )E(ω2 )E(ω3 ). (5.4)

The last of these phenomena can occur for all three fields at the same
frequency and has a degeneracy factor of 3

P (3) (ω) = 3χ (3) (ω, −ω, ω)E(ω)E ∗ (−ω)E(ω). (5.5)

This equation gives rise to an intensity-dependent refractive index also

known as the optical Kerr effect,

n(I) = n + n2 I, (5.6)

where the nonlinear coefficient is related to the third-order susceptibility

n2 = 3[χ (3) ]/n2 ε0 c. (5.7)
Two-photon absorption is a nonlinear effect by which two-photons ar-
rive within a coherence time of each other and can be simultaneously
absorbed exciting an electron in a material at twice the photon energy.
This process gives rise to either induced or saturable absorption depend-
ing on the sign of the imaginary part of the third-order susceptibility:

α(I) = α + α2 I, (5.8)

where the two-photon attenuation coefficient is related to the third-order

susceptibility as
α2 = 12π [χ (3) ]/n2 ε0 cλ. (5.9)

5.2 Resonator Enhanced χ (3) Nonlinear Effects

5.2.1 Enhanced Nonlinear Phase Shift

If an all-pass microresonator is constructed with a material that pos-

sesses a third-order nonlinearity manifested as an intensity-dependent
refractive index, then the single-pass phase shift acquires a power-
dependent term, φ = φ0 + 2π Ln2 P2 /λAeff where φ0 , is a linear phase
offset. The derivative of the effective phase shift with respect to input
power gives a measure of the power-dependent accumulated phase. This
derivative can be expressed as
dΦ dΦ dφ dP3 1+r 2
2π Ln2
= →
dP1 dφ dP3 dP1 φ=m2π ,a=1 λAeff1−r
 2  2
2π Ln2 2 π 2
= F = F , (5.10)
λAeff π Pπ π
5.2 Resonator Enhanced χ (3) Nonlinear Effects 125

where Pπ = λAeff /2n2 L is the threshold power required to achieve a non-

linear phase shift of π radians. The effect of the resonator is to introduce
two separate enhancements for which the combined action on resonance
yields an overall nonlinear response enhanced quadratically by the fi-
nesse [151]. The dual effect can be understood intuitively noting that an
increased interaction length develop from the light recirculation, and an
increased field intensity develops from the coherent buildup of the opti-
cal field. The increased interaction length in a microresonator of course
comes with the penalty of a reduction in bandwidth.

5.2.2 Nonlinear Pulsed Excitation

In the previous sections, resonator enhancement of nonlinearity was de-

rived in a steady-state basis. The steady-state analysis presented earlier
breaks down when the bandwidth of the optical field incident on a mi-
croring resonator is of the order of or greater than the cavity bandwidth.
To simulate the time-dependent nature of the resonator response, a
recirculating sum of successively delayed and interfered versions of the
incident pulse must be performed numerically. The circulating field after
M passes is built-up from successive and increasingly delayed coupler

A3,M (t) = it (r a)m eiφm (t) A1 (t − mTR ) , (5.11)

where the phase cannot be treated as a constant due to the time-

dependence of the nonlinear phase shift. The phase shift at each pass
is computed as
φm=0 (t) = 0 (5.12)

φm (t) − φm−1 (t)

 2π R
= φ0 + γ dz A3,M (t) e−αz

a2 − 1  2
= φ0 + 2
γ2π R A3,M (t) , (5.13)
ln a
where γ is the self-phase modulation coefficient. Finally, the transmitted
intensity is computed as
A2,M (t) = r A1 (t) − t 2 a (r a)m−1 eiφm (t) A1 (t − mTR ) . (5.14)

To preserve pulse fidelity and achieve finesse-squared enhancement, it

is necessary to operate the device with pulses of widths that are greater
than or equal to the cavity lifetime.
126 5. Nonlinear Optics with Microresonators

We next examine the trade-off between bandwidth and nonlinear re-

sponse for enhanced phase accumulation in a single microresonator. The
bandwidth of a resonator is primarily governed by its radius and finesse,

∆ν = c/ (2π RnF ) . (5.15)

For a single add–drop resonator, this bandwidth corresponds to the width

of the narrow-band add or drop transmission windows. For a single all-
pass resonator, this bandwidth corresponds to the frequency interval
over which the phase varies sensitively and in a nearly linear manner over
π radians. Outside this interval, the sensitivity falls and the phase signif-
icantly departs from linear behavior such that a pulse with a larger band-
width can become severely distorted by higher order dispersive terms.
There is an exact trade-off for linear properties. A resonator’s lifetime,
group delay, and interaction length may be increased at the expense of
bandwidth in direct proportion to the finesse. Such trade-offs can be cir-
cumvented fortuitously for certain nonlinear properties. The strength of
the enhanced self-phase modulation may be characterized by how much
power is required to achieve a nonlinear phase shift of π radians in a
structure composed of a few resonators. To good approximation, the
threshold power required to achieve a π nonlinear phase shift in just
a single all-pass resonator is given by
Pπ ≈ . (5.16)
4F 2 n2 R
The ratio of the reduced threshold power to the reduced bandwidth
for a resonator of a given finesse is a form of figure of merit and is re-
lated to the threshold pulse energy. The minimum pulse energy required
to achieve the π nonlinear phase shift is obtained when the pulse width
is of the order of the inverse of the resonator bandwidth. This process
is easily understood because a longer pulse width with the same peak
power will carry more energy but not be any more effective at accumulat-
ing nonlinear phase. A shorter pulse width will not allow the resonator
sufficient time to buildup in intensity and thus will experience a weak-
ened nonlinear response in addition to being severely distorted. For a
high contrast dielectric waveguide, the effective area in which the power
is confined may be as small as λ2 /8n2 , where n is the refractive index of
the guiding layer. The threshold energy required to achieve a π nonlinear
phase shift is accordingly reduced in linear proportion to finesse:

λ3 π ln(2)
Eπ = . (5.17)
16F nn2 c
To reduce the parameter space, we make some practical choices (n = 3,
n2 = 1.5 × 10−17 m2 /W) corresponding to AlGaAs [187, 188] or chalco-
genide [189, 190] glass waveguides operating near 1.55 µm. Figure 5.1
5.2 Resonator Enhanced χ (3) Nonlinear Effects 127

Fig. 5.1. The inherent trade-off between bandwidth and energy required
to achieve a π nonlinear phase shift in a single add–drop microresonator.
The diagonal lines correspond to constant resonator diameter for AlGaAs
or chalcogenide-based systems near 1.55µm. Increasing finesse is directly
proportional to decreasing energy.

displays the trade-off between the energy requirement for a π radian

nonlinear phase shift per resonator and the bandwidth for resonators of
varying diameter. It is of technological interest to note that a π nonlinear
phase shift is obtainable with a 1ps, 1pJ pulse by use of a single, ultra-
compact microresonator of moderate finesse. Being able to accomplish
useful tasks with low loaded finesse relaxes tolerances on the design and
fabrication of microresonators that require λ/nF precision. Fortunately,
although low-finesse devices do not make good high resolution add–drop
filters or sensors, they still can possess strong nonlinear effects due to
these favorable scaling laws.

5.2.3 Kerr Effect in Solid State Materials below Mid-Gap

The Kramers–Kronig relation connects the refractive and absorptive spec-

tra of a linear optical material with a causal response

c ∞ α (ω )
n (ω) − 1 = dω 2 . (5.18)
π 0 ω − ω2
Changes in each of these quantities due to an external perturbation are
causal and similarly related. However, in a strict mathematical sense, the
Kramers–Kronig relations are invalid for degenerate third-order processes
such as self-induced changes in refractive index because of the existence
of a pole in the upper half complex frequency plane [178]. Nevertheless,
it has been demonstrated that application of the Kramers–Kronig rela-
tion to the two-photon absorption spectrum of a material correctly pre-
dicts the magnitude and dispersion of the Kerr effect in solids [191,192].
128 5. Nonlinear Optics with Microresonators

The spectra of nonlinear refraction and two-photon absorption are thus

related as 
c ∞ α2 (ω )
n2 (ω) = dω 2 . (5.19)
π 0 ω − ω2
A two-band model is generally sufficient for accurately predicting the
third-order, nonlinear absorptive and refractive properties of semicon-
ductors. Under this model, the two-photon absorption coefficient is given
2|pvc | ω
2 πe 4
2 Eg − 1
α2 (ω) = √
. (5.20)
5 m0 c 2 n20 Eg3 ω 5

For most direct gap semiconductors, many parameters are constant such
that to good approximation, the two-photon absorption spectrum can be
simplified to be only dependent on the bandgap energy and the linear
refractive index as


1.42 × 10−7 2 Eg − 1
α2 (ω) ≈
5 m/W, (5.21)
n20 Eg3 2 ω

where Eg is assumed to be given in eV. Equations 5.19 and 5.21 can then
be used to calculate the nonlinear refractive index spectrum. This pro-
cedure results in a bandgap scaling law of Eg−4 for the magnitude of n2 .
This scaling explains why chalcogenide and AlGaAs materials possessing
bandgaps much smaller than silica possess 100–1000 times higher Kerr
nonlinearities. Moreover, operation just below the half-gap ensures that
one- and two-photon absorption processes are negligible; yet, a reason-
ably high and ultra-fast nonlinear Kerr coefficient is retained. Figure 5.2
displays the predicted two-photon absorption and nonlinear refractive
index for a two-band model of AlGaAs. Although ignoring Urbach tail
absorption and competing nonlinearities, this model provides excellent
intuition for identifying the ideal spectral regimes in which to operate.

5.2.4 Experimental Enhancement of the Kerr Effect

Heebner et al. constructed and characterized a 10-µm-diameter micro-

ring resonator for the purpose of verifying the enhancement of the Kerr
nonlinearity in AlGaAs [193]. Many proposed applications of the Kerr
effect require nonlinear phase shifts near π radians to become practical,
and such was the goal of this experimental work.
The microring resonator was constructed in AlGaAs and probed at a
photon energy (hc/λ = 0.8 eV) below the half-gap (Eg /2 = 0.97 eV) of the
material. The motivation for this choice was to maximize the ultra-fast
5.2 Resonator Enhanced χ (3) Nonlinear Effects 129

Fig. 5.2. Plots of the two-photon absorption coefficient and associated

Kerr nonlinear refractive index of SiO2 , GaAs, and Al0.36 Ga0.64 As versus
wavelength. The plot for two-photon absorption is derived from a two–
band model [192]. The Kramers–Kronig relation applied to the third-order
susceptibility is subsequently used to generate the plot for the nonlinear
refractive index in each case. The dip on the log plot of n2 corresponds to
a reversal of sign for the nonlinear refractive index, and gray line-strokes
correspond to negative nonlinear refractive indices.

bound (Kerr) nonlinearities resulting from virtual transitions while mini-

mizing the two-photon contribution to carrier generation [194]. Although
this effect is a relatively weak nonlinearity requiring relatively high cir-
culating intensities, the deposition of heat resulting from two-photon
absorption can be substantially minimized. In addition, the lower limit
of the response time is not dictated by the carrier recombination lifetime
but only by the cavity lifetime and hence the structural design.
Due to an increased phase sensitivity (or increased effective path
length) near resonance, the resonator need only be detuned by 2π /F
radians to achieve a net effective phase shift of π . Furthermore, the
power threshold requirement to achieve this detuning via an intensity-
dependent refractive index change is reduced due to a circulating inten-
sity buildup of B = 2F /π so that
π λAeff
Pthreshold ≈ , (5.22)
2n2 LF 2
where Aeff is the effective mode area and L is the circumference. For
a typical AlGaAs channel waveguide with nonlinear coefficient of n2 =
130 5. Nonlinear Optics with Microresonators

10−17 m2 /W, an effective mode area of 0.5-µm2 ring diameter of 10µm

and finesse of 10, the threshold peak power is about 40 W at 1.55 micr on.
Note that in comparison with standard single-mode silica fiber with a
threshold power-length product of 500 Wm, this situation corresponds
to 1200 Wm. This nearly six-order-of-magnitude reduction is attributed
to a stronger nonlinear coefficient, a tighter mode area, and resonant en-
To demonstrate this enhancement, a device was fabricated in a man-
ner that would display the nonlinear phase shift interferometrically by
coupling an all-pass microring resonator to one arm of a Mach–Zehnder
interferometer [195]. First, the vertical structure was grown via molec-
ular beam epitaxy on a GaAs substrate providing a 1-µm-thick guiding
layer. The arrangement of the layers and their composition are shown
in Fig. 5.3. The narrow gap region was extended along the propagation
direction by 4 µm into a racetrack geometry to increase the guide-to-ring
coupling. Due to the 80-nm-gap feature, the device was patterned using
e-beam lithography. An intermediate oxide-chromium mask was used be-
cause the resist itself was not sufficiently robust to serve as a high-quality
mask for deep anisotropic etching. Lift-off of the bilayer PMMA resist
yielded a 40 nm chromium mask for reactive ion etching of the underly-
ing 800 nm oxide layer. Finally, a highly anisotropic chlorine-based ICP
etch transferred the pattern directly into the AlGaAs with highly verti-
cal sidewalls. Figure 5.4 displays a scanning electron microscrope (SEM)
image of a resulting device.
Figure 5.5 displays the spectral transmission data for an output
port of a nearly balanced Mach–Zehnder interferometer containing an

Fig. 5.3. Design of a resonator structure fabricated to demonstrate the en-

hancement of the Kerr effect in AlGaAs. The vertical structure is formed
by molecular beam epitaxy and the horizontal structure by nanolithogra-
phy. All dimensions are in µm. (After Ref. [193], ©2002, Optical Society
of America.)
5.2 Resonator Enhanced χ (3) Nonlinear Effects 131

Fig. 5.4. SEM image of an all-pass microring resonator coupled to one

arm of a Mach–Zehnder interferometer. (After Ref. [193], ©2002, Optical
Society of America.)

Fig. 5.5. Measured transmission spectrum (left axis) of the device shown
in Fig. 5.4 with a theoretical fit. The resonator-induced phase shift as
inferred from the interferogram is also shown in the plot (right axis).
(After Ref. [193], ©2002, Optical Society of America.)

all-pass resonator in one arm as shown in Fig. 5.4. In this configuration,

the overcoupled phase response of the 5-µm-radius racetrack resonator
could be inferred from the amplitude response of the composite res-
onator enhanced Mach–Zehnder (REMZ). A measured bandwidth of 240
GHz and free spectral range of 2.3 THz results in a finesse of about 10
(Q = 810), dictated primarily by the coupling. Scattering-limited losses
are estimated at 11% per round-trip. Also shown in the figure is the phase
response obtained from a fit to the data displaying increased sensitivity
near each of two resonances. Next, the intensity-dependent transmission
of the REMZ was measured using a 10 Hz, 30 ps Nd:YAG pumped opti-
cal parametric generator (Ekspla) source at 1.545 µm. The low average
power of the source ensured that thermal effects could be ignored. Opti-
cal self-switching was clearly observed in some samples. Figure 5.6 dis-
plays transverse slices of the imaged outputs of the REMZ in Fig. 5.4.
Three traces are shown each with increasing pulse energies. Incomplete
extinction resulted from higher losses in the arm containing the ring.
132 5. Nonlinear Optics with Microresonators

Fig. 5.6. Demonstration of all-optical switching between ports of a Mach–

Zehnder interferometer resulting from the accumulation of a π -radian
phase shift in an AlGaAs microring resonator (device shown in Fig. 5.4).
(After Ref. [193], ©2002, Optical Society of America.)

The actual pulse energies injected into the resonators were of the order
of 1 nanojoule. The peak power associated with the pulse is thus of the
order of the 40 W threshold requirement as predicted for a finesse of 10.
Because the relative distribution of output power is clearly seen to shift
from one output guide to the other as the pulse energy is increased a
phase shift advance of approximately π radians is inferred.
The experiment shows that previous demonstrations of large Kerr
nonlinear phase shifts in 5 mm or longer waveguide lengths [187,196] can
be compressed into devices 100 or more times shorter through the use of
microring resonators. At photon energies below the half-gap, the bound-
electron Kerr nonlinearity is essentially instantaneous whereas carrier
generation due to one- and two-photon absorption is negligible. Ultra-
compact devices constructed from these building-blocks thus have the
potential to be engineered into ultra-fast nonlinear photonic devices that
generate negligible heat. For example, a device similar to the one pre-
sented but with a finesse of 100 could still support 12.5-ps pulses and
switch with energies as low as 4 pJ. Additional optimization of the guid-
ing confinement area and nonlinear response can produce devices with
a 1-THz bandwidth and 1-pJ energy threshold.

5.2.5 Nonlinear Saturation

An implicit assumption in the derivations thus far is the fact that the
resonator does not get power-detuned or pulled away from its initial
detuning. One might call this the non-pulling pump approximation (NPPA).
The power-dependent pulling away from resonance decreases the nonlin-
ear enhancement such that the process may be described mathematically
as a saturation of the effective nonlinearity. Figure 5.7 displays the re-
sults of a simulation involving the interaction of a resonant pulse with a
5.2 Resonator Enhanced χ (3) Nonlinear Effects 133

a) b)
18 r

Circulating power, Watts

Incident & transmitted

16 po
14 300 i ng
power, Watts

12 250 cul
10 200 nt
8 o na
150 R es
4 100
2 50
0 0
0 100 200 0 5 10 15 20
Time, ps Incident power, Watts
c) d) π

Effective phase shift, rad

Effective phase shift, rad

d pre
π π

2 2


0 0
0 100 200 0 5 10 15 20
Time, ps Incident power, Watts

Fig. 5.7. Simulations demonstrating the dynamic accumulation of nonlin-

ear phase shift across a resonant pulse interacting with a nonlinear res-
onator. A 50-ps (10TC ), 16-W peak power pulse (corresponding to twice
the reduced threshold power) interacts with a 5-µm diameter nonlinear
resonator with r = 0.905 (B = 20). (a) incident and transmitted pulses,
(b) circulating power versus incident power in steady state showing the
nonlinear pulling (dashed line corresponds to resonant slope), (c) accu-
mulated effective phase shift, and (d) effective phase shift versus incident
power (the dashed line corresponds to resonant slope). Notice that the
effective phase shift accumulation at the center of the pulse falls short
of reaching π radians.

nonlinear resonator. The phase accumulation (c) across the pulse tracks
the pulse intensity but falls short of reaching π radians even though
its peak power corresponds to twice the value predicted by the NPPA.
Figures 5.7(b) and (d) illustrate why this is the case. As the incident inten-
sity rises, the resonator is indeed pulled off resonance and the circulating
intensity and effective phase shift are pulled away from their respective
NPPA predicted slopes (dashed lines).
A greater nonlinear phase shift may be extracted from the resonator
by employing a small amount of initial detuning [197]. This can be pre-
dicted first by examining the phase transfer function vs. detuning in the
linear regime as in Fig. 5.8. This is faithfully represented in the nonlinear
regime [Fig. 5.9], where an initially detuned pulse pulls itself through res-
onance and in the process acquires a nonlinear phase shift of π radians.
134 5. Nonlinear Optics with Microresonators

Effective phase shift, Φ

π π

-4π -2π 0 2π 4π
Normalized detuning, φ

Fig. 5.8. Linear effective phase shift showing the optimum amount of
detuning such that the dynamic range of the effective phase shift is effi-
ciently implemented near resonance.

a) b)
18 r
Circulating power, Watts

Incident & transmitted

16 po
300 g
14 tin
power, Watts

12 250
10 200 ant
8 son
150 Re
4 100
2 50
0 0
0 100 200 0 5 10 15 20
Time, ps Incident power, Watts
c) d) π
Effective phase shift, rad

Effective phase shift, rad


π π

2 2

0 0
0 100 200 0 5 10 15 20
Time, ps Incident power, Watts

Fig. 5.9. Simulations demonstrating the dynamic accumulation of non-

linear phase shift across an optimally detuned pulse interacting with a
nonlinear resonator. All parameters with the exception of the detuning
(φ0 = −π /F ) are the same as in Fig. 5.7. Notice that the effective phase
shift accumulation at the center of the pulse reaches π radians.
5.2 Resonator Enhanced χ (3) Nonlinear Effects 135

The value of the detuning chosen (φ0 = −π /F ) corresponds to the point

at which a symmetric pull-through resonance results in a π radian phase
shift. If the detuning is increased some more, the circulating versus inci-
dent intensity relation exhibits a very sharp upward sloping curve. The
curve in fact can become √ steeper than that predicted by the NPPA. At
a detuning of (φ0 = − 3π /F ), the curve becomes infinitely steep over
a narrow range, which corresponds to an operating point just below the
threshold of optical bistability. Operation in this regime will be examined
more closely in the next section. For certain applications, this might be
more or less useful than the previous case. Figure 5.10(c) demonstrates
the phase accumulation across such a pulse. The phase shift actually can
exceed π radians but is confined to a narrower range near the peak of the
pulse. Devices operating in this regime of high differential phase gain are
referred to as transphasors in analogy to the very high differential gains
exhibited by transistors in electronics [82].

a) b)
18 r
Circulating power, Watts

Incident & transmitted

16 po
300 g
14 in
power, Watts

12 250
10 200 ntc
8 so
150 Re
4 100
2 50
0 0
0 100 200 0 5 10 15 20
Time, ps Incident power, Watts
c) d) π
Effective phase shift, rad

Effective phase shift, rad


π π

2 2

0 0
0 100 200 0 5 10 15 20
Time, ps Incident power, Watts

Fig. 5.10. Simulations demonstrating the dynamic accumulation of non-

linear phase shift across a pulse detuned just below the limit of bistability
interacting with a nonlinear
√ resonator. All parameters with the exception
of the detuning (φ0 = − 3π /F ) are the same as in Fig. 5.7. Notice the
very sharp jump in phase.
136 5. Nonlinear Optics with Microresonators

5.2.6 Multistability

The inclusion of nonlinearity changes the input–output relations in a

qualitatively different manner. As shown in Fig. 5.11, multistable branches
in the input–output relationships of a resonator become possible [80,
198]. The circulating intensity is a function of detuning, which in turn is
a function of the circulating intensity and so on ad infinitum. Mathemat-
ically, the circulating intensity can be expressed either as an infinitely
nested expression or more compactly as an implicit relation between the
incident and the circulating intensities.

1 − r2
I3 = I1
. (5.23)
2π (2π R)
1 − 2r cos φ0 + λ n2 I3 + r 2

100 a) Input/output pulse 500 b) Circulating pulse

Circulating power, Watts
Output power, Watts

80 400

60 300

40 200

20 100

0 0
0 50 100 150 0 50 100 150
Time, ps Time, ps

6 c) Accumulated phase shift d) Circulating vs incident power

Circulating power, Watts
Accumulated phase, rad

4 400


2 200


0 0
0 50 100 150 0 20 40 60
Time, ps Incident power, Watts

Fig. 5.11. Simulation demonstrating bistability in a 5-µm-radius nonlin-

ear resonator with r = 0.818, φ0 = −3π /F , Pπ = 1644.6 W, and a
50-ps, and 66-W peak power pulse. (a) input/output pulse, (b) circulat-
ing pulse, (c) accumulated phase, (d) incident versus circulating bistable
power relation. Notice the fast switch between stable states, when the
input pulse power reaches the hysteresis “jump up” and “jump down”
points at 53 and 30 W.
5.2 Resonator Enhanced χ (3) Nonlinear Effects 137

This implicit relation is easily converted into a single-valued function if

the circulating intensity is considered the independent variable and the
incident intensity the dependent variable. Multistability is always present
in such a lossless nonlinear resonator above some threshold intensity.
One way to reduce this threshold intensity is to red-detune the resonator
slightly so that the resonator is initially off resonance but forms two
stable circulating intensity levels
√ for a low intensity. The onset of multi-
stability takes place at φ0 = − 3π /F . Optical bistability in a resonator
allows for the construction of all-optical flip-flops and other devices ex-
hibiting dynamic optical memory.

5.2.7 Fabry–Perot, Add–Drop, and REMZ Switching

The placement of a nonlinear all-pass resonator within one arm of an

interferometer, for example, a Mach–Zehnder as in Fig. 5.12, allows for
the conversion of phase modulation into amplitude modulation serving
as a basis for a nonlinear all-optical switch [150,151]. One might ask why
one would want to use this indirect method of implementing a microres-
onator as an amplitude switching device when a nonlinear add–drop
resonator accomplishes the same task in a more direct manner. The prop-
erties of an add–drop filter are analogous to that of a traditional Fabry–
Perot interferometer which offers light the choice of two output ports
and can display nonlinear amplitude switching between them. A careful
examination of a nonlinear add–drop resonator as a switching device,
however, reveals a fundamental limitation. The switching curve cannot

a) c)
E3 E4 EB2
E1 E2 EA1 EB1

b) d)

E3 E4 EB2
E1 E2
Fig. 5.12. (a) An all-pass microresonator, (b) an add–drop microresonator
(or Fabry–Perot), (c) a Mach–Zehnder interferometer (MZI), and (d) an
REMZ interferometer.
138 5. Nonlinear Optics with Microresonators

be made to perform a complete switch within the phase-sensitive region

near resonance because the center of the phase-sensitive region (at a res-
onance frequency) directly coincides with a minimum or maximum of
the transmission spectrum. Only half of the sensitive region is usable
as a switch, the other half being wasted. It would be advantageous to
shift the peak in phase sensitivity away from the transmission minimum
such that it coincides with the linear portion of the transmission curve.
However, there is simply no way to accomplish this shift in an add–drop
resonator or Fabry–Perot. The missing degree of freedom that can pro-
vide this capability can be found in a device formed from a nonlinear
all-pass resonator coupled to one arm of an interferometer. An REMZ, for
example, can be used to set independently the peaks of nonlinear phase
sensitivity and transmission. Figure 5.13 displays the linear transmis-
sion characteristics against detuning for a REMZ with offset bias phase
between arms of φB = 0, π /2, π , and 3π /2. Optimized switching char-
acteristics are obtained when the peak of the phase sensitivity coincides
with the linear portion of the switching curve φB = π /2,3π /2. The char-
acteristic switching curve of a Mach–Zehnder is cosine-squared. To build
a effective switch it is important to switch in a region of this curve where
the maximum change in transmission is brought about by the minimum
change in accumulated phase. This is at the point of maximum switching
sensitivity, related to the phase sensitivity as
dT dΦ
= − 12 sin(Φ − φB ) . (5.24)
dφ dφ

Figure 5.14 compares the switching characteristics for an add–drop

resonator, an unbiased REMZ, and a properly biased REMZ. For the add–
drop resonator, the buildup and finesse are each lower by a factor of
four in comparison with an all-pass resonator with the same coupling
strength. In a REMZ, however, only half of the power enters the arm con-
taining the resonator and its phase contributes in a manner that another
factor of two is lost. Still, it results in a net improvement in switching
threshold by a factor of four that can be observed by comparing the
curves in the figure. It might be argued that this analysis is not valid
since the finesse is not maintained equal in the comparison. When the
finesse is maintained equal, the curves are in fact equivalent. More sig-
nificant, however, is the improvement in the shape of the switching curve
when the REMZ is properly biased. Proper biasing is achieved by tuning
the peak of the nonlinear phase sensitivity to the 50% transmission op-
erating point of the unloaded Mach–Zehnder.

5.2.8 Reduced Nonlinear Enhancement via Attenuation

As shown, the presence of attenuation mechanisms in the resonator lead
to a reduction in peak buildup, a paradoxical increase in phase sensitivity,
5.2 Resonator Enhanced χ (3) Nonlinear Effects 139

Fig. 5.13. A comparison of the linear transmission spectra for an REMZ

interferometer with varying phase shift bias differences between arms.

and a reduction in transmission. The dependence of the nonlinear en-

hancement on the round-trip attenuation is thus non-trivial.
The peak buildup (and phase sensitivity) with no loss is termed B0
for simplicity in the following comparisons all taken on resonance. The
buildup varies with the round-trip amplitude transmission, a as,

1 − r2
B(a) = → 1 B0 , (5.25)
(1 − r a)2 r =a 4
B(α2π R) ≈ B0 − 12 B20 (α2π R) + 3 3
16 B0 (α2π R)
+ O(3), (5.26)
140 5. Nonlinear Optics with Microresonators

Fig. 5.14. A comparison of the nonlinear switching characteristics of an

add–drop resonator (Fabry–Perot), an unbiased, and a properly biased
REMZ interferometer. Here, the coupling parameter is held constant for
all three cases of resonators. If the finesse is held constant, the nonlinear
response of the add–drop resonator matches that of the unbiased REMZ.

where the second expression is a Taylor expansion for loss near zero. The
buildup drops with decreasing a until reaching critical coupling (r = a)
where it drops to 1/4 its lossless value. In contrast, the phase sensitivity
increases from the lossless value without bound at critical coupling,

a(1 − r 2 )
S(a) = → ∞, (5.27)
(a − r )(1 − r a) r =a
1 3 2
S(α2π R) ≈ B0 + 16 B0 (α2π R) + O(3). (5.28)

The nonlinear enhancement is equal to the product of the buildup and

phase sensitivity. Additionally, since the intensity drops continuously
when traversing the resonator, a correction factor is introduced, C(a) =
ln a−2 , such that the effective nonlinear phase shift induced by a much
weaker single-pass nonlinear phase shift is

∆ΦNL (a) = B(a)S(a)C(a)∆φNL . (5.29)

Dividing this quantity by the the shift obtained for the lossless case re-
sults in a normalized nonlinear enhancement
∆ΦNL (a)
N (a) = . (5.30)
∆ΦNL (a = 1)
N (α2π R) ≈ 1 − 12 B0 (α2π R) + 14 B20 (α2π R)2 + O(3). (5.31)

The lowest order variation in phase sensitivity with loss is quadratic.

Thus, for small losses, the nonlinear enhancement is reduced primarily
because of the reduction in peak buildup. Near critical coupling, the phase
sensitivity increase without bound results in a growing enhancement as
5.2 Resonator Enhanced χ (3) Nonlinear Effects 141



Normalized nonlinear
1.0 enhancement

Net transmission

0 0.02 0.04 0.06 0.08 0.10
Single-pass loss (1-a2)
Fig. 5.15. Variation in net transmission and normalized enhancement
with loss for a resonator with r 2 = 0.9. For zero loss, both are equal to
unity. The transmission steadily drops to zero at critical coupling (r = a),
here at 10% loss. By contrast, the normalized enhancement is somewhat
impervious to loss and never dips below a certain value due to an in-
creasing phase sensitivity. Although the enhancement diverges at critical
coupling, it is of little use to operate there since the transmission drops
to zero.

well but at the expense of rapidly decreasing attenuation to zero at critical

coupling. For comparison, the transmission varies as

(a − r )2
T (a) = → 0, (5.32)
(1 − r a)2 r =a
T (α2π R) ≈ 1 − B0 (α2π R) + 2 B20 (α2π R)2 + O(3). (5.33)

Figure 5.15 shows the variation of normalized nonlinear enhancement

and transmission with respect to loss. Notice that although the transmis-
sion steadily drops with increasing attenuation, the normalized enhance-
ment is somewhat more stable and never dips below a certain value.1 For
a nonlinear device, both enhancement and transmission are generally im-
portant and how their importance is weighted will affect how much loss
can be tolerated.

0.592 in the high-finesse limit.
142 5. Nonlinear Optics with Microresonators

5.2.9 Nonlinear Figures of Merit (FOMs)

As will be examined in the next chapter, nonlinear photonic devices hold
great promise for the implementation of densely integrated all-optical
signal processing. Fundamental restrictions imposed by material proper-
ties have not allowed the promise to be fulfilled. Most refractive optical
processing devices require a π radian nonlinear phase shift for success-
ful operation that may in turn demand impractically high optical powers.
In many material systems, strong nonlinear refractive indices are often
accompanied by strong linear and nonlinear absorption that limit the
achievable phase shift in many cases far below the target of π radians.
The limitations to the achievable nonlinear phase shift imposed by linear
and nonlinear absorption thus merit examination.
Linear absorption limits the interaction length to an effective length
given by 
1 − e−αL
Lα = . (5.34)
Because the effective length is independent of intensity, in theory, any de-
sired nonlinear phase shift may still be obtained by making the intensity
high enough. In practice of course, this may not be practical because all
materials possess some threshold for optical damage [178]. Several non-
linear figures of merit are useful for comparing the relative strength of
a nonlinear coefficient to absorption. A common definition is the nonlin-
ear coefficient divided by the linear attenuation (M1 = 4π n2 /λα). In gen-
eral, this parameter is material property with some fixed value at a given
wavelength. Most highly nonlinear materials are inherently absorptive
because they operate via enhancement near some atomic or molecular
resonance. For this reason, M1 is useful for comparing the nonlinearities
of different material systems provided that interaction length is not a
limitation. Silica single-mode fiber, for example, has a very high M1 de-
spite a low intrinsic nonlinearity because its attenuation is extremely low
(5.3 × 10−8 m2 /W). Clever arrangements exist whereby M1 can be modi-
fied from its traditionally fixed value granted by nature. The techniques
of electromagnetically induced transparency (EIT), for instance, hold the
promise of maintaining a strong nonlinearity and canceling the linear ab-
sorption via quantum interference. In the case of a nonlinear resonator,
although the attenuation is increased in proportion to the effective num-
ber of round-trips or finesse, the third order nonlinearity is increased in
quadratic proportion.
Two-photon absorption imposes a stricter limitation on the achiev-
able nonlinear phase shift. In the presence of linear and nonlinear absorp-
tion, the reduced wave equation for nonlinear phase evolution (self-phase
modulation) takes the form:

A = iγ 1 + iM−1 2 |A|2 A − αA. (5.35)
∂z 2
5.2 Resonator Enhanced χ (3) Nonlinear Effects 143

Where a second FOM is introduced proportional to the ratio of the

nonlinear refractive index to the two-photon absorption coefficient, M2 =
4π n2 /λα2 . Equation 5.35 has exact solutions for the transmitted inten-
sity and nonlinear phase shift after propagating for a distance L. The
fractional intensity remaining after propagating is

T = . (5.36)
1 + α2 ILα
Because two-photon absorption increases in proportion to the nonlinear
refractive index, the achievable nonlinear phase shift saturates logarith-
mically with increasing intensity:
M2 M2 M1

φNL = ln (1 + 2γILα /M2 ) = ln 1 + 1 − e−2γL/M1 I , (5.37)

2 2 M2

→ γILα = 1 − e−2γL/M1 I, (5.38)
M2 →∞ 2
→ γIL. (5.39)
M1 →∞

Within a resonator, two-photon absorption2 is likewise enhanced in pro-

portion to the the nonlinear refractive index,3 i.e., quadratically with fi-
nesse. As a result, M2 would not be modified by the use of a resonator.
In general, however, both linear and nonlinear absorptive processes
are detrimental to switching devices based on the intensity-dependent
refractive index. When considering both linear and nonlinear absorption,
constructing a device that delivers a π radian nonlinear phase shift using
a resonator can exceed in performance; an equivalent device is formed
from a simple waveguide. A third FOM, proportional to the nonlinear

∆φNL |E2 |2
phase shift multiplied by the transmission [199], M3 = π e |E |2
is deemed more useful in such a comparison. This FOM is defined in
such a way that a π phase change over a 1/e intensity falloff results
in M3 = 1. This FOM can be favorably modified by use of a resonator.
Figure 5.16 compares M3 for a waveguide of length L, 10L, F L, with that
of a resonator. There are of course many other ways of characterizing
the trade-offs between attenuation and nonlinearity. Perhaps the most
useful definition is the interferometric contrast between a pulse that has
acquired a given phase with some some unavoidable attenuation and a
reference copy of the same pulse.
It is expected that higher photon number absorptive processes of order N
would be enhanced by a value of F N .
Two-photon absorption is not always deleterious. A nonlinear refractive re-
sponse in a semiconducting material can be obtained through the generation
of carriers usually leading to a stronger, albeit slower, response limited by
the carrier recombination time. We shall investigate this phenomena further
in the next chapter.
144 5. Nonlinear Optics with Microresonators

Nolinear device figure of merit, M3

10 L
10-1 FL



10-9 10-8 10-7 10-6 10-5 10-4 10-3 10-2

Nonlinear refractive index, n2I

Fig. 5.16. Comparison of the nonlinear device figure of merit (M3 ) for a
waveguide of length L, 10L, F L, with that of a resonator. (After Ref. [199],
©2002, Optical Society of America.)

5.2.10 Inverted Effective Nonlinearity

Most refractive nonlinear materials behave in such a way that increas-

ing intensity brings about a positive change in index. Negative nonlin-
ear refractive indices are rare but nevertheless can be found in thermal
(effectively) nonlinear phenomena and in semiconductors well above the
half-gap. Neither case is useful for high-speed all-optical switching. Many
proposed applications, such as nonlinear management, would greatly
benefit from a negative nonlinear material. A negative effective nonlin-
earity may be achieved by use of a nonlinear microring resonator. In the
undercoupled regime, the sign of the input–output phase relationship is
inverted and thus allows for the possibility of inverting the sign of the in-
trinsic nonlinearity of the resonator medium. Figure 5.17 demonstrates a
pulse propagating through a single resonator and acquiring a phase shift
of negative π /2 radians.

5.3 Resonator-Enhanced Free Carrier Refraction

The increased circulating intensity in a semiconductor microresonator

can also be used to enhance charge carrier-based nonlinear optical
phenomena. Carriers may be excited through a variety of processes,
5.3 Resonator-Enhanced Free Carrier Refraction 145

a) 0.10 b) 30


Power, Watts
20 pulse
0.02 5 pulse

0.00 0
0 5 10 15 20 25 0 20 40 60 80 100 120 140
Incident power, Watts Time, ps
c) d)
Effective phase shift, rad

Effective phase shift, rad



π π
20 5 10 15 20 25 20 20 40 60 80 100 120 140
Incident power, Watts Time, ps

Fig. 5.17. A resonator can display negative effective nonlinearity even if

the intrinsic material nonlinearity is positive, which is possible only in the
undercoupled regime and thus requires a lossy resonator or imbalanced
add–drop resonator. (a) Transmission versus incident power. (b) Power
versus time. (c) Effective phase shift versus incident power. (d) Effective
phase shift versus time. The parameters used in these simulations are
r = 0.9, a = 0.875, φ0 = −0.0115π , P = 20W, Pπ = 3183 W, and TP =
20TC = 35 ps.

including two-photon absorption. The presence of free carriers modifies

the refractive index of a probe signal via the plasma effect.4
The recovery time associated with refractive index changes brought
about by carrier generation depends on the lifetime of carriers in the
device. Since the width of microring waveguides is usually <1 µm, diffus-
ing carriers may quickly find a boundary where surface recombination
becomes an important consideration. The carrier lifetime in this regime
can be approximated by the time taken to reach the surface. Holes have
much lower mobility than electrons, and in intrinsic materials, the carrier
lifetime is dominated by the hole lifetime. This phenomenon is known as
“ambipolar diffusion” [201,202]. Several methods can be used to shorten

This effect has been implemented in silicon microrings through linear absorp-
tive effects as well [200].
146 5. Nonlinear Optics with Microresonators

carrier lifetime: a bias to sweep the carriers out of the waveguide core
and a p-doped core, among others [203].
If the pump pulse is longer than the cavity lifetime, a quasi-steady-
state analysis can be used to solve for the interaction between the pump
and the probe pulses. To find the change in refractive index due to two-
photon absorption, we must first find the change in index from carriers,
i.e., the contribution of free carriers to refractive index change.
The refractive index n is given as
n2 (ω) = (ω) = b + 4π χf , (5.40)
where  is the relative permittivity of the medium, and b = 1 + 4π χb ;
χb is the contribution of bound charges to the susceptibility, and χf is
the contribution of free charges. The equation of motion for carriers in a
harmonic electric field, E exp(iωt), is [181]
m∗ ẍ = −eE exp(iωt), (5.41)
where m∗ is the effective mass of the carrier and e is the electron charge.
We assume that the damping term is negligible. The solution to this equa-
tion is
x= exp(iωt). (5.42)
m∗ ω2
The corresponding contribution to polarization may be written as
4π Nfree e2 E
P = − (5.43)
m∗ ω2
= 4π χf E, (5.44)
where Nfree is the free carrier density. We define the plasma frequency
ωp as
4π Nfree e2
ω2p = , (5.45)
which gives
4π χf = − 2 , (5.46)
n2 = (ω) = b − 2 . (5.47)
Since there is no imaginary term above, the refractive index is real. 5 Then,
n2 = b − (5.48)
dn ωp
⇒ 2n = −2 2 . (5.49)
dωp ω
Realistic, noninfinite carrier mobilities lead to free-carrier-induced changes in
the absorption as well.
5.4 Enhanced Four-Wave Mixing Efficiency in Microring Resonators 147

Since ωp  ω for optical frequencies,

dn 1 2ωp
= − (5.50)
dωp b − ω2p /ω2 ω2

1 2ωp ω3p
= −√ + . (5.51)
b ω2 b ω 4
Now that we know the dependence of refractive index on carriers (via
the plasma frequency), we have to consider the dependence of carrier
concentration on intensity for two-photon absorption — the effect of in-
terest. The absorption cross section for two-photon absorption increases
linearly with laser intensity, so the atomic transition rate increases as the
square of intensity [178]. If we define a two-photon absorption coefficient
α2 we can write the rate equation for the free carrier density Nfree as
dNfree (t) α2 2 Nfree (t)
= I (t) − , (5.52)
dt 2ω pump τfree
where ω is the pump beam frequency, Ipump is the pump beam intensity,
and τfree is the free carrier lifetime. From Eqs. 5.45, 5.51, and 5.52,
∝ B3 . (5.53)
Since the change in refractive index scales as B3 , the switching power can
be reduced dramatically for moderate intensity buildup. We will return to
this in experimental demonstrations of optical logic later in the following

5.4 Enhanced Four-Wave Mixing Efficiency in Microring

We next consider the case of four-wave mixing in a microresonator
adapted from Absil et al. [56]. Intuitively, we can expect that the inten-
sity buildup in microrings boosts four-wave mixing as B4 , since all four
waves involved experience an intensity buildup. Here, we show that this
is indeed the case.
Four-wave mixing is due to χ (3) . In the presence of a pump beam
E1 (ω1 ) and a signal E2 (ω2 ), a converted signal E3 (ω3 = 2ω1 − ω2 ) is
∂E1 α1
=− E1 (5.54)
∂z 2
∂E2 α2
=− E2 (5.55)
∂z 2
∂E3 α3
=− E3 + γE12 E2∗ exp(i∆βz), (5.56)
∂z 2
148 5. Nonlinear Optics with Microresonators

where ∆β = β3 − (2β1 − β2 ), ∆β = 0 due to dispersion in the material.

The equation in E3 is a nonhomogeneous differential equation, and has
the solution [204]
α3 α3
E3 = E3 (t = 0) exp − z dz exp z γE12 E2∗ exp(i∆βz). (5.57)
2 0 2

However, this just represents the response in the absence of feedback

(via resonance). If the effect of dispersion is small enough to ensure that
ω3 is on resonance, then E3 is enhanced by 1/(1 − ar ), and the field
coupled out is
t α3 α3
E3,out = exp − L dz exp z γE12 E2∗ exp(i∆βz). (5.58)
1 − ar 2 0 2

Since we are concerned with the intensity, we consider the squared mag-
nitude of the right-hand-side; comparing with the expressions for add–
drop buildup we see that |t/(1 − ar )|2 ∝ B, noting that the pump and
signal intensities are enhanced as B, we get the result that conversion
efficiency associated with four-wave mixing is enhanced as B4 .

5.5 Summary

In this chapter we introduced the nonlinear optical properties of

microresonators. The finesse-scaling laws associated with resonant en-
hancement of third-order nonlinearities were derived. Experimental
demonstrations of the enhancement of the Kerr effect were presented.
In the following chapter we describe the results of all-optical switching
and logic experiments based on two-photon absorption-induced changes
in the refractive index of microrings.
6. All-Optical Switching and Logic using

In this chapter we present the results of all-optical switching and logic

experiments carried out in III–V microring resonators. All-optical data
processing devices are key components for future integrated photonic
circuits. One category of such devices is switching and logic. Architec-
tures have been proposed based on intensity-dependent all-optical logic
switches and gates constructed from nonlinear planar waveguide de-
vices such as Mach-Zehnder interferometers and Distributed Bragg grat-
ings [205, 206]. Microring resonators can be used to build elementary
switches and logic gates with the added advantages of compact size [207]
and reduced switching threshold [56, 151].
The transmittance of a microring coupled to a single optical bus,
shown in Fig. 6.1, can be altered by changing either the absorption or the
refractive index of the microring, or both. For GaAs/AlGaAs microrings
operated at the 1.55-µm wavelength, the absorption and index changes
can be accomplished via two-photon absorption (TPA) of a high-intensity
pump beam tuned to one of the microring’s resonance modes. The sub-
sequently induced change in the transmittance is then used to switch a
low-intensity probe beam tuned to the vicinity of a different resonance

6.1 All-Optical Switching

Through the implementation of refractive index changes near critical cou-

pling, one can go from low to high transmission if initially on-resonance,
or from high to low transmission if initially off-resonance.

6.1.1 Theory

The interaction between the pump and the probe beams as they circulate
in the microring can be described by a coupled set of nonlinear propaga-
tion equations in the slowly varying envelope approximation. Let Ar (z, t)
and Br (z, t) represent the field envelope in the ring of the pump and the
probe signals, respectively, and designate z as the linear coordinate along
150 6. All-Optical Switching and Logic using Microresonators

Fig. 6.1. Optical micrograph of a vertically coupled GaAs/AlGaAs micror-

ing resonator.

the ring path. Assuming |Br |  |Ar |, the propagation of the pump and
the probe beams in the presence of TPA-induced nonlinearity is given by
dAr n0 dAr
+ = −α0 Ar − (α2 + in2 k0 )Ip Ar
dz c dt
−(∆αfc + i∆nf c k0 )Ar , (6.1)
dBr n0 dBr
+ = −α0 Br − (α2 + in2 k0 )Ip Br
dz c dt
−(∆αfc + i∆nf c k0 )Ar , (6.2)
In the above, Ip ∝ |Ar |2 is the pump beam intensity; n0 is the waveguide
effective index; α0 is the linear absorption coefficient; α2 and n2 are the
nonlinear TPA loss and refraction coefficients, respectively; and ∆αfc and
∆nfc are the changes in the absorption and refractive index induced by
free carriers generated from the TPA process. The contributions ∆αfc and
∆nfc are proportional to the free-carrier density, Nfc , via ∆αfc = σa Nfc
and ∆nfc = σr Nfc , with σa and σr being the absorption cross section
and refraction volume, respectively. The free-carrier density generated
by TPA evolves according to
dNfc α2 2 Nfc
= I − , (6.3)
dt 2ω p τfc
where τfc is the carrier relaxation time. Coupling among the input pump
Ai (t), the output pump Ao (t), and the circulating pump beam Ar (z, t) in
the ring is described by the set of equations,

Ao (t) = r Ai (t) − i 1 − r 2 Ar (t − TR )eiφ , (6.4)

Ar (t) = r Ar (t − TR )e − i 1 − r Ai (t),
2 (6.5)
where r is the lumped self-coupling coefficient, TR is the cavity transit
time, and φ is the round-trip phase. A similar relationship exists for the
probe signals Bi (t), Bo (t), and Br (t). Equations. 6.2, 6.3, and 6.5 are time-
difference equations that are numerically integrated to simulate the pas-
sage of the pump and the probe beams through the microring resonator.
6.1 All-Optical Switching 151

6.1.2 Linear Device Characteristics

The microresonators used in the experiment consist of 10-µm-radius

GaAs/AlGaAs microrings vertically coupled to a straight bus waveguide.
The waveguiding structure has high lateral index contrast (3.37:1.5)
to minimize the bending loss. A high-index, epitaxially-grown AlGaAs
mid-layer allows for efficient coupling between the ring and the bus
waveguides. Figure 6.2 shows the measured spectral response of the mi-
croresonator at the 1562-nm resonance and the theoretical fit. At res-
onance, an extinction notch of 8.5 dB in the signal is evidence of near
critical coupling. For the device shown, assuming negligible propagation
loss in the bus waveguide, a coupling coefficient of 3% and a round-trip
power loss of 6% were obtained for the microring, giving a 3-dB band-
width of 0.16 nm and a quality factor of 9800. The finesse of the device
as determined from the measured 3-dB bandwidth and free spectral range
was 68. The device has a computed field enhancement factor of 3.7 and
a cavity lifetime of 42 ps.




0.2 measaured
model fit

0.4 0.2 0 0.2 0.4
Wavelength detuning, nm
Fig. 6.2. Measured spectral response and its theoretical fit of a 10-µm-
radius GaAs/AlGaAs microring resonator at the 1562-nm resonance. The
discrepancy between measurement and model on both sides of the reso-
nance peak is due to the Fabry–Perot modulations in the bus waveguide.
152 6. All-Optical Switching and Logic using Microresonators

6.1.3 Simulations

A discretized form of Eqs. 6.2, 6.3, and 6.5 enables numerical simula-
tion of the nonlinear propagation of a 300-ps pump pulse through the
microring. The wavelength of the pulse was set at 1561.97 nm, that
is, 0.1-nm blue detuned with respect to the cavity resonance mode at
1562.07 nm. The parameters used in the simulation are α0 = 10.2 cm−1 ,
α2 = 24 cm/GW, n2 = 1.5 × 10−13 cm2 /W, σa = 1.5 × 10−16 cm2 ,
σr = 10−20 cm3 , and τfc = 50 ps. The free carrier lifetime τfc was an ad-
justed parameter in the simulations to fit measured transient responses.
Figure 6.3 shows the evolution of the normalized pump intensity, Ip ,
inside the microring and the output pump intensity. The circulating
pump intensity in the ring is normalized to its peak value and is actually
12 times higher than the peak input intensity, due to coherent buildup

Input pump
Circulating pump
Normalized Intensity


0.4 Output pump


0 200 400 600 800
Time, ps
Fig. 6.3. Simulated nonlinear propagation of a 300-ps pump pulse
through a 10-µm-radius GaAs-AlGaAs microring resonator. The output
intensity is normalized with respect to the peak input intensity. The
pump intensity circulating inside the resonator is normalized with re-
spect to its peak value, that is 12 times higher than the peak intensity of
the input pump. The input pump is a real experimental pulse captured
by the oscilloscope to be used in the simulation.
6.1 All-Optical Switching 153

in the resonator. As can be observed in Fig. 6.3, the circulating pump in-
tensity is lagging the input pump intensity by approximately the cavity
charging time. It was observed in the simulation that as the intensity
in the ring builds up, carriers generated from TPA cause a net decrease in
the refractive index of about 2 × 10−4 , which has the effect of shifting the
resonance mode toward shorter wavelengths by nearly 0.1 nm. Conse-
quently, the output pump is observed to rise initially with the field in
the ring, but then dip to a minimum as the microring resonance is pulled
closer to the pulse wavelength. We note that this dip coincides with the
peak of the pump intensity in the microring. At the falling edge of the
input pulse, the field in the microring begins to discharge, and the output
pump intensity is observed to rise again as the cavity resonance mode re-
turns to its initial position. Next, a low-intensity continuous wave (CW)
probe signal tuned to the resonance mode at 1550.92 nm is launched into
the microring. By tuning the probe beam with respect to the resonance
and looking at the transmitted probe, one can quantify the refractive in-
dex change and its sign. This is demonstrated in the three-dimensional
plot shown in Fig. 6.4. When the probe beam is tuned to 1550.6 nm, far
from resonance, it is completely transmitted to the throughput port or
in the “ON” state. As the probe beam is tuned closer to resonance, the
Normalized Intensity





400 1551.4
Time, ps 200 1551
0 1550.6 Probe wavelength, nm

Fig. 6.4. Three-dimensional plot of the simulation results for the through-
put probe intensity versus wavelength and time. The dc level has been
removed for clarity.
154 6. All-Optical Switching and Logic using Microresonators

probe experiences an increased attenuation as the pump pulse causes

the resonance modes of the microring to shift to shorter wavelengths,
effectively pulling the probe beam into resonance. When the probe wave-
length is tuned to resonance at 1550.92 nm and prior to the arrival of the
pump pulse, the transmitted probe signal is attenuated strongly. When
the pump pulse arrives, it shifts the resonance away from the probe beam,
and as a result, we see a rapid increase in the transmission of the probe
signal as it switches on. In Fig. 6.8, we plot the input pump intensity and
the switched probe for the two cases when the probe beam is initially
blue-detuned off resonance at 1550.75 nm, and when it is initially tuned
to resonance at 1550.92 nm. These represents the ideal scenarios for the
“ON-OFF” and “OFF-ON” switching, respectively.

6.1.4 Nonlinear Device Characteristics

The experimental setup for the pump and the probe experiment is shown
in Fig. 6.6. An externally modulated laser diode was used to produce

Normalized Intensity





0 200 400 600 800
Time, ps
Fig. 6.5. Simulated pump-probe interactions. (a) Probe beam initially off
resonance. (b) Probe beam initially on resonance. The probe beam in-
tensities are normalized with respect to the maximum transmittance at
6.1 All-Optical Switching 155

Fig. 6.6. Experimental setup for the pump-probe technique to demon-

strate all-optical switching using a GaAs-AlGaAs 10-µm-radius microring
resonator. PG: pulse generator. FPC: fiber polarization controller. MZI:
Mach-Zehnder interferometer. EDFA: Erbium doped fiber amplifier. D: de-
tector. BPF: band-pass filter.

a 300-ps pump pulse signal with a 20-MHz repetition rate. The pump
beam was then amplified using an erbium doped fiber amplifier (EDFA) to
compensate for the external modulator and fiber to waveguide coupling
losses. The average input power was 8 mW at the device input waveguide,
giving an estimated peak intensity of 4.8 GW/cm2 in the microring at reso-
nance. The pump wavelength was set at 1562.0 nm, slightly blue-detuned
from the resonance. A 3-mW CW probe signal tuned to the next higher
resonance at 1550.9 nm was coupled to the pump beam via a 50/50 fiber
coupler and fed to the device input using a conically tipped fiber to min-
imize the mode-mismatch between the fiber and the waveguide. At the
output, the pump and the probe signals were collected using another
conically tipped fiber, optically amplified, then separated by a band-pass
filter, and detected using a 40-GHz detector and a 50-Ghz oscilloscope. In
Fig. 6.7, we plot a three-dimensional graph for the measured throughput
probe intensity as a function of time and probe wavelength. It is in very
good agreement with the simulated 3-dimensional graph shown earlier
156 6. All-Optical Switching and Logic using Microresonators

Normalized Intensity




Time, ps 400
200 1551
0 Probe wavelength, nm

Fig. 6.7. Three-dimensional plot of the measured throughput probe in-

tensity versus wavelength and time. The dc level has been removed for

in Fig. 6.4. Fig. 6.8 show the time traces of the input pump, the output
pump, and the transmitted probe power for the switch-on and switch-
off cases. The measured responses of both the pump and the probe sig-
nals are in excellent agreement with the simulated responses shown in
Fig. 6.5. The pump pulse is strongly distorted and absorbed due to crit-
ical coupling and TPA in the microring. The on and off switching times
of the probe beam are slightly less than 100 ps. Also, it is observed that
the probe beam quickly returns to the original state after the pump beam
has passed. It is postulated that this fast recovery time of the order of
50 ps is due to surface-state recombination at the microring sidewalls.

6.2 Thresholding and Pulse Reshaping

The nonlinear transmission transfer function of a notch filter microring
resonator can be used for thresholding. If an input signal is tuned ex-
actly to a critically coupled microring resonance, it is initially absorbed
by the resonator and no output is transmitted. If the signal intensity
is increased, the self-induced phase shift in the resonator will result in
a detuning in the microring resonance wavelength. If such detuning is
6.2 Thresholding and Pulse Reshaping 157

Input pump

Output probe
Normalized Intensity


Output pump


0 200 400 600 800
Input pump

Output probe
Normalized Intensity



Output pump


0 200 400 600 800
Time, ps
Fig. 6.8. Measured time traces of the pump and the probe beam intensities
with (a) probe beam initially off resonance (high transmission) and (b)
probe beam initially on resonance (low transmission).
158 6. All-Optical Switching and Logic using Microresonators

Normalized Intensity




-100 -50 0 50 100
Time, ps
Fig. 6.9. Time traces of the input and output pulses of a GaAs-AlGaAs
microring resonator. Input and output intensities are normalized by their
respective peak values. The delayed output pulse is shifted in time by
25 ps to overlap with the input pulse.

enough, the signal will be out of resonance and no longer absorbed by

the microresonator and totally transmitted to the output port. This effect
can be used for thresholding applications such as digital level restoration.
Figure 6.9 shows the input and output time traces of a GaAs-AlGaAs mi-
croring notch filter excited by a 35-ps laser pulse carrying approximately
50 pJ of energy. The input signal has a rise time of 40 ps, whereas the
rise time of the output pulse was measured to be 20 ps. The output pulse
was delayed with respect to the input pulse by 20 ps corresponding to
the charging time of the resonator. We plot both pulses to show the pulse

6.3 Time-Division Demultiplexing

A routing switch connects the input port to one of several output ports
and physically moves photons from one port to another. Routing can
be based on either the intensity of the input signals or an external
6.3 Time-Division Demultiplexing 159

control beam. Typically, the control beam is in a different physical format

than the data. For example, the pump beam might have a different
wavelength or polarization than the data beam. Nonlinear directional
Couplers [208], distributed-feedback waveguides [209], are examples of
passive all-optical routing and demultiplexing switches.
When a microring resonator is used in the add–drop configuration,
the pump and probe method can be used to spatially route an incoming
signal to the through port or drop port of the device depending on the
presence or absence of the control beam [210]. If the incoming signal
consists of a train of different data pulses or channels multiplexed in
time, the pump pulse can be applied at the appropriate time slots to pick
out an individual data pulse or channel and drop it at the drop port,
while allowing the rest of the pulses to pass on via the through port. The
device in this case functions as an optical time-division demultiplexer. In
a similar manner, data pulses can also be multiplexed onto the incoming
signal through the add port of the device.

6.3.1 Linear Device Characteristics

The microresonator used in the experiment consists of a single, 10-µm ra-

dius GaAs/AlGaAs microring vertically coupled to two straight waveguide
buses as shown in Fig. 6.10. The wavelength spectrum of the both the


20 micron

Input Through
Fig. 6.10. Optical micrograph of a GaAs vertically coupled OCDF micror-
ing resonator.
160 6. All-Optical Switching and Logic using Microresonators

Fig. 6.11. Normalized measured spectral response and for the throughput
and drop ports for the OCDF microring resonator.

through and the drop ports of the device are shown in Fig. 6.11. The res-
onator has a bandwidth of 0.22 nm, a finesse of 45, and a quality factor
of 7000.

6.3.2 Nonlinear Device Characteristics

The response time of the microring was measured as follows. A semi-

conductor CW source, tunable in the 1.5 − 1.6-µm range, provided a
low-power (signal) beam with typical coupled power of 1 mW. A second
semiconductor laser diode is gain-switched at 8.5 GHz to produce 25-ps
pulses and is fed to an external modulator to reduce the repetition rate
to 125 MHz. It was subsequently amplified by an EDFA to 10-mW aver-
age power. The CW signal (probe) beam was blue-tuned to the microring
resonance at 1555 nm, whereas the control beam was tuned to the next
higher microring resonance at 1544 nm. Wavelength tuning was achieved
by both tuning the input laser sources wavelengths and by thermally tun-
ing the microring resonator using a thermoelectric cooler. The tempera-
ture sensitivity of the microring was 0.11 nm/◦ C. Both input beams were
coupled together using a 10:90 fiber coupler and were fed to the device
input waveguide using a conically tipped single-mode fiber. The dropped
and through beams were collected using conically tipped fibers, spec-
trally filtered from the pump beam using a band-pass filter centered at the
probe beam wavelength, and analyzed as functions of time with a 45-GHz
6.3 Time-Division Demultiplexing 161

Fig. 6.12. Time traces for (a) the input pump pulse at 1545 nm and (b) the
output dropped signal initially blue-tuned to the 1555-nm resonance.

detector and a 40-GHz digitizing oscilloscope. Thus, measurements were

instrument limited.
Fig. 6.12 shows both the input pump pulse at 1545 nm and the output
dropped signal whose wavelength was slightly tuned off resonance at
1555 nm, thus initially undropped. When the 25-ps pump beam arrives,
it is partially absorbed by the microring material via TPA, and free carriers
are generated causing a change in both the absorption and the refractive
index coefficients. This leads to a shift in the microring resonance toward
shorter wavelengths temporarily overlapping the CW signal pulse and
resulting in high transmission at the drop port. It is observed that the
dropped signal exhibits a switching window of 30-ps pulse width and
approximately 8-dB switching contrast, demonstrating that the device is
capable of switching data up to a rate of 30 Gb/s.
In the second part of the experiment, the use of the microring res-
onator as a spatial pulse router is demonstrated. Specifically, the micror-
ing routes incoming data to either the drop port or the through port,
depending on the presence or absence of a control pulse. The signal
(probe) beam was a 5-GHz, 40-ps RZ data stream tuned to the microring
resonance at 1555 nm. For the control beam, an external modulator was
used to chop a CW beam that was tuned to the next microring resonance
at 1544 nm. This resulted in a train of 150-ps control pulses at 80-MHz
repetition rate. The signals detected at the drop and through ports are
shown in Fig. 6.13.
162 6. All-Optical Switching and Logic using Microresonators

Fig. 6.13. Time traces for (a) the input 5-GHZ RZ data stream slightly de-
tuned to the resonance at 1555 nm, (b) the control signal at the 1545 nm
resonance, (c) unswitched data at the throughput port, and (d) demulti-
plexed data at the drop port.

It can be observed from the figure that the drop port signal is the
demultiplexed data from the input data stream, whereas the remaining
unswitched data are passed on to the through port where further de-
multiplexing can be performed. The cross talk at the drop port was
measured to be less than 8 dB. This cross talk is primarily due to the
asymmetry between the input and the output coupling coefficients re-
sulting from fabrication imperfections. This asymmetry in the coupling
coefficients is the reason for incomplete extinction at resonance for the
6.4 All-Optical Logic 163

throughput spectrum shown in Fig. 6.11. Higher contrast and therefore

less cross talk can be obtained by better matching of the input and out-
put coupling coefficients and reducing the round-trip loss to result in a
critically coupled OCDF resonator. The pulse energy required for switch-
ing was approximately 50 pJ/pulse at the input of the waveguide. The
actual energy coupled to the resonator is much less due to scattering
and waveguide tapering losses and is difficult to measure. Engineering
the material bandgap and increasing the resonator field enhancement
will lower the switching power. It should be noted that the device can be
also used in a similar manner to multiplex data onto the incoming data
stream via the add port of the device.

6.4 All-Optical Logic

We next consider the implementation of nonlinear switching mecha-

nisms in microresonators for the construction of all-optical logic gates.
By injecting two signals in a microresonator copropagating at two differ-
ent wavelengths or counter-propagating at the same wavelength, we can
switch on or switch off an auxiliary signal. Careful preparation can en-
able AND/NAND or OR logical operations when the injected signal pulses
coincide spatially and temporally.

6.4.1 AND/NAND

In this section, all-optical AND and NAND logic gate operation via nonlin-
ear absorption is demonstrated in single semiconductor micro-racetrack
resonators [211]. Operation near critical coupling allows a high switch-
ing contrast [57]. The device is pumped with two counter-propagating
data streams tuned to one of its resonance wavelengths and is probed at
the next higher resonance wavelength. The performance of both InP and
GaAs devices are compared in terms of switching energy and speed.
The nonlinear optcal mechanism implemented in the gates is again
the change in refractive index from free carriers generated by TPA [212].
The number of free carriers generated inside the resonator is quadrati-
cally proportional to the pump intensity inside the ring, which is in turn
proportional to the resonator intensity buildup factor. In addition, the
phase change needed for switching is reduced by the resonator finesse.
The input optical energy needed for switching is, thus, reduced by the
third power of the intensity buildup. On the other hand, thermal nonlin-
earity is only proportional to the intensity inside the resonator.
The guiding layer of the InP and GaAs devices are designed to have
bandgap energies at 1380 and 800 nm, respectively. The 1550-nm pump
pulses are absorbed through TPA inside the resonator, and free carriers
164 6. All-Optical Switching and Logic using Microresonators

are generated. This results in a temporal decrease in the refractive in-

dex of the resonator waveguide and, thus, a blue shift in its resonance
wavelengths. In a critically coupled ring, when a probe beam is initially
tuned to a resonance, it is highly attenuated at the output representing
a logical “0.” If “A” and “B” are the input data streams, then when either
“A” or “B” is “1,” the phase shift acquired by the resonator is not enough
to switch the probe out of resonance. However, when both signals are
“1,” the number of generated carriers is four times higher because of the
quadratic dependence of the TPA. Moreover, the nonlinear transmission
of the resonator enhances the switching and brings the probe beam out of
resonance to “1” or high transmission. This is functionally representative
of AND logic operation. If the probe beam is initially out of resonance,
then the device can be operated as a NAND logic gate.
The InP devices [76] were tested first. These consisted of a later-
ally coupled micro-racetrack resonator with a 10-µm radius and a 3-µm
straight coupling section [Fig. 6.14(a)]. The waveguide core consists of
GaInAsP and possesses a cross section of 0.5 × 0.5 µm2 . Figure 6.14(b)
shows the measured spectral response at the throughput port of the

Fig. 6.14. (a) Scanning electron microscope picture of the InP device and
(b) its measured (solid line) and fitted (dashed line) spectral response at
the throughput port.
6.4 All-Optical Logic 165

Fig. 6.15. Experimental setup for optical logic AND gate using the micro-
racetrack resonator. PG: Pattern generator. PC: Polarization controller.
BPF: Band-pass filter.

racetrack resonator. For the resonance at 1550 nm, the extinction is 20

dB, the round-trip power loss is 40% and the coupling coefficient is 46%.
The ring resonator has a 3-dB bandwidth of 1.8 nm, a free spectral range
of 10 nm, a finesse of 6, and an estimated of 1.5.
Fig. 6.15 displays a schematic diagram for the experimental setup
used to demonstrate the logic gate operation. A pump beam source, con-
sisting of a CW external cavity laser diode tuned at one of the resonator
resonant wavelengths at 1550 nm, was externally modulated by a pattern
generator. The pump pulses were return-to-zero (RZ) data generated at
0.5 GHz with 500-ps pulse duration. They were split by a 50/50 coupler
into two counter-propagating data streams channels. Each channel was
separately amplified by two EDFAs and filtered with a band-pass filter
then injected into one port of the device using conically tipped fibers.
The pump pulse energy was 18 pJ at the input of the device. A variable
optical delay line was used to adjust the arrival of one data channel with
respect to the other at the resonator. The probe beam, a CW signal tuned
to the resonator resonance at 1560 nm, was amplified separately, and
coupled with one data channel using a 50/50 coupler. The probe power
was 10 mW at the device input. The output probe was collected using a
fiber circulator, band-pass filtered at the probe beam wavelength, opti-
cally amplified and fed into a 40-GHz detector and a 50-GHz oscilloscope.
166 6. All-Optical Switching and Logic using Microresonators

Fig. 6.16. Time traces showing an AND logic gate using the InP microres-
onator: (a) “A” and (b) “B” are the two input pumps tuned to the resonance
at 1550 nm and (c) “F = A AND B” is the output probe signal tuned to the
next higher resonance at 1560 nm.

Fig. 6.16 displays the time traces of the input and output data patterns
illustrating AND gate operation. The probe beam is initially tuned to res-
onance and, hence, suffers low transmission representing a logical “0.”
The estimated pump intensity required to switch the probe beam out of
resonance is 0.4 cm, when the pump beam is tuned to the max slope of
the resonator where the switching sensitivity is maximum [212]. We have
adjusted the intensity and wavelength of the pump beams, “A” and “B,”
such that either one alone cannot switch the probe beam out of reso-
nance. When both “A” and “B” are logical “1”, the intensity is four times
higher and is enough to switch the probe out of resonance to the ON or
“1” logic state.
The speed of the device was investigated using a mode-locked laser at
5.6GHz and externally modulated at 140MHz. The pump pulse possessed
a pulse width of 35 ps and energy of 20 pJ. A switching window of 100 ps
6.4 All-Optical Logic 167

(limited by the carrier lifetime) and a 13-dB switching contrast (limited

by the noise floor of the receiver) were measured. From measured data
and a simulation model, we estimate a 0.9-nm temporal tuning of the
microring resonator.
Next, similar GaAs devices [57] were tested. The device had the same
configuration as the one shown in Fig. 6.14(a). It had a resonance at 1548
nm with an extinction of 19.5 dB, a round-trip power loss of 20%, a cou-
pling coefficient of 25%, a 3-dB bandwidth of 0.78 nm, a free spectral
range of 11 nm, and a finesse of 14. We used a similar setup to the
one shown in Fig. 6.15. To obtain high-speed data, a gain-switched laser
diode at 8.4 GHz and externally modulated at 140 MHz was used for the
data source. The pump pulses had energy of 80 pJ per pulse at the in-
put waveguide of the device. Again, the actual pulse energy coupled to
the device is hard to measure. The pump data were tuned to the device
resonance at 1548 nm, whereas the probe was tuned to its resonance at
1559 nm. The probe beam had a power of 10 mW at the input waveguide.
Figure 6.17(a) and (b) display the time traces for the inputs “A” and “B.”
Figure 6.17(c), displays the output probe “F” when the probe beam was ini-
tially tuned to resonance. It shows that the output is only “1” when both
“A” and “B” are “1,” demonstrating AND gate operation. Figure 6.17(d)
displays the output “F” when the probe beam was initially 0.4 nm blue-
tuned to resonance. Initially, the probe beam is highly transmitting or
“1.” When both “A” and “B” are “1”, the generated carriers shift the res-
onance wavelength enough to bring the probe beam in resonance and,
thus, impart low transmission resulting in a logical “0.” In both cases,
the output “F” can be observed to follow the inputs instantaneously. We
measured the switching window of the device to be 35 ps limited by the
carrier lifetime and the detection system.
To characterize the performance of the devices, two additional mea-
surements were performed: the TPA coefficient and the carrier lifetime. In
straight waveguides, the single-beam transmittance technique [213] was
used to measure the TPA coefficient at approximately 18 and 43 cm/GW
for GaAs and InP, respectively. Based on these values, the assumption of
constant round-trip loss inside the resonator is valid for the pump in-
tensity that we used for the logic gate experiment. In other words, the
nonlinear loss contribution is neglected.
The carrier lifetime of the materials was measured by pumping a
straight waveguide section and probing the induced absorption at a dif-
ferent wavelength. Since the materials are intrinsic and bulk, ambipolar
diffusion is expected. The mobility of GaAs was estimated as 500 cm2 /Vs
and of InP as 210 cm2 /Vs, consistent with 780 and 310 cm2 /Vs, as cal-
culated from published data [201]. A measured quadratic dependence of
the carrier lifetime on the straight waveguide width is a signature of the
diffusion process. Faster response might be achieved by shortening the
168 6. All-Optical Switching and Logic using Microresonators

Fig. 6.17. Time traces showing logic operation using the GaAs microres-
onator: (a) and (b) are the inputs “A” and “B” tuned to the resonance
at 1548 nm; (c) output “F” when the probe was initially in resonance at
1559 nm; and (d) output “F” when the probe was initially blue-tuned out
of resonance at 1558.6 nm.

carrier lifetime, e.g., by sweep-out [202], or doping the guiding layer with
p-type carriers 1 Engineering the bandgap through tensile strain can also
enhance the hole mobility and, thus, shorten the carrier lifetime.
The generated free carrier density was calculated from the carrier life-
time measurements and compared with the estimated value from the
index change. The refractive index change was estimated by measuring
the shift in the microresonator resonance wavelength and was estimated
to be approximately −10−3 . The free carrier density calculated by ab-
sorption is two orders of magnitude higher than that estimated. This
This results in a reduced n-type minority carrier lifetime, which is much
shorter than the ambipolar lifetime.
6.4 All-Optical Logic 169

discrepancy is likely due to the difficulty in estimating the input power

at the coupling section of the microring because of coupling losses and
tapering waveguide losses.

6.4.2 NOR

In this section, all-optical NOR logic gate operation is demonstrated.

Again, the employed mechanism is nonlinear absorption albeit this time
in symmetric GaAs-AlGaAs microring resonators whose resonances are
closely matched [214]. Two input pump data streams are tuned close
to one resonance of the symmetric microrings to switch a probe beam
tuned to another resonance by two-photon absorption. The use of two
rings provides for better cascading in photonic logic circuits because of
the higher number of available ports.
All-optical AND and NAND logic gates based on single-ring devices
were discussed previously [211]. However, multiple ring-based logic gates
are better suited for cascading in photonic logic circuits because they pro-
vide more ports, which provide for better separation of single channels
and prevent backfeeding or crossfeeding of optical power, and minimize
the need for additional external components such as circulators, isola-
tors, and band-pass filters.
An optical micrograph of the NOR gate is shown in Fig. 6.18. It consists
of two symmetric 9.6-µm-radius microring resonators vertically coupled
to a throughput bus waveguide. Each ring has a drop port separated by
250 µm from the throughput port. These extra input and output ports
eliminate the need for any external fiber splitter and circulator, as were
required for the single microring logic gate [211]. The 250-µm-separation
was chosen to match the pitch distance of a conically tipped fiber array
used to couple light in and out of the device. The throughput and drop
waveguides are tapered from 2.5 µm at the cleaved facets to 0.8 µm at the
coupling section to the resonators over a length of 200 µm. The 0.8-µm
width at the coupling sections of the microrings maintains single-mode
behavior of the bus waveguide. The microring level was first fabricated
and then flip-bonded to a new substrate via polymer wafer bonding. The
growth substrate was then etched away, and the bus waveguides were
Fig. 6.19 displays the spectra of the throughput and both drop ports
of the device. The resonance wavelengths of the two microrings were de-
signed to coincide with each other. The resonators have 3 dB bandwidths
of 0.22 and 0.30 nm; finesse of 47 and 35; quality factors of 7100 and
5200, respectively; and a free spectral range of 10.42 nm. Assuming sym-
metric coupling for each resonator, a coupling coefficient of 5.3% and a
round-trip loss of 2.3% are estimated for the first microring resonator,
and a coupling coefficient of 7.2% and a round-trip loss of 2.9% are esti-
mated for the second microring resonator. The small differences in the
170 6. All-Optical Switching and Logic using Microresonators

Fig. 6.18. Optical micrograph of the NOR logic gate. The two microring
resonators are symmetric and of 9.6-µm radii. The etched features ap-
pear lighter, whereas the waveguides appear as dark lines and curves.
Inset, magnified picture of the microring resonators underneath the bus
waveguides. A, B, pump inputs; C, probe input; F, probe output.

two microrings arise from bus-waveguide alignment asymmetry. Because

of lithographic nonuniformity, there is a 0.2-nm mismatch between the
resonators for the 1557-nm peak, which results in a higher than expected
switching energy.
For a pump pulse that is much longer than the cavity lifetime, the
quasi-steady-state analysis can again be used to solve for the interac-
tion between the pump and the probe electric fields circulating inside a
microring resonator. As the guiding layer of the GaAs microrings is de-
signed to have bandgap energy at 800 nm, the 1550-nm pump pulses
are partially absorbed through TPA inside the resonator, and free car-
riers are generated according to Eq. 6.3. The TPA coefficient was found
to be 18-cmGW, based on transmittance measurements of straight GaAs
waveguides, and the effective carrier lifetime was found by pump-probe
measurements of similar straight and tapered waveguide sections to be
40 ps. The effective carrier lifetime is determined by carrier diffusion and
surface recombination at the sidewalls. The change in refractive index
6.4 All-Optical Logic 171

Fig. 6.19. Normalized spectrum for (a) the throughput port, (b) the drop
port of the first microring resonator, and (c) the drop port of the second
microring resonator.

resulting from the generated free carriers causes the microring resonance
peaks to shift and thus can be used to switch a probe beam on or off.
Fig. 6.20 displays a spectral schematic for one of the microrings in the
NOR logic gate operation. The pump beam (A or B) is slightly blue-tuned to
the first peak, whereas the probe beam (C) is initially blue-detuned to the
second (or any other) peak in the spectrum (shown as a solid curve) and
hence passes unaffected to the through port. When the pump beam (A or
B) is switched on, it induces free carriers in the ring resonator by TPA, and
the resultant refractive-index change causes the spectrum to blueshift
(shown as a dashed curve). This spectral shift causes the second peak to
coincide with the probe wavelength, resulting in its being dropped from
the output. The first peak moves past the pump wavelength, causing it to
be slightly red-detuned. Note that the pump wavelength is initially only
slightly blue-detuned [0.5 full width at half maximum (FWHM)], whereas
the probe wavelength has a larger gap (1 FWHM). This exact wavelength
placement is critical to device operation, as the pump wavelength has to
overlap the first peak at all times for any reasonable intensity buildup
within the ring.
172 6. All-Optical Switching and Logic using Microresonators

Fig. 6.20. Spectral schematic of NOR logic gate operation. Solid curves,
spectral characteristics of intensity inside the ring when the pump beam
is off. Dashed curves, the blueshifted spectral characteristics when the
pump is on. The positions of the pump and the probe wavelengths are
indicated by dotted vertical lines.

We estimate that the intensity in the ring needed to produce a spectral

shift of 0.30 nm (1 FWHM) is 2.0×108 W/cm2 . The pulse energy in the bus
waveguide is estimated to be 6 pJ for a pulse width of 35 ps. Compared
with that of conventional two-armed interferometers such as Michelson
and Mach-Zehnder types, the switching energy is reduced by a factor
proportional to the third power of the microring cavity finesse.
The input and output ports and signals are as shown in Fig. 6.18.
The pump pulses are provided by a laser diode tuned close to the device
resonance at 1546 nm and gain switched at 8.5 GHz. The repetition rate
is reduced by use of an external modulator at 140 MHz. The pulses are
further split by a 50/50 fiber splitter, and one arm is delayed by an optical
delay to provide the two independent input data pulses, A and B. Each
data stream is amplified separately by an EDFA and then filtered to reduce
the amplified spontaneous emission noise. The CW probe beam (C) is
provided by a tunable external cavity laser diode that is tuned to the
device resonance at 1557 nm. Input pump data and the probe beam are
then coupled to the device by use of two conically tipped fiber arrays, and
the output light is collected in the same manner. The output probe signal
is then spectrally filtered and optically amplified by another EDFA. When
either beam A or beam B is switched on, it induces free carriers in the
corresponding ring by TPA and causes the resonance peaks to blueshift
and overlap the probe wavelength, resulting in its being dropped from
the output (F; Fig. 6.18). Thus the logic function can be written as F = A
In Fig. 6.21 we plot the time traces for input data streams A and B and
output probe F, showing all possible binary combinations. The switching
6.5 Summary 173

Fig. 6.21. Time traces for the input pump data streams (a) A and (b) B at
1546 nm and (c) the output probe signal at 1557 nm, showing the NOR
logic operation.

contrast is 6 dB and is limited by the scattering loss in the rings, which

results in noncritical coupling. The measured switching energy is esti-
mated to be 20 pJ per pulse at the input of the waveguide, which is
higher than expected for an ideal case calculated by use of device and
material parameters. We believe that two factors contribute to the higher
than estimated switching energy: the slight mismatch between the two
resonators, and the fiber-to-waveguide coupling efficiency. The 0.2-nm
resonance mismatch between the two microring resonators results in
nonoptimal placement of the pump wavelength with respect to the peaks.
Also, the fiber coupling efficiency was measured based on experiments
in which a single fiber was coupled to a single test waveguide. Aligning
a fiber array to multiple waveguides is inherently more difficult and re-
sulted in lower coupling efficiency. Decreasing the pump pulse widths to
lower values will result in distortion of the output pulses. Specifically,
the output pulse width will be determined by carrier diffusion times and
not by the input pulse width anymore. Assuming ideal input pulses, the
rate of free-carrier generation by TPA, and hence the index fall time, is
limited by the photon cavity lifetime (time for intensity to build up in
the ring). However, the carrier removal rate, and hence the index rise or
recover time, is limited by diffusion and surface recombination.

6.5 Summary
In this chapter we gave an overview of the theoretical framework along
with experimental demonstrations of all-optical switches and logic de-
vices in III–V microring resonators by TPA using the pump-probe method.
174 6. All-Optical Switching and Logic using Microresonators

The dominant nonlinear effect was observed experimentally and con-

firmed by simulation to be carrier-induced change in the refractive index.
Also demonstrated was the use of a single microring resonator notch fil-
ter in pulse reshaping. Moreover, using a microring resonator in the OCDF
configuration, we demonstrated time-division demultiplexing and spatial
pulse routing. As the switching speed of these devices is currently limited
by the carrier lifetime and not the microring cavity lifetime, the perfor-
mance may be improved further by a variety of known methods for reduc-
ing carrier lifetimes.2 The pump energy required for switching can also
be reduced by more efficient absorption mechanisms such as operating
near the band edge at the expense of increased losses, which ultimately
limit device cascadability. Consequently, although much weaker, the best
promise for cascadable all-optical logic continues to be the Kerr effect.

Methods for reducing carrier lifetimes in III–V semiconductors include pro-
ton/ion damage and dc biasing to sweep out free carriers.
7. Distributed Microresonator Systems

7.1 Introduction

In the previous chapters, the linear and nonlinear transfer characteristics

of a single (or a few) microresonator were examined in detail. In this chap-
ter, the linear and nonlinear propagation characteristics of a distribution
of microresonators are examined. Due to their inherent simplicity, we
begin with a treatment of the linear pulse propagation characteristics of
distributed all-pass resonators, move on to an examination of their non-
linear propagation characteristics, including the prediction of solitons,
and make some comments on the limitations of these structures. Finally,
we conclude with a generalized treatment of periodically distributed mi-
croresonators. These structures may be considered to be a new family of
artificial media [215, 216] with analogies to photonic crystals [217–219].
Note: Both in this chapter and in the literature cited here, the terms SCIS-
SOR (side-coupled integrated spaced sequence of resonators) and CROW
(coupled-resonator optical waveguide) are used, which refer, respectively,
to parallel and serial cascaded rings.

7.2 Linear Propagation in Distributed Microresonators

A pulse propagating though a single all-pass resonator acquires a

frequency-dependent phase shift that serves to delay and/or distort the
pulse shape. By arranging a sequence of resonators coupled to an ordi-
nary waveguide as in Fig. 7.1 (SCISSOR or a parallel-cascade), the effective
propagation constant of the guide can be modified. The modified effective
propagation constant can be defined as the accumulated phase per unit
length and is composed of the propagation constant of the waveguide
itself plus a contribution from the effective phase shift of the resonators.
For a resonator spacing of L, the effective propagation constant becomes

keff (ω) = n0 ω/c + Φ (ω) /L. (7.1)

A plot of the dispersion relation (effective propagation constant against

radian frequency) for various values of the self-coupling parameter r is
176 7. Distributed Microresonator Systems

E3 E4 R

E1 E2 L
Fig. 7.1. A microresonator-structured, fully transmissive waveguide con-
sisting of a SCISSOR. E1 is the incident field, E4 is the field injected into
the disk, E3 is the field after one pass around the resonator, and E2 is the
transmitted field.
Wavevector, keff −k0

kR + 2π

r2 = 0.00
r2 = 0.25
r2 = 0.75
kR − 2π
L r2 = 0.95

ωR −ncR ωR ωR + ncR
0 0
Frequency, ω
Fig. 7.2. The dispersion relation for light propagation in a SCISSOR with
differing values of the self-coupling coefficient r . For generality, the
waveguide contribution of constant slope k0 has been subtracted from
the effective propagation constant keff .

shown in Fig. 7.2. The deviation in the curve from the light line of the
ordinary waveguide takes the form of periodic changes in the group ve-
locity and group velocity dispersion with a periodicity of c/n0 R. Here,
the material and waveguide dispersion are assumed to be negligible. In
fact, it will be shown later that the dispersive nature of the resonators
in general dominates the intrinsic dispersion by many orders of mag-
nitude. A pulsed waveform can be decomposed into the product of
a slowly varying envelope A (t) and a carrier wave with frequency ω0
as E (t) ≡ 12 A (t) exp (−iω0 t) + c.c.. The relationship of the carrier fre-
quency to some resonance frequency controls the central operating point
on the dispersion relation curve and thus sets the normalized detuning
φ0 = (ω0 − ωR ) TR , where TR = FSR−1 is the resonator transit time and
ωR is the closest resonance frequency. The transfer function of a single
resonator can be expanded into two embedded Taylor’s series: one for
7.2 Linear Propagation in Distributed Microresonators 177

the effective phase shift (expanded about the normalized detuning φ0 ),

and one for the exponential (expanded about the effective phase shift of
the carrier Φ0 ):
⎧ ⎡ ⎤n ⎫
⎨ -∞ n

- 1 d m
Φ ⎬
H (ω) = eiΦ = eiΦ0 1 + ⎣ (φ − φ )
m ⎦ . (7.2)
⎩ n! m! dφm φ
0 ⎭
n=1 m=1 0

Using this formal expansion, the transmitted field is related to the inci-
dent field with the assumption that the effective phase shift induced by
each resonator is distributed over the spacing L so that the effective prop-
agation constant is independent of propagation distance at the macro-
scopic scale. The field at some point zj+1 separated a small distance δz
from the field at another zj is given by a similar equation that distributes
the resonator contribution and includes that of the waveguide:

nω Φ0

i + δz
Ej+1 (ω) = e c L ×
⎧ ⎡ ⎤n ⎫
⎨ -∞ n
i ⎣ n0 -∞
1 δz dm Φ ⎬
1+ ∆ωδz + (φ − φ0 ) E (ω) .
⎩ n! c m
m! L dφ φ0 ⎭ j
n=1 m=1
Taking the Fourier transform of this equation results in a difference equa-
tion relating the pulse envelopes at the two points:

Aj+1 (t) = Aj (t)

⎡ ⎤n
-∞ n ∞
i ⎣ n0 ∂ 1 δz dm Φ ∂ m⎦
+ i δz + i Aj (t) . (7.4)
n! c ∂t m=1 m! L dωm φ0 ∂t
Finally, the differential approximation Aj+1 (t) − Aj (t) /δz → dA/dz
is made and δz is allowed to go to zero.1 This procedure yields a linear
propagation equation for the pulse envelope:
⎡ ⎤

dA ⎣ n0 ∂ 1 1 dm Φ ∂ m⎦
= − +i i A. (7.5)
dz c ∂t m=1
m! L dωm φ0 ∂t

The different terms in this equation are isolated and examined in what

7.2.1 Group Velocity Reduction

The increased phase sensitivity on resonance is related to an increased
group delay per resonator. This extra delay distributed among the res-
onators is responsible for a slower group velocity. The inverse of the
Implicit in this assumption is that each resonator is not strongly driven; i.e.,
the effective phase shift Φ per resonator is small with respect to unity.
178 7. Distributed Microresonator Systems

group velocity is proportional to the frequency derivative of the propa-

gation constant,
dkeff n0 1 dΦ n0 2π R 1 − r2
k eff = = + = 1+
dω c L dω c L 1 − 2r cos φ0 + r 2
n0 4R
→ 1+ F . (7.6)
φ0 =0,r ≈1 c L

The group velocity 1/k eff is observed to be composed of contributions

from propagation in the waveguide and discrete delays introduced by the
resonators. The component of the group velocity reduction introduced
by the resonators is proportional to the finesse and can dominate the
waveguide component for moderate values of the finesse.

7.2.2 Group Velocity Dispersion

Although propagation in the waveguide is assumed to be dispersionless,

strong dispersive effects are induced by the resonator contribution. It
has been argued that of all photonic devices, microresonators currently
provide the highest dispersion per unit volume [207]. The group velocity
dispersion (GVD) is proportional to the second frequency derivative of
the effective propagation constant,
dk2eff 1 d2 Φ TR2 −2r 1 − r 2 sin φ0
eff = = = 2
dω2 L dω2 L (1 − 2r cos φ0 + r 2 )

3 3F 2 TR2
→√ ∓ . (7.7)
φ0 =±π /F 3,r ≈1 4π 2 L

Although the GVD coefficient is zero on resonance, appreciably strong

normal (positive) or anomalous (negative) values of the dispersion can
be obtained on the red (lower) or blue (higher) side of resonance,
√ respec-
tively. The dispersion maxima occur at detunings φ0 = ±π /F 3 where
the magnitude of the group velocity dispersion is proportional to the
square of the finesse. This induced structural dispersion can be many
orders of magnitude greater than the material dispersion of typical op-
tical materials [220]. For example, a 10 ps optical pulse propagating in
a sequence of resonators with a finesse of 10π , free spectral range of
10 THz (∼5 µm diameter), and a spacing of 10 µm experiences a group
velocity dispersion coefficient k 2
eff of roughly 100 ps per mm. In general,
this structural dispersion can be as much as seven orders of magnitude
greater than material dispersion in conventional materials such as silica
fiber (20 ps2 per km). Figure 7.3 shows the pulse evolution for a 10 ps
pulse propagating through 100 resonators, each tuned to the anomalous
dispersion maxima.
7.3 Nonlinear Propagation in Distributed Microresonators 179

Power, mW

0 100
0 20 40 60 200
80 100 Time, ps
z (resonator #)
Fig. 7.3. A weak pulse tuned to the dispersion maxima disperses while
propagating in a SCISSOR. A 10-ps FWHM hyperbolic secant pulse tuned
for maximum anomalous GVD (B = 0.13) enters the system consisting
of 100 resonators each with a 5-µm diameter and finesse of 10π , spaced
by 10 µm. Note that the peak power is reduced by a factor of about 4
after propagating only 1 mm as a consequence of the strong induced
dispersion. (After Ref. [129], ©2002, Optical Society of America.)

7.2.3 Higher Order Dispersion

Higher orders of dispersion may be derived from Eq. 7.5, with each subse-
quently possessing a maximum that is proportional to the cavity lifetime
F TR to the nth power. Specifically, the third-order dispersion coefficient
  + ,
1 d3 Φ TR3 −2r 1 − r 2 1 + r 2 cos φ0 − 3r + r cos 2φ0
eff ≡ = 3
L dω3 L (1 − 2r cos φ0 + r 2 )
4 F 3 TR3
→ − . (7.8)
φ0 =0,r ≈1 π3 L

It is important to note that all orders of dispersion in the Taylor expan-

sion become significant when the pulse bandwidth is nearly as wide as the
resonance bandwidth or equivalently when the pulse is nearly as short
as the cavity lifetime.

7.3 Nonlinear Propagation in Distributed Microresonators

In addition to inducing a strong group delay and dispersion, a resonator

may enhance a weak nonlinearity. If the resonator possesses a nonlinear
refractive index, i.e., Kerr nonlinearity, then the internal phase shift will
be intensity dependent. For simplicity the nonlinearity of the waveguide
180 7. Distributed Microresonator Systems

is neglected in what follows since it is expected to be small. The intensity-

dependent contribution of the resonators to the internal phase shift is
given by γ2π R|E3 |2 , where γ represents the strength of the intrinsic
nonlinear propagation constant. This parameter is of course tradition-
ally fixed for a given material system. The more useful parameter γ/Aeff
can be as low as 0.002 m−1 W−1 for standard single-mode silica fiber or
as high as 102 m−1 W−1 in an air-clad GaAs or chalcogenide glass based
waveguide.2 Near resonance, the effective phase shift is sensitively de-
pendent upon the internal phase shift that is in turn dependent on an en-
hanced circulating intensity. The combined action of these effects gives
rise to a dually-enhanced effective nonlinear propagation constant γeff ,
calculated from the derivative of the effective phase shift with respect to
the input intensity:
1 dΦ 1 dΦ dφ d|E3 |2 γ2π R 1 − r2
γeff ≡ = =
L d|E1 | 2 L dφ d|E3 | d|E1 |
2 2 L 1 − 2r cos φ0 + r 2
8R 2
→ γ F . (7.9)
φ0 =0,r ≈1 πL

As can be observed from this equation, the increased phase sensitivity (or
group velocity reduction) and the buildup of intensity contribute equally
to enhance quadratically the nonlinear propagation constant with respect
to the finesse [151]. To properly account for the all the third-order Kerr
nonlinear contributions of the spectral components of three fields, a dou-
ble convolution of the three interacting fields is performed in the spectral
domain. In the time-domain, the double convolution operation is reduced
to simple multiplication. This allows for the straightforward addition of
a nonlinear contribution [97] to the internal phase shift term in the linear
propagation equation (Eq. 7.5):
⎧ ⎫
! "
dA ⎨ n0 ∂ ∂ m⎬
1 1 dm Φ 2
= − +i γ2π RB|A| + iTR A.
dz ⎩ c ∂t m! L dφm φ ∂t ⎭
m=1 0

For two nonlinearly interacting resonant pulses, the results here derived
for the self-phase modulation (SPM) effect similarly apply to the effect of
cross-phase modulation (XPM) although with an extra degeneracy factor
of two [178].

For the purpose of quoting the material nonlinearities, we default to the more
useful parameter γ whereby the nonlinear propagation constant is scaled by
the effective mode area such that γP L is the nonlinear phase shift acquired
for a power level of P over a distance L.
7.3 Nonlinear Propagation in Distributed Microresonators 181

7.3.1 SCISSOR Solitons

Next, the nonlinear propagation equation that retains only the lowest
order dispersive and nonlinear terms in Eq. 7.10 is examined.

The time

coordinate is shifted to the reference frame of the pulse t = t − k eff z . It
is found that, in this limit, the pulse evolution is governed by a nonlinear
Schrödinger equation (NLSE) with effective GVD and SPM parameters:

∂ 1 ∂2
A = −i k A + iγeff |A|2 A. (7.11)
∂z 2 eff
∂t 2
Soliton solutions for this equation exist provided that the enhanced non-
linearity and induced dispersion are of opposite sign [221–223]. Although
the sign of the enhanced nonlinearity is predetermined by the sign of the
intrinsic nonlinearity in the overcoupled regime, the sign of the induced
dispersion is as shown previously in Eq. 7.7, assumes the opposite sign
of the normalized detuning from resonance. Fig. 7.4 shows the frequency
dependence of the lowest order GVD k eff and enhanced nonlinearity γeff
and also indicates the optimum detuning for soliton propagation.
The fundamental soliton solution for this equation is [224]
  1 2
A z, t = A0 sech t /TP ei 2 γeff |A0 | z , (7.12)
  the amplitude and pulse width are related according to |A0 | =
keff  /γeff TP2 , below which the pulse is severely distorted by all orders
of dispersion. The finite response time of the resonator places a lower
bound on the pulse width TP
. A scaling
factor B is defined
to be the ra-
√ 2
tio of the pulse bandwidth 2 arcsech 1/ 2 /π TP to the resonator

Soliton condition
0.5 γeff
(2FT/π)2/L γ (2F/π)2 (2πR/L)

-4π -2π 0 π 2π 4π
F F 3F F F
Normalized detuning, φ0
Fig. 7.4. A SCISSOR soliton is created from a balance between resonator
enhanced nonlinearity and resonator-induced anomalous dispersion.
182 7. Distributed Microresonator Systems

bandwidth (1/F TR ), such that B = 2 arcsech 1/ 2 /π 2 F TR /TP . A

nonlinear strength parameter Γ = 4/π 2 F 2 γ |A0 |2 R is also defined.
With these definitions, a simple relation holds between Γ and B for the
fundamental soliton operating at the anomalous dispersion peak3 :
Γ = √
√ B2 ≈ B2. (7.13)
2 3 arcsech2 1/ 2

Higher order dispersive and nonlinear terms become increasingly signif-

icant when either B and/or Γ approach unity.
Time-domain simulations confirm the validity of this approximation.4
Figure 7.3 shows the pulse evolution of a low-power 10 ps FWHM hy-
perbolic secant pulse tuned for maximum anomalous GVD (B = 0.13)
in an AlGaAs or chalcogenide glass-based system. The system consists
of 100 resonators spaced by 10 µm each with a 5-µm diameter and fi-
nesse of 10π . As can be observed, the temporal pulse profile is greatly
dispersed. Figure 7.5 shows the pulse evolution for the same system,
but with a peak power of 125 mW, corresponding to the fundamental

Power, W

0 100
0 20 40 60 200
80 100 Time, ps
z (resonator #)
Fig. 7.5. A pulse with amplitude corresponding to the fundamental soli-
ton propagates in a SCISSOR without dispersing. The same parameters
were used as in Fig. 7.3, but with a peak power of 125 mW (Γ = 0.0196)
in a chalcogenide glass-based system. (After Ref. [129], ©2002, Optical
Society of America.)
The values of k eff and γeff are, respectively, lowered by factors of 3/4 and 9/16
from their given maximum values when operating at dispersion extremum
The simulations used to study pulse evolution in a sequence of waveguide-
coupled resonators are carried out using an iterative method in which each
iteration consisted of linear and nonlinear phase accumulation during one
round-trip within the resonator followed by interference at the coupler. Tra-
ditional beam or pulse propagation split-step Fourier methods are unneces-
sary as both nonlinear phase accumulation and structural dispersion resulting
from a discrete impulse response are more readily treated in the time-domain.
7.3 Nonlinear Propagation in Distributed Microresonators 183

SCISSOR soliton (Γ = 0.0196). As can be observed, the pulse shape is

well preserved upon propagation. Many of the familiar characteristics of
fundamental solitons such as robustness, reshaping, pulse compression,
and pulse expansion have been observed in simulations to carry over
from the continuous-medium case. The strengths of dispersion and opti-
cal nonlinearity are two key parameters that control the length scale over
which solitons evolve with a F 2 reduction. Higher order solitons, satisfy-
ing Γ ≈ NS2 B 2 , where NS is an integer are readily observed in simulations,
but they are unstable because of higher order dispersive nonlinear effects
present in this system.5
Dark solitons consist of an intensity dip in an otherwise uniform
continuous-wave field. They can also be supported if the enhanced non-
linearity and induced dispersion are of the same sign (on the other side of
resonance). The fundamental dark soliton possesses a hyperbolic tangent
field distribution [97],
A z, t = A0 tanh t /TP eiγeff |A0 | z . (7.14)

Fig. 7.6 shows the propagation of the fundamental dark SCISSOR soliton
tuned to the normal dispersion peak.
Power, mW

40 0
60 200 100
80 300
z (resonator #) 100 400 Time, ps
Fig. 7.6. A negative pulse in a uniform intensity background with pa-
rameters corresponding to the fundamental dark soliton propagates in
a SCISSOR without dispersing. The incident field distribution was a hy-
perbolic tangent with twice the pulse width of the bright soliton and a
background power that was one fourth that of its peak power in Fig. 7.5.
(After Ref. [129], ©2002, Optical Society of America.)

Additionally, higher order dispersive and/or nonlinear effects render the scat-
tering of solitons to be inelastic. Under these conditions, the term solitary
wave is more appropriate.
184 7. Distributed Microresonator Systems

7.3.2 Induced Self-Steepening

In the previous section, frequency dependence of γeff was neglected. One

of the effects resulting from the frequency-dependent nature of γeff is an
intensity-dependent group velocity. This effect leads to the phenomenon
of self-steepening (SS) of a pulse where the peak of a pulse travels slower
than (+SS) or faster than (−SS) its wings. The self-steepening coefficient
s may be derived6 from Eq. 7.10, but it is more readily obtained from the
frequency derivative of the nonlinear coefficient:

γeff 2 dB 3F TR
s= = →√ ∓ . (7.15)
γeff B dω φ0 =±π /F 3,r ≈1 π

Although this effect is not easily isolated from the induced GVD in a se-
quence of resonators to form pulses with steep edges, it plays an impor-
tant role in the breakup of higher order solitons. The known phenomenon
of soliton decay [97, 225] involves the breakup of an NS order breathing
soliton into NS fundamental solitons of differing pulse amplitudes and
widths. Fig. 7.7 shows a situation in which a second-order SCISSOR soliton

Power, W

0 100
0 20 40 60 200
80 100 Time, ps
z (resonator #)
Fig. 7.7. A higher order breathing soliton is unstable under the
influence of the resonator-induced intensity-dependent group velocity
(self-steepening). Here a second-order soliton splits into two stable fun-
damental solitons upon propagation in a SCISSOR. The incident field dis-
tribution was the same as in Fig. 7.5 but with four times the peak power.
(After Ref. [129], ©2002, Optical Society of America.)
So as to correctly expand Eq. 7.10, the B term also must be expanded gener-
ating more time derivative terms within the square brackets. Thus, the self-
steepening contribution will not only consist of two m = 2 terms but also of
one m = 1 term. For terms such that m > 1, the time derivatives implicitly
appear to the far left of each term when the square brackets are expanded.
7.3 Nonlinear Propagation in Distributed Microresonators 185

with a launched peak power of 500 mW undergoes decay and splits into
two stable fundamental SCISSOR solitons. The solitons are well isolated
in time and uncorrupted by a background or pedestal. One of them pos-
sesses a higher peak power and narrower width than the original demon-
strating the potential for pedestal-free optical pulse compression. The
effects of induced self-steepening in a sequence of resonators can take
place for picosecond and even nanosecond pulses because unlike in the
case of intrinsic self-steepening, the relative strength of SS to SPM is not
governed by how close the pulse width is to becoming a single optical
cycle, 2π /ω0 but rather how close the pulse width is to becoming a sin-
gle cavity lifetime F TR . For the 10 ps pulse propagating in a SCISSOR
with the above parameters, the nondimensional self-steepening parame-
ter (s/TP ) is 0.173. To observe the same effect with traditional intrinsic
self-steepening, a six-cycle or 30-femtosecond pulse would be required.

7.3.3 Pulse Compression

More complicated interaction can exist between structural dispersion and

enhanced self-phase modulation in a SCISSOR structure. The soliton or-
der NS , proportional to the injected amplitude, is a convenient measure
of the relative strengths of the two processes. If the relative strengths of
the processes are balanced (NS = 1), it was shown previously that soliton-
like pulses may be propagated. If, however, the nonlinearity dominates
the dispersion (NS > 1), pulse compression can result. Because the en-
hanced nonlinearity and induced dispersion are each proportional to the
finesse squared [129], the soliton order for a SCISSOR can be set by sim-
ply choosing the ratio of the pulse width to the microring transit time
(TR ), 
8π n2 I0 R TP
NS = 3/4 . (7.16)
3 λ TR
Figure 7.8 demonstrates the result of a simulation in which a 5-ps Fourier
transform-limited pulse is injected into a ten-resonator SCISSOR tuned
above resonance at the peak of anomalous dispersion. The injected NS =
5 pulse fractures into multiple solitons of differing amplitudes along with
some dispersing waves. The soliton with the largest amplitude emerges
compressed by a factor of approximately five. It is worth noting that
although this method can be used to compress a pulse on a very short
distance scale, the amount of pulse compression is ultimately limited by
the bandwidth of SCISSOR.
The process of temporal imaging is related closely to pulse compres-
sion. With temporal imaging [226], intra-pulse structure can be preserved
and magnified (or demagnified). An imaging condition must be satisfied
in a setup that consists of an initial dispersive segment, a directly im-
posed quadratic phase, and a final dispersive segment of opposite sign
186 7. Distributed Microresonator Systems

Input Output
Power, W

50 Output
40 (1ps)
20 Input
10 (5ps)
0 10 20 30 40 50
Time, ps
Fig. 7.8. A highly compact microresonator-based 5× pulse compressor. A
25 W transform-limited input pulse (NS = 5) of 5-ps width is compressed
to 1 ps. In the process, some energy is shed in the form of other non-
dispersing pulses that walk away from each other linearly in time owing
to an intensity-dependent group velocity. These extra pulses might be
eliminated through the use of a saturable absorbing material that may
even be microresonator-based. Here, 10 AlGaAs or chalcogenide-based
microresonators of 10-µm-diameter form a SCISSOR. The resonators pos-
sess a finesse of 5π , coherent intensity buildup of 10, and nonlinear en-
hancement of 100.

to that of the initial. Owing to the large dispersive and nonlinear proper-
ties of microresonators, temporal imagers might be fabricated on an in-
tegrated photonic chip. Each of the three sections might be composed of
differently tuned SCISSOR elements. Sign selection of the dispersive seg-
ments is easily accomplished in SCISSORs tuned below (normally disper-
sive) or above (anomalously dispersive) resonances. The quadratic phase
chirp providing the temporal lensing may be accomplished by enhanced
cross-phase modulation in a resonant SCISSOR.

7.3.4 Nonlinear Detuning and Multistability

Equation 7.9 was implicitly derived with the assumption of low intensity.
Next, the intensity dependence of γeff is examined. It is found that the
7.3 Nonlinear Propagation in Distributed Microresonators 187

circulating intensity and γeff are in fact interdependent. The circulating

intensity depends on the buildup factor, which in turn depends on the cir-
culating intensity from the nonlinear detuning contribution. As a result
nonlinear resonators can possess memory and multistable branches in
the input–output relationships within certain operating regimes [80,198].
When operated near resonance, the onset of multistability occurs when
the circulating power is high enough to generate a single-pass nonlinear
phase shift of 2π radians. Saturation resulting from intensity-dependent
detuning pulling the resonator off resonance generally takes place well
before this effect. Working on the lower branch of the multistable relation
for positive detunings, the saturation of the effective nonlinear propaga-
tion constant is well fitted by a (1 + I/Is )−1 type of saturation model given
explicitly as
γ 2πL R B2φ0
γeff → , (7.17)
r ≈1 γ2π RB2φ
1 + 2π −Φ0 0 |A|2

where the saturating intensity near resonance is |AS |2 = π /γeff L. The

saturating intensity is lower for higher detunings from resonance. A gen-
eralized, nonlinear Schrödinger equation incorporating every effect dis-
cussed so far takes the following form:

∂ ∂
A + k eff A =
∂z ∂t
1 ∂2 1 ∂ 3 ∂ γeff |A|2
− i k A + k A + i 1 + is A. (7.18)
2 eff ∂t 2 6 eff ∂t 3 ∂t 1 + |A|2
|AS |

7.3.5 Nonlinear Frequency Mixing

The characteristics of frequency mixing processes such as harmonic

generation and four-wave mixing can also be enhanced via waveguide-
coupled resonators. As a general rule, the scaling of the enhancement
of these processes can be inferred by including contributions from each
intensity involved (lying within a resonance) and the interaction length.
Each contributes a factor proportional to the finesse.
Four-wave mixing is a third-order nonlinear process that annihilates
two photons at one frequency and generates two photons at a higher
and lower frequency. Four-wave mixing can give rise to modulation insta-
bility whereby a low contrast amplitude ripple grows via the amplifica-
tion of sidebands at the expense of the central frequency. In a dispersive
medium, four-wave mixing is stifled due to phase mismatch. However, in
an anomalous dispersive medium, positive self-phase modulation gener-
ates new frequency components that compensate for the mismatch [178].
If the process is allowed to continue, the modulation depth increases until
188 7. Distributed Microresonator Systems

Power, W

0 20 30 20 10 0
40 60 70 60 50 40
z (resonator #) Time, ps
Fig. 7.9. Demonstration of modulation instability in a SCISSOR. The input
field consists of 800 mW of CW power with a 1% power ripple. The SCIS-
SOR parameters have been chosen such that the peak of the instability
gain is at the input modulation frequency of 100 GHz. Note that the mod-
ulation frequency given by Eq. 7.19 need not be a resonance frequency
of the structure. (After Ref. [129], ©2002, Optical Society of America.)

a train of solitons stabilizes. Figure 7.9 shows the increase in modulation

depth for a seeded 1% amplitude ripple of 100 GHz with a propagation
distance for a sequence of 60 resonators [129]. The peak of the effective
instability gain gm = 2γeff |A0 |2 occurs at some modulation frequency
Ωm = ± 2γeff |A0 |2 / k eff  (7.19)

provided that this value does not exceed the resonance bandwidth. The
gain is enhanced by the square of the finesse. Up until this point, atten-
tion has been restricted to pulses whose bandwidth is of the order or less
than that of a single resonance peak. By copropagating pulses with car-
rier frequencies lying within differing resonance peaks, four-wave mixing
processes can be enhanced with frequency separations of pump and sig-
nal equal to an integer number of free spectral ranges (FSRs). Because
the efficiency of idler generation depends on the pump intensity, signal
intensity, and grows quadratically with length, the efficiency scales as the
fourth power of the finesse [56]. We expect such effects to be important
in systems that have low intrinsic dispersion such that the FSR is inde-
pendent of frequency so that the three enhancement linewidths coincide
with signal, pump, and idler frequencies. Finally, the efficiency of har-
monic generation processes may be increased. The efficiency of second
harmonic generation (SHG), for example, would be enhanced cubically
with the finesse.
7.4 Limited Depth of Phase 189

7.4 Limited Depth of Phase

Within a free spectral range, a single resonator can impart only a maxi-
mum phase depth of 2π radians. This limitation has important implica-
tions for the maximum delay, chirp, and a nonlinear phase that can be
imposed on a pulse per resonator. As the imparted phase nears only π /2
radians, higher order effects become increasingly significant such that
the system can no longer be treated perturbatively. The extent of group
velocity reduction that can be achieved in a SCISSOR is limited by how
high the finesse can be made. A SCISSOR with an ultra-high value of fi-
nesse can be used to slow a pulse appreciably but that pulse must be long
enough such that it is at least of the order of the finesse times the tran-
sit time of a single resonator. Thus, the maximum delay per resonator
is fixed and equal to one pulse width at best. The same is true for the
induced group velocity dispersion. A high GVD coefficient (proportional
to F 2 ) can be obtained by making the finesse very large. However, the in-
creasing finesse places an increasing restriction on the pulse bandwidth
∆ω (proportional to 1/F ). As a result, the imposed spectral chirp per
resonator 1/2k 2
eff ∆ω L is independent of finesse and only dependent on
the scaling factor B. If the requirement is to broaden a pulse by N pulse
widths, then the minimum number of resonators needed (occurring at
B ∼ 1) is roughly N. This is an important point: An ultra-high finesse is
not required for designing dispersive devices based on resonators. How-
ever, while reducing the resonator size and increasing the finesse in in-
verse proportion maintains the same resonator bandwidth and thus the
same linear properties, the nonlinear properties are enhanced. This is of
fundamental importance since a low threshold power and small number
of resonators is desirable practically. As a result of saturation, it is very
difficult to achieve an effective nonlinear phase shift of π radians from
a single resonator when operating on resonance. It is achieved only in
the limit as the resonator’s internal phase shift is power-detuned com-
pletely by π radians resulting in an external phase shift of π radians as
well. As a result of this saturation, one completely loses the advantage
of resonant enhancement. Achieving a phase shift of π /2 is, however,
h much easier to attain before the saturation takes place and requires a
power-detuning of only ∆φ = π /F . A effective nonlinear phase shift of
π may be obtained readily from a single resonator taking advantage of
enhancement by ensuring that the resonator is red-detuned initially by
π /F and allowing the resonator to be power-detuned though resonance
for a total value of π radians [227].
190 7. Distributed Microresonator Systems

7.5 Attenuation in Distributed Microresonators

Attenuation in microresonators is in general detrimental. Internal atten-

uation reduces the net transmission, buildup, and (in general) nonlinear
response. It also broadens the resonance limiting the achievable finesse.
Attenuation in microresonators typically arises from three mechanisms:
intrinsic absorption, radiation loss [133, 135], and scattering. Intrinsic
absorption can typically be rendered insignificant over millimeter-sized
propagation distances by choosing an appropriate material system at a
given wavelength. Additionally, since the circulating intensity can greatly
exceed the incident intensity, intensity-dependent absorption processes
such as two-photon absorption may be significant in resonators. Two-
photon absorption may be minimized (if desired) by proper selection of
a material with a bandgap that is greater than twice the incident photon
energy [189, 228]. Whispering-gallery modes of microdisks and micror-
ings suffer from bending or radiation loss that is increasingly important
for small resonators with low refractive index contrast. Scattering can
take place in the bulk or on the edges. Edge scattering is typically the
dominant loss mechanism that results from roughness on the microres-
onator edges that in practice cannot be made perfectly smooth [229]. The
surface perturbations phase-match the guided mode to radiating modes
outside the disk structure [120] or into the contra-directional mode [121].
In the planar waveguide approximation, the attenuation constant associ-
ated with scattering loss is derived as [98]


αscatt = σ 2 n21,eff − n22,eff k20 2
E(x=−d/2) 2
+ E(x=+d/2) . (7.20)

Where σ is the RMS edge roughness. Applying this result to a microdisk

or microring results in the expression [122]


αscatt = σ 2 n21,eff − n22,eff k20 2
E(r =R1 ) + E 2
(r =R2 ) . (7.21)

Because the strength of the scattering loss coefficient increases with

the square of the refractive index difference, it can be expected that
high-contrast microring and microdisk microresonators will possess high
scattering losses. Figure 7.10(a) shows a finite-difference time-domain
(FDTD) [112] simulation of a 5-resonator SCISSOR with 50-nm sidewall
roughness displaying strong scattering losses and weak circulating in-
tensity. In Fig. 7.10(b), a lower sidewall roughness of 30 nm results in
negligible scattering loss and strong intensity buildup.
Gain may be implemented where possible to combat attenuation. More
interestingly, a dispersion decreasing system may be used to propagate a
SCISSOR soliton in an attenuating structure. In this case, the pulse width
is kept constant as the amplitude decreases via an exponential decrease
7.6 Slow and Fast Light in SCISSORs 191

Fig. 7.10. FDTD method of solving Maxwell’s equations for a SCISSOR

structure composed of five microresonators. A TE field of wavelength
1.55 µm is launched into the 0.4-µm-wide waveguide evanescently side-
coupled to a disk with diameter of 5.1 µm. The refractive index of the
air-clad disk and guide is 2. Exclusive coupling to the m = 16 azimuthal
whispering-gallery mode (WGM) is achieved by careful selection of para-
meters. In (a), strong scattering losses result due to roughness associated
with a 50-nm grid. In (b), scattering losses are made negligible by using
a 30-nm grid. Consequently, a buildup factor of 16 and finesse of 25 are
achieved in this structure. (After Ref. [129], ©2002, Optical Society of

in dispersion down the length of the structure. In general, however, low-

loss propagation through a SCISSOR constructed from N resonators can
be ensured if the single-pass attenuation satisfies α2π R  1/NF .

7.6 Slow and Fast Light in SCISSORs

In recent years there has been a flurry of activity aimed at the develop-
ment of techniques that can lead to a significant modification of the group
velocity of propagation of a light pulse through a material medium [230].
Proposed applications of these procedures include the development of
optical delay lines [220] and the “storage” of light pulses [231, 232] with
possible applications in optical communications and quantum informa-
tion. Most of this research has made use of the response of resonant
media [233], and much of it has made use of the concept of electromag-
netically induced transparency [231, 234] and other quantum coherence
effects [235, 236].
192 7. Distributed Microresonator Systems

7.6.1 Slow Light

Many properties of the SCISSOR system are in fact analogous to those

found in atomic systems. In both cases, light is coupled into and out of
discrete resonators without loss or dispersion, but with delay. It is known
from studies of slow light propagation in atomic systems displaying elec-
tromagnetically induced transparency (EIT) that the width of the induced
transparency window in atomic systems√ with N interacting atoms is re-
duced by a factor that scales as 1/ N [237]. This fundamental limitation
on bandwidth results from the fact that near the frequency of maximum
transmission, the transmission decreases quadratically with detuning.
Such a limitation is absent in a fully transmissive SCISSOR geometry used
to propagate solitons, but a fundamental limitation is imposed ultimately
by fourth-order dispersion. This point is next examined in detail.
The effective group index associated with a SCISSOR takes on its max-
imum value when the optical wave is tuned to a cavity resonance (φ = 0),
and it can be expressed in any of the forms
2π R 1 + r 2π R 4R
ng = n 1 + =n 1+ B0 = n 1 + F (7.22)
L 1−r L L
Although the steep slope of the dispersion-relation curve near resonance
is responsible for the reduced group velocity, the transition from the flat
sections of the dispersion curve to the steep section is necessarily curved
and introduces GVD. On resonance, the lowest-order GVD parameter k eff
(and all other even orders) was shown previously to be zero. However,
the distance over which pulses can propagate is limited ultimately by
broadening induced by third-order dispersion. It is better to propagate
off resonance and sacrifice some enhancement in order to gain much
in terms of the maximum possible propagation √ distance. Specifically, by
tuning slightly above resonance (φ0 = π /F 3), third-order dispersion
can be eliminated. The lowest order dispersion that is introduced by op-
erating off resonance can be compensated for by the enhanced nonlin-
ear response of the structure. The negative, lowest order GVD occurring
at this operating point can be balanced precisely by the nonlinearity to
form a SCISSOR soliton. In the following paragraphs, this conjecture will
be proven.
An all-pass resonator is inherently a phase-only filter and possesses
a field amplitude transmission function that is a pure phasor that can be
expanded at resonance about φ0 = 0:
! "
d2 Φ d3 Φ
i Φ0 + dφ φ+ 12

iΦ φ2 + 16 φ3 +...
e =e dφ2 dφ3
. (7.23)

On resonance, the delay associated with the resonator is given by the

value of the first normalized frequency derivative term (TD = π F TR ).
Because no amplitude filtering function is associated with the operation
7.6 Slow and Fast Light in SCISSORs 193

of this device, the normalized bandwidth is determined by the phase error

induced by higher order dispersive terms. On resonance, the even-order
terms such as the group-delay dispersion vanish and the next nonzero
higher order limiting term is third-order dispersion. The dominant phase
error on resonance is thus
 1 d3 Φ 
 11 2 3
|Φerror | =  3
φ = F φ . (7.24)
 6 dφ 3 62 π 

Assuming that the phase error of 1 radian is the maximum tolerable error,

3 1 1
the usable bandwidth is restricted to ∆ν = 2 F TR ≈ F TR . A SCISSOR,
composed of a sequence of N all-pass resonators, possesses an effective
phase shift equal to the single device phase shift multiplied by N. It fol-
lows that the net accumulated delay for such a system is simply equal to
N times the single-device delay. Fortunately, the net accumulated phase
errors do not scale in the same fashion. That is, the phase-error limitation
on usable bandwidth (governed on resonance by third-order √ dispersion)
is not simply inversely proportional to N, but rather as ∆ν/ N. As dis-
cussed previously, soliton propagation at the maximally dispersive fre-
quency allows for the cancellation of group velocity dispersion and the
elimination of third-order dispersion. This leaves fourth-order dispersion
as the limiting phase error. The dominant phase error in this regime is
 1  d4 
Φ  1 27 3 2 

|Φerror | =    φ =4
F φ  4
 24  dφ 4  24 128 π 

4 1/8

2 1 1
with a similar bandwidth restriction of ∆ν = 35 F TR ≈ F TR for a
√ resonator. It is easily shown that a less restricting scaling law of
∆ν/ N applies to propagation in the SCISSOR soliton regime.
Next, the analysis derived in the previous paragraph is tested via rig-
orous simulations. Fig. 7.11 compares three approaches to attempting
to propagate slow light in a SCISSOR with a group velocity of approxi-
mately c/(100n). In Fig. 7.11(a), a weak 100-ps pulse tuned to resonance
is delayed greatly, but broadens and acquires ripples associated with neg-
ative third-order dispersion. In Fig. 7.11(b), the pulse frequency is tuned
above resonance to the extremum of the lowest order GVD. At this fre-
quency, the third-order GVD necessarily vanishes, and pulse distortion
of the sort shown in part (a) is decreased noticeably. However, the pulse
broadens considerably as a result of non-vanishing second-order (lowest
order) dispersion. In Fig. 7.11(c), the same pulse but with a peak power
corresponding to that of the fundamental SCISSOR soliton is observed to
propagate with a preserved pulse shape. The group velocity reduction in
this case is 75× as opposed to 100× in part (a), but the high fidelity of
pulse propagation makes this strategy seem to be superior.
194 7. Distributed Microresonator Systems

Fig. 7.11. Numerical results showing the advantage of using optical non-
linearity on the propagation of slow light through a SCISSOR structure.
The SCISSOR consists of 100, 10-µm diameter resonators spaced by
10π µm with r = 0.98 corresponding to a group velocity reduction of
100×. (a) A weak, 100-ps resonant pulse propagates at a group velocity of
(c/n)/100 through the SCISSOR and is corrupted by resonator induced
third-order dispersion. (b) The same pulse, but with carrier frequency
tuned to the anomalous GVD maximum propagates with a group velocity
of (c/n)/75 but is greatly broadened. (c) A 6.4-mW peak power, 100-ps
pulse tuned to the anomalous GVD maximum propagates as the funda-
mental SCISSOR soliton with a group velocity of (c/n)/75 and is well pre-
served. Here, parameters typical of a GaAs or chalcogenide-glass based
waveguiding structure, (γ/Aeff = 60m− 1W− 1) were employed.

A sequence of resonators might someday be useful in studying the

properties of slow light in the regime where acoustic and optical group
velocities are of the same order of magnitude [238, 239]. However, in
order to slow the group propagation to this level in silica, an ultra-high
finesse of about 105 is required.
7.6 Slow and Fast Light in SCISSORs 195

7.6.2 Tunable Optical Delay Lines

Because the group delay can in practice be controlled by detuning res-

onators thermally [152], electro-optically, by carrier injection [65], elec-
troabsorption [240], or other means, tunable optical delay lines may be
constructed from microresonators. The net group delay TD in a SCISSOR
increases linearly with the number of resonators,

TD = N . (7.26)

The net group delay is inversely proportional to the average group veloc-
ity that can be minimized by coupling the resonators in a manner that
yields the maximum possible finesse. Under this condition, however, the
bandwidth may be restricted severely. For example, reducing the group
velocity in a SCISSOR by six orders of magnitude results in a bandwidth of
less than 10 MHz for 10-µm-diameter resonators. As a result, designing
a SCISSOR for the minimum attainable group velocity (in practice limited
by attenuation) is not desirable. The fractional group delay, defined as
the group delay normalized to one pulse width TP , is a convenient figure
of merit characterizing a resonator-based delay line. A fractional group
delay near unity corresponds to a delay by one pulse width. This delay is
the greatest that can be achieved in a single resonator without the intro-
duction of higher order dispersive effects. These higher order dispersive
effects result from frequency content in the spectral wings experiencing
a lower delay. Fractional group delays greater than unity can be obtained
by use of more resonators. However, dispersive effects accumulate and
eventually severely distort pulses. For a desired net fractional delay, these
phase errors might be reduced by relaxing the fractional delay per res-
onator and by increasing the number of resonators. Alternatively, the
phase errors might be corrected by some means of dispersion compen-
sation, thereby increasing the system complexity. In principle, there is an
advantage to using nonlinear soliton propagation in a SCISSOR.
Figure 7.12 displays the results of a simulation of a tunable delay line
based on a SCISSOR. Eight bits of a 160-Gb/s pulse train are delayed by
one and four bit-slots. For simplicity, the delay associated with propaga-
tion in the side-coupling waveguide is removed. In (a), six resonators are
sufficient to delay the pulse-train by a single bit-slot while suffering minor
third-order phase distortion. In (b), a four bit-slot delay is achieved via the
use of 26 resonators, but the pulses emerge greatly dispersed. By tuning
the resonator and pulse energy to parameters near the SCISSOR soliton
for the structure, pulse fidelity can be improved as demonstrated in (c).
Note that a delay of four bit-slots corresponds to a fractional delay of 12
for the duty cycle of 1/3 employed. For comparison, the same pulse train
is sent through the SCISSORs with the resonators tuned far away from
resonance. The delay accumulated for both weak and strong pulses in
196 7. Distributed Microresonator Systems

Fig. 7.12. A microresonator-based tunable delay line. Shown are eight

bits of a 160-Gb/s (duty cycle = 1/3) pulse train. In (a), six resonators
achieve a delay of a single bit-slot with some minor distortion result-
ing from third-order dispersion. In (b) 26 resonators achieve a delay of
four bit-slots, although with increased distortion. In (c), propagation near
parameters associated with a SCISSOR soliton results in noticeably less
distortion. Also shown in (b) and (c) are the same pulse trains when the
resonator is detuned. In all cases, a finesse of 10π and 4 µm diameters
are used. The bit sequence is 10100111.

the off-resonance state, where pulses effectively bypass the resonators,

is negligible. This demonstrates the possibility for tunable delay. Most
importantly, this simulation demonstrates that a unit fractional delay
does not require an ultra-high finesse, but rather it is achieved when the
pulse bandwidth is of the order of the resonator bandwidth. In order that
sequences of resonators forming the SCISSOR overlap in their delay band-
widths, their effective optical circumferences must be made reproducibly
7.7 Generalized Periodic Resonator Systems 197

to within λ/F . If the reproducibility does not conform to this standard,

a Doppler-like broadening of the Lorentzian resonance linewidths will
result in lowered group delay and broader bandwidth.

7.6.3 Fast Light

Lossy resonators forming a SCISSOR structure can be implemented to

propagate light superluminally [241]. If the coupling strength is chosen
to be weaker than the round-trip loss, the resonator is said to be un-
dercoupled. The dispersion relation for the SCISSOR in this regime is
qualitatively different from that found in the overcoupled regime. Fig-
ure 7.13(a) displays the transmission for a single resonator in the three
regimes. In addition, Fig. 7.13(b) displays the resonator contribution to
the dispersion relation for all three regimes and displays the reversal
of sign near resonance. This negative slope implies that a pulse exiting
from a resonator will emerge with its center advanced in time with re-
spect to the incident pulse. Causality is maintained because the discrete
impulse response of the resonator does not display any advanced im-
pulses. A pulse displays superluminal propagation but quickly attenuates
in a multi-resonator undercoupled SCISSOR. Gain (assumed flat across
the pulse bandwidth) may be incorporated into the straight waveguide
to offset the losses associated with the undercoupled configuration. The
group index near resonance in the undercoupled regime is given by [242]

2π R a(1 − r 2 )
ng = n 1 − , (7.27)
L (r − a)(1 − r a)

where a = e−απ R . Figure 7.13(c) demonstrates superluminal propaga-

tion of a linear, resonant pulse through undercoupled resonators situated
near an amplifying waveguide. The exiting pulse experiences a negative
time delay or, equivalently, a time advance. Moreover, the theory also pre-
dicts superluminal propagation for a SCISSOR in which each resonator is
constructed of an amplifying medium in which the round-trip gain is
greater than the coupling strength7 ; however, simulations suggest that
propagation is highly unstable in this regime.

7.7 Generalized Periodic Resonator Systems

We now compare and contrast the geometry and propagation characteris-

tics of periodic SCISSORs with that of other well-known systems in optical
physics that possess photonic bandgaps. In certain respects, the SCISSOR
soliton is analogous to gap [243], Bragg [244], and discrete solitons [245],
This occurs in the equivalent undercoupled regime accompanying gain.
198 7. Distributed Microresonator Systems

Fig. 7.13. (a) The transmission for a single resonator with r = 0.9 in the
overcoupled (a = 0.96), critically coupled (a = 0.9), and undercoupled
(a = 0.84) regimes. (b) The dispersion relation for the same three cases.
Note the change in the sign of the slope of the curve near resonance in the
undercoupled case. In the critically coupled case, the curve undergoes a
π /L phase jump (where the transmission is zero) whereas in the overcou-
pled case, the second half of curve has been cut and displaced down by
2π /L for generality. The frequency units of δω correspond to the nonat-
tenuating resonance bandwidth. (c) Numerical simulation demonstrating
superluminal propagation of 36-ps pulses through a SCISSOR structure
composed of twenty lossy undercoupled resonators with 10-µm diame-
ter, spaced by 10π µm. Gain has been added to the straight waveguide
section to maintain pulse power.

which result from nonlinear pulse propagation within or near the pho-
tonic bandgap [217–219] of a distributed feedback structure. Many pho-
tonic crystal systems possess similar phase [246] and intensity enhancing
properties [247–250] but in general possess bandgaps. Because each con-
stituent resonator of a SCISSOR is an all-pass filter, feedback is present
7.7 Generalized Periodic Resonator Systems 199

within each resonator but not among resonators. Alternatively, there is

no intended mechanism for light to couple into the counterpropagating
modes of the microresonator or guide.8 As a result, there is no frequency
at which light is restricted from propagating and thus the structure can-
not possess a photonic bandgap (PBG). Nevertheless, the SCISSOR struc-
ture displays enhanced nonlinear optical response for much the same
reason that a PBG structure can produce enhanced nonlinearity.
We will thus take a step back and examine the relationships between
periodically distributed microresonator systems and photonic crystals.
We will accomplish this by comparing and contrasting the light propaga-
tion characteristics (dispersion relations) associated with different con-
nection geometries for sequences of microring resonators. We restrict
our attention to four archetypal building blocks from which sequences
can be constructed [252]. For each of these four microring-based units, a
direct analog in the form of a traditional Fabry–Perot cavity is depicted in
Fig. 7.14. With the insight provided by these Fabry–Perot analogues, it is
clear that these geometries identify the four ways of connecting four-port
resonators9 to form sequences.
A sequence based on Fig. 7.14(a), (cavity version) represents the tra-
ditional multilayered structure or Bragg grating. The microring-based
version of such a multilayer is a sequence of directly coupled micror-
ings [253] often termed a coupled resonator optical waveguide (CROW)
[254, 255]. Another way to couple the rings is indirectly through a com-
mon waveguide as depicted in Fig. 7.14(b). Such is the case in a SCISSOR.
The other two ways (c and d) are characterized by indirect coupling of
rings via two common waveguides. Note that these four structures are
identical in terms of their fundamental building block and differ only in
the manner in which they are interconnected. Of the cavity versions, only
geometry (a) forms a building block that is naturally sequenced as in a
multilayered structure. The other three geometries possess interchanged
ports that do not sequence naturally in their cavity embodiments and can
be constructed only with greater complexity. Microring resonators are,
however, more naturally suited to sequencing in any of these geometries
because the two input and two output ports occupy spatially distinct
channels. Furthermore, their planar nature is compatible with current
fabrication technologies.

Counterpropagating waves in a microring resonator can be coupled via either
surface roughness or a small perturbation, such as a notch. When side coupled
to a guide, such a notched resonator can approximate a general second-order
Chebyschev reflection response, because a single notched microring with one
guide is equivalent to two coupled rings between two guides [150, 251].
Strictly speaking, the first configuration is constructed from two-port res-
onators and is a special case.
200 7. Distributed Microresonator Systems

Ring version Cavity version

Bj Bj+1 Bj Bj+1
Aj Aj+1 Aj Aj+1

Aj Aj+1 Aj Aj+1

Bj Bj+1 Bj Bj+1
Aj Aj+1 Aj Aj+1

Bj Bj+1 Bj Bj+1
Aj Aj+1 Aj Aj+1

Fig. 7.14. Illustrations of four archetypal microring resonator unit cells

for (a) a CROW, (b) a single-channel SCISSOR, (c) a double-channel SCIS-
SOR, and (d) a twisted double-channel SCISSOR — along with their cavity
versions. These unit cells can serve as building blocks for constructing
sequences of resonators that serve as novel photonic guidance architec-

7.7.1 Bloch-Matrix Formalism

In previous sections, the frequency analysis of periodically sequenced all-

pass resonators was straightforward because of the serial feed-forward
nature of the transmission. That is, after the fields leave a resonator unit
of a SCISSOR, they never return to it and phase accumulation is sim-
ply cumulative. Two other archetypes do not share this property (CROW
and double-channel SCISSOR). Thus, a more generalized description of
periodically sequenced microresonator systems is required. We adopt a
matrix formalism used to analyze photonic bandgap structures based on
Bloch’s theorem with the assumption that the resonator sequences are
infinite and periodic. Using this formalism, the dispersion relations for
the four archetypal sequenced-resonator geometries are readily derived.
7.7 Generalized Periodic Resonator Systems 201

Table 7.1. Port-to-port relations for the building blocks in Fig. 7.14.
M21 M22
φ2 φ2
(φ2 +φ1 ) (φ2 −φ1 ) r ei 2 −e−i 2
−ei 2 +r 2 e−i 2 ∗ ∗
(a) t 2 t2 M12 M11
(1−r e−iφ )eiθ
(b) r −e−iφ
0 0 0
φ φ
(1−r1 r2 e−iφ )eiθ −t1 t2 ei 2 −t1 t2 e−i 2 (1−r1 r2 eiφ )e−iθ
(c) r1 −r2 e−iφ r2 −r1 eiφ r1 −r2 e−iφ r2 −r1 eiφ
φ φ
(r1 −r2 eiφ )eiθ −t1 t2 ei 2 eiθ −t1 t2 ei 2 eiθ (r2 −r1 eiφ )eiθ
(d) 1−r1 r2 eiφ 1−r1 r2 eiφ 1−r1 r2 eiφ 1−r1 r2 eiφ

We begin with the scattering matrix equations for a single coupler.

E3 = r E1 + itE2 ,
E4 = itE1 + r E2 .

Neglecting loss allows the simple relation r 2 + t 2 = 1 to hold for each

coupler. Using these relations for each coupling point as required, we can
calculate the matrix M that relates the fields at the two right ports of a
coupled resonator, Aj+1 and Bj+1 , to the fields at the two left ports Aj
and Bj . For a generic four-port optical device, this takes the form
Aj+1 M11 M12 Aj
= . (7.28)
Bj+1 M21 M22 Bj

Table 7.1 shows the components Mij that represent the port-to-port re-
lations associated with each of the building blocks in Fig. 7.14.
We are now ready to consider an infinite sequence of resonators, each
of the same building block and connected to its neighbors, with spatial
periodicity L. Bloch’s theorem states that we can find solutions for the
fields that at periodic intervals in the infinite lattice are simply related by
a phase factor:  
Aj+1 Aj
= eikeff L . (7.29)
Bj+1 Bj

That is, eikeff L must be an eigenvalue of the matrix formed from Mij ; for
this to hold, we require

M11 − eikeff L M12
det = 0. (7.30)
M21 M22 − eikeff L

The quadratic formula is applied to the expanded form of this equation to

obtain the dispersion relation (keff vs. ω) based on the matrix coefficients
that in general are frequency dependent:
202 7. Distributed Microresonator Systems

ei2keff L − (M11 + M22 ) eikeff L + (M11 M22 − M12 M21 ) = 0, (7.31)

⎡  1/2 ⎤
1 (M + M ) (M − M )
keff = arg ⎣ ⎦.
11 22 11 22
± + M12 M21 (7.32)
L 2 4

In the following sections, we apply this technique to derive dispersion

relations for the four archetypal infinite periodic lattices that result from
connecting the unit cells in Fig. 7.14.

7.7.2 Directly Coupled Resonators

A sequence of directly coupled resonators, or serial-cascaded rings [253–

255], with no auxiliary waveguides is depicted in Fig. 7.15. One interpre-
tation of this geometry results from considering every other resonator to
act as two auxiliary waveguides that connect the unit cells of Fig. 7.14(a)
together in sequence. Consequently, the optical properties are exactly
analogous to a multilayered structure.
To allow for a richness of greater complexity, we consider a struc-
ture with two length scales where the resonator circumference alternates
between 2π R1 and 2π R2 . In what follows we will restrict ourselves to
the case where light propagates clockwise in the larger resonators and
counterclockwise in the smaller resonators. Implementing Eq. 7.32 and
Table 7.1 entry (a), the dispersion relation for an infinite sequence is
given by
!!    "
−i −1 φ1 + φ2 r φ2 − φ1
keff = arg cos + cos ±
2(R1 + R2 ) t2 2 t2 2
 1/2 ⎤
−1 φ1 + φ2 r φ2 − φ1 2 ⎦ , (7.33)
cos + 2 cos −1
t2 2 t 2

where the two phase degrees of freedom φj = 2π Rj nω/c are propor-

tional to the two characteristic length scales. The two solutions repre-
sent the forward- and backward-going waves throughout the structure.
Due to distributed feedback, photonic bandgaps emerge where the Bragg
condition is satisfied (2π Rj = (m + 1/2)λ/n). The dispersion relation,
normalized group index ((c/n)dkeff /dω), and group velocity dispersion
(d2 keff /dω2 ) are plotted in Fig. 7.15 for a sequence of resonators at wave-
lengths near 1.55 µm. The shaded areas represent bandgaps while the
bands are labelled according to the nearest resonance order (m). The
structure may be used to support Bragg soliton propagation [244], par-
ticularly near the band edges where the group velocity dispersion is
strong [256] and the nonlinearity is enhanced [257].
7.7 Generalized Periodic Resonator Systems 203

R1 R2

Coupled-resonator optical waveguide

a) Low finesse
Wavelength (µm)

1.522 R1 32, R2 48 common band

R2 band 47

1.571 R1 band 31

R2 band 46
10 8 6 4 2 0 - Lπ π
-2L 0 π
L -10 -5 0 5 10
Norm. group index (ng/n) Bloch vector (keff) GVD (ps2/mm)
b) High finesse
Wavelength (µm)

1.522 R1 32, R2 48 common band

1.546 R2 band 47

R1 band 31
R2 band 46
50 40 30 20 10 0 - Lπ π
-2L 0 π
L -400 0 400
Norm. group index (ng/n) Bloch vector (keff) GVD (ps2/mm)

Fig. 7.15. The dispersion relation, normalized group index, and GVD
for (a) a low-finesse and (b) a high-finesse coupled resonator optical
waveguide. The dispersion relation is analogous to that of a multilayered
structure with alternating layer indices and/or thicknesses. Bandgaps
are always of the direct type and result from distributed Bragg reflec-
tion. Parameters include a refractive index of n = 3.1, alternating radii of
R1 = 2.5 µm and R2 = 1.5R1 . Resonances mR1 = 31 and 32 and mR2 = 46,
47 and 48 are shown. In (a), a high coupling strength, t 2 = 0.75, results
in narrow bandgaps, whereas in (b), a low coupling strength, t 2 = 0.1814,
results in wide bandgaps. These qualitative features are directly opposite
those found in the double-channel SCISSOR. (After Ref. [252], ©2004, Op-
tical Society of America.)
204 7. Distributed Microresonator Systems

7.7.3 Single-Channel SCISSORs

Next we revisit the single-channel SCISSOR [129, 224, 252] from the
perspective of the Bloch formalism. Because no mechanism for contra-
directional coupling is present, light of all frequencies is simply trans-
mitted in a feed-forward sequential manner from resonator to resonator
“pausing” for localized feedback at each. As a result, no photonic band-
gaps can exist in this geometry. The optical properties are in fact inde-
pendent of whether all the spacings between neighboring resonators are
the same; only the average density of resonators in a given length (or sim-
ply the total number) dictates the optical properties of the structure. For
an infinite periodic SCISSOR, the only surviving matrix coefficient from
Table 7.1 is M11 .
The dispersion relation (Eq. 7.32) takes the form
n 1 r − eiφ n Φ (ω)
keff = ω + arg = ω+ , (7.34)
c L 1 − re iφ c L

where φ = 2π Rnω/c, and Φ(ω) is the effective phase shift of a single

resonator. The dispersion relation, normalized group index, and GVD are
plotted in Fig. 7.16 for a series of 2.5-µm radius resonators at wavelengths
near 1.55 µm. To avoid confusion, the dispersion relation is shown only
for forward-going propagation defined by the arrows in the schematic.
For resonance frequencies, light experiences group delays of the order
of the cavity lifetime (2F /π )2π nR/c at each resonator. From a macro-
scopic point of view, the discrete delays may be considered to be distrib-
uted along the channel resulting in an effective group velocity that can
be slower than that in the common waveguide. On resonance, the group
index scales directly with finesse as ng = (1 + (4R/L)F ) n for r ≈ 1.
Due to the frequency-dependent nature of the effective group velocity
peaking at resonance, the lowest order dispersion vanishes on resonance
but can be very strongly normal (positive) or anomalous (negative) below
or above each resonance, respectively. Additionally, as we have shown
earlier, nonlinear processes such as self-phase modulation are enhanced
dramatically. Other coupled-cavity structures have been shown to en-
hance nonlinearities [258] and support soliton propagation. [245] The
SCISSOR structure is interesting in its ability to slow, disperse, and/or in-
tensify light pulses while maintaining a transmission spectrum free from
bandgaps — a feature that, to the best of our knowledge, does not bear
direct analogy with any traditional artificial medium. Fundamentally, this
results from disabling distributed feedback throughout the structure and
localizing it to the all-pass resonators. Recently, it has been shown that
photonic crystals can also be used to synthesize distributed all-pass res-
onant structures [259].
7.7 Generalized Periodic Resonator Systems 205

Single-channel SCISSOR
a) Low finesse
Wavelength (µm)




10 8 6 4 2 0 - Lπ π
-2L 0 π
L -10 -5 0 5 10
Norm. group index (ng/n) Bloch vector (keff) GVD (ps2/mm)
b) High finesse
Wavelength (µm)




50 40 30 20 10 0 - Lπ π
-2L 0 π
L -400 0 400
Norm. group index (ng/n) Bloch vector (keff) GVD (ps2/mm)

Fig. 7.16. The dispersion relation, normalized group index, and GVD for
(a) a low-finesse and (b) a high-finesse single-channel SCISSOR. Note that
the dispersion relation does not display photonic bandgaps. Neverthe-
less, at the resonances (λm = 2π nR/m), the group index (and the in-
tensity buildup) is maximized. Parameters include a refractive index of
n = 3.1, and a radius of R = 2.5 µm. Resonances mR = 31 and 32
at 1.571 µm and 1.522 µm are shown. In (a), a high coupling strength,
t 2 = 0.75, results in a wide bandwidth, whereas in (b), a low coupling
strength, t 2 = 0.1814, results in a narrow bandwidth. To avoid redun-
dancy, and because the forward and backward traveling waves do not cou-
ple, only the dispersion relation for the forward-traveling wave is shown.
(After Ref. [252], ©2004, Optical Society of America.)
206 7. Distributed Microresonator Systems

7.7.4 Double-Channel SCISSORs

The addition of a second common coupling channel to a SCISSOR as de-
picted in Fig. 7.17 enables distributed feedback. The unit cell for this
structure is an add–drop resonator oriented as in Fig. 7.14(c). In contrast
to the single-channel case, the spacing between resonators is now impor-
tant because resonances can develop not only within the resonators but
also among them. Photonic bandgaps emerge around each type of reso-
nance in the dispersion relation of an infinite periodic structure [260].
One type of bandgap emerges when the periodicity of the structure
satisfies the Bragg condition (2L = mB λ/n, with mB an integer), and an-
other type emerges when the circumference of the microresonator is an
integer multiple of the wavelength (2π R = mR λ/n, with mR an integer).
Compared with a multilayered structure or CROW, the output port con-
nections (drop and through) at each unit cell are reversed. This reversal
couples what were distributed Bragg reflections to the feed-forward di-
rection and diverts resonance frequencies to the retro-reflected direction.
In the case of a multilayered structure or CROW, high reflectivity can be
achieved by distributed feedback from many partially reflecting unit cells
and 100% transmission is always obtained at a resonant frequency. In con-
trast, in a double-channel SCISSOR, 100% reflection is achieved readily
at any of the resonator’s resonant frequencies irrespective of the cou-
pling parameter r , provided that the coupling coefficients are matched.
This result is attributed to the well-known phenomena of critical cou-
pling, whereby incident light is attenuated completely at the resonance
frequency due to complete destructive interference of the bypassed in-
coming field with the outcoupled circulating field. For this condition to
be met, the sum of the tap coupling due to the second channel (and losses
if present in the microring) must equal the cross-coupling from the ex-
citation channel. Implementing Eq. 7.32, the dispersion relation for the
double-channel SCISSOR structure takes the form

1 1 1 − r1 r2 e−iφ iθ 1 1 − r1 r2 eiφ −iθ
keff = arg e + e
L 2 r1 − r2 e−iφ 2 r2 − r1 eiφ
⎛ 2
⎝ 1 1 − r1 r2 e−iφ iθ 1 1 − r1 r2 eiφ −iθ
± e − e
2 r1 − r2 e−iφ 2 r2 − r1 eiφ

⎞1/2 ⎤
1 − r12 1 − r22
+  ⎠ ⎥ ⎦ , (7.35)
r2 − r1 eiφ r1 − r2 e−iφ

where the two phase degrees of freedom, φ = 2π Rnω/c and θ = Lnω/c,

are proportional to the two characteristic length scales. The two solu-
tions represent the bottom channel forward- and top channel backward-
going waves in the photonic structure. The dispersion relation, group
index, and GVD are plotted in Fig. 7.17 for 2.5-µm-radius resonators
7.7 Generalized Periodic Resonator Systems 207

Double-channel SCISSOR
a) Low finesse
Wavelength (µm)

1.522 Bragg gap 48 + resonator gap 32

Bragg gap 47

1.571 resonator gap 31

Bragg gap 46
10 8 6 4 2 0 - Lπ π
-2L 0 π
L -10 -5 0 5 10
Norm. group index (ng/n) Bloch vector (keff) GVD (ps2/mm)
b) High finesse
Wavelength (µm)

1.522 Bragg gap 48 + resonator gap 32

1.546 Bragg gap 47

1.571 res. gap 31

Bragg gap 46
50 40 30 20 10 0 - Lπ π
-2L 0 π
L -400 0 400
Norm. group index (ng/n) Bloch vector (keff) GVD (ps2/mm)

Fig. 7.17. The dispersion relation, normalized group index, and GVD
for (a) a low-finesse and (b) high-finesse double-channel SCISSOR. Note
that unlike the dispersion relation for the single-channel SCISSOR, the
double-channel variety displays photonic bandgaps. Two qualitatively
different bandgaps manifest themselves. For spectral components sat-
isfying the Bragg condition (λmB = 2nL/mB ), the bandgap is direct and
results from distributed Bragg reflection. At the resonances of the rings
(λmR = 2π nR/mR ), the bandgap is indirect and results from strong
resonator-mediated back-coupling. Parameters were chosen such that
one Bragg gap was coincident with one resonator gap within the figure:
refractive index n = 3.1, radii R = 2.5 µm, and spacing L = 1.5π R.
Resonator resonances mR = 31 and 32 and Bragg resonances mB = 46,
47 and 48 are shown. The coincident resonator (mR = 32) and Bragg
(mB = 48) resonance results in a wide direct gap. In (a) a high coupling
strength, t 2 = 0.75, results in wide bandgaps, whereas in (b) a low cou-
pling strength, t 2 = 0.1814, results in narrow bandgaps. (After Ref. [252],
©2004, Optical Society of America.)
208 7. Distributed Microresonator Systems

spaced by 1.5π R at wavelengths near 1.55 µm. Of the two types of gaps
mentioned above, those associated with the periodicity of the structure
(inter-resonator spacing) are called “Bragg gaps” and are always direct.
The indirect gaps are termed “resonator gaps” and are those associated
with the internal resonances of the resonators. A comparison of the qual-
itative features of the dispersion relation reveals that the bandgaps are
wider for low-finesse resonators. The interpretation is simple: In the high-
finesse case, the band over which the individual resonators are reflecting
is narrow whereas in the low-finesse case, it is wide. This directly carries
over to the widths of the bandgaps in the infinitely periodic structure
and is in stark contrast to the situation of a multilayered structure where
high reflectivity results in a wider bandgap. Due to the efficient excita-
tion of the resonators near stop-gaps, this structure is ideal for explor-
ing nonlinear effects [261] particularly within the bandgap, such as gap
solitons [243, 260].

7.7.5 Twisted Double-Channel SCISSORs

A “twist” on the double-channel SCISSOR that possesses qualitatively dif-

ferent optical properties is constructed by interchanging the top ports
(add and drop) of each unit cell, as in Fig. 7.14(d). For optical fiber ring
resonators, this configuration may be implemented by twisting each of
the resonators to form figure-8 loops. In an integrated geometry, it is eas-
ier to build two resonators that are nearly 100% coupled. This “twisted”
double-channel SCISSOR is depicted in Fig. 7.18. As a result of the port
interchange, there is no longer any mechanism for contra-directional cou-
pling, but codirectional coupling across the channels is now mediated
by the resonators with localized feedback enhancement. Thus, light in-
jected into one of the ports only couples to either of the channels in the
forward-going direction. The structure, therefore, behaves as a resonator-
enhanced directional coupler. In Table 7.1, the matrix coefficients are
given taking the coefficients characterizing the couplings between the
waveguides and resonators to be r1 and r2 , and assuming that the twisted
structure is implemented by use of two resonators that are 100% coupled
to each other. For an infinite periodic structure, the resulting dispersion
relation (Eq. 7.32) is

r1 −r2 eiφ r2 −r1 eiφ
1 ⎢ 1−r1 r2 e iφ + 1−r1 r2 e iφ
keff = arg ⎣eiθ +
L 2


⎞1/2 ⎤
⎜ r1 − r2 eiφ − r2 − r1 eiφ + 4 1 − r12 1 − r22 eiφ ⎟ ⎥
±⎝  ⎠ ⎥,
2 ⎦
4 1 − r1 r2 eiφ
7.7 Generalized Periodic Resonator Systems 209

Twisted double-channel SCISSOR

a) Low finesse
Wavelength (µm)




10 8 6 4 2 0 - Lπ π
-2L 0 π
L -10 -5 0 5 10
Norm. group index (ng/n) Bloch vector (keff) GVD (ps2/mm)
b) High finesse
Wavelength (µm)




50 40 30 20 10 0 - Lπ π
-2L 0 π
L -400 0 400
Norm. group index (ng/n) Bloch vector (keff) GVD (ps2/mm)

Fig. 7.18. The dispersion relation, normalized group index, and GVD for
(a) a low-finesse and (b) a high-finesse twisted double-channel SCISSOR.
Bandgaps are absent in the dispersion relation, which resembles that of
the single-channel SCISSOR, but with the presence of a second branch.
The two branches correspond to the two decoupled forward-traveling
normal modes. Near the microring resonances, (λm = 2π nR/m) the two
branches are strongly coupled, as in the case of a directional coupler.
Parameters used are the same as in Fig. 7.17 except that there are two
resonators each half in circumference and 100% coupled. Resonances
mR = 31 and 32 at 1.571 µm and 1.522 µm are shown. In (a), a high cou-
pling strength, t 2 = 0.75, results in wide-bandwidth channel-to-channel
coupling whereas in (b), a low coupling strength, t 2 = 0.1814, results
in narrow-bandwidth channel-to-channel coupling. To avoid redundancy,
and because the two forward- and two backward-traveling waves do not
couple, only the two forward-traveling dispersion relation branches are
shown. (After Ref. [252], ©2004, Optical Society of America.)
210 7. Distributed Microresonator Systems

where φ = 2π Rnω/c and θ = Lnω/c. The solutions represent the two

coupled forward-going waves in the photonic structure. The dispersion
relation, normalized group index, and GVD are plotted in Fig. 7.18 for
the symmetric case of r1 = r2 . Two branches are present, each one cor-
responding to one of two normal modes of the structure.

7.7.6 Bandgap Engineering in Distributed Feedback Structures

Fig. 7.19 qualitatively compares the transmission properties of unit struc-
tures and infinite periodic structures for a double-channel SCISSOR and
a multilayered structure. Although a multilayered structure or CROW al-
lows propagation on resonance and takes advantage of maximum coher-
ent buildup of intensity within the resonators, propagation is not allowed
at resonances of the double-channel SCISSOR around which bandgaps
We next compare the qualitative features of the transmission for a
finite double-channel SCISSOR and CROW with the corresponding infi-

a) Fabry- b) Multi-layer c) Add-drop d) Coupled- e) Add-drop f) Double-

Perot Stack resonator resonator resonator channel
optical (re-oriented) SCISSOR




Fig. 7.19. A qualitative comparison of the transmission properties of

three structures possessing bandgaps: (a) a Fabry–Perot, (b) a multi-
layered stack, (c) an add–drop resonator, (d) a CROW, (e) an add–drop
resonator (reoriented), and (f) a double-channel SCISSOR. Note that the
qualitative features of the transmission peaks and valleys are equiva-
lent for multilayered stacks and CROWs but reversed for double-channel
SCISSORs. With increasing finesse, the bandgap widths increase in both
multilayered stacks and CROWs while they decrease in double-channel
SCISSORs. (After Ref. [252], ©2004, Optical Society of America.)
7.7 Generalized Periodic Resonator Systems 211

nite periodic structures. The finite structures are termed parallel and
serial resonator coupled structures respectively [22, 55, 63, 165, 262]. Fig-
ure 7.20 displays the transmission spectra for low-finesse one, five, and
infinite unit-celled structures. Figure 7.21 displays the transmission spec-
tra for equivalent high-finesse structures. The displayed plots show that
the dispersion relation can provide a heuristic guide to the location and
width of the transmission dips in a finite structure. It is evident that
the infinite double-channel SCISSOR and CROW possess complementary
transmission characteristics; that is, band and bandgap locations are in-
terchanged. Typically, waveguide or fiber Bragg gratings typically possess
low reflectivity (and thus low finesse) per unit cell and thus display small

Fig. 7.20. Transmission spectra for a low-finesse (t 2 = 0.75) finite

double-channel SCISSOR and finite CROW. Parameters are the same as
in Figs. 7.15 and 7.17. Here, 5-unit cells are used to approximate the
structures. For comparison, the transmission for a single resonator is
shown in each case and shaded regions correspond to one-dimensional
photonic bandgaps in the corresponding infinite structure. The labels B
and R correspond to the double-channel SCISSOR’s Bragg and resonator
gaps, whereas the labels R1 and R2 correspond to the alternating CROW
resonances. (After Ref. [252], ©2004, Optical Society of America.)
212 7. Distributed Microresonator Systems

Fig. 7.21. Transmission spectra for a high-finesse (t 2 = 0.1814) finite

double-channel SCISSOR and finite 1D-CROW. Parameters are the same
as in Figs. 7.15 and 7.17. Here, 5-unit cells are used to approximate the
structures. For comparison, the transmission for a single resonator is
shown in each case and shaded regions correspond to one-dimensional
photonic bandgaps in the corresponding infinite structure. The labels B
and R correspond to the double-channel SCISSOR’s Bragg and resonator
gaps, whereas the labels R1 and R2 correspond to the alternating CROW
resonances. (After Ref. [252], ©2004, Optical Society of America.)

ripple in the passbands. Here, the deep ripples in the transmission bands
result from the abruptly terminated ends of the structure as encountered
in unapodized high-reflectivity multilayered stacks. First, note that the
ripple depth across the passbands is controlled by the transmission of a
single resonator and results from the “splitting” of individual interacting
resonances. Second, note that in going from low to high finesse, the width
of the single resonator transmission dips (and peaks) shrink faster than
the corresponding gaps (and bands) in the infinite structure. Analysis of
the bandwidths show that although the single resonator dip (and peak)
bandwidth scales as F −1 ≈ t 2 /π ,the widths of infinite resonator gaps
(and bands) scale as F∞ ≈ t/2 ∝ F −1 . This explains why the passband
7.7 Generalized Periodic Resonator Systems 213

ripples are of similar depth in the low-finesse cases whereas are strong
only in the high-finesse case for the CROW.

7.7.7 Slow Light, Group Velocity Dispersion, and Nonlinearities

The dispersion relations associated with the microresonator-based pho-

tonic structures considered here reveal greatly reduced group velocities
(vg = dω/dkeff ) near structural resonances [224, 263]. The reduction of
group velocity may be understood intuitively by noting that near a struc-
tural resonance light spends extra time circulating localized within the
resonators or in propagating back and forth between distributed compo-
nents of the periodic structure. There are several criteria by which one
can compare different optical mechanisms that induce slow light effects.
Furthermore, different applications demand different criteria. For some
applications, the slowest attainable group velocity may be desirable. Many
optical systems have demonstrated such slow light capability [230] but
typically only over bandwidths that are too narrow for practical use in
optical communications or logic. For such applications, a large fractional
group delay (L/τ)dkeff /dω is a better measure. This dimensionless quan-
tity is interpreted as the number of pulse widths τ by which a pulse
or train of pulses can be delayed, while maintaining the integrity of the
pulses. Many instances of slow light phenomena, although impressive in
their ability to slow the speed of pulse propagation to a small fraction
of c, would fail to delay an incoming pulse by more than a fraction of
a pulse width without seriously distorting it, simply because of a large
accompanying group velocity dispersion or narrow spectral window.
Microresonator-based structures have been proposed [220] for use
as optical delay lines because, when designed properly, they can ex-
hibit a large fractional delay. Additionally, because the group velocity
can be made tunable by shifting resonances thermally [264] or electri-
cally [65, 77], controllable optical delay lines can be constructed. The
shortest pulse for which the group delay is meaningful, and thus that
for which the fractional delay is largest, is one with a pulse width that
matches the cavity lifetime. Shorter pulses become distorted due to high-
order dispersive effects, whereas longer pulses that experience the same
physical delay exhibit a lower fractional delay.
These considerations remind us that a vanishingly small group ve-
locity is often not the sole important feature. An optical delay line
or buffer based on microresonator structures will in general be useful
only if slow group velocity is accompanied by high-transmission and
low pulse-distorting dispersion. Near resonances of a resonances of a
single-channel SCISSOR, the fractional group delay can approach unity.
In distributed feedback structures, the edges of two bandgaps can be en-
gineered so that they are close in frequency but not overlapping resulting
in a flat, defect-like band generated between them. Khurgin [265] has
214 7. Distributed Microresonator Systems

proposed a combination of distributed microresonator-based structures

for improved performance.
Alternatively, the strong group velocity dispersion encountered at
certain frequencies can be put to use for the construction of photonic
devices that perform dispersion compensation in an ultra-compact geom-
etry. Each of the archetypal structures in this can display group velocity
dispersion values that greatly exceed 20 ps2 per millimeter. These are at
least six orders of magnitude higher than what is conventionally mea-
sured in typical single-mode silica fiber (20 ps2 per kilometer). The situa-
tion is similar to that encountered in photonic crystal defect guides that
exhibit ultra-strong dispersive effects [246]. Just as an intrinsically high
group velocity reduction is not as important in many applications as a
large fractional group delay, many photonics applications require strong
fractional group delay dispersion irrespective of the actual group veloc-
ity dispersion. Microresonator-based structures have been implemented
as dispersion compensators [266] because, when designed properly, can
exhibit a fractional group delay dispersion near unity (of either sign) per
Coupled microresonator structures have been shown to possess prop-
agation characteristics associated with extended nonlinear Schrödinger
equations (NLSEs). The phenomena of slow light, light trapping [267], soli-
ton propagation [224], soliton compression [129], soliton switching [268],
gap solitons [260], self-steepening, modulation instability, and four-wave
mixing are all directly transferrable to the much more compact scale af-
forded by a microresonator-based photonic structures.
The enhancement of nonlinear effects in microresonators is attributed
to two different, although related, effects. First, because the group veloc-
ity is reduced in such a way that light circulates through a longer path
length within a photonic structure, the interaction length (or time) for a
nonlinear process is increased. This is directly related to the well-known
enhancement of phase sensitivity in ring resonators [151] and photonic
crystal-based [247, 269] structures. Second, because the light circulating
within the resonators is built-up coherently to a higher intensity than that
initially injected into the structure, stronger nonlinear effects are possi-
ble. As a result of these two effects, nonlinear processes like self-phase
modulation are enhanced in proportion to the square of the resonator
finesse [151]. This scaling law holds for single resonators and localized
feedback structures. For example, in the case of the single-channel SCIS-
SOR, the phase sensitivity and buildup are coincident and equal, and the
enhancement of accumulated nonlinear phase shifts is directly propor-
tional to the square of the group index, (c/vg )2 . This general result is
found in many nonlinear photonic systems displaying slow light effects
resulting from structural resonances.
7.8 Summary 215

7.8 Summary

Passive, nonlinear resonators are still a relatively widely untapped area of

research. Currently, many single microresonator systems with excellent
optical properties have been constructed. In many of these cases, ex-
tending the fabrication techniques to construct long sequences of such
devices to yield large-scale integration of photonic devices [207] is achiev-
able. The aim of this chapter was to describe some of the linear and
nonlinear propagation characteristics of sequences of microresonators.
Figure 7.22 summarizes the detuning dependence of the linear and non-
linear propagation parameters for a single-channel SCISSOR. The appli-
cation of thermal or electrical fields to the resonators makes it possible
to control the detuning and/or coupling coefficients. We envision that
such structures could be used as artificial media to study and apply
NLSE pulse propagation effects on an integrated chip where the prop-
agation parameters may be chosen and/or modified in realtime. Other
applications might include a testbed for studies of slow-light phenom-
ena, variable optical delay lines [220], clean pulse compression on a
chip without pedestal formation via the soliton decay mechanism, and
soliton-based optical switching and routing with low energy pulses. Al-
though all of these concepts have been implemented in various geome-
tries and material systems, the SCISSOR system has the potential for pro-
viding a highly compact, integrated optical platform for such phenom-
ena. Furthermore, it has the potential of creating familiar phenomena in

0.5 (d)
keff (2F/π)2 γ 2πR/L
keff γeff
(c) (e)
(2FT/π)3/L (2F/π) γ 2πRT/L
-4π -2π 0 2π 4π
Normalized detuning, φ0
Fig. 7.22. Functional dependence of (a) the group velocity reduction, (b)
group velocity dispersion, (c) third-order dispersion, (d) self-phase mod-
ulation coefficient, and (e) self-steepening coefficient on the normalized
detuning φ for a SCISSOR. The parameters have been scaled such that the
curves are universal and fit within the same plot limits. (After Ref. [129],
©2002, Optical Society of America.)
216 7. Distributed Microresonator Systems

regions of parameter space that typically do not manifest it such as the

case of self-steepening for pulses greater than picoseconds. As manufac-
turing techniques continue to improve, guiding structures based on dis-
tributed microresonators will become important photonic architectures.
8. Fabrication Techniques for Microresonators

The theories presented in this book have been around for a very long
time. Only recently, due to advances in fabrication technology, have the
realization of microresonators become a possibility. Indeed their great
potential is tempered only by the challenges associated with their con-
struction. In this chapter, we present the key fabrication issues involved
with microresonators. Due to the authors’ expertise with III–V semicon-
ductor fabrication techniques, detailed process flows are presented for
AlGaAs- and InP-based material systems.

8.1 Materials
Material systems being researched for microresonator fabrication include
various glasses and polymers (on a silica-on-silicon substrate), silicon-
on-silica (using readily available silicon-on-insulator wafers), gallium
arsenide, and indium phosphide.
The ideal material, is of course, one based on silicon, since that can
leverage the considerable knowledge accumulated by the electronics in-
dustry in the mass-production of silicon microprocessors. The drawback
of silicon is that it is difficult to make optically active, and thus has not
at the time of this writing been used for monolithic optical integration
(hybrid integration, as demonstrated by the researchers at UCSB, is al-
ways a possibility). III–V semiconductors are optically active, and there-
fore suitable for monolithic integration, but present their own problems.
The processing of III–V semiconductors is an expensive proposition, be-
cause the raw material is expensive, fragile, and requires much more
careful processing than glasses and polymers. Also, III–V semiconduc-
tor wafers are still not readily available in 6” diameters (the first wafers
were announced by Showa Denko in April 2002), making mass production
difficult. The electronics industry, in contrast, has moved to 12” silicon
wafers. This means that equipment has to be custom designed for small
wafer diameters; standard commercial equipment cannot be used readily.
Finally, single-mode waveguides made from both silicon and III–V
semiconductors tend to have extremely small cross sections (usually less
than 1 µm2 ). Together with the high effective indices of such waveguides,
218 8. Fabrication Techniques for Microresonators

this makes the problem of coupling to off-chip waveguides (fibers) ex-

tremely difficult. Although the problem of fabrication costs (for III–Vs)
may be beyond a solution currently, several groups are achieving vari-
ous degrees of success at tackling the problem of coupling between op-
tical fibers and high-effective-index waveguides, notably the work from
Michal’s group at Cornell.
Currently, the most successful effort toward the commercial real-
ization of microring resonator-based filters (by the now-Infinera-owned
Little Optics) relies on a glass-based system. These materials offer an
excellent solution to the problem of costs (the raw material — usually
silica-coated-silicon substrates — is inexpensive), the equipement needed
is essentially off-the-shelf equipment used by the semiconductor micro-
electronics industry, the waveguides themselves are made with glass, and
there is a vast body of knowledge from the optical fiber industry on how
to get glasses with low optical losses. The waveguide stack can be de-
posited easily in a chemical-vapor-deposition system, and processing can
be done with dielectric etching chemistries similar to those used by the
semiconductor industry.
All this said, III–V semiconductors remain of great importance because
some of the most innovative new devices have been demonstrated in
this material system. These devices span the breadth of devices needed
to realize monolithically integrated all-optical circuits, such as filters,
routers, amplifiers, lasers, and logic gates. Glass-based systems may yet
prove as versatile if not more so; doped glasses may provide high-optical-
nonlinearities needed for photonic logic, and erbium doping may help in
the realization of amplifiers.
One aspect independent of material system is the geometry of the
device. Since we require the resonator to be coupled to one or more bus
waveguides, these waveguides must either be fused with, or placed in
close proximity to, the resonator. The bus guides may be placed in-plane
with the resonator (lateral coupling) or out-of-plane (vertical coupling).
Future photonic circuits will almost certainly utilize a combination of
the two approaches.

8.2 III–V Semiconductors for Active and Passive Microrings

III–V semiconductors have high refractive indices (>3) compared with
competing material systems such as glass/silica (∼1.5), silicon oxynitride
(1.45 – 2.1), and LiNbO3 (∼2.2). This allows the fabrication of very high
(lateral) index contrast waveguides that allows the bend radius to be as
small as 1 µm before bending losses cause unacceptable deterioration in
Lattice matched III–V ternaries and quaternaries can be grown on both
indium phosphide and gallium arsenide with a wide variety of bandgaps
8.3 Growing a Waveguide Stack 219

and refractive indices [270–275]. Using an appropriate choice of materi-

als, we can make slab waveguides in both materials; patterning shallow-
etched ridges produces loosely confined waveguides, and deeply etched
pedestals give us strongly confined waveguides. Since III–V semiconduc-
tors have been the subject of studies since the 1960s, a vast body of
knowledge exists on fabrication technologies such as chemo-mechanical
polishing [276] and selective wet-chemical etching [277–279].
Most III–V compound semiconductors have a direct bandgap. There-
fore, the waveguide core and/or cladding can be active, allowing us to
make optically active semiconductor devices. Modern optical communi-
cation systems operate around 1550 nm because silica optic fibers have
the lowest loss around 1550 nm. InP-based materials can be used to make
amplifiers, lasers, and related devices, making it the system of choice for
active integrated semiconductor optical devices for the C band.
So III–V semiconductors — InP and its quaternaries, and GaAs and its
ternaries — are ideal for the fabrication of microring resonators if other
fabrication requirements (smooth, vertical, deep, anisotropic etching, and
lithography) are met. Recent demonstrations of devices in this material
system at LPS have been driven by the development of technologies for
in-house wafer growth by solid-source molecular beam epitaxy [280,281];
smooth, vertical dry-etching of both GaAs and InP [282–286]; and high-
resolution electron-beam lithography (via a collaboration with the Cornell
NanoScale Facility, using the Leica VB6 HR).
Commercial acceptance of III–V semiconductors for passive filters is
hampered currently by the high materials cost, high equipment costs, and
high optical losses in the material when compared with glass. However,
III–V semiconductors are important for the development of ultra-compact
monolithically integrated all-optical systems because they are optically
active, so this material system is an important target for research and

8.3 Growing a Waveguide Stack

8.3.1 Polymers

Typically, this refers to waveguides with polymer cores or refractive in-

dex ∼1.5 or more. This requirement is placed on the material because the
refractive index of silica (which usually forms the lower cladding), is typ-
ically 1.45—1.51 depending on the method used (thermal/wet growth,
or CVD). Given a silica-coated silicon wafer, the waveguide stack can be
grown quite simply by spinning-on the appropriate polymer layers, fol-
lowed by a cure step. The cure step is usually a high-temperature bake to
drive off solvents and stress that may affect the polarization dependence
of the waveguides, but this can usually be compensated for in the design.
220 8. Fabrication Techniques for Microresonators

8.3.2 Glass

Glass waveguide stacks usually are grown with chemical vapor deposi-
tion, with silica forming the lower cladding. As with polymers, residual
stresses from high temperature steps can result in stresses that affect
the polarization dependence. Low-temperature deposition is possible to
eliminate such stresses. Another problem arises from losses due to the
vibration modes of OH ions. This can be eliminated by the use of deuter-
ated silane [287].

8.3.3 III–V Semiconductors (InP and GaAs)

These are grown with metal-organic chemical vapor deposition (MOCVD),

or molecular-beam epitaxy (MBE). These methods are significantly more
costly than those used for polymers and glasses. Some of the raw ma-
terials needed (arsine, phosphine, phosphorous, arsenic, etc.) are very
dangerous (inhalation hazard, flammability, etc.) and require careful han-
dling and sophisticated safety equipment.
The waveguide core/upper and lower cladding/substrate in these sys-
tems are GaAs/AlGaAs/GaAs, AlGaAs/AlGaAs/GaAs, InP/GaInAsP/InP,
and GaInAsP/GaInAsP/InP, where the compositions of the ternaries and
quaternaries are suitably adjusted for low/high index depending on their
use in the core/cladding. The in-plane cladding is usually air, silica, spin-
on-glass, polyimide, or other low-index polymer to allow for tight con-

8.3.4 Silicon Oxynitride

Silica, silicon oxynitride (SiON), and silicon nitride (SiN) are typically de-
posited with CVD (Low-Pressure CVD, Plasma-Enhanced CVD, etc.). By
varying the oxygen content of SiON, it is possible to get a continuous
variation in refractive index from around 1.45 (silica) to 2 (silicon ni-
tride). So it makes for an attractive material system from the point of
view of design. (Note: we have deliberately omitted Si:O:N ratios, since
those depend on the deposition parameters.)

8.4 Feature Definition

By feature definition we mean the process of forming the transverse

waveguiding structure from the planar waveguide stack. This process
will typically consist of a lithography step followed by an etch step.
8.4 Feature Definition 221

8.4.1 Polymers, Glass, SiON, SiN

1. Mask definition (Lithography): For these material systems, the mask

definition is achieved by optical contact and/or projection lithogra-
phy. Projection lithography has the advantage of greater throughput
— since the mask is not in contact with the substrate, it does not need
cleaning after each lithography step. Projection lithography also pro-
vides smoother waveguide sidewalls along curves for systems where
the sidewall roughness is determined by the mask resolution, since
the discretization on the mask is reduced by the magnification of the
projection system. The mask itself could be the photoresist used in
the lithography, or a hard mask of silica, aluminum, etc., defined by
lift-off or dry-etching.
2. Etching: These material systems are dry-etched using one or all of
sulfur hexafluoride, Cx Hy Fz (fluorinated organic gases like freon,
tetrafluoromethane, etc.), hydrogen, oxygen, nitrogen, and argon.
Anisotropic dry-etching is used to obtain vertical sidewalls required
to prevent polarization mixing. Such etching may be performed in
a parallel-plate reactive-ion-etching (RIE) chamber, inductively cou-
pled plama RIE (ICP-RIE), electron-cyclotron resonance RIE (ECR), etc.
Anisotropy is assured by the use of low-pressure (usually <10−6 T)
and high substrate bias. When organic chemistries are used, vertical-
ity is assisted by the formation of polymers from the plasma, passi-
vating the sidewalls and preventing additional etching. Nitrogen and
argon are usually unreactive and are used for sputtering material
off the surface (similar to ion-milling), and they stabilize the plasma
and/or as a dilutant to regulate the reaction rate.

8.4.2 III–V Semiconductors (InP and GaAs)

1. Mask definition: In cases where small gaps (usually around 0.1 µm

for lateral coupling) are needed, the mask is defined by electron-beam
lithography. Although state-of-the-art lithography systems can pro-
vide such resolutions, most research institutions use e-beam litho-
graphy instead; even 20-year-old e-beam lithography machines can
provide such resolution. Elsewhere (vertical coupling), i-line or g-line
optical lithography is used. The mask itself could be the e-beam resist
or photoresist used in the lithography, or a hard mask of silica, SiN,
titanium, chromium, nickel, etc., or a suitable combination of them,
defined by lift-off or dry-etching.
2. Etching: Both GaAs and InP-based materials can be etched with resist
masks if halogenated chemistries are used (poly(methyl methacry-
late) is usually used for e-beam lithography; photoresists are too
numerous to list here). InP-based materials that do not contain alu-
minum also can be etched with a combination of methane and
222 8. Fabrication Techniques for Microresonators

hydrogen (with or without argon). In this case, anisotropy is assured

by the passivation of the sidewalls with a polymer formed from the
plasma. Verticality is obtained by a cleaning step used to remove the
bulk of the polymer buildup from the sidewalls periodically, so that
passivation is maintained without compromising the sidewall angle.
The cleaning step involves an oxygen plasma, which affects resist
masks, so methane-based chemistries require the use of hard masks
not affected by oxygen plasmas. For this reason, masks of chromium,
nickel, titanium, silica, SiN, or combinations of these materials, are
used. Such masks may also be used in dry-etching with halogenated
chemistries where the etch depth is needed is large enough that ero-
sion of the resist mask in the plasma becomes a concern. For details,
see [286] and the references therein.
Detailed examinations of various patterning methods are beyond the
scope of this volume. For that, we refer the reader to any of the excellent
device processing books on the market, such as the ones by Wolf and
Tauber, Campbell, and Pearton.

8.5 Multilayer Processing

Multilayer processing is specific to the case of vertically coupled mi-

croresonators. Stacking multiple waveguides needs one or of:
1. Epilayer transfer using polymer-assisted, or direct wafer bonding fol-
lowed by growth-substrate removal [172], or
2. Planarization and regrowth/redeposition [21].
Each method has its advantages (and disadvantages). Polymer wafer
bonding is messy and introduces a material of different coefficient of
thermal expansion and different chemical stability from the substrates.
The different chemical stability can limit the number of high-temperature
processing steps that can follow a polymer-assisted epilayer transfer. Fur-
thermore, if volatile by-products are produced during the bonding/curing
process, they can move the two substrates apart and cause angular mis-
alignment, making later cleaving difficult. In the case where active devices
are being made, the use of polymer wafer bonding will typically limit con-
tacts to the same side (top), since polymer dielectric films used for bond-
ing are poor conductors of electricity as they serve the added purpose of
forming the side and top/bottom cladding. For a comprehensive review,
see Niklaus et al. [166].
In the case of direct wafer bonding, both the transfer substrate and the
epilayer have to be very clean and free of any interfacial impurities that
can cause the formation of voids. Furthermore, the presence of particles
or a surface with undulations can destroy waveguides (or other desirable
features) on the epilayer being transferred (Fig. 8.1). When successful,
8.5 Multilayer Processing 223

Fig. 8.1. (a) Raburn’s fixture for wafer bonding, made from graphite. The
solid black portions are the fasteners used to apply pressure between
the two substrates. (b) A successfully fabricated double-bonded device.
(c) Bonded interface. (Continued on next page.)
224 8. Fabrication Techniques for Microresonators

Fig. 8.1. (Continued from previous page.) (d) A crushed waveguide.

(Adapted from Ref. [289], copyright Maura Raburn, 2003, used with per-

this method affords an excellent way of making both active and passive
devices, and do not limit the number of waveguiding layers to two, as
is the case with polymer wafer bonding. In fact, Raburn et al. have used
this method to bond three waveguiding layers [172, 288, 289]. Raburn
also discusses the various factors that influence successful direct wafer
bonding, and achieves bond-area fractions as high as 95% in [289].
Planarization and regrowth/redeposition is another technique where
an arbitrary number of waveguiding layers can be used. Thus far, most
reports have used two layers, but there is no impediment to three or more
(other than, perhaps, a lack of need with present designs). This method
usually limits the side-cladding to materials in the same family, such
as III–Vs or glasses. It is possible to achieve planarization around high-
index glasses/SiON/polymers with low-index glasses/silica/polymers, so
that even microring and microdisk resonators can be buried, as demon-
strated by Little, Laine, and Haus [143]. However doing the same with
III–V semiconductors is more difficult. Not only does regrowth of semi-
conductors place stringent demands on surface purity, the index contrast
available between the core and side-cladding (planarization material), is
typically quite small (usually <0.3). This means that it is difficult to make
buried microring and microdisk resonators, as the bending losses will
8.6 Laterally Coupled III–V Passive Microresonators 225

be debilitatingly high. To get around this, Choi et al. [64] use a buried
bus waveguide, and an air-clad microring/disk resonator. However, this
results in a new problem — the microring and bus waveguide are no
longer well matched, so it is difficult to get good coupling between the

8.6 Laterally Coupled III–V Passive Microresonators

InP and GaAs-based laterally coupled microresonators have been demon-

strated by numerous groups, notably Rafizadeh et al. [51], Absil et al. [55]
(both GaAs-based), Rommel et al. [74], and Grover et al. [76].
As an example, let’s examine the process flow for InP-based micror-
ing resonators used by Grover et al. [76]. For their InP-based devices, they
used wafers on which the waveguiding stack was grown with solid-source
molecular beam epitaxy at the Laboratory for Physical Sciences (for ex-
ample, [281]). The same process has been used successfully by the group
at the Institute of Optics, Rochester, NY, to make GaAs-AlGaAs devices
with BCl3 -Cl2 -Ar etching [195].
First, the blank epitaxially-grown wafer (Fig. 8.2a) is coated with
900 nm of silicon dioxide by plasma-enhanced chemical vapor deposi-
tion. A Leica VB6 e-beam lithography machine is subsequently used to
pattern a bilayer film of poly(methyl methacrylate) (PMMA) as follows:
1. Spin on 100K MW 5% PMMA in Anisole at 4000 RPM for 60 s, which
gives a 113-nm-thick film.
2. Bake for 15 min. at 170 ◦ C.
3. Spin on 495K MW PMMA A4 in Anisole at 2000 RPM for 60 s, which
gives a 204-nm-thick film.
4. Bake for 15 min. at 170 ◦ C.
5. Write pattern in Leica VB6 HR using a dosage of 736 µC/cm2 for small
features (waveguides and microrings) and 610 µC/cm2 for large fea-
tures (device labels).
6. Develop with 2-Methyl-2-Pentanone (methyl isobutyl ketone, or MIBK)
diluted with 2-Propanol as 1 : 3 for 2 min. 30 s.
7. Rinse with 2-Propanol and blow-dry with nitrogen.
The two layers of PMMA are deliberately chosen such that the lower res-
olution PMMA resides on the bottom so that when developed, there is an
overhang of the higher resolution PMMA above it where the pattern was
written. This helps substantially in the later liftoff step, as the solvent can
make its way into the gaps between the metal and the lower portion of
the lip in order to dissolve both layers of PMMA. After the e-beam expo-
sure and development, a 50-nm-thick layer of chromium is deposited and
warm 1-Methyl-2-Pyrrolidone is used to lift off the PMMA resist bilayer.
226 8. Fabrication Techniques for Microresonators

Growth substrate

Upper/lower cladding

(a) Epitaxially grown wafer.

Resist and/or hard mask

(b) Pattern etch mask for resonator and bus


Coupling gap
Bus Resonator Bus

(c) Etch resonator and bus waveguides. This

may be followed by encapsulation in a poly-
mer, spin-on glass, polyimide, etc., to protect
the waveguides.
Fig. 8.2. Process flow for the fabrication of laterally coupled microres-
8.7 Polymer-Bonded, III–V Vertically Coupled Passive Microresonators 227

The pattern is then transferred to the underlying silicon dioxide by dry-

etching in a trifluoromethane-oxygen plasma to obtain the hard mask
for patterning the bus and resonator [Fig. 8.2b]. The InP and quaternary
layers are etched by a methane-hydrogen plasma to a depth of 3.2 µm
with the Cr-SiO2 mask. The mask is subsequently stripped with buffered
hydrofluoric acid [Fig. 8.2c]. Finally, the device is prepared for testing by
thinning the chip in bromine-methanol to enable good cleavage, followed
by cleaving into bars, and mounting on brass mounts.
A scanning electron micrograph of a waveguide cross-section is shown
in Fig. 8.3, and a fabricated device is shown in Fig. 8.4. The waveguides
are fused over a short region (<1 µm) because of imperfect liftoff during
the mask definition. The radius of the curved sections is 2.25 µm, the
straight sections are 10 µm long, the waveguide width is 0.5 µm, and the
width of the coupling gap is 0.2 µm. The bus waveguide is tapered to a
width of 3 µm away from the coupling region to allow increased coupling
of light to and from optical fibers. The waveguides have been etched to
a depth of 3.2 µm chosen to minimize losses resulting from leakage to
the substrate. The pedestal design provides a high lateral index contrast
allowing small bends without substantial bending loss [122]. Earlier SEM
images of GaAs-AlGaAs microresonators made by the UMD group are
shown in Fig. 8.5.

8.7 Polymer-Bonded, III–V Vertically Coupled Passive

Vertically coupled III–V microresonators can be fabricated with poly-
mer wafer-bonding, direct-wafer bonding, and regrowth. We will exam-
ine the latter two methods in the following sections; here, we describe
polymer-bonded vertically coupled microresonators, as demonstrated by
Absil et al. [58] and Grover et al. [62]. The specific case we examine here
is for InP following Grover [174].
The process begins with a SiO2 -coated waveguide stack grown in re-
verse order (top layers first) over an etch-stop. The etch-stop will later
enable substrate removal by selective chemical etching. The general pat-
terning method proceeds as follows: We use a 10× i-line stepper for
photolithography and use the resist mask to transfer the features into
a 400-nm-thick SiO2 layer with a trifluoromethane-oxygen plasma in a
reactive ion etching system. Next, we use the patterned SiO2 as a mask to
define the waveguides in GaInAsP-InP with a methane–hydrogen–argon
plasma-based reactive ion etching process for smooth and vertical side-
walls [286].
First alignment keys are etched to a depth of 2 µm to reach the GaInAs
layer; these keys are used later for processing both sides without re-
quiring infrared back-side alignment (especially useful if the equipment
228 8. Fabrication Techniques for Microresonators


Fig. 8.3. Scanning electron micrographs of waveguide cross-section. (a)
0.5-µm-wide waveguide in the filter section, (b) 3-µm-side waveguide at
the input and output(s). The dark bands are the quaternary layers that
comprise the core of the waveguide.
8.7 Polymer-Bonded, III–V Vertically Coupled Passive Microresonators 229

Fig. 8.4. Scanning electron micrograph of a laterally coupled microring

notch filter.

available does not have IR backside illumination). Afterward, the wave-

guide layer is etched to a depth of 0.9 µm, and the sample is bonded to a
GaAs (or InP) transfer substrate using ∼1–2-µm-thick benzocyclobutene
(BCB), following the process described by Sakamoto et al. [290]. GaAs is
preferred as the transfer substrate due to better mechanical properties
230 8. Fabrication Techniques for Microresonators

(a) Single-microring resonator.

(b) Serially coupled double-microring resonator.

(c) Serially coupled triple-microring resonator.

Fig. 8.5. GaAs-AlGaAs rings made by the group at the University of Mary-
land, College Park. Copyright Philippe P. Absil, 2000, used with permis-

than InP. All but 100 µm of the growth substrate is removed by chemo-
mechanical polishing [276]. For the remaining growth substrate, we use
H3 PO4 :HCl (1:1), which selectively etches the remaining InP, stopping
at the GaInAs layer [277]. Finally, the etch-stop layer is removed with
8.7 Polymer-Bonded, III–V Vertically Coupled Passive Microresonators 231

Growth substrate

Etch stop

Upper/lower cladding


Midlayer (coupling layer, cladding)

(a) Epitaxially grown wafer.

Resist and/or hard mask

(b) Pattern etch mask for bus waveguides.

Fig. 8.6. Process flow for the fabrication of polymer-bonded, vertically
coupled microresonators. (Continued on next page.)

H2 O2 :H2 SO4 :H2 O (1:1:10), which selectively etches GaInAs over InP. This
step clears the other side of the epilayer for processing. The etched-
through alignment keys are used to align the top layer to the lower layer.
The top layer is then etched to a depth of 0.9 µm. A 200-nm mid-layer
is left between the waveguides to decrease losses caused by leakage
to the substrate. Finally, the sample is encapsulated in BCB to ensure
232 8. Fabrication Techniques for Microresonators

Bus Bus

(c) Etch bus waveguides.

Bus Bus

Polymer glue

Transfer substrate
(d) Flip patterned chip and bond to transfer
Fig. 8.6. (Continued from previous page.) Process flow for the fabrication
of polymer-bonded, vertically coupled microresonators. (Continued on
next page.)

refractive-index profile symmetry, thinned, and cleaved producing high-

quality facets.
A microring made on the chip is shown in Fig. 8.7. The rough fea-
ture near the top is etch residue, left behind after removal of SiO2 us-
ing a CHF3 -O2 plasma. (Note: RG subsequently found that there is no
etch residue if the SiO2 is removed with buffered hydrofluoric acid, and
adopted it as the method for such removal.)
The use of BCB provides a low-refractive-index (1.53–1.55) layer be-
tween the GaAs transfer substrate and the epilayer, making tightly
8.7 Polymer-Bonded, III–V Vertically Coupled Passive Microresonators 233

Polymer glue

Transfer substrate
(e) Remove growth substrate and etch stop to
expose “other”side of epilayer for processing.

Polymer encapsulation


Bus Bus

Polymer glue

Transfer substrate
(f) Pattern etch mask for resonator, etch res-
onator, and spin-on encapsulation to protect
the device and to get a symmetric refractive
index structure.
Fig. 8.6. (Continued from previous page.) Process flow for the fabrication
of polymer-bonded, vertically coupled microresonators.

confined waveguides possible on both device layers. Also, the BCB en-
capsulation provides the symmetric refractive-index structure that is
required for identical microrings on both layers, allowing multiring res-
onators with microrings on both layers. Leaving a thin layer of InP cou-
pling layer (0.2 µm) reduces the effect of layer-to-layer misalignment
[58]. This design also provides added mechanical strength to the struc-
ture for the BCB encapsulation. Finally, the alignment scheme, using
a stepper and etched-through alignment keys, provides a high-quality
layer-to-layer alignment with errors smaller than 0.25 µm (measured
from on-chip verniers; Fig. 8.8). The fabricated single-microring optical
channel-dropping filter is shown in Fig. 8.9. Residuals from removal of
234 8. Fabrication Techniques for Microresonators

Fig. 8.7. Dry etched microring.

the growth substrate caused the upper level to appear dirty. A cross sec-
tion for a similar device made by Absil et al. [58], showing waveguides on
either side of the epilayer, is shown in Fig. 8.10.

8.8 Active III–V Laterally Coupled Microresonators

Here is an example of the process flow for active laterally coupled mi-
croresonators, adapted from [174]. The waveguide stack is MBE-grown
p-InP/GaInAsP-MQW/n-InP, with the central MQW region composed of a
8.8 Active III–V Laterally Coupled Microresonators 235

Fig. 8.8. Alignment accuracy measured from on-chip verniers.

Fig. 8.9. Optical micrograph of a fabricated single-microring add–drop

filter with 5-µm-radius ring on the lower level.

superlattice of quantum wells designed for maximizing the electro-optic

coefficient of the waveguide. The fabrication is similar to that for passive
laterally coupled microresonators, except that the waveguides are etched
to a depth of 4.5 µm. Since the device needed planarization for deposi-
tion of contacts, it was coated with a thin layer of SiO2 (∼70 nm) to help
with the adhesion of the BCB planarization layer. Prior to spin-coating
the BCB, an adhesion promoter (AP3000 from Dow Chemical) was spun
on at 2000 RPM for 30 s. The BCB used was Cyclotene 3022-46 from Dow
236 8. Fabrication Techniques for Microresonators

Fig. 8.10. Optical micrograph of the cross section of a polymer-bonded

vertically coupled waveguide device. Copyright Philippe P. Absil, 2000,
used with permission.

Chemical, spun-on at 4000 RPM for 60 s. This was followed by etchback

of benzocyclobutene and SiO2 to expose the tops of the waveguides. For
the top (p-side) contact, Ti-Pt-Au was deposited. The chip was thinned
via an abrasive (alumina) to enable good cleavage and the bottom (n-side)
contact of Pd-Sn-Au-Pd-Au was deposited. Although the n-contact usually
requires annealing [291], the BCB cannot tolerate the high temperature
(∼400◦ C) required; this can be resolved by using other planarization ma-
terials like spin-on glasses or polyimides. Finally, the chip was cleaved
into bars and mounted on oxygen-free copper mounts with a silver epoxy.
In the device described here, Grover et al. deposited the top-side elec-
trode globally (on both the bus and the resonator). This led to unwanted
electro-absorption in the bus waveguide due to the QCSE when a re-
verse bias was applied. Thus, for improved performance, the bus and
ring should be isolated electrically. The cross-section schematic for a de-
vice where the bus and resonator are isolated electrically is illustrated in
Fig. 8.11. A scanning electron micrograph of a fabricated device prior
to planarization is shown in Fig. 8.12. The microring is in the shape of a
racetrack. The waveguides are fused over the short coupling region (1 µm)
because of imperfect liftoff during the mask definition. The radius of the
curved sections is 2.25 µm, whereas the straight sections are 1 µm long,
and the waveguide width is 0.5 µm. The designed width of the coupling
gap was 0.1 µm; here, since the bus and ring are fused, the coupling re-
gion has a width of 1.1 µm. The bus waveguide is tapered to a width of
3 µm away from the coupling region to allow easier coupling of light to
and from optical fibers. The waveguides have been etched to a depth of
8.8 Active III–V Laterally Coupled Microresonators 237

Coupling gap





i-GaInAsP/InP MQW SL core

InP cladding

Growth substrate

Fig. 8.11. Cross section schematic of an active laterally coupled resonator.

Fig. 8.12. Scanning electron micrograph of fabricated device prior to pla-

narization and metallization.
238 8. Fabrication Techniques for Microresonators

Fig. 8.13. Cross section of input–output waveguide in fabricated device.

The ridge in the center is the waveguide, the material on either side is the
BCB planarization layer, and the contact is on the top.

∼4.5 µm. The tall, grass-like structures on either side of the waveguide
are probably due to contaminants in the reaction chamber. The “grass” is
not very close to the sidewalls, possibly because both the Cr-SiO2 -mask
and the particles that caused micro-masking may have had the same sign
of charge.
Fig. 8.13 shows the cross section of the input–output waveguide of a
fabricated device with the BCB planarization and Ti-Pt-Au top contact. As
can be observed in the picture, the planarization layer began to detach
from the waveguide sidewall in this particular device. This problem is
serious and can severely limit the yield and reliability in a manufacturing
environment, as it can create a short circuit around the diode. However, it
can be addressed by using adhesion promoters, planarization materials
that have better adhesion (say, polyimides), or by increasing the thickness
of the SiO2 deposited to improve the adhesion of the benzocyclobutene.
If the SiO2 layer were thick enough, it would prevent the diode from being

8.9 Other Configurations

Grover et al. chose to use the lateral coupling approach for the device de-
scribed in this chapter because of the simplicity of fabrication. Real-world
applications typically will require the use of vertical coupling so that the
bus and the ring can be of different bandgaps allowing for active-passive
8.9 Other Configurations 239

device integration. Integration of these components could be achieved by

making electro-optic modulators on the same side as the bus and chang-
ing the bandgap of ring with quantum-well intermixing. Alternatively,


Fig. 8.14. Two possible configurations for active, vertically coupled mi-
croring resonator filters. The active core could enable devices like lasers,
amplifiers, and electroabsorption and electro-optic modulators.
240 8. Fabrication Techniques for Microresonators

Fig. 8.15. (a) Rabiei’s process for the fabrication of large polymer micror-
ings. (Continued on next page.)

with the direct-wafer-bonding approach, two wafer bonding steps could

be used to make the electro-optically tuned components on the third
layer. Two possible configurations for active vertically coupled rings are
shown in Fig. 8.14. In each case, the p- and n- regions can be interchanged.
The p-contacts in each case require a planarization step. The planariza-
tion and passivation layers are not shown for clarity. In Fig. 8.14(a), the
wafer bonding is achieved with a suitable glue, such as BCB (not shown).
8.9 Other Configurations 241

SU-8 SU-8
Spin coating, UV 15 Protection Layer
Baking, RIE Teflon

Teflon Si
Patterning RIE
Spin coating, UV Teflon
Cure, Baking, RIE
SU-8 SU-8
UV 15 Protection Layer
UV 15 Protection Layer

Si Si

Spin coating Spin coating, UV cure,


UFC 170
UV 15 Protection Layer
SU-8 SU-8
UV 15 Protection Layer
Pattering, UV Exposure

SU-8 SU-8
RIE Etching
UV 15 Protection Layer
Teflon UFC170
SU-8 SU-8
UV 15 Protection Layer
Spin coating,
Baking, RIE

Fig. 8.15. (Continued from previous page.) (b) Rabiei’s process for the fab-
rication of small polymer microrings. Adapted from ref. [292], copyright
2003 Payam Rabiei, used with permission.

In Fig. 8.14(b), Grover et al. show the case for direct-wafer-bonding. An

alternative method of achieving a conducting interface would be to pla-
narize the bus level, etchback, and bond with a conducting material, such
as Pb-Sn [167] Pd [168], Au-Sn [169], or Pd-In [170, 171].
242 8. Fabrication Techniques for Microresonators

8.10 Polymer Microrings

Rabiei et al. use a multitude of polymers to fabricate active and passive

polymer microrings [292]. In addition, the group at the University of Mary-
land, College Park, has demonstrated BCB microrings [293]. In Fig. 8.15,
we show the process flow for two of Rabiei’s devices; his active rings are
fabricated in a similar manner, except with contacts on either side (above
and below) the resonators.

8.11 Summary

This chapter examined the key elements associated with the fabrication
of optical microresonators and associated devices. There are, of course,
many demonstrations of these devices from various groups, each em-
ploying slightly different approaches. For example,
1. The group at University of Illinois, Urbana-Champaign [74], in collab-
oration with researchers from Sarnoff, uses MMI-coupling for laterally
coupled microresonators.
2. The group at Twente University [294] uses SiON waveguides with a
combination of low-pressure and plasma-enhanced chemical vapor
deposition, as does the group at Politecnico di Milano [295].
3. Little Optics (now Infinera) uses glass-based materials to make buried
bus (and ring) resonators [37, 287].
The fabrication of microrings, microdisks, and other optical devices em-
ploying high index contrasts such as photonic crystals continues to be
extremely challenging. At the time of writing, it has been over 15 years
since the first microdisk lasers were demonstrated, yet they are only now
finding their way into commercial products.

1. Lord Rayleigh, “The problem of the whispering gallery,” Philosophical Mag-

azine, vol. xx, pp. 1001–1004, 1910.
2. E. A. J. Marcatili, “Bends in optical dielectric guides,” The Bell System
Technical Journal, vol. 48, pp. 2103–2132, 1969.
3. H. P. Weber and R. Ulrich, “A thin-film ring laser,” Applied Physics Letters,
vol. 19, pp. 38–40, 1971.
4. R. Ulrich and H. P. Weber, “Unidirectional thin-film ring laser,” Applied
Physics Letters, vol. 20, pp. 38–40, 1972.
5. R. Ulrich and H. P. Weber, “Solution-deposited thin films as passive and
active light-guides,” Applied Optics, vol. 11, pp. 428–434, 1972.
6. J. Haavisto and G. A. Pajer, “Resonance effects in low-loss ring waveguides,”
Optics Letters, vol. 5, pp. 510–512, 1980.
7. L. F. Stokes, M. Chodorow, and H. J. Shaw, “All-single-mode fiber resonator,”
Optics Letters, vol. 7, pp. 288–290, 1982.
8. R. G. Walker and C. D. W. Wilkinson, “Integrated optical ring resonators
made by silver ion-exchange in glass,” Applied Optics, vol. 22, pp. 1029–
1035, 1983.
9. A. Mahapatra and J. M. Connors, “High finesse ring resonators fabrication
and analysis,” in Integrated Optical Circuit Engineering, Innsbruck, Austria,
10. J. M. Connors and A. Mahapatra, “High finesse ring resonators made by
silver ion exchange in glass,” Journal of Lightwave Technology, vol. LT-5,
pp. 1686–1689, 1987.
11. K. Honda, E. M. Garmire, and K. E. Wilson, “Characteristics of an integrated
optics ring resonator fabricated in glass,” Journal of Lightwave Technology,
vol. LT-2, pp. 714–719, 1984.
12. A. Naumaan and J. T. Boyd, “Ring resonator fabricated in phosphosilicate
glass films deposited by chemical vapor deposition,” Journal of Lightwave
Technology, vol. LT-4, pp. 1294–1303, 1986.
13. K.-H. Tietgen, “Tunable integrated optical ring resonator,” in Topical Meet-
ing on Integrated and Guided Wave Optics, Kissimee, FL, 1984.
14. A. Mahapatra and W. C. Robinson, “Integrated-optic ring resonators made
by proton exchange in lithium niobate,” Applied Optics, vol. 24, pp. 2285–
2286, 1985.
15. K. Oda, N. Takato, and H. Toba, “A wide-FSR waveguide double-ring res-
onator for optical FDM transmission systems,” Journal of Lightwave Tech-
nology, vol. 9, pp. 728–736, 1991.
16. T. Kominato, Y. Ohmori, N. Takato, H. Okazaki, and M. Yasu, “Ring res-
onators composed of GeO2 -doped silica waveguides,” Journal of Lightwave
Technology, vol. 10, pp. 1781–1788, 1992.
244 Bibliography

17. W. J. Wang, S. Honkanen, S. I. Najafi, and A. Tervonen, “New integrated

optical ring resonator in glass,” Electronics Letters, vol. 28, pp. 1967–1968,
18. S. T. Chu, B. E. Little, P. Wugen, T. Kaneko, and Y. Kokubun, “Cascaded
microring resonators for crosstalk reduction and spectrum cleanup in add-
drop filters,” IEEE Photonics Technology Letters, vol. 11, pp. 1423–1425,
19. S. T. Chu, B. E. Little, W. Pan, and Y. Kokubun, “A cross-grid array of mi-
croresonators for very large scale integrated photonic circuits,” in Confer-
ence on Lasers and Electro-Optics, San Francisco, CA, 1999.
20. S. T. Chu, P. Wugan, S. Suzuki, B. E. Little, S. Sato, and Y. Kokubun, “Temper-
ature insensitive vertically coupled microring resonator add/drop filters by
means of a polymer overlay,” IEEE Photonics Technology Letters, vol. 11, pp.
1138–1140, 1999.
21. B. E. Little, S. T. Chu, W. Pan, D. Ripin, T. Kaneko, Y. Kokubun, and E. P. Ippen,
“Vertically coupled glass microring resonator channel dropping filters,”
IEEE Photonics Technology Letters, vol. 11, pp. 215–217, 1999.
22. S. T. Chu, B. E. Little, W. Pan, T. Kaneko, and Y. Kokubun, “Second order filter
response from parallel coupled glass microring resonators,” IEEE Photonics
Technology Letters, vol. 11, pp. 1426–1428, 1999.
23. S. T. Chu, W. Pan, S. Sato, T. Kaneko, B. E. Little, and Y. Kokubun, “Wave-
length trimming of a microring resonator filter by means of a UV sensitive
polymer overlay,” IEEE Photonics Technology Letters, vol. 11, pp. 688–690,
24. Y. Kokubun, S. Kubota, and S. T. Chu, “Polarisation-independent vertically
coupled microring resonator filter,” Electronics Letters, vol. 37, pp. 90–92,
25. Y. Yanagase, S. Suzuki, Y. Kokubun, and S. T. Chu, “Box-like filter response
by vertically series coupled microring resonator filter,” in European Con-
ference on Optical Communications, Amsterdam, The Netherlands, 2001.
26. J. Bismuth, P. Gidon, F. Revol, and S. Valette, “Low-loss ring resonators
fabricated from silicon based integrated optics technologies,” Electronics
Letters, vol. 27, pp. 722–724, 1991.
27. S. Suzuki, K. Shuto, and Y. Hibino, “Integrated-optic ring resonators with
two stacked layers of silica waveguide on Si,” IEEE Photonics Technology
Letters, vol. 4, pp. 1256–1258, 1992.
28. T. Kominato, Y. Hibino, and K. Onose, “Silica-based finesse-variable ring
resonator,” IEEE Photonics Technology Letters, vol. 5, pp. 560–562, 1993.
29. S. Suzuki, K. Oda, and Y. Hibino, “Integrated-optic double-ring resonators
with a wide free spectral range of 100 GHz,” Journal of Lightwave Technol-
ogy, vol. 13, pp. 1766–1771, 1995.
30. P. Katila, P. Heimala, and J. Aarnio, “Thermo-optically controlled ring res-
onator on silicon,” Electronics Letters, vol. 32, pp. 1005–1006, 1996.
31. J. S. Foresi, B. E. Little, G. Steinmeyer, E. Thoen, S. T. Chu, H. A. Haus, E. P.
Ippen, L. C. Kimerling, and W. Greene, “Si/SiO2 micro-ring resonator optical
add/drop filters,” in Conference on Lasers and Electro-Optics, Baltimore,
MD, 1997.
32. B. E. Little, J. S. Foresi, G. Steinmeyer, E. R. Thoen, S. T. Chu, H. A. Haus,
E. P. Ippen, L. C. Kimerling, and W. Greene, “Ultra-compact Si-SiO2 micror-
ing resonator optical channel dropping filters,” IEEE Photonics Technology
Letters, vol. 10, pp. 549–551, 1998.
Bibliography 245

33. D. R. Lim, B. E. Little, K. K. Lee, M. Morse, H. H. Fujimoto, H. A. Haus, and L. C.

Kimerling, “Micron-sized channel dropping filters using silicon waveguide
devices,” in Optical Devices for Fiber Communication, Boston, MA, 1999.
34. D. J. W. Klunder, F. S. Tan, T. van der Veen, H. F. Bulthuis, H. J. W. M. Hoek-
stra, and A. Driessen, “Design and characterization of waveguide-coupled
cylindrical microring resonators in Si3 N4 ,” in IEEE Lasers and Electro-Optics
Society, 13th Annual Meeting, Rio Grande, Puerto Rico, 2000.
35. D. J. W. Klunder, E. Krioukov, F. S. Tan, T. van der Veen, H. F. Bulthuis,
G. Sengo, C. Otto, H. J. W. M. Hoekstra, and A. Driessen, “Vertically and
laterally waveguide-coupled cylindrical microresonators in Si3 N4 on SiO2
technology,” Applied Physics B (Lasers and Optics), vol. B73, pp. 603–608,
36. S. Suzuki, Y. Hatakeyama, Y. Kokubun, and S. T. Chu, “Precise control of
wavelength channel spacing of microring resonator add-drop filter array,”
Journal of Lightwave Technology, vol. 20, pp. 745–750, 2002.
37. B. E. Little, “A VLSI photonics platform,” in Optical Fiber Communication,
Atlanta, GA, 2003.
38. P. Rabiei and W. H. Steier, “Micro-ring resonators using polymer materials,”
in Conference on Lasers and Electro Optics, Baltimore, MD, 2001.
39. P. Rabiei, W. H. Steier, C. Zhang, C.-G. Wang, and H. J. Lee, “Polymer micro-
ring modulator with 1 THz FSR,” in Conference on Lasers and Electro-Optics,
Long Beach, CA, 2002.
40. P. Rabiei, W. H. Steier, C. Zhang, and L. R. Dalton, “Polymer micro-ring filters
and modulators,” Journal of Lightwave Technology, vol. 20, pp. 1968–1975,
41. W. Y. Chen, R. Grover, T. A. Ibrahim, V. Van, and P.-T. Ho, “Compact single-
mode benzocyclobutene microracetrack resonators,” in Integrated Photon-
ics Research, Washington, DC, 2003.
42. S. L. McCall, A. F. J. Levi, R. E. Slusher, S. J. Pearton, and R. A. Logan,
“Whispering-gallery mode microdisk lasers,” Applied Physics Letters, vol.
60, pp. 289–291, 1992.
43. D. Y. Chu, M. K. Chin, N. J. Sauer, Z. Xu, T. Y. Chang, and S. T. Ho, “1.5 µm
InGaAs/InAlGaAs quantum-well microdisk lasers,” IEEE Photonics Technol-
ogy Letters, vol. 5, pp. 1353–1355, 1993.
44. C. R. Abernathy, S. J. Pearton, J. D. MacKenzie, J. R. Mileham, S. R. Bharatan,
V. Krishnamoorthy, K. S. Jones, M. Hagerott-Crawford, R. J. Shul,
S. P. Kilcoyne, J. M. Zavada, D. Zhang, and R. M. Kolbas, “Growth and
fabrication of GaN-InGaN microdisk laser structures,” Solid-State Electron-
ics, vol. 39, pp. 311–313, 1996.
45. T. Baba, P. Fujita, A. Sakai, M. Kihara, and R. Watanabe, “Lasing charac-
teristics of GaInAsP-InP strained quantum-well microdisk injection lasers
with diameter of 2 − 10 µm,” IEEE Photonics Technology Letters, vol. 9, pp.
878–880, 1997.
46. L. Dai, B. Zhang, R. A. Mair, K. Zeng, J. Lin, H. Jigang, A. Botchkarev, W. Kim,
H. Morkoc, and M. A. Khan, “Optical properties and resonant modes in
GaN/AlGaN and InGaN/GaN multiple quantum well microdisk cavities,” in
Semiconductor Lasers III, Beijing, China, 1998.
47. R. A. Mair, K. C. Zeng, J. Y. Lin, H. X. Jiang, B. Zhang, L. Dai, A. Botchkarev,
W. Kim, H. Morkoc, and M. A. Khan, “Optical modes within III–nitride mul-
tiple quantum well microdisk cavities,” Applied Physics Letters, vol. 72, pp.
1530–1532, 1998.
246 Bibliography

48. D. V. Tishinin, P. D. Dapkus, A. E. Bond, I. Kim, C. K. Lin, and J. O’Brien,

“Vertical resonant couplers with precise coupling efficiency control fabri-
cated by wafer bonding,” IEEE Photonics Technology Letters, vol. 11, pp.
1003–1005, 1999.
49. L. Djaloshinski and M. Orenstein, “Disk and ring microcavity lasers and
their concentric coupling,” IEEE Journal of Quantum Electronics, vol. 35,
pp. 737–744, 1999.
50. N. B. Rex, R. K. Chang, and L. J. Guido, “Threshold minimization and direc-
tional laser emission from GaN microdisks,” in Laser Resonators, San Jose,
CA, 2000.
51. D. Rafizadeh, J. P. Zhang, S. C. Hagness, A. Taflove, K. A. Stair, S. T. Ho,
and R. C. Tiberio, “Waveguide-coupled AlGaAs/GaAs microcavity ring and
disk resonators with high finesse and 21.6 nm free spectral range,” Optics
Letters, vol. 22, pp. 1244–1246, 1997.
52. D. Rafizadeh, J. P. Zhang, S. C. Hagness, A. Taflove, K. A. Stair, S. T. Ho, and
R. C. Tiberio, “Temperature tuning of microcavity ring and disk resonators
at 1.5 µm,” in IEEE Lasers and Electro-Optics Society, 10th Annual Meeting,
San Francisco, CA, 1997.
53. J. V. Hryniewicz, P. P. Absil, B. E. Little, and P.-T. Ho, “Semiconductor micror-
ing resonators for integrated photonics,” in Integrated Photonics Research,
Québec City, Canada, 2000.
54. J. V. Hryniewicz, P. P. Absil, B. E. Little, R. A. Wilson, L. G. Joneckis, and
P.-T. Ho, “Microring resonator notch filters,” in Conference on Lasers and
Electro-Optics, San Francisco, CA, 2000.
55. J. V. Hryniewicz, P. P. Absil, B. E. Little, and P.-T. Ho, “Higher order filter
response in coupled microring resonators,” IEEE Photonics Technology Let-
ters, vol. 12, pp. 320–322, 2000.
56. P. P. Absil, J. V. Hryniewicz, B. E. Little, P. S. Cho, R. A. Wilson, L. G. Joneckis,
and P.-T. Ho, “Wavelength conversion in GaAs micro-ring resonators,” Op-
tics Letters, vol. 25, pp. 554–556, 2000.
57. P. P. Absil, J. V. Hryniewicz, B. E. Little, R. A. Wilson, L. G. Joneckis, and P.-T.
Ho, “Compact microring notch filters,” IEEE Photonics Technology Letters,
vol. 12, pp. 398–400, 2000.
58. P. P. Absil, J. V. Hryniewicz, B. E. Little, F. G. Johnson, and P.-T. Ho, “Ver-
tically coupled microring resonators using polymer wafer bonding,” IEEE
Photonics Technology Letters, vol. 13, pp. 49–51, 2001.
59. R. Grover, P. P. Absil, V. Van, J. V. Hryniewicz, B. E. Little, O. S. King, F. G.
Johnson, L. C. Calhoun, and P.-T. Ho, “Vertically coupled GaAs-AlGaAs
and GaInAsP-InP microring resonators,” in Optical Fiber Communication,
Anaheim, CA, 2001.
60. P. P. Absil, J. V. Hryniewicz, B. E. Little, F. G. Johnson, and P.-T. Ho, “Wave-
length selective mirror using notched microring resonators,” in IEEE Lasers
and Electro-Optics Society, 14th Annual Meeting, San Diego, CA, 2001.
61. V. Van, T. A. Ibrahim, P. P. Absil, R. Grover, J. V. Hryniewicz, B. E. Little,
F. G. Johnson, and P.-T. Ho, “Periodically-coupled GaAs/AlGaAs microring
array filter with wide free spectral range,” in Integrated Photonics Research,
Monteray, CA, 2001.
62. R. Grover, P. P. Absil, V. Van, J. V. Hryniewicz, B. E. Little, O. S. King, L. C.
Calhoun, F. G. Johnson, and P.-T. Ho, “Vertically coupled GaInAsP-InP mi-
croring resonators,” Optics Letters, vol. 26, pp. 506–508, 2001.
63. R. Grover, V. Van, T. A. Ibrahim, P. P. Absil, L. C. Calhoun, F. G. Johnson,
J. V. Hryniewicz, and P.-T. Ho, “Parallel-cascaded semiconductor micror-
ing resonators for high-order and wide-FSR filters,” Journal of Lightwave
Technology, vol. 20, pp. 900–905, 2002.
Bibliography 247

64. K. Djordjev, S.-J. Choi, S.-J. Choi, and P. D. Dapkus, “High-Q vertically cou-
pled InP microdisk resonators,” IEEE Photonics Technology Letters, vol. 14,
pp. 331–332, 2002.
65. K. Djordjev, S.-J. Choi, S.-J. Choi, and P. D. Dapkus, “Microdisk tunable
resonant filters and switches,” IEEE Photonics Technology Letters, vol. 14,
pp. 828–830, 2002.
66. E. A. J. Marcatili, “Improved relations describing directional control in elec-
tromagnetic wave guidance,” The Bell System Technical Journal, vol. 48, pp.
2161–2188, 1969.
67. M. Heiblum and J. H. Harris, “Analysis of curved optical waveguides by
conformal transformation,” IEEE Journal of Quantum Electronics, vol. QE-
11, pp. 75–83, 1975.
68. M. Heiblum and J. H. Harris, “Correction to ‘Analysis of curved optical
waveguides by conformal transformation’,” IEEE Journal of Quantum Elec-
tronics, vol. 12, pp. 313–313, 1976.
69. D. Marcuse, “Bend loss of slab and fiber modes computed with diffraction
theory,” IEEE Journal of Quantum Electronics, vol. 29, pp. 2957–2961, 1993.
70. G. Griffel, J. H. Abeles, R. J. Menna, A. M. Braun, J. C. Connolly, and M. King,
“Low-threshold InGaAsP ring lasers fabricated using bi-level dry etching,”
IEEE Photonics Technology Letters, vol. 12, pp. 146–148, 2000.
71. D. G. Rabus and M. Hamacher, “MMI-coupled ring resonators in GaInAsP-
InP,” IEEE Photonics Technology Letters, vol. 13, pp. 812–814, 2001.
72. D. G. Rabus, M. Hamacher, and H. Heidrich, “Active and passive microring
resonator filter applications in GaInAsP/InP,” in Indium Phosphide and
Related Materials, Nara, Japan, 2001.
73. D. G. Rabus, M. Hamacher, U. Troppenz, and H. Heidrich, “High-Q channel
dropping filters using ring resonators with integrated SOAs,” IEEE Photonics
Technology Letters, vol. 14, pp. 1442–1444, 2002.
74. S. L. Rommel, J.-H. Jang, W. Lu, G. Cueva, L. Zhou, I. Adesida, G. A. Pajer,
R. W. Whaley, A. Lepore, Z. Schellanbarger, and J. H. Abeles, “Effect of H2
on the etch profile of InP/InGaAsP alloys in Cl2 /Ar/H2 inductively coupled
plasma reactive ion etching chemistries for photonic device fabrication,”
Journal of Vacuum Science & Technology B, vol. 20, pp. 1327–1330, 2002.
75. R. Grover, T. A. Ibrahim, T. N. Ding, L.-C. Kuo, S. Kanakaraju, L. C. Calhoun,
and P.-T. Ho, “Ultracompact single-mode GaInAsP-InP microracetrack res-
onators,” in Integrated Photonics Research, Washington, DC, 2003.
76. R. Grover, T. A. Ibrahim, T. N. Ding, Y. Leng, L.-C. Kuo, S. Kanakaraju,
K. Amarnath, L. C. Calhoun, and P.-T. Ho, “Laterally coupled InP-based
single-mode microracetrack notch filter,” IEEE Photonics Technology Let-
ters, 2003.
77. R. Grover, T. A. Ibrahim, S. Kanakaraju, M. L. Lucas, L. C. Calhoun, and P.-T.
Ho, “Tunable GaInAsP-InP optical microresonator notch filter,” submitted
to IEEE Photonics Technology Letters, 2003.
78. A. Szoke and V. Daneu, “Bistable optical element and its applications,”
Applied Physics Letters, vol. 15, pp. 376–379, 1969.
79. S. L. McCall, H. M. Gibbs, G. G. Churchill, and T. N. C. Venkatesan, “Optical
transistor and bistability,” Bulletin of the Americal Physical Society, vol. 20,
pp. 636, 1975.
80. H. M. Gibbs, S. L. McCall, and T. N. C. Venkatesan, “Differential gain and
bistability using a sodium-filled Fabry–Perot interferometer,” Physical Re-
view Letters, vol. 36, pp. 1135–1138, 1976.
81. J. H. Marburger and F. S. Felber, “Theory of a lossless nonlinear Fabry–Perot
interferometer,” Physical Review A, vol. 17, pp. 335, 1978.
248 Bibliography

82. D. A. B. Miller and S. D. Smith, “Two-beam optical signal amplification and

bistability in InSb,” Optics Communications, vol. 31, pp. 101–104, 1979.
83. D. A. B. Miller, “Refractive Fabry–Perot bistability with linear absorption:
Theory of operation and cavity optimization,” IEEE Journal of Quantum
Electronics, vol. 17, pp. 306–311, 1981.
84. R. M. Shelby, M. D. Levenson, D. F. Walls, A. Aspect, and G. J. Milburn, “Gen-
eration of squeezed states of light with a fiber-optic ring interferometer,”
Physical Review A, vol. 33, pp. 4008–4025, 1986.
85. R. M. Shelby, M. D. Levenson, and S. H. Permutter, “Bistability and other
effects in a nonlinear fiber-optic ring resonator,” Journal of the Optical
Society of America B, vol. 5, pp. 347–357, 1988.
86. V. B. Braginsky and V. S. Ilchenko, “Properties of optical dielectric microres-
onators,” Soviet Physics Doklady, vol. 32, pp. 306–307, 1987.
87. V. B. Braginsky, M. L. Gorodetsky, and V. S. Ilchenko, “Quality-factor and
nonlinear properties of optical whispering-gallery modes,” Physics Letters
A, vol. 137, pp. 393–397, 1989.
88. D. W. Vernooy, A. Furusawa, N. P. Georgiades, V. S. Ilchenko, and H. J. Kim-
ble, “Cavity QED with high-Q whispering gallery modes,” Physical Review
A, vol. 57, pp. R2293–R2296, 1998.
89. J. Popp, M. H. Fields, and R. K. Chang, “Q-switching by saturable absorption
in microdroplets: elastic scattering and laser emission,” Optics Letters, vol.
22, pp. 1296–1298, 1997.
90. F. C. Blom, D. R. van Dijk, H. J. Hoekstra, A. Driessen, and T. J. A. Popma,
“Experimental study of integrated-optics microcavity resonators: Toward
an all-optical switching device,” Applied Physics Letters, vol. 71, pp. 747–
749, 1997.
91. F. C. Blom, H. Kelderman, H. J. W. M. Hoekstra, A. Driessen, T. J. A. Popma,
S. T. Chu, and B. E. Little, “A single channel dropping filter based on a
cylindrical microresonator,” Optics Communications, vol. 167, pp. 77–82,
92. A. T. Rosenberger, “Nonlinear optical effects in the whispering-gallery
modes of microspheres,” in Operational Characteristics and Crystal Growth
of Nonlinear Optical Materials, R. B. Lal and D. O. Frazier, Eds., vol. 3793,
pp. 179–186. SPIE Proceedings, 1999.
93. D. Braunstein, A. M. Khazanov, G. A. Koganov, and R. Shuker, “Lowering
of threshold conditions for nonlinear effects in a microsphere,” Physical
Review A, vol. 53, pp. 3565–3572, 1996.
94. S. M. Spillane, T. J. Kippenberg, and K. J. Vahala, “Ultralow-threshold Raman
laser using a spherical dielectric microcavity,” Nature, vol. 415, pp. 621–
623, 2002.
95. P. Michler, A. Kiraz, C. Becher, W. V. Schoenfeld, P. M. Petroff, L. Zhang,
E. Hu, and A. Imamoglu, “A Quantum Dot Single-Photon Turnstile Device,”
Science, vol. 290, pp. 2282–2285, 2000.
96. A. Kiraz, P. Michler, C. Becher, B. Gayral, A. Imamoglu, L. Zhang, E. Hu, W. V.
Schoenfeld, and P. M. Petroff, “Cavity Quantum Electrodynamics Using a
Single InAs Quantum Dot in a Microdisk Structure,” Applied Physics Letters,
vol. 78, pp. 3932–3934, 2001.
97. G. Agrawal, Nonlinear Fiber Optics, Academic Press, San Diego, CA, 3rd
edition, 2001.
98. P. K. Tien, “Light waves in thin films and integrated optics,” Applied Optics,
vol. 10, pp. 2395–2413, 1971.
99. H. Kogelnik and V. Ramaswamy, “Scaling rules for thin-film optical
waveguides,” Applied Optics, vol. 13, pp. 1857–1862, 1974.
Bibliography 249

100. S. M. Saad, “Review of numerical methods for the analysis of arbitrarily-

shaped microwave and optical dielectric waveguides,” IEEE Transactions
on Microwave Theory and Techniques, vol. 33, pp. 894–899, 1985.
101. E. A. J. Marcatili, “Dielectric rectangular waveguide and directional coupler
for integrated optics,” The Bell System Technical Journal, vol. 48, pp. 2071–
2102, 1969.
102. A. Kumar, K. Thyagarajan, and A. K. Ghatak, “Analysis of rectangular-core
dielectric waveguides: an accurate perturbation approach,” Optics Letters,
vol. 8, pp. 63–65, 1983.
103. J. Fox, Ed., Integrated circuits for the millimeter through optical frequency
range. Polytechnic Press, 1970.
104. G. B. Hocker and W. K. Burns, “Mode dispersion in diffused channel
waveguides by the effective index method,” Applied Optics, vol. 16, pp.
113–118, 1977.
105. J.-S. Lee and S.-Y. Shin, “On the validity of the effective-index method for
rectangular dielectric waveguides,” Journal of Lightwave Technology, vol.
11, pp. 1320–1324, 1993.
106. C. M. Kim, B. G. Jung, and C. W. Lee, “Analysis of dielectric rectangular
waveguide by modified effective-index method,” Electronics Letters, vol.
22, pp. 296–298, 1986.
107. Y. Cai, T. Mizumoto, and Y. Naito, “Perturbation feedback method for
analysing rectangular dielectric waveguides,” Electronics Letters, vol. 26,
pp. 521–523, 1990.
108. K. S. Chiang, “Analysis of rectangular dielectric waveguides: Effective-index
method with built-in perturbation correction,” Electronics Letters, vol. 28,
pp. 388–390, 1992.
109. J. E. Goell, “A circular-harmonic computer analysis of rectangular dielectric
waveguides,” The Bell System Technical Journal, vol. 48, pp. 2133–2160,
110. J. A. Fleck, J. R. Morris, and M. D. Feit, “Time-dependent propagation of
high energy laser beams through the atmosphere,” Applied Physics, vol.
10, pp. 129–160, 1976.
111. K. Yee, “Numerical solution of initial boundary value problems involving
Maxwell’s equations in isotropic media,” IEEE Transactions on Antennas
and Propagation, vol. AP-14, pp. 302–307, 1966.
112. A. Taflove and S. C. Hagness, Computational Electrodynamics, The Finite-
Difference Time-Domain Method, Artech House, Boston, MA, 2000.
113. A. Yariv, “Coupled-mode theory for guided-wave optics,” IEEE Journal of
Quantum Electronics, vol. 9, pp. 919–933, 1973.
114. J. E. Sipe and G. I. Stegeman, “Comparison of normal mode and total field
analysis techniques in planar integrated optics,” Journal of the Optical
Society of America, vol. 69, pp. 1676–1683, 1979.
115. H. A. Haus, W. P. Huang, S. Kawakami, and N. A. Whitaker, “Coupled-mode
theory of optical waveguides,” Journal of Lightwave Technology, vol. 5, pp.
16–23, 1987.
116. D. G. Hall, “A comment on the coupled-mode equations used in guided-
wave optics,” Optics Communications, vol. 82, pp. 82–83, 1991.
117. H. A. Haus, Waves and Fields in Optoelectronics, Prentice-Hall, Englewood
Cliffs, NJ, 1984.
118. F. de Fornel, Evanescent Waves: From Newtonian Optics to Atomic Optics,
Springer-Verlag, New York, NY, 1st edition, 2001.
250 Bibliography

119. M. K. Chin, C. Youtsey, W. Zhao, T. Pierson, Z. Ren, S. L. Wu, L. Wang, Y. G.

Zhao, and S. T. Ho, “GaAs microcavity channel-dropping filter based on a
race-track resonator,” IEEE Photonics Technology Letters, vol. 11, pp. 1620–
1622, 1999.
120. B. E. Little and S. T. Chu, “Estimating surface-roughness loss and output
coupling in microdisk resonators,” Optics Letters, vol. 21, pp. 1390–1392,
121. B. E. Little, J. P. Laine, and S. T. Chu, “Surface-roughness-induced contradi-
rectional coupling in ring and disk resonators,” Optics Letters, vol. 22, pp.
4–6, 1997.
122. V. Van, P. P. Absil, J. V. Hryniewicz, and P.-T. Ho, “Propagation loss in
single-mode GaAs-AlGaAs microring resonators: Measurement and model,”
Journal of Lightwave Technology, vol. 19, pp. 1734–1739, 2001.
123. J. Sarma and K. A. Shore, “Electromagnetic theory for optical disc res-
onators,” IEE Proceedings Journal (Optoelectronics), vol. 132, pp. 325–330,
124. A. W. Snyder and J. D. Love, “Tunnelling leaky modes on optical
waveguides,” Optics Communications, vol. 12, pp. 326–328, 1974.
125. L. Collot, V. Lefevre-Seguin, M. Brune, J.-M. Raimond, and S. Haroche, “Very
high-Q whispering-gallery mode resonances observed on fused silica mi-
crospheres,” Europhysics Letters, vol. 23, pp. 327–334, 1993.
126. M. K. Chin and S. T. Ho, “Design and modelling of waveguide-coupled single-
mode microring resonators,” Journal of Lightwave Technology, vol. 16, pp.
1433–1446, 1998.
127. W. Berglund and A. Gopinath, “WKB analysis of bend losses in optical
waveguides,” Journal of Lightwave Technology, vol. 18, pp. 1161–1166,
128. M. Kuznetsov and H. A. Haus, “Radiation loss in dielectric waveguide struc-
tures by the volume current method,” IEEE Journal of Quantum Electronics,
vol. 19, pp. 1505–1514, 1983.
129. J. E. Heebner, R. W. Boyd, and Q. Park, “SCISSOR solitons and other propaga-
tion effects in microresonator modified waveguides,” Journal of the Optical
Society of America B, vol. 19, pp. 722–731, 2002.
130. S. Arnold, C. T. Liu, W. B. Whitten, and J. M. Ramsey, “Room-temperature
microparticle-based persistent spectral hole burning memory,” Optics Let-
ters, vol. 16, pp. 420–422, 1991.
131. A. Serpengüzel, S. Arnold, and G. Griffel, “Excitation of resonances of mi-
crospheres on an optical fiber,” Optics Letters, vol. 20, pp. 654–656, 1995.
132. N. Dubreuil, J. C. Knight, D. K. Leventhal, V. Sandoghdar, J. Hare, and
V. Lefevre, “Eroded monomode optical fiber for whispering-gallery mode
excitation in fused-silica microspheres,” Optics Letters, vol. 20, pp. 813–
815, 1995.
133. M. L. Gorodetsky, A. A. Savchenkov, and V. S. Ilchenko, “Ultimate Q of
optical microsphere resonators,” Optics Letters, vol. 21, pp. 453–455, 1996.
134. V. V. Vassiliev, V. L. Velichansky, V. S. Ilchenko, M. L. Gorodetsky, L. Holl-
berg, and A. V. Yarovitsky, “Narrow-line-width diode laser with a high-Q
microsphere resonator,” Optics Communications, vol. 158, pp. 305–312,
135. M. L. Gorodetsky and V. S. Ilchenko, “Optical microsphere resonators: op-
timal coupling to high-Q whispering-gallery modes,” Journal of the Optical
Society of America B, vol. 16, pp. 147–154, 1999.
136. V. Lefevre-Seguin, “Whispering-gallery mode lasers with doped silica mi-
crospheres,” Optical Materials, vol. 11, pp. 153–165, 1999.
Bibliography 251

137. M. Cai, O. Painter, and K. Vahala, “Observation of critical coupling in a fiber

taper to a silica-microsphere whispering-galley mode system,” Physical
Review Letters, vol. 85, pp. 74–77, 2000.
138. J. P. Rezac and A. T. Rosenberger, “Locking a microsphere whispering-
gallery mode to a laser,” Optics Express, vol. 8, pp. 605–610, 2001.
139. D. K. Armani, T. J. Kippenberg, S. M. Spillane, and K. J. Vahala, “Ultra-high-Q
toroid microcavity on a chip,” Nature, vol. 421, pp. 925–928, 2003.
140. J. C. Knight, N. Dubreuil, V. Sandoghdar, J. Hare, V. Lefevre-Seguin, J. M.
Raimond, and S. Haroche, “Mapping whispering-gallery modes in mi-
crospheres with a near-field probe,” Optics Letters, vol. 20, pp. 1515–1517,
141. J. C. Knight, N. Dubreuil, V. Sandoghdar, J. Hare, V. Lefevre-Seguin, J. M.
Raimond, and S. Haroche, “Characterizing whispering-gallery modes in
microspheres by direct observation of the optical standing-wave pattern in
the near field,” Optics Letters, vol. 21, pp. 698–700, 1996.
142. M. L. M. Balistreri, D. J. W. Klunder, F. C. Blom, A. Driessen, H. W. J. M.
Hoekstra, J. P. Korterik, L. Kuipers, and N. F. van Hulst, “Visualizing the
whispering gallery modes in a cylindrical optical microcavity,” Optics Let-
ters, vol. 24, pp. 1829–1831, 1999.
143. B. E. Little, J. P. Laine, and H. A. Haus, “Analytic theory of coupling from
tapered fibers and half-blocks into microsphere resonators,” Journal of
Lightwave Technology, vol. 17, pp. 704–715, 1999.
144. S. C. Hagness, D. Rafizadeh, S. T. Ho, and A. Taflove, “FDTD microcav-
ity simulations: design and experimental realization of waveguide-coupled
single-mode ring and whispering-gallery-mode disk resonators,” Journal
of Lightwave Technology, vol. 15, pp. 2154–2165, 1997.
145. M. Borselli, K. Srinivasan, P. E. Barclay, and O. Painter, “Rayleigh scattering,
mode coupling, and optical loss in silicon microdisks,” Applied Physics
Letters, vol. 85, pp. 3693–3696, 2004.
146. P. Rabiei, “Calculation of losses in micro-ring resonators with arbitrary
refractive index or shape profile and its applications,” Journal of Lightwave
Technology, vol. 23, pp. 1295–1301, 2005.
147. S. G. Johnson, M. Ibanescu, M. A. Skorobogatiy, O. Weisberg, J. D.
Joannopoulos, and Y. Fink, “Pertubation theory for Maxwell’s equations
with shifting material boundaries,” Physical Review E, vol. 65, pp. 066611,
148. S. G. Johnson, M. L. Povinelli, M. Soljacic, A. Karalis, S. Jacobs, and
J. D. Joannopoulos, “Roughness losses and volume current methods in
photonic-crystal waveguides,” Applied Physics B, vol. 81, pp. 283–293, 2005.
149. C. Fabry and A. Perot, “A Multipass Interferometer,” Ann. Chim. Phys, vol.
16, pp. 115, 1899.
150. B. E. Little, S. T. Chu, and H. A. Haus, “Track changing by use of the phase
response of microspheres and resonators,” Optics Letters, vol. 23, pp. 894–
896, 1998.
151. J. E. Heebner and R. W. Boyd, “Enhanced all-optical switching by use of a
nonlinear fiber ring resonator,” Optics Letters, vol. 24, pp. 847–849, 1999.
152. C. K. Madsen and J. H. Zhao, Optical Filter Design and Analysis, John Wiley
& Sons, New York, NY, 1999.
153. J. E. Heebner, V. Wong, A. Schweinsberg, R. W. Boyd, and D. J. Jackson,
“Optical transmission characteristics of fiber ring resonators,” IEEE Journal
of Quantum Electronics, vol. 40, pp. 726–730, 2004.
154. La difusion del la lumiere pa les atoines, vol. 2, 1927.
252 Bibliography

155. R. Kronig, “On the theory of the dispersion of X-rays,” Journal of the Optical
Society of America, vol. 12, pp. 547–557, 1926.
156. M. Beck, I. A. Walmsley, and J. D. Kafka, “Group delay measurements of
optical components near 800nm,” IEEE Journal of Quantum Electronics,
vol. 27, pp. 2074–2081, 1991.
157. R. H. J. Kop, P. deVries, R. Sprik, and A. Lagendijk, “Kramers-Kronig rela-
tions for an interferometer,” Optics Communications, vol. 138, pp. 118–126,
158. L. J. Wang, “Causal ‘all-pass’ filters and Kramers-Kronig relations,” Optics
Communications, vol. 213, pp. 27–32, 2002.
159. A. S. Sedra and K. C. Smith, Microelectronic Circuits, Oxford University
Press, New York, NY, fourth edition, 1998.
160. B. E. Little, S. T. Chu, H. A. Haus, J. Foresi, and J. P. Laine, “Microring res-
onator channel dropping filters,” Journal of Lightwave Technology, vol. 15,
pp. 998–1005, 1997.
161. G. Barbarossa and A. M. Matteo, “Novel double-ring optical-guided-wave
Vernier resonator,” IEE Proceedings (Optoelectronics), vol. 144, pp. 203–208,
162. M. A. Popovic, T. Barwicz, M. R. Watts, P. T. Rakich, L. Socci, E. P. Ippen,
F. X. Kärtner, and H. I. Smith, “Multistage high-order microring-resonator
add-drop filters,” Optics Letters, vol. 31, 2006.
163. B. E. Little, S. T. Chu, J. V. Hryniewicz, and P. P. Absil, “Filter synthesis
for periodically coupled microring resonators,” Optics Letters, vol. 25, pp.
344–346, 2000.
164. G. Griffel, “Vernier effect in asymmetrical ring resonator arrays,” IEEE
Photonics Technology Letters, vol. 12, pp. 1642–1644, 2000.
165. A. Melloni, “Synthesis of a parallel-coupled ring-resonator filter,” Optics
Letters, vol. 26, pp. 917–919, 2001.
166. F. Niklaus, P. Enoksson, E. Kälvesten, and G. Stemme, “Low-temperature full
wafer adhesive bonding,” Journal of Micromechanics and Microengineering,
vol. 11, pp. 100–107, 2001.
167. P. A. Moskowitz, H. L. Yeh, and S. K. Ray, “Thermal dry process soldering,”
Journal of Vacuum Science & Technology A, vol. 4, pp. 838–840, 1985.
168. I. H. Tan, C. Reaves, A. L. Holmes Jr., E. L. Hu, J. E. Bowers, and S. DenBaars,
“Low-temperature Pd bonding of III–V semiconductors,” Electronics Letters,
vol. 31, pp. 588–589, 1995.
169. G. R. Dohle, T. J. Drabik, J. J. Callahan, and K. P. Martin, “Low tempera-
ture bonding of epitaxial lift off devices with AuSn,” IEEE Transactions on
Components, Packaging, and Manufacturing Technology, Part B, vol. 19, pp.
575–580, 1996.
170. W. S. Wong, A. B. Wengrow, Y. Cho, A. Salleo, N. J. Quitoriano, N. W. Cheung,
and T. Sands, “Integration of GaN thin films with dissimilar substrate
materials by Pd-In metal bonding and laser lift-off,” Journal of Electronic
Materials, vol. 28, pp. 1409–1413, 1999.
171. W. S. Wong, T. Sands, N. W. Cheung, M. Kneissl, D. P. Bour, P. Mei, L. T.
Romano, and N. M. Johnson, “Inx Ga1−x N light emitting diodes on Si sub-
strates fabricated by Pd-In metal bonding and laser lift-off,” Applied Physics
Letters, vol. 77, pp. 2822–2824, 2000.
172. M. A. Raburn, B. Liu, K. Rauscher, Y. Okuno, N. Dagli, and J. E. Bowers, “3-D
photonic circuit technology,” IEEE Journal of Selected Topics in Quantum
Electronics, vol. 8, pp. 935–942, 2002.
Bibliography 253

173. P. P. Absil, Microring resonators for wavelength division multiplexing and

integrated photonics applications, Ph. D. dissertation, University of Mary-
land, College Park, 2000.
174. R. Grover, Indium phosphide based optical micro-ring resonators, Ph. D. dis-
sertation, Department of Electrical and Computer Engineering, University
of Maryland, College Park, 2003.
175. K. Djordjev, S.-J. Choi, S.-J. Choi, and P. D. Dapkus, “Active semiconductor
microdisk devices,” Journal of Lightwave Technology, vol. 20, pp. 105–113,
176. S. Adachi and K. Oe, “Linear electro-optic effects in zincblende-type semi-
conductors: Key properties of InGaAsP relevant to device design,” Journal
of Applied Physics, vol. 56, pp. 74–80, 1984.
177. S. Adachi and K. Oe, “Quadratic electro-optic (Kerr) effects in zincblende-
type semiconductors: Key properties of InGaAsP relevant to device design,”
Journal of Applied Physics, vol. 56, pp. 1499–1504, 1984.
178. R. W. Boyd, Nonlinear Optics, Academic Press, San Diego, CA, 1992.
179. D. A. B. Miller, D. S. Chemla, and S. Schmitt-Ring, “Relation between elec-
troabsorption in bulk semiconductors and in quantum wells: The quantum-
confined Franz-Keldysh effect,” Physical Review B, vol. 33, pp. 6976–6982,
180. J. P. Donnelly, H. Q. Le, E. A. Swanson, S. H. Groves, A. Darwish, and
E. P. Ippen, “Nondegenerate four-wave mixing wavelength conversion in
low-loss passive InGaAsP-InP quantum well waveguides,” IEEE Photonics
Technology Letters, vol. 8, pp. 623–625, 1996.
181. J. D. Jackson, Classical Electrodynamics, John Wiley & Sons, New York, NY,
third edition, 1999.
182. S. L. Chuang, Physics of Optoelectronic Devices, John Wiley & Sons, New
York, NY, 1995.
183. D. A. B. Miller, D. S. Chemla, T. C. Damen, A. C. Gossard, W. Wiegmann, T. H.
Wood, and C. A. Burrus, “Band-edge electroabsorption in quantum well
structures: the quantum-confined Stark effect,” Physical Review Letters,
vol. 53, pp. 2173–2176, 1984.
184. M. Fetterman, C.-P. Chao, and S. R. Forrest, “Fabrication and analysis of
high-contrast InGaAsP-InP Mach-Zehnder modulators for use at 1.55 µm
wavelength,” IEEE Photonics Technology Letters, vol. 8, pp. 69–71, 1996.
185. J. E. Zucker, I. Bar-Joseph, G. Sucha, U. Koren, B. I. Miller, and D. S.
Chemla, “Electrorefraction in GaInAsP/InP multiple quantum well het-
erostructures,” Electronics Letters, vol. 24, pp. 458–460, 1988.
186. M. J. Bloemer and K. Myneni, “Electro-optic properties near the absorp-
tion edge of GaAs/AlGaAs multiple-quantum-well waveguides,” Journal of
Applied Physics, vol. 74, pp. 4849–4859, 1993.
187. S. T. Ho, C. E. Soccolich, M. N. Islam, W. S. Hobson, A. F. J. Levi, and R. E.
Slusher, “Large nonlinear phase shifts in low-loss AlGaAs waveguides near
half-gap,” Applied Physics Letters, vol. 59, pp. 2558–2560, 1991.
188. G. I. Stegeman, A. Villeneuve, J. Kang, J. S. Aitchison, C. N. Ironside,
K. Al-Hemyari, C. C. Yang, C. H. Lin, H. H. Lin, G. T. Kennedy, R. S. Grant,
and W. Sibbett, “AlGaAs below the half-gap: The silicon of nonlinear op-
tical materials,” International Journal of Nonlinear Optical Physics, vol. 3,
pp. 347–371, 1994.
189. S. Spalter, H. Y. Wang, J. Zimmermann, G. Lenz, T. Katsufuji, S. W. Cheong,
and R. E. Slusher, “Strong self-phase modulation in planar chalcogenide
glass waveguides,” Optics Letters, vol. 27, pp. 363–365, 2002.
254 Bibliography

190. J. M. Harbold, F. O. Ilday, F. W. Wise, J. S. Sanghera, V. Q. Nguyen, L. B.

Shaw, and I. D. Aggarwal, “Highly nonlinear As-S-Se glasses for all-optical
switching,” Optics Letters, vol. 27, pp. 119–121, 2002.
191. M. Sheik-Bahae, D. J. Hagan, and E. W. van Stryland, “Dispersion and band-
gap scaling of the electronic Kerr effect in solids associated with two-
photon absorption,” Physical Review Letters, vol. 65, pp. 96–99, 1990.
192. M. Sheik-Bahae, D. C. Hutchings, D. J. Hagan, and E. W. van Stryland, “Dis-
persion of bound electronic nonlinear refraction in solids,” IEEE Journal of
Quantum Electronics, vol. 27, pp. 1296–1306, 1991.
193. J. E. Heebner, N. N. Lepeshkin, A. Schweinsberg, G. W. Wicks, R. W. Boyd,
R. Grover, and P.-T. Ho, “Enhanced linear and nonlinear optical phase re-
sponse of AlGaAs microring resonators,” Optics Letters, vol. 29, pp. 769–
771, 2004.
194. V. Mizrahi, K. W. DeLong, G. I. Stegeman, M. A. Saifi, and M. J. Andrejco,
“Two photon absorption as a limitation to all-optical switching,” Optics
Letters, vol. 14, pp. 1140–1142, 1989.
195. J. E. Heebner, Nonlinear Optical Whispering Gallery Microresonators for
Photonics, Ph. D. dissertation, University of Rochester, 2003.
196. A. Villeneuve, C. C. Yang, P. G. J. Wigley, G. I. Stegeman, J. S. Aitchison, and
C. N. Ironside, “Ultrafast all-optical switching in semiconductor nonlinear
directional couplers at half the band gap,” Applied Physics Letters, vol. 61,
pp. 147–149, 1992.
197. Y. Chen, G. Pasrija, B. Farhang-Boroujeny, and S. Blair, “Engineering the
nonlinear phase shift,” Optics Letters, vol. 28, pp. 1945–1947, 2003.
198. H. M. Gibbs, Optical bistability: controlling light with light, Academic Press
Inc., New York, 1985.
199. S. Blair, J. E. Heebner, and R. W. Boyd, “Beyond the absorption-limited
nonlinear phase shift with microring resonators,” Optics Letters, vol. 27,
pp. 357–359, 2002.
200. V. R. Almeida, C. A. Barrios, R. R. Panepucci, M. Lipson, M. A. Foster, D. G.
Ouzounov, and A. L. Gaeta, “All-optical switching on a silicon chip,” Optics
Letters, vol. 29, pp. 2867–2869, 2004.
201. S. M. Sze, Physics of Semiconductor Devices, John Wiley & Sons, second
edition, 1981.
202. M. B. Yairi and D. A. B. Miller, “Equivalence of diffusive conduction and giant
ambipolar diffusion,” Journal of Applied Physics, vol. 91, pp. 4374–4381,
203. A. Miller, “Transient grating studies of carrier diffusion and mobility in
semiconductors,” in Nonlinear Optics in Semiconductors II, E. Garmire and
A. Kost, Eds., vol. 59 of Semiconductors and Semimetals, pp. 287–312. Aca-
demic Press, San Diego, CA, 1999.
204. E. Kreyszig, Advanced Engineering Mathematics, John Wiley & Sons, New
York, NY, eighth edition, 1999.
205. G. I. Stegeman and C. T. Seaton, “Nonlinear integrated optics,” Journal of
Applied Physics, vol. 58, pp. R57–R78, 1985.
206. S. D. Smith, “Optical bistability, photonic logic, and optical computation,”
Applied Optics, vol. 25, pp. 1550–1564, 196.
207. B. E. Little and S. T. Chu, “Toward very large-scale integrated photonics,”
Optics & Photonics News, vol. 11, pp. 24–29, 2000.
208. C. C. Yang, A. Villeneuve, G. I. Stegeman, C. H. Lin, and H. Lin, “Measure-
ments of two-photon absorption coefficient and induced nonlinear refrac-
tive index in GaAs/AlGaAs multiquantum well waveguides,” Electronics
Letters, vol. 29, pp. 37, 1993.
Bibliography 255

209. C. Coriasso, D. Campi, C. Carcciatore, L. Faustini, L. Rigo, and A. Stano, “All

optical switching and pulse routing in a distributed-feedback waveguide
device,” Optics Letters, vol. 23, pp. 183–185, 1998.
210. T. A. Ibrahim, V. Van, and P.-T. Ho, “All-optical time division demultiplexing
and spatial pulse routing using a GaAsAlGaAs microring resonator,” Optics
Letters, vol. 27, pp. 803–805, 2002.
211. T. A. Ibrahim, R. Grover, L.-C. Kuo, S. Kanakaraju, L. C. Calhoun, and P.-T.
Ho, “All-optical AND/NAND logic gates using semiconductor microres-
onators,” IEEE Photonics Technology Letters, vol. 15, pp. 1422–1424, 2003.
212. V. Van, T. A. Ibrahim, P. P. Absil, F. G. Johnson, R. Grover, and P.-T. Ho,
“Optical signal processing using nonlinear semiconductor microring res-
onators,” IEEE Journal of Selected Topics in Quantum Electronics, vol. 8, pp.
705–713, 2002.
213. J. H. Bechtel and W. L. Smith, “Two-photon absorption in semiconductors
with picosecond laser pulses,” Physical Review B, vol. 13, pp. 3515–3522,
214. T. A. Ibrahim, K. Amarnath, L. C. Kuo, R. Grover, V. Van, and P.-T. Ho, “Pho-
tonic logic NOR gate based on two symmetric microring resonators,” Optics
Letters, vol. 29, pp. 2779–2781, 2004.
215. G. L. Fisher, R. W. Boyd, R. J. Gehr, S. A. Jenekhe, J. A. Osaheni, J. E. Sipe, and
L. A. Weller-Brophy, “Enhanced nonlinear optical response of composite
materials,” Physical Review Letters, vol. 74, pp. 1871–1875, 1995.
216. N. N. Lepeshkin, A. Schweinsberg, G. Piredda, R. S. Bennink, and R. W. Boyd,
“Enhanced nonlinear optical response of one-dimensional metal-dielectric
photonic crystals,” Physical Review Letters, vol. 93, pp. 123902, 2004.
217. E. Yablanovitch, “Inhibited spontaneus emission in solid-state physics and
electronics,” Physical Review Letters, vol. 58, pp. 2059–2062, 1987.
218. S. John, “Strong localization of photons in certain disordered dielectric
superlattices,” Physical Review Letters, vol. 58, pp. 2486–2489, 1987.
219. J. D. Joannopoulos, R. D. Meade, and J. N. Winn, Photonic Crystals, Princeton
University Press, Singapore, 1995.
220. G. Lenz, B. J. Eggleton, C. K. Madsen, and R. E. Slusher, “Optical delay lines
based on optical filters,” IEEE Journal of Quantum Electronics, vol. 37, pp.
525–532, 2001.
221. V. E. Zakharov and A. B. Shabat, “Interaction between solutions in a stable
medium,” Soviet Physics JETP, vol. 37, pp. 823–828, 1973.
222. A. Hasegawa and F. Tappert, “Transmission of stationary nonlinear optical
pulses in dispersive dielectric fibers,” Applied Physics Letters, vol. 23, pp.
142–144, 1973.
223. L. F. Mollenauer, R. H. Stolen, and J. P. Gordon, “Experimental observation of
picosecond pulse narrowing and solitons in optical fibers,” Physical Review
Letters, vol. 45, pp. 1095–1099, 1980.
224. J. E. Heebner, R. W. Boyd, and Q. Park, “Slow light, induced dispersion, en-
hanced nonlinearity, and optical solitons in a resonator-array waveguide,”
Physical Review E, vol. 65, pp. 036619, 2002.
225. E. Golovchenko, E. Dianov, A. Prokhorov, and V. Serkin, “Decay of optical
solitons,” JETP Letters, vol. 42, pp. 87–91, 1985.
226. C. V. Bennett and B. H. Kolner, “Upconversion time microscope demonstrat-
ing 103x magnification of femtosecond waveforms,” Optics Letters, vol. 24,
pp. 783–785, 1999.
227. Y. Chen and S. Blair, “Nonlinear phase shift of cascaded microring res-
onators,” Journal of the Optical Society of America B, vol. 20, pp. 2125–
2132, 2003.
256 Bibliography

228. G. Lenz, J. Zimmermann, T. Katsufuji, M. E. Lines, H. Y. Wang, S. Spalter,

R. E. Slusher, S. W. Cheong, J. S. Sanghera, and I. D. Aggarwal, “Large Kerr
effect in bulk Se-based chalcogenide glasses,” Optics Letters, vol. 25, pp.
254–256, 2000.
229. D. Rafizadeh, J. P. Zhang, R. C. Tiberio, and S. T. Ho, “Propagation loss
measurements in semiconductor microcavity ring and disk resonators,”
Journal of Lightwave Technology, vol. 16, pp. 1308–1314, 1998.
230. R. W. Boyd and D. J. Gauthier, Slow and Fast Light, Progress in Optics,
Elsevier Science, Amsterdam, 2002.
231. C. Liu, Z. Dutton, C. H. Behroozi, and L. V. Hau, “Observation of coherent
optical information storage in an atomic medium using halted light pulses,”
Nature, vol. 409, pp. 490–493, 2001.
232. D. F. Phillips, M. Fleischhauer, A. Mair, R. L. Walsworth, and M. D. Lukin,
“Storage of light in atomic vapor,” Physical Review Letters, vol. 86, pp.
783–786, 2001.
233. S. Chu and S. Wong, “Linear pulse-propagation in an absorbing medium,”
Physical Review Letters, vol. 48, pp. 738–741, 1982.
234. S. E. Harris, “Electromagnetically induced transparency,” Physics Today,
vol. 50, pp. 36–42, 1997.
235. M. S. Bigelow, N. N. Lepeshkin, and R. W. Boyd, “Observation of ultraslow
light propagation in a ruby crystal at room temperature,” Physical Review
Letters, vol. 90, pp. 113903, 2003.
236. M. S. Bigelow, N. N. Lepeshkin, and R. W. Boyd, “Superluminal and slow light
propagation in a room-temperature solid,” Science, vol. 301, pp. 200–202,
237. M. D. Lukin, M. Fleischhauer, A. S. Zibrov, H. G. Robinson, V. L. Velichansky,
L. Hollberg, and M. O. Scully, “Spectroscopy in dense coherent media: Line
narrowing and interference effects,” Physical Review Letters, vol. 79, pp.
2959–2962, 1997.
238. A. B. Matsko, Y. V. Rostovtsev, H. Z. Cummins, and M. O. Scully, “Using slow
light to enhance acousto-optical effects: Application to squeezed light,”
Physical Review Letters, vol. 84, pp. 5752–5755, 2000.
239. A. B. Matsko, Y. V. Rostovtsev, M. Fleischhauer, and M. O. Scully, “Anom-
alous stimulated brillouin scattering via ultraslow light,” Physical Review
Letters, vol. 86, pp. 2006–2009, 2001.
240. K. Djordjev, S. Choi, S. Choi, and P. D. Dapkus, “Vertically coupled InP
microdisk switching devices with electroabsorptive active regions,” IEEE
Photonics Technology Letters, vol. 14, pp. 1115–1117, 2002.
241. R. Y. Chiao and A. M. Steinberg, Progress in Optics XXXVII, Elsevier Science,
Amsterdam, 1997.
242. J. E. Heebner, R. W. Boyd, and Q. Park, “‘Slow’ and ‘fast’ light in resonator-
coupled waveguides,” Journal of Modern Optics, vol. 49, pp. 2629–2636,
243. W. Chen and D. L. Mills, “Gap solitons and the nonlinear optical response
of superlattices,” Physical Review Letters, vol. 58, pp. 160–163, 1987.
244. B. J. Eggleton, R. E. Slusher, C. M. de Sterke, P. A. Krug, and J. E. Sipe, “Bragg
grating solitons,” Physical Review Letters, vol. 76, pp. 1627–1630, 1996.
245. D. N. Christodoulides and N. K. Efremidis, “Discrete temporal solitons along
a chain of nonlinear coupled microcavities embedded in photonic crystals,”
Optics Letters, vol. 27, pp. 568–570, 2002.
246. M. Notomi, K. Yamada, A. Shinya, J. Takahashi, C. Takahashi, and
I. Yokohama, “Extremely large group-velocity dispersion of line-defect
waveguides in photonic crystal slabs,” Physical Review Letters, vol. 87, pp.
253902, 2001.
Bibliography 257

247. M. Soljacic, S. G. Johnson, S. Fan, M. Ibanescu, E. Ippen, and J. D. Joannopou-

los, “Photonic-crystal slow-light enhancement of nonlinear phase sensitiv-
ity,” Journal of the Optical Society of America B, vol. 19, pp. 2052–2059,
248. M. D. Rahn, A. M. Fox, M. S. Skolnick, and T. F. Krauss, “Propagation of
ultrashort nonlinear pulses through two-dimensional AlGaAs high-contrast
photonic crystal waveguides,” Journal of the Optical Society of America B,
vol. 19, pp. 716–721, 2002.
249. S. Mingaleev and Y. Kivshar, “Nonlinear transmission and light localization
in photonic-crystal waveguides,” Journal of the Optical Society of America
B, vol. 19, pp. 2241–2249, 2002.
250. J. E. Sipe, N. A. R. Bhat, P. Chak, and S. Pereira, “Effective field theory for
the nonlinear optical properties of photonic crystals,” Physical Review E.,
vol. 69, pp. 016604, 2004.
251. B. E. Little, S. T. Chu, and H. A. Haus, “Second-order filtering and sensing
with partially coupled traveling waves in a single resonator,” Optics Letters,
vol. 23, pp. 1570–1572, 1998.
252. J. E. Heebner, P. Chak, S. Perera, J. E. Sipe, and R. W. Boyd, “Distributed and
localized feedback in microresonator sequences for linear and nonlinear
optics,” Journal of the Optical Society of America B, vol. 21, pp. 1818–1832,
253. R. Orta, P. Savi, R. Tascone, and D. Trinchero, “Synthesis of multiple-ring-
resonator filters for optical systems,” IEEE Photonics Technology Letters,
vol. 7, pp. 1447–1449, 1995.
254. A. Yariv, X. Yong, R. K. Lee, and A. Soberer, “Coupled-resonator optical
waveguide: a proposal and analysis,” Optics Letters, vol. 24, pp. 711–713,
255. Y. Xu, R. K. Lee, and A. Yariv, “Scattering theory analysis of waveguide-
resonator coupling,” Physical Review E, vol. 62, pp. 7389–7404, 2000.
256. S. Mookherjea, D. S. Cohen, and A. Yariv, “Nonlinear dispersion in a
coupled-resonator optical waveguide,” Optics Letters, vol. 27, pp. 933–935,
257. W.-Y. Chen, R. Grover, T. A. Ibrahim, V. Van, W. N. Herman, and P.-T. Ho,
“High-finesse laterally-coupled single-mode benzocyclobutene micro-ring
resonators,” IEEE Photonics Technology Letters, vol. 16, pp. 470–472, 2004.
258. Y. Xu, R. K. Lee, and A. Yariv, “Propagation and second-harmonic genera-
tion of electromagnetic waves in a coupled-resonator optical waveguide,”
Journal of the Optical Society of America B, vol. 17, pp. 387–400, 2000.
259. Z. Wang and S. Fan, “Compact all-pass filters in photonic crystals as the
building block for high-capacity,” Physical Review E., vol. 68, pp. 066616,
260. S. Pereira, J. E. Sipe, J. E. Heebner, and R. W. Boyd, “Gap solitons in a two-
channel SCISSOR structure,” Optics Letters, vol. 27, pp. 536–538, 2002.
261. P. Chak, S. Pereira, and J. Sipe, “Lorentzian model for nonlinear switching
in a microresonator structure,” Optics Communications, vol. 213, pp. 163–
171, 2002.
262. G. Griffel, “Synthesis of optical filters using ring resonator arrays,” IEEE
Photonics Technology Letters, vol. 12, pp. 810–812, 2000.
263. A. Melloni, F. Morichetti, and M. Martinelli, “Optical slow wave structures,”
Optics and Photonics News, vol. 14, pp. 44–48, 2003.
264. C. K. Madsen, G. Lenz, A. J. Bruce, M. A. Cappuzzo, L. T. Gomez, and R. E.
Scotti, “Integrated all-pass filters for tunable dispersion and dispersion
slope compensation,” IEEE Photonics Technology Letters, vol. 11, pp. 1623–
1625, 1999.
258 Bibliography

265. J. Khurgin, “Expanding the bandwidth of slow-light photonic devices based

on coupled resonators,” Optics Letters, vol. 30, pp. 513–515, 2005.
266. G. Lenz, B. J. Eggleton, C. R. Giles, C. K. Madsen, and R. E. Slusher, “Dis-
persive properties of optical filters for WDM systems,” IEEE Journal of
Quantum Electronics, vol. 34, pp. 1390–1402, 1998.
267. P. Chak, J. Sipe, and S. Pereira, “Depositing light in a photonic stop gap by
use of Kerr nonlinear microresonators,” Optics Letters, vol. 28, pp. 1966–
1968, 2003.
268. S. Pereira, P. Chak, and J. E. Sipe, “Gap-soliton switching in short microres-
onator structures,” Journal of the Optical Society of America B, vol. 19, pp.
2191–2202, 2002.
269. S. Blair, “Nonlinear sensitivity enhancement with one-dimensional pho-
tonic bandgap microcavity arrays,” Optics Letters, vol. 27, pp. 613–615,
270. S. Adachi, “Material parameters of In1−x Gax Asy P1−y and related binaries,”
Journal of Applied Physics, vol. 53, pp. 8775–8792, 1982.
271. S. Adachi, “Refractive indices of III–V compounds Key properties of In-
GaAsP relevant to device design,” Journal of Applied Physics, vol. 53, pp.
5863–5869, 1982.
272. S. Adachi, “Optical properties of InGaAsP: discussion,” in Properties of
Indium Phosphide, London, UK, 1989, EMIS Datareviews Series, IEE, Inspec.
273. S. Adachi, Physical Properties of III–V Semiconductor Compounds, John
Wiley & Sons, New York, NY, 1992.
274. S. Adachi, Properties of Aluminum Gallium Arsenide, EMIS Datareviews
Series. IEE, Inspec, London, UK, 1993.
275. R. J. Deri and M. A. Emanuel, “Consistent formula for the refractive index
of Alx Ga1−x As below the band edge,” Journal of Applied Physics, vol. 77,
pp. 4667–4672, 1995.
276. M. V. Sullivan and G. A. Kolb, “The chemical polishing of gallium arsenide
in bromine-methanol,” Journal of the Electrochemical Society, vol. 110, pp.
585–587, 1963.
277. S. B. Phatak and G. Kelner, “Material selective etching in the system In-
GaAsP/InP,” Journal of the Electrochemical Society, vol. 126, pp. 287–292,
278. G. C. DeSalvo, W. F. Tseng, and J. Comas, “Etch rates and selectivi-
ties of citric acid/hydrogen peroxide on GaAs, Al0.3 Ga0.7 As, In0.3 Ga0.8 As,
In0.53 Ga0.47 As, In0.52 As0.48 As and InP,” Journal of the Electrochemical Soci-
ety, vol. 139, pp. 831–835, 1992.
279. H. Fourre, F. Diette, and A. Cappy, “Selective wet etching of lattice-matched
InGaAs/InAlAs on InP and metamorphic InGaAs/InAlAs on GaAs using suc-
cinic acid/hydrogen peroxide solution,” Journal of Vacuum Science & Tech-
nology B, vol. 14, pp. 3400–3402, 1996.
280. F. G. Johnson, O. S. King, F. Seiferth, D. R. Stone, R. D. Whaley, M. Dagenais,
and Y. J. Chen, “Solid source molecular beam epitaxy of low threshold
1.55 µm wavelength GaInAs/GaInAsP/InP semiconductor lasers,” Journal
of Vacuum Science & Technology B, vol. 14, pp. 2753–2756, 1996.
281. F. G. Johnson, O. S. King, F. Seiferth, S. Horst, D. R. Stone, R. D. Whaley,
M. Dagenais, and Y. J. Chen, “Solid source MBE growth and regrowth of
1.55 µm wavelength GaInAsP/InP ridge lasers,” Journal of Crystal Growth,
vol. 175, pp. 46–51, 1997.
282. J. V. Hryniewicz, Y. J. Chen, S. H. Hsu, C. H. D. Lee, and G. A. Porkolab,
“Ultrahigh vacuum chemically assisted ion beam etching system with a
three grid ion source,” Journal of Vacuum Science & Technology A, vol. 15,
pp. 616–621, 1997.
Bibliography 259

283. S. Agarwala, S. C. Horst, O. S. King, R. A. Wilson, D. R. Stone, M. Dagenais,

and Y. J. Chen, “High-density inductively coupled plasma etching of
GaAs/AlGaAs in BCl3 /Cl2 /Ar: A study using a mixture design experiment,”
Journal of Vacuum Science & Technology B, vol. 16, pp. 511–514, 1998.
284. S. Agarwala, O. S. King, S. C. Horst, R. A. Wilson, D. R. Stone, M. Dagenais, and
Y. J. Chen, “Response surface study of inductively coupled plasma etching
of GaAs/AlGaAs in BCl3 /Cl2 ,” Journal of Vacuum Science & Technology A,
vol. 17, pp. 52–55, 1999.
285. R. D. Whaley, B. Gopalan, M. Dagenais, R. D. Gomez, F. G. Johnson,
S. Agarwala, O. King, and D. R. Stone, “Use of atomic force microscopy
for analysis of high performance InGaAsP/InP semiconductor lasers with
dry-etched facets,” Journal of Vacuum Science and Technology B, vol. 16,
pp. 1007–1011, 1998.
286. R. Grover, J. V. Hryniewicz, O. S. King, and V. Van, “Process development
of methane-hydrogen-argon-based deep dry-etching of InP for high aspect-
ratio structures with vertical facet-quality sidewalls,” Journal of Vacuum
Science & Technology B, vol. 19, pp. 1694–1698, 2001.
287. F. G. Johnson, O. S. King, J. V. Hryniewicz, L. G. Joneckis, S. T. Chu, and D. M.
Gill, “Use of deuterated gases for the vapor deposition of thin films for low-
loss optical devices and waveguides,” United States Patent, vol. 6,614,977,
288. M. A. Raburn, B. Liu, P. Abraham, and J. E. Bowers, “Double-bonded InP-
InGaAsP vertical coupler 1 : 8 beam splitter,” IEEE Photonics Technology
Letters, vol. 12, pp. 1639–1641, 2000.
289. M. Raburn, Three-dimensional wafer bonded indium phosphide photonic
waveguide devices, Ph.D. thesis, Department of Electrical and Computer
Engineering, University of California, Santa Barbara, 2003.
290. S. R. Sakamoto, C. Ozturk, Y. T. Byun, J. Ko, and N. Dagli, “Low-loss
substrate-removed (SURE) optical waveguides in GaAs-AlGaAs epitaxial lay-
ers embedded in organic polymers,” IEEE Photonics Technology Letters, vol.
10, pp. 985–987, 1998.
291. R. Williams, Modern GaAs Processing Methods, Artech House, Boston, MA,
292. P. Rabiei, Electro-optic and thermo-optic polymer micro-ring resonators and
their applications, Ph.D. thesis, University of Southern California, 2003.
293. Y. Chen and S. Blair, “Nonlinearity enhancement in finite coupled-resonator
slow-light waveguides,” Optics Express, vol. 12, pp. 3353–3366, 2004.
294. A. Driessen, D. H. Geuzebroek, H. J. W. M. Hoekstra, H. Kelderman, E. J.
Klein, D. J. W. Klunder, C. G. H. Roeloffzen, F. S. Tan, E. Krioukov, and
C. Otto, “Microresonators as building blocks for VLSI photonics,” in Sum-
mer School on Microresonators as building blocks for VLSI Photonics, Ettore
Majorana Centre for Scientific Culture, International School of Quantum
Electronics, Erice, Sicily, Italy, 2003.
295. F. Morichetti, R. Siano, A. Boletti, and A. Melloni, “Optical integrated
receiver for DQPSK systems,” in Transparent Optical Networks, 2005.
296. J. E. Heebner, T. C. Bond, and J. S. Kallman, “Generalized formulation
for performance degradations due to bending and edge scattering loss in
microdisk resonators,” Optics Express, vol. 15, pp. 4452–4473, 2007.

Q beam propagation, 25
– definition, 89
– disks, 48 cavity QED, 6
– rings, 48 coupling
– spheres, 48 – coupled wave formalism, 30
Q switching, 6 – optimization of, 31
– perturbation solutions, 27
active LiNbO3 , 3 – scattering matrix, 31
active resonators – TE planar waveguides, 29
– carrier injection, 110 coupling of waveguides, see coupling
– electro-optic, 110 CROW, 202
– need for, 107 – bandgap engineering, 210
– thermal tuning, 110
add–drop rings, 84 delay lines, tunable, 195
– all-optical switching, 137 dispersion relations, 18
– finesse, 89 – WGM, 43
– free spectral range, 89 distributed microresonator systems,
– intensity buildup, 85 175
– resonance width, 87 – linear propagation, 175
all-optical switching, 6, 149
– add–drop, see add–drop rings effective index method, 23
– all-pass, see all-pass rings electro-optic effect, 110
– Fabry–Perot, 137
– pump reshaping, 156 fabrication
– pump-probe experiments, 154, 159 – feature definition, 221
– pump-probe simulations, 152 – lateral active III–V, 234
– response time measurement, 160 – lateral passive III–V, 225
– ring-enhanced MZ, 137 – multilayer, 222
– routing, 161 – polymer rings, 242
– theory, 149 – vertical passive III–V, 227
– thresholding, 156 – wafer bonding, 222
– time-division mux/demux, 158 fast light, 191, 197
all-pass rings, 74, 124 figures of merit, 142
– all-optical switching, 137 filters, 97
– attenuation in, 83 – parallel-cascade of rings, 99
– finesse, 77 – practical limitations to cascading,
– group delay, 79 103
– group delay dispersion, 79 – serial-cascade of rings, 99
– intensity buildup, 75 – vernier effect, 103
– phase shift in, 78 finesse
– enhanced nonlinearities, 124

262 Index

finite element method, 25 Mach–Zehnder interferometer, 130

finite-difference time-domain, 26 Marcatili’s method, 22
four-wave mixing, 123 material systems
– resonator-enhanced, 147 – fibers, 3
– SCISSOR, 187 – glass, 3, 220
Franz–Keldysh effect, 111 – III–V semiconductors, see III–V
free spectral range semiconductors
– add–drop rings, 89 – ion-exchanged glass, 3
– PMMA on quartz, 3
GaAs-AlGaAs, see III–V semiconduc- – polymers, 3, 242
tors – polyurethane on glass, 3
GaInAsP-InP, see III–V semiconductors – silica, 3
Goell’s method, 24 multi-ring devices, 175
multistability, 136
III–V semiconductors
– active, see active resonators nonlinear figures of merit, 142
– advantages, 218 nonlinear optics
– direct bandgap, 219 – all-pass rings, see all-pass rings
– feature definition, 221 – bandwidth vs. nonlinear response,
– GaAs-AlGaAs, 107, 127, 128, 151, 125
159, 163, 167, 169 – challenges of, 6
– GaInAsP-InP, 107, 116, 164 – four-wave mixing, see four-wave
– history, 4 mixing
– lateral active, 116, 234 – in microrings, 5
– lateral coupling, 5, 105, 107 – inverted effective nonlinearity, 144
– lateral passive, 225 – Kerr effect, see Kerr effect
– multilayer fabrication, 222 – multistability, 136
– vertical coupling, 4, 105, 107 – phase shift, 124
– vertical passive, 227 – potential of, 6
– waveguide fabrication, 220 – pulse fidelity, 125
– waveguides, 218 – pulsed excitation, 125
intensity enhancement, 6 – resonant enhancement, 124
inverted effective nonlinearity, 144 – saturable absorption, see saturable
Kerr effect, 123, 127 – two-photon absorption, see
– experimental enhancement, 128 two-photon absorption
nonlinear saturation, 132
lateral coupling, 105 nonlinear susceptibility, 123
– relevance, 107 – four-wave mixing, see four-wave
loss mixing
– bending, 35 – Kerr effect, see Kerr effect
– by the volume current method, 50 – multistability, 136
– effect on nonlinear enhancement, – resonator enhanced χ (3) , 124
138 – saturable absorption, see saturable
– far field scattered power, 53 absorption
– mechanisms, 35 – two photon absorption, see
– normalized formulation for two-photon absorption
scattering loss, 61 normalized dispersion relations, 19
– radiation, 42 – WGM, 46
– scattering loss, 50
– spectral density formulation, 52 optical delay lines, 213
– substrate leakage, 35 optical delay lines, tunable, 195
– TE scattering loss, 56 optical logic, 6, 149, 163
Index 263

– (N)AND, 163 saturable absorption, 123

– NOR, 169 scattering loss, see loss
– speed improvement, 173 SCISSOR, 197
optical microresonators – bandgap engineering, 210
– applications, 1 – Bloch matrix, 200
– compared to Fabry–Perots, 2 – depth of phase, 189
– definition, 1, 2 – double-channel, 206
– fabrication, 217 – fast light, 191, 197
– history, 3 – four-wave mixing, 187
– integrated, 3 – group-velocity relations, 177
– materials, 217 – higher order dispersion, 179
– size matters, 2, 4 – induced self-steepening, 184
optical multistability, 6 – linear propagation, 175
– multistability, 186
parallel-cascaded rings, 99, 197 – nonlinear propagation, 179
– bandgap engineering, 210 – pulse compression, 185
– distributed resonator system, 175 – single-channel, 204
– group-velocity relations, 177 – slow light, 191
– higher order dispersion, 179 – solitons, 181
– induced self-steepening, 184 serial-cascaded rings, 99, 197, 202
– nonlinear propagation, 179 – bandgap engineering, 210
– solitons, 181 slow light, 191, 213
paraxial waveguiding equation, 12 solitons, 181
passive ring resonators, 105 – dark, 183
photonic integrated circuits, 107 squeezed light, 6
planar slab waveguide, 13 susceptibility, see nonlinear suscepti-
polymers, 219, 242 bility

quantum optics, 6 total internal reflection, 9

quantum-confined Stark effect, 111 two-beam a.c. amplification, 6
two-photon absorption, 124, 128, 142,
Raman interactions, 6 144
rectangular dielectric waveguides, 20
resonance, 1 vertical coupling, 105
resonator-enhanced MZ, 130
resonators, 71 wafer bonding, 222
– Q, 89 waveguide confinement, 9
– – disks, 48 waveguides, 9
– – rings, 48 – dispersion relations, 18
– – spheres, 48 – losses in, see loss
– add–drop rings, see add–drop rings – normalized dispersion relations,
– all-pass rings, see all-pass rings 19
– characterization, 89 – planar slab, 13
– Fabry–Perot, 71 – rectangular dielectric waveguides,
– finesse, 77 20
– Gires–Tournois, 73 whispering gallery modes, 1, 38
– high-order filters, 97 – dispersion relations, 43
– physical significance, finesse and Q, – normalized dispersion relations, 46
91 – radiation loss, 42
– ring, 73 – TE, 41
– TM, 39
Springer Series in
Volume 1
1 Solid-State Laser Engineering
By W. Koechner, 5th revised and updated ed. 1999, 472 figs., 55 tabs., XII, 746 pages

Published titles since Volume 85

85 Sensing with Terahertz Radiation
By D. Mittleman (Ed.), 2003, 207 figs., 14 tabs., XVI, 337 pages
86 Progress in Nano-Electro-Optics I
Basics and Theory of Near-Field Optics
By M. Ohtsu (Ed.), 2003, 118 figs., XIV, 161 pages
87 Optical Imaging and Microscopy
Techniques and Advanced Systems
By P. Török, F.-J. Kao (Eds.), 2003, 260 figs., XVII, 395 pages
88 Optical Interference Coatings
By N. Kaiser, H.K. Pulker (Eds.), 2003, 203 figs., 50 tabs., XVI, 504 pages
89 Progress in Nano-Electro-Optics II
Novel Devices and Atom Manipulation
By M. Ohtsu (Ed.), 2003, 115 figs., XIII, 188 pages
90/1 Raman Amplifiers for Telecommunications 1
Physical Principles
By M.N. Islam (Ed.), 2004, 488 figs., XXVIII, 328 pages
90/2 Raman Amplifiers for Telecommunications 2
Sub-Systems and Systems
By M.N. Islam (Ed.), 2004, 278 figs., XXVIII, 420 pages
91 Optical Super Resolution
By Z. Zalevsky, D. Mendlovic, 2004, 164 figs., XVIII, 232 pages
92 UV-Visible Reflection Spectroscopy of Liquids
By J.A. Räty, K.-E. Peiponen, T. Asakura, 2004, 131 figs., XII, 219 pages
93 Fundamentals of Semiconductor Lasers
By T. Numai, 2004, 166 figs., XII, 264 pages
94 Photonic Crystals
Physics, Fabrication and Applications
By K. Inoue, K. Ohtaka (Eds.), 2004, 209 figs., XV, 320 pages
95 Ultrafast Optics IV
Selected Contributions to the 4th International Conference
on Ultrafast Optics, Vienna, Austria
By F. Krausz, G. Korn, P. Corkum, I.A. Walmsley (Eds.), 2004, 281 figs., XIV, 506 pages
96 Progress in Nano-Electro Optics III
Industrial Applications and Dynamics of the Nano-Optical System
By M. Ohtsu (Ed.), 2004, 186 figs., 8 tabs., XIV, 224 pages
97 Microoptics
From Technology to Applications
By J. Jahns, K.-H. Brenner, 2004, 303 figs., XI, 335 pages
98 X-Ray Optics
High-Energy-Resolution Applications
By Y. Shvyd’ko, 2004, 181 figs., XIV, 404 pages
99 Mono-Cycle Photonics and Optical Scanning Tunneling Microscopy
Route to Femtosecond Ångstrom Technology
By M. Yamashita, H. Shigekawa, R. Morita (Eds.) 2005, 241 figs., XX, 393 pages
100 Quantum Interference and Coherence
Theory and Experiments
By Z. Ficek and S. Swain, 2005, 178 figs., XV, 418 pages
101 Polarization Optics in Telecommunications
By J. Damask, 2005, 110 figs., XVI, 528 pages
102 Lidar
Range-Resolved Optical Remote Sensing of the Atmosphere
By C. Weitkamp (Ed.), 161 figs., XX, 416 pages
103 Optical Fiber Fusion Splicing
By A.D. Yablon, 2005, 137 figs., XIII, 306 pages
104 Optoelectronics of Molecules and Polymers
By A. Moliton, 2005, 229 figs., 592 pages
105 Solid-State Random Lasers
By M. Noginov, 2005, 131 figs., XII, 238 pages
106 Coherent Sources of XUV Radiation
Soft X-Ray Lasers and High-Order Harmonic Generation
By P. Jaeglé, 2005, 150 figs., approx. 264 pages
107 Optical Frequency-Modulated Continuous-Wave (FMCW) Interferometry
By J. Zheng, 2005, 137 figs., XVIII, 254 pages
108 Laser Resonators and Beam Propagation
Fundamentals, Advanced Concepts and Applications
By N. Hodgson and H. Weber, 2005, 497 figs., approx. 790 pages
109 Progress in Nano-Electro Optics IV
Characterization of Nano-Optical Materials and Optical Near-Field Interactions
By M. Ohtsu (Ed.), 2005, 123 figs., XIV, 206 pages
110 Kramers–Kronig Relations in Optical Materials Research
By V. Lucarini, J.J. Saarinen, K.-E. Peiponen, E.M. Vartiainen,
2005, 37 figs., X, 162 pages
111 Semiconductor Lasers
Stability, Instability and Chaos
By J. Ohtsubo, 2005, 169 figs., XII, 438 pages
112 Photovoltaic Solar Energy Generation
By A. Goetzberger and V.U. Hoffmann, 2005, 139 figs., XII, 234 pages
113 Photorefractive Materials and Their Applications 1
Basic Effects
By P. Günter and J.P. Huignard, 2005, 169 figs., XIV, 426 pages
114 Photorefractive Materials and Their Applications 2
By P. Günter and J.P. Huignard, 2005, 370 figs., XVIII, 646 pages
115 Photorefractive Materials and Their Applications 3
By P. Günter and J.P. Huignard, 2005, 316 figs., X, 366 pages
116 Spatial Filtering Velocimetry
Fundamentals and Applications
By Y. Aizu and T. Asakura, 2006, 112 figs., XII, 212 pages
117 Progress in Nano-Electro-Optics V
Nanophotonic Fabrications, Devices, Systems, and Their Theoretical Bases
By Motoichi Ohtsu, 2006, 122 figs., 3 tables, 188 pages
118 Mid-infrared Semiconductor Optoelectronics
By A. Krier, 2006, 443 figs., 751 pages
119 Optical Interconnects
The Silicon Approach
By L. Pavesi and G. Guillot, 2006, 260 figs., 384 pages
120 Relativistic Nonlinear Electrodynamics
Interaction of Charged Particles with Strong and Super Strong Laser Fields
By H.K. Avetissian, 2006, 23 figs., XIII, 333 pages
121 Thermal Process Using Attosecond Laser Pulses
When Time Matters
By M. Kozlowski and J. Marciak-Kozlowska, 2006, 46 figs., XII, 232 pages
122 Modeling and Analysis of Transient Processes in Open Resonant Structures
New Methods and Techniques
By Y.K. Sirenko, S. Ström, N.P. Yashina, 2006, 110 figs., XIV, 362 pages
123 Wavelength Filters in Fibre Optics
By H. Venghaus, 2006, 210 figs., XXIV, 454 pages
124 Light Scattering by Systems of Particles
Null-Field Method with Discrete Sources - Theory and Programs
By A. Doicu, T. Wriedt and Y.A. Eremin, 2006, 119 figs., XIII, 324 pages
125 Electromagnetic and Optical Pulse Propagation 1
Spectral Representations in Temporally Dispersive Media
By K.E. Oughstun, 2006, 74 figs., XVI, 460 pages
126 Quantum Well Infrared Photodetectors
Physics and Applications
By H. Schneider, 2006, 153 figs., XVI, 250 pages
127 Integrated Ring Resonators
The Compendium
By D.G. Rabus, 2007, 243 figs., 258 pages
128 High-Power Diode-Lasers
Technology and Applications
By F. Bachmann, P. Loosen, and R. Poprave, 2006, 543 figs., VI, 554 pages
129 Laser Ablation and its Applications
By C. Phipps, 2007, 300 figs., XX, 588 pages
130 Concentrator Photovoltaics
By A. Luque and V. Andreev, 2007, 250 figs., XIV, 345 pages
131 Surface Plasmon Nanophotonics
By M.L. Brongersma and P.G. Kik, 2007, VII, 271 pages
132 Ultrafast Optics V
By S. Watanabe and K. Midorikawa, 2007, 339 figs., 562 pages
133 Frontiers in Surface Nanophotonics
Principles and Applications
By D.L. Andrews and Z. Gaburro, 2007, 89 figs., 176 pages
134 Strong Field Laser Physics
By T. Brabec, 2008, 150 figs., approx. 500 pages
135 Optical Nonlinearities in Chalcogenide Glasses and their Applications
By A. Zakery and S.R. Elliot, 2007, 78 figs., 202 pages
136 Optical Measurement Techniques
Innovations for Industry and the Life Sciences
By K.E. Peiponen, R. Myllyl Ɨ and A.V. Priezzhev, 2008, 60 figs., approx. 300 pages

137 Modern Developments in X-Ray and Neutron Optics

By A. Erko, M. Idir, T. Krist and A.G. Michette, 2008, 150 figs., approx. 584 pages
138 Optical Microresonators
Theory, Fabrication, and Applications
By J. Heebner, R. Grover, and T. Ibrahim, 2008, 100 figs., 280 pages