EditorinChief
William T. Rhodes Ferenc Krausz
Georgia Institute of Technology LudwigMaximiliansUniversität München
School of Electrical and Computer Engineering Lehrstuhl für Experimentelle Physik
Atlanta, GA 303320250, USA Am Coulombwall 1
Email: bill.rhodes@ece.gatech.edu 85748 Garching, Germany
and
MaxPlanckInstitut für Quantenoptik
HansKopfermannStrass 1
Editorial Board 85748 Garching, Germany
Email: ferenc.krausz@mpq.mpg.de
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 303320250, USA Linköping University
Email: adibi@ece.gatech.edu 58183 Linköping, Sweden
Email: bom@ifm.liu.se
Toshimitsu Asakura
HokkaiGakuen University Herbert Venghaus
Faculty of Engineering HeinrichHertzInstitut
11, Minami26, Nishi 11, Chuoku für Nachrichtentechnik Berlin GmbH
Sapporo, Hokkaido 0640926, Japan Einsteinufer 37
Email: asakura@eli.hokkaisu.ac.jp 10587 Berlin, Germany
Email: venghaus@hhi.de
Theodor W. Hänsch
MaxPlanckInstitut für Quantenoptik Horst Weber
HansKopfermannStrasse 1 Technische Universität Berlin
85748 Garching, Germany Optisches Institut
Email: t.w.haensch@physik.unimuenchen.de Strasse des 17. Juni 135
10623 Berlin, Germany
Takeshi Kamiya Email: weber@physik.tuberlin.de
Ministry of Education, Culture, Sports
Science and Technology
Harald Weinfurter
National Institution for Academic Degrees LudwigMaximiliansUniversität München
3291 Otsuka, Bunkyoku Sektion Physik
Tokyo 1120012, Japan Schellingstrasse 4/III
Email: kamiyatk@niad.ac.jp 80799 München, Germany
Email: harald.weinfurter@physik.
unimuenchen.de
John Heebner • Rohit Grover • Tarek Ibrahim
Optical Microresonators
Theory, Fabrication, and Applications
ABC
John Heebner Tarek Ibrahim
Lawrence Livermore National Laboratory RA3353
7000 East Ave Intel Corporation
Livermore, CA 94550 5200 NE Elam Young Parkway
heebner1@llnl.gov Hillsboro, OR 97124
tarek.a.ibrahim@intel.com
Rohit Grover
MS F1558
Intel Corporation
3585 SW 198th Ave
Aloha, OR 97007
rohit.grover@intel.com
9 8 7 6 5 4 3 2 1
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
ix
x Contents
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
1. Introduction
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 scalesize of microresonators currently achievable by state
oftheart 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 alloptical
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
Since the early eﬀorts outlined above, there have been numerous
works in various doped and undoped silicabased 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, temperatureinsensitive
operation and so on. Oda’s work with TiO2 doped silicaglass rings rep
resents the ﬁrst demonstration of serially cascaded rings, with increased
free spectral range over singlering 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 GaInAsPInP and III–Nitrides using the whisper
inggallery; 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 PingTong Ho’s group at the Laboratory for Phys
ical Sciences (LPS), College Park, MD, have demonstrated both laterally
and vertically coupled rings in GaAsAlGaAs acting as multi ring devices,
switches, routers, and mux/demux operation [53–61]. The GaInAsPInP
material system has proven problematic for passive microrings because
of processing diﬃculties resulting in high device losses. Nevertheless,
the ﬁrst vertically coupled passive InPbased 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].
n
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
1
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 corecladding interfaces of a planar slab waveguide of
thickness d. The dotted line represents the direction of the total wave
vector k1 . The wavevector 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 roundtrip between the
two corecladding 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 roundtrip 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
k
2d n21 k20 − β2m − 4 arctan ⎝ ⎠ = m2π
2 0
(2.6)
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
(2.8)
n2 n2 k2 − β2m
1 0
12 2. Optical Dielectric Waveguides
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 timeindependent Helmholtz equation
∂2U
+ ∇2T U + k2 U = 0, (2.12)
∂z2
∂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 timeindependent 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.
For TE polarized modes, Maxwell’s vector curl equations reduce from six
to three equations involving three ﬁeld components (Ey , Hx , Hz ):
∂Ey
= iωµ0 Hz (2.16)
∂x
∂Hx ∂Hz
− = −iωεEy (2.17)
∂z ∂x
that reduces to a single wave equation for the transverse electric ﬁeld
component:
∂ 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
∂2
− γx2 Ey = 0 (−d/2 < x < d/2) (2.20)
∂x 2
resulting in ﬁeld proﬁles for the three regions here for evenorder modes.2
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)
2
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:
neﬀ
Hx = − Ey (2.24)
Z0
i ∂Ey
Hz = − . (2.25)
k0 Z0 ∂x
The power per unit length carried by the mode can be obtained by inte
grating the timeaveraged Poynting vector (S = 21 {E × H∗ }) across the
waveguide dimension.
2
For oddorder modes, the cosine is replaced with a sine and the two cladding
amplitudes are oppositely signed.
2.3 The Planar Slab Waveguide 15
+∞
1
P/L =− dxEy Hx∗ (2.26)
2
−∞
+∞
2
neﬀ
= dx Ey (2.27)
2Z0
−∞
2
neﬀ Ey0 2
= d+ . (2.28)
4Z0 γx
The power per unit length thus reduces to an intuitive expression equal
2
ing half the timeaveraged 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.
(a)
(b)
Fig. 2.3. (a) An even and (b) an odd mode of a slab waveguide. The dotted
lines represent the waveguide boundaries.
resulting in the following transverse electric ﬁeld proﬁles for the three
regions for evenorder modes.3
3
For oddorder modes, the cosine is replaced with a sine and the two cladding
amplitudes are oppositely signed.
2.3 The Planar Slab Waveguide 17
1
Hy = Hy0
e−γx (x−d/2) (x > d/2) (2.37)
n41 n2eﬀ −n22
1+ n42 n21 −n2eﬀ
neﬀ Z0
Ex = Hy (2.40)
n2
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
n21
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 timeaveraged
Poynting vector across the waveguide dimension.
+∞
1
P/L = dxEx Hy∗ (2.42)
2
−∞
+∞
neﬀ n2
= dx Ex 2 (2.43)
2Z0 n2eﬀ
−∞
⎡ ⎤
n41
neﬀ Ex0  2 ⎢ n21 n22 2 n42
−1 n22 2 ⎥
= ⎢ + + ⎥.
4Z0 ⎣ n2 d n2eﬀ γx n41 n2eﬀ −n22 n2eﬀ γx ⎦
(2.44)
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
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.
Useful relations between the real and normalized parameters are given
by
neﬀ = n22 + b n21 − n22 (2.61)
√
V 1−b
kx = (2.62)
d
20 2. Optical Dielectric Waveguides
√
V b
γx = (2.63)
d
2
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,
b
V 1 − b − 2 arctan = mπ . (2.66)
1−b
This transcendental relation between V and b is plotted in Fig. 2.4. Note
that singlemode operation is guaranteed for V < π .
The previous sections were intended to refresh the reader with the prop
erties of onedimensional planar waveguides. However, the guiding struc
tures associated with ring resonators are generally twodimensional (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 singlemoded. 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.
n3
Rib
waveguide n1 a2 a 1
n2
b
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)
b
Fig. 2.6. Eﬀective index approximation for a rib waveguide.
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
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
(2.67)
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 secondorder 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 retroreﬂections (which of course are highly nonparaxial).
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 shortpulsed 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.
2.5 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 lefttoright (or vice versa).
∇ × 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
∗
(2.72)
∇ · E∗ ∗ ∗
k × Hj = −iω µ0 Hj · Hk − ε Ej · Ek − Pper,j · Ek
∗
(2.73)
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)
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.
Substituting the ﬁeld proﬁle expressions derived earlier for a core sepa
ration of s yields:
E 2
d
n21 − n2eﬀ −γ (s+x−d/2)
0
κ21 = k0 n21 − n22 dx cos (kx x) e x
4Z0 P/L n 2
− n 2
0 1 2
(2.82)
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
(2.83)
30 2. Optical Dielectric Waveguides
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
2
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.
where
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
Normalized coupling, κ / k1
d
0.3
s/d=0 s
0.2
s/d=1
0.1
increasing
gap/width ratio
0
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 guidetodisk
coupling.
be high. This implies that, in order to make the guides singlemoded 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 singlemode 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 lateraltapering guide with high lateral index contrast that is ini
tially highly multimoded 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 singlemoded by
employing low vertical index contrast.
A ﬁniteelement 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 resonatortostraight waveguide coupling,
numerical techniques are generally required. The threedimensional 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 roundtrip, 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
betamismatch 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 1µm tall and 0.5µm 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
intensity.
36 2. Optical Dielectric Waveguides
n3
Core, n2
GaInAsPInP: n1
n1 ∼ n3 ≈ 3.17 (InP)
n2 ∼ 3.3 − 3.5
Weak conﬁnement
(a)
Air
Strong conﬁnement
Air n3 Air
Core, n2
n1
Strong conﬁnement
GaInAsPInP:
n1 ∼ n3 ≈ 3.17 (InP)
n2 ∼ 3.3 − 3.5
(b)
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.
ncore = 1.5
Minimum radius for
10
ncore = 2.0
1 ncore = 3.5
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 “tightconﬁning” or “strongconﬁning” because of the corresponding
mode shape. Although it may seem like a good idea to use everhigher
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 lowindexcontrast waveguides to
maintain singlemode 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.
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,
(1)
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)
8
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
)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 )
(2.113)
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)
2
n k0) r ∂ϕ )0 r
n2 k
−iZ0 ∂
Eϕ = Hz . (2.116)
n2 k)0 ∂r
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
9
As opposed to a closedboundary structure, i.e., dielectric guiding region with
perfectly conducting walls.
2.7 Whispering Gallery Modes 43
Fig. 2.16. Plot of the electric ﬁeld associated with the sixth order TM (axial
Eﬁeld) 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 superevanescent 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).
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
(2.125)
where the following relations hold:
iπ i mπ (1)
Km (−iz) = e 2 Hm (z) (2.126)
2
iπ i mπ (1)
−iKm (−iz) = e 2 Hm (z) (2.127)
2
(1)
Km (−iz) Hm (z)
= (1)
. (2.128)
Km (−iz) −iHm (z)
TE WGMs
9
8 n1/n2 = 3.5
Hz
7 Eφ 3.0
6 Er 2.5
Log10 F
5 2.0
4 1.7 100
3 90
80
2 1.5 70
1 1.35 60
50
0 1.2530 40
m = 2 5 10 20
0 20 40 60 80 100
Normalized radius, (n1ω/c)R
TM WGMs
9
8 n1/n2 = 3.5
Ez 3.0
7 Hφ
2.5
6 Hr
Log10 F
2.0
5
4 1.7
100
3 90
80
2 1.5 70
1 1.35 60
50
1.25 40
0 30
m = 2 5 10 20
0 20 40 60 80 100
Normalized radius, (n1ω/c)R
120
Disk Single Azimuthal #
moded m=8
100 ring
Radial potential, µm2
1st mode
20
λ=1438nm, Q=3247
0
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,
1µm 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.
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 0.65µm radius would cut oﬀ the second
and thirdorder radial modes. Its potential energy distribution, shown as
a dashed line, only supports the lowest order mode.
∂
∇×J = − ∇×D+∇×∇×H (2.134)
∂t
= +iω∇ × (εE) + ∇ × ∇ × H (2.135)
= +iω∇ε × E + iωε∇ × E + ∇ × ∇ × H (2.136)
2
= +iω∇ε × E − ω µ0 εH + ∇ × ∇ × H (2.137)
= +iω∇ε × E. (2.138)
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.
Smeas
1
C(s) = ds ∆R(s )∆R(s − s). (2.150)
Smeas
0
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:
2
s
−π
C(s) = σ 2 e Sc . (2.151)
When
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 ,
2
C(fs ) = σ 2 Sc e−π (Sc fs ) . (2.152)
M+ 12
2
Sc −π
Sc
2π R M
= σ2 dM e (2.154)
2π R
M− 12
2
Sc −π
Sc
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.
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)
z
cos(θ ) ≈ . (2.159)
r
Incorporating these approximations results in a volume integral written
completely in cylindrical coordinates:
+d/2
∞
2π
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
(2.160)
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)
2
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:
π
dPs dPs
Ps = dϕ sin θ dθ = 2π sin θ dθ . (2.165)
dΩ dΩ
0
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
For TM WGM modes, the modal electric ﬁeld is perpendicular to the plane
of the disk (directed along the zaxis) 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 zcomponent),
Nθ = − sin θ Nz , (2.167)
resulting in
+d/2
2π
∞
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)
+d/2
2π )e−iMϕ
R+∆RM (z
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:
+d/2
NθM = iωε0 R(n21 − n22 ) sin θ dz ∆RM (z )Ez (R, z )eikz cos θ
−d/2
2π
dϕ ei(m−M)ϕ eikR sin(θ) cos(ϕ−ϕ ) . (2.170)
0
2π
dϕ ei(m−M)ϕ eikR sin(θ) cos(ϕ−ϕ ) = 2π im−M Jm−M (kR sin θ) ei(m−M)ϕ
0
(2.171)
56 2. Optical Dielectric Waveguides
and retaining only the square modulus of the polar radiation vector com
ponent yields
2
M 2
Nθ = 2π ωε0 R(n21 − n22 ) sin2 θ Jm−M (kR sin θ)2
2
+d/2
ikz cos θ
dz ∆RM (z )Ez (R, z )e . (2.172)
−d/2
Next the scattered power is calculated assuming that the ﬁeld and corru
gation are zindependent:
∞
 k40 (n21 − n22 )2
Ps = 2π R 2 2
∆RM Ez (R)2
M=−∞
8Z0
2
+d/2
π
3 2 ikz cos θ
dθ sin θ Jm−M (kR sin θ) dz e . (2.173)
0
−d/2
∞
2
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=−∞
π
d d cos θ
dθ sin3 θ Jm−M (kR sin θ)2 sinc2 . (2.174)
λ λ
0
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)
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)
+d/2
2π
∞
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)
+d/2
2π
∞
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
+d/2
2π
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
+d/2
2π R+∆RM (z )e−iMϕ
M
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
+d/2
2π
1 1
M
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 )
(2.185)
+d/2
2π R+∆RM (z )e−iMϕ
M
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.
+d/2
1 1
M
Nθ,r = −iωR 2
− 2 cos θ dz ∆RM (z )Dr (R, z )eikz cos θ
n1 n2
−d/2
2π
+ ,
dϕ cos ϕ − ϕ ei(m−M)ϕ eikR sin(θ) cos(ϕ−ϕ ) (2.187)
0
+d/2
M
Nθ,ϕ = −iωε0 R(n21 − n22 ) cos θ dz ∆RM (z )Eϕ (R, z )eikz cos θ
−d/2
2π
+ ,
dϕ sin ϕ − ϕ ei(m−M)ϕ eikR sin(θ) cos(ϕ−ϕ ) (2.188)
0
+d/2
1 1
M
Nϕ,r = −iωR − 2 dz ∆RM (z )Dr (R, z )eikz cos θ
n21 n2
−d/2
2π
+ ,
dϕ − sin ϕ − ϕ ei(m−M)ϕ eikR sin(θ) cos(ϕ−ϕ ) (2.189)
0
2.8 Scattering Losses Resulting From Edge Roughness 59
+d/2
M
Nϕ,ϕ = −iωε0 R(n21 − n22 ) dz ∆RM (z )Eϕ (R, z )eikz cos θ
−d/2
2π
+ ,
dϕ cos ϕ − ϕ ei(m−M)ϕ eikR sin(θ) cos(ϕ−ϕ ) . (2.190)
0
2π
dϕ e±i(ϕ−ϕ ) ei(m−M)ϕ eikR sin(θ) cos(ϕ−ϕ ) =
0
and retaining only the square modulus of the radiation vector compo
nents yields
2
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
4
2
+d/2
ikz cos θ
dz ∆RM (z )Dr (R, z )e (2.192)
−d/2
2
M 2
Nθ,ϕ = 2π ωε0 R(n21 − n22 ) cos2 θ
2
M 2 1 1
Nϕ,r = 2π ωR − 2
n21 n2
Jm−M+1 (kR sin θ) + Jm−M−1 (kR sin θ)2
4
2
+d/2
ikz cos θ
dz ∆RM (z )Dr (R, z )e (2.194)
−d/2
60 2. Optical Dielectric Waveguides
2
M 2
Nϕ,ϕ = 2π ωε0 R(n21 − n22 )
Assuming that the ﬁeld and corrugation are zindependent, the scattered
power is given as
∞
 k40 (n21 − n22 )2 2
Ps = 2π R 2 ∆RM
M=−∞
8Z0
π 0
Jm−M+1 (kR sin θ) − Jm−M−1 (kR sin θ)2
dθ sin θ cos2 θ
4
0
Jm−M+1 (kR sin θ) + Jm−M−1 (kR sin θ)2 Dr (R)2
+ sin θ
2
4 n2 n2 ε 1 2 0
Jm−M+1 (kR sin θ) + Jm−M−1 (kR sin θ)2
+ sin θ cos2 θ
4
1
Jm−M+1 (kR sin θ) − Jm−M−1 (kR sin θ)2
2
+ sin θ Eϕ (R)
4
2
+d/2
ikz cos θ
dz e . (2.196)
−d/2
Sc
2
−π 2π
r e RM
⎩ Pg 2Z n4 n4 ε2
0 1 2 0 M=−∞
π
2
dθ sin θ cos2 θ Jm−M (kR sin θ)
0
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
sinc2
(kR sin θ) λ λ
2 ∞
1 λd Eϕ (R) 
2
Sc
−π 2π RM
+ e
Pg 2Z0 M=−∞
π
(m − M)2 Jm−M (kR sin θ)2
dθ sin θ cos2 θ 2
(kR sin θ)
0
2
2 d d cos θ
+ sin θ Jm−M (kR sin θ) sinc2 . (2.197)
λ λ
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
∞
2
 −π Sc λ
λ/n1 n1 2π R M
e
M=−∞
π
3 d 2 d cos θ 2
dθ sin θ Jm−M (kR sin θ) sinc (2.198)
λ λ
0
2 2
1 n1 2π R n22 σ Sc
= 4π 3 1−
QsTEr mλ n21 λ/n1 λ/n1
∞
2
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
π
2
dθ sin θ cos2 θ Jm−M (kR sin θ)
0
(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
Sc
TEϕ = 4π 1−
Qs mλ n21 λ/n1
λ/n1
2 ∞
2
1 d Eϕ (R)  Sc
−π λ/n λ
n1 2π R M
e 1
Pg 4kZ0 M=−∞
π
(m − M)2 Jm−M (kR sin θ)2
dθ sin θ cos2 θ 2
(kR sin θ)
0
2 d d cos θ
+ sin θ Jm−M (kR sin θ) sinc2 . (2.200)
λ λ
π
d d cos θ
TM
Pm−M = dθ sin3 θ Jm−M (kR sin θ)2 sinc2
⎛ λ λ ⎞
0
d/λ 2 2 2
λ λ
= dτ
1− τ Jm−M ⎝kR 1 − τ ⎠ 2
sinc τ
d d
−d/λ
∞
π
2
TEr
Pm−M = dθ sin θ cos2 θ Jm−M (kR sin θ)
0
(m − M)2 Jm−M (kR sin θ)2 d 2 d cos θ
+ sin θ 2
sinc
(kR sin θ) λ λ
⎡ ⎛ ⎞
d/λ 2 2 2
⎢ λ λ
= dτ ⎣ τ ⎝
Jm−M kR 1 − d τ
⎠
d
−d/λ
⎤
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)
−∞
(2.203)
π
TEϕ (m − M)2 Jm−M (kR sin θ)2
Pm−M = dθ sin θ cos2 θ 2
(kR sin θ)
0
2 d 2 d cos θ
+ sin θ Jm−M (kR sin θ) sinc
⎡ λ λ
2
⎢ 2
λ 2
d/λ
⎢ 2 (m − M) Jm−M kR 1 − d τ
⎢ λ
= dτ ⎢ τ 2
⎢ d
2
⎣ 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)
dλ
−∞
64 2. Optical Dielectric Waveguides
π
d 2
TEr
Pm−M → dθ sin θ cos2 θ Jm−M (kR sin θ)
dλ λ
0
(m − M)2 Jm−M (kR sin θ)2
+ sin θ 2
(2.206)
(kR sin θ)
π
TEϕ d (m − M)2 Jm−M (kR sin θ)2
Pm−M → dθ sin θ cos2 θ 2
dλ λ (kR sin θ)
0
2
+ sin θ Jm−M (kR sin θ) . (2.207)
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.)
n1/n2=1.5
106
n1/n2=3.0
104
Finesse
102
100
σ=1nm
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 roundtrip. 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
America.)
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, thumbtacklike (d < λ) microresonators, they are
equivalent:
1 16π 3 1 2 2
TM/TE
= 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
6
10
m=
100 Scatterin
50 g loss li
25 mited
4
10 15
Finesse
5
2
10
0
10
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
2.9 Summary
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 roundtrip 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 = → ,
m=1
1 − r1 r2 eiφ r1 =r2 ≡r 1 − r 2 e+iφ
(3.3)
and t̃ represents the complex transmittivity:
∞
m−1

φ φ
φ
im 2 imφ −t1 t2 eim 2 − 1 − r 2 eim 2
t̃ = −t1 t2 e r1 r2 e = → .
m=1
1 − r1 r2
r1 =r2 ≡r 1 − r2
(3.4)
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.
 m−1 r − eiφ
r̃ = r − t 2 eimφ r eimφ = . (3.5)
m=1
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 staircaselike 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).
Fig. 3.3. Two ring resonator devices and their freespace embodiments.
where the lumped self and crosscoupling 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
αring
E3 = e − 2 2π R ik2π R
e E4 ≡ aeiφ E4 . (3.8)
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 crosscoupling
Fig. 3.5. A plot of the buildup factor versus detuning for an allpass ring
resonator.
3.2 AllPass Ring Resonators 77
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 allpass 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
2π π 2π
F= → ≈ 2 . (3.13)
2r r ≈1 1−r t
2 arccos
1+(r )2
E2 a − r e−iφ
= ei(π +φ) . (3.14)
E1 1 − r ae+iφ
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 singlepass 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 AllPass Ring Resonators 79
Fig. 3.7. A plot of the eﬀective phase shift versus the singlepass phase
shift or normalized detuning for an allpass ring resonator. Note the in
creasing sensitivity near resonance for increasing selfcoupling parame
ter r 2 .
dΦ
Φ =
dφ
1 − r 2 a2
=
(
1+a2 ) 2
1 − 2r a cos(φ) 2 + r 2 a2 + sin2 (φ) (1 − a2 ) r 2 − (1 − a2 )
1+r
→ . (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.
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 roundtrips light traverses in the resonator.
Because the group delay associated with an allpass 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)
dω2
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:
2
− 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
3
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 timedivision
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 roundtrip
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 pulsetrain spectrum may completely lie within multiple res
onance bandwidths and can thus take advantage of the phaseenhancing
properties of the resonator. However, unless changes (on–oﬀ switches)
3
This is analogous to the Raleigh range for a Gaussian beam.
3.2 AllPass 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 informationcarrying 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
timedomain. 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
4
Output pulse
2
Input pulse
0
0 2 4 6 8 10 12 14
Time, ps
10 Gaussian response
8
Power, arb. units
6 Circulating pulse
0
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
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.
(g) (h)
Spectrum, arb.
Spectrum, arb.
effective effective
phase phase
a b c d e f
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 AllPass Ring Resonators 83
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 allpass, phaseonly ﬁlter. In particular, near resonance, the inter
nal attenuation is increased such that dips appear in the transmission
spectrum:
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 singlepass 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 roundtrip. If
the attenuation is comparable to the crosscoupling, 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 counterintuitive eﬀects may take place in this regime
such as the inversion of the phase sensitivity. Overcoupling occurs when
the roundtrip loss does not exceed the coupling strength (r < a) or
(t 2 > αL) and is the conventional mode of operation for an allpass res
onator. Figure 3.10 displays the eﬀect of attenuation on the transmission
and buildup for the overcoupled, criticallycoupled, 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
allpass nature of a normally lossy ring resonator. Of course if the round
trip gain exceeds the net roundtrip 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
0.9
a = 0.75 6
a = 0.96
0.8
a = 0.50 a = 0.96
0.7 5
a = 0.20
Buildup factor
Transmission
0.6
4
a = 0.00 a = 0.90
0.5
3
0.4
0.3 2
a = 0.75
0.2
1
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 allpass resonator with r = 0.75 and varying
loss. The singlepass ﬁeld transmission a is displayed for each curve in
the ﬁgure.
(3.22)
3.3 Add–Drop Ring Resonators 85
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
φ, B φ, B
Input T,Φ Input T1 , Φ1
.75
3π
.50 Φ2 2π
Φ Φ1
π
.25
0
0 π
(b) 20 (d) (f) 20
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 allpass 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.
1−(1−r1 r2 a)2
2 arccos 2r1 r2 a
∆ω = . (3.29)
TR
88 3. Optical Microresonator Theory
(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,
1/2
(1 − r 2 )2
φ = (3.31)
r2
2(1 − r 2 )
⇒ ∆ω = , (3.32)
r TR
λ20
∆λ ≈ ∆ω, (3.34)
2π c
where λ0 is the freespace wavelength, and we have assumed λ0 ∆λ.
Then,
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
If αdis L 1, we get
2 αdis L
∆ω =
TR 2
αdis L cαdis
= = . (3.37)
TR neﬀ
λ20
FSRwavelength = . (3.40)
Lneﬀ
3.3.4 Finesse F
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 roundtrip is:
dEn 2
= −αdis LEn 2 . (3.47)
dn
We can now relate this to the power loss, as each roundtrip takes
time TR . Since the power loss is energy lost per unit time, dEn 2 /dt =
(1/TR )dEn 2 /dn. Therefore,
En 2
Q = ω0
−dEn 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)
2π
or
ω0 Lneﬀ
Q = F (3.51)
2π c
neﬀ L
≈ F (3.52)
λ0
= 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
1
E4 (t) ∼ L−1 (3.56)
(1 − A )/TR − A s
1 − A
∼ exp − t . (3.57)
A TR
E4 2
Q = ω0
−dE4 2 /dt
A T R
= ω0
2 (1 − A )
ω0 TR
= , (3.58)
αdis L
To ﬁnd the physical meaning of the ﬁnesse and Q, we consider the num
ber of roundtrips 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
roundtrips required to reduce the energy to 1/e of its initial value, we
get
92 3. Optical Microresonator Theory
1
exp (−αdis NL) = (3.59)
e
1
⇒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 roundtrips 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 allpass and add–drop
resonators are related in the following manner:
π Q
Ballpass = F = m (3.63)
2
Q
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 microresonatorbased 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.
−π / F
1.0
−π/2F
Imaginary( E2 / E1 )
0.5
−4π/F
0.0 φ = 0
+4π/F
0.5
+π/2F
1.0
+π/F
1.5
1.5 1.0 0.5 0.0 0.5 1.0 1.5
Real( E2 / E1 )
1.0
Imaginary( E2 / E1 )
−π/F
0.5 −π/2F −4π/F
0.0 φ=0
+π/2F +4π/F
0.5
+π/F
1.0
1.5
1.5 1.0 0.5 0.0 0.5 1.0 1.5
Real( E2 / E1 )
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)
m=1
∞
m−1 r − e+iωTR
H(ω) = r − 1 − r 2 e+iωTR r e+iωTR = . (3.71)
m=1
1 − r e+iωTR
Fig. 3.15. (a) Plots of the real and imaginary components of the com
plex ﬁeld transfer function for an allpass resonator with r = 0.9. (b)
Plots of the amplitude square modulus (transmission) and phase. Note
that although an allpass 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
Fig. 3.18. A comparison of the rolloﬀ for a ﬁrst and thirdorder optical
ring resonator ﬁlter.
3.5 Higher Order Filters 99
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 ﬁlterorder is determined from the order of P2 ; in
a ﬁrstorder ﬁlter, the polynomial P2 is of the form (s − p), i.e., a ﬁrst
order polynomial; the singlering OCDF response shown above, therefore,
qualiﬁes as a ﬁrstorder
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 rolloﬀ than a
ﬁrstorder ﬁlter. At the same time, by using interference eﬀects between
multiple ﬁlters, a ﬂatter top compared with lower order ﬁlters can be
obtained. Sharper rolloﬀ is desirable in applications such as bandpass
ﬁlters for better isolation between channels. A ﬂattop minimizes signal
distortion in similar applications.
We go over parallelcascaded ring ﬁlters here. Serially coupled ring
ﬁlters have been analyzed elsewhere [15,152,160,161] and demonstrated
by several groups [15,18,55,162]. Parallelcascaded 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
2
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 inputport ﬁeld for the mth resonator, Edm is the drop
port ﬁeld, Etm is the throughport ﬁeld, and Eam is the addport ﬁeld.
To show that the response of a multiring cascade has higher order
behavior, we only need to show that a doublering 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 rolloﬀ 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 xaxis has been normalized to the bandwidths of the two ﬁlters.
3.5 Higher Order Filters 101
FSRring
FSRbus
FSRmulri−ring
Wavelength, a. u.
Fig. 3.20. Vernier eﬀect with a multiring 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
0.1
Normalized power
0.01
0.1
0.01
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
ring
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
FSR.
Then
2RP
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 rolloﬀ near a resonance is determined by
the denominator. Since both T11 and T21 are polynomials of ﬁrstorder in
Φ = exp(−iωTR ) ≈ 1 − i(ω − ω0 )TR , or s = −i(ω − ω0 ), the denominator
3.6 Summary 103
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 dropport response of a triple
ring parallelcascade 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
4.1.1 GaAsAlGaAs
4.1.2 GaInAsPInP
Fig. 4.2. Single and multiple ring GaAsAlGaAs laterally coupled micror
ing add/drop ﬁlters (left), and single and doublering ﬁlter dropport re
sponses (right). 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 ﬁnetune 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
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
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 electrooptic eﬀect. The second term is quadratic in Edc
and causes the quadratic electrooptic eﬀect. For III–V semiconductors,
(2)
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 electrooptic eﬀect since the
refractive index can be varied as the square of the electric ﬁeld.
Near the bandedge, the quadratic eﬀect is caused primarily by the
Franz–Keldysh eﬀect in bulk and the quantumconﬁned Stark eﬀect
(QCSE) in multiplequantumwell (MQW) waveguides. The two eﬀects are
related; Miller, Chemla, and SchmittRing 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
quantumconﬁned eﬀect is stronger than the bulk eﬀect.
Although both the linear and the quadratic electrooptic eﬀect can be
described by the nonlinear susceptibility, they have traditionally been de
scribed in terms of the linear and quadratic electrooptic 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 outofplane polarization, Γ is the conﬁnement factor, r11
is the linear electrooptic coeﬃcient (contracted notation), and s1111 is an
element of the quadratic electrooptic coeﬃcient tensor.
Fig. 4.6. Electrical characteristics of diode made with wafer used for
electrooptically tuned ring resonators. The diode area is 100×2000 µm2 .
4.2 Active Resonators 113
1 3 2
∆n = n sE , and (4.5)
2
2π L
2π = ∆n . (4.6)
λ
Using λ = 1.55 µm, L = 0.2 cm, E = 10 V/µm (5 V), and n = 3.22 (eﬀective
index of slab) gives
(a)
(b)
Fig. 4.9. (a) Setup for impedance measurement of slab waveguide, and
(b) voltage across R2 and L2 versus frequency.
For the tuning measurements, the device was tested without antireﬂec
tion coatings. Ampliﬁed spontaneous emission (ASE) from two cascaded
erbiumdoped 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 inplane polarization. So the
response shown in Fig. 4.13 is for the outofplane 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. 4.12. EDFA spectrum and polynomial ﬁt used to normalize the raw
spectral data for the active resonator.
1 ♦
+
♦
+ ♦ ♦♦
♦
+
+
♦
++
+
♦
++
+
++
+
♦
♦
+♦
♦
♦
+
+
+♦
♦
♦
+
+ ♦
♦
+♦
♦
+
+
++
+
♦
++
+
+
♦+
+
♦
♦
♦
++
♦
+
♦
+ ♦
♦♦♦
+♦
+
♦
+
+
♦+
+
+
♦+
+ ♦
♦
♦♦
♦++
+
♦
+
♦+
+
+
♦
♦
♦+
♦
+ ♦
♦♦
♦ ♦
♦
♦
♦+
♦
♦+ ♦
♦+♦
+
+
+
+♦
+ +
♦
+
♦♦
♦+
+
♦+
+ ♦
♦♦♦♦
♦
♦
+
+♦
+
+
+
♦
♦+
♦
♦+
+ ♦
♦
♦♦
+
+♦
♦
+
+
+♦
+
+
♦
++
♦+
♦+
+
++ ♦
♦♦
♦+
++
+
Normalized power
♦
♦♦+ ♦♦
♦
♦♦++
♦
♦ + ♦
♦♦
♦ ++
♦
♦+
+ ♦♦+
++
+
++
+
♦ +
+ ♦♦
♦+
++
♦
♦+ ♦
♦
♦+++
♦ +
+ ♦♦
♦
♦++
+ ♦
♦ +
0.1 ♦ + ♦ + 0V ♦
♦♦++ ♦ ♦+
♦
+
+
6V +
♦
♦ + ♦ +
♦ +♦ ♦+ +
♦
♦ +♦ + +
+
♦
♦+♦♦++
♦♦
♦
♦♦+
+
♦
+
+ +
+
+
+
+
0.01
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
1553.4
2.8878 ♦
♦ 1553.3
2.8876
♦ 1553.2
2.8874 ♦ ♦
1553.1
2.8872 ♦ ♦
♦ ♦ 1553
2.887
♦ ♦ ♦ ♦
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 steplike 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 redshifted 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 electrooptic coef
ﬁcient for the waveguide can be obtained. The value of the coeﬃcient
for the waveguide is a lower bound for the quadratic electrooptic 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 curveﬁt 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 electrooptic
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 electrooptic 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 oﬀresonance 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 modemismatch 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 highﬁnesse 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
30
25
Loss, cm−1
20
15
10
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
loss.
120 4. Microring Filters: Experimental Results
bandwidth is a little over 0.5 nm. The simulated dropport response for
an add/drop ﬁlter at two voltages is shown in Fig. 4.16. The increased
loss aﬀects the bandwidth and decreases the dropport 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. Ultracompact
tunable microring notch ﬁlters have been demonstrated in a pin 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 electrooptic, and that the quadratic
electrooptic coeﬃcient is at least 2.3×10−15 cm2 /V2 , which is consistent
with values reported in literature. The quadratic electrooptic 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 wavelengthdivision
multiplexing applications.
1
Normalized power
0V 7.5 V
0.1
0.01
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
0.8
♦
0.75
♦
Bandwidth, nm
0.7
♦
0.65 ♦
♦
0.6
♦
♦
0.55 ♦
♦ ♦
♦ ♦ ♦ ♦ ♦ ♦
0.5
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
The last of these phenomena can occur for all three ﬁelds at the same
frequency and has a degeneracy factor of 3
n(I) = n + n2 I, (5.6)
α(I) = α + α2 I, (5.8)

M
A3,M (t) = it (r a)m eiφm (t) A1 (t − mTR ) , (5.11)
m=0
Fig. 5.1. The inherent tradeoﬀ 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 chalcogenidebased systems near 1.55µm. Increasing ﬁnesse is directly
proportional to decreasing energy.
For most direct gap semiconductors, many parameters are constant such
that to good approximation, the twophoton absorption spectrum can be
simpliﬁed to be only dependent on the bandgap energy and the linear
refractive index as
3/2
ω
1.42 × 10−7 2 Eg − 1
α2 (ω) ≈
5 m/W, (5.21)
n20 Eg3 2 ω
Eg
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 halfgap ensures that
one and twophoton absorption processes are negligible; yet, a reason
ably high and ultrafast nonlinear Kerr coeﬃcient is retained. Figure 5.2
displays the predicted twophoton absorption and nonlinear refractive
index for a twoband 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.
Fig. 5.5. Measured transmission spectrum (left axis) of the device shown
in Fig. 5.4 with a theoretical ﬁt. The resonatorinduced phase shift as
inferred from the interferogram is also shown in the plot (right axis).
(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 halfgap, the bound
electron Kerr nonlinearity is essentially instantaneous whereas carrier
generation due to one and twophoton absorption is negligible. Ultra
compact devices constructed from these buildingblocks thus have the
potential to be engineered into ultrafast 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 12.5ps 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 1THz bandwidth and 1pJ energy threshold.
An implicit assumption in the derivations thus far is the fact that the
resonator does not get powerdetuned or pulled away from its initial
detuning. One might call this the nonpulling pump approximation (NPPA).
The powerdependent 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
350
16 po
14 300 i ng
power, Watts
at
12 250 cul
cir
10 200 nt
8 o na
150 R es
6
4 100
2 50
0 0
0 100 200 0 5 10 15 20
Time, ps Incident power, Watts
c) d) π
π
n
tio
Effective phase shift, rad
dic
d pre
are
π π
u
sq
2 2
e
ess
Fin
0 0
0 100 200 0 5 10 15 20
Time, ps Incident power, Watts
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
2π
Effective phase shift, Φ
π π
2π
F
0
4π 2π 0 2π 4π
F F F F
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
we
Incident & transmitted
350
16 po
300 g
14 tin
power, Watts
ula
12 250
circ
10 200 ant
8 son
6
150 Re
4 100
2 50
0 0
0 100 200 0 5 10 15 20
Time, ps Incident power, Watts
c) d) π
π
n
tio
Effective phase shift, rad
dic
pre
red
π π
ua
sq
2 2
sse
e
Fin
0 0
0 100 200 0 5 10 15 20
Time, ps Incident power, Watts
a) b)
18 r
Circulating power, Watts
we
Incident & transmitted
350
16 po
300 g
14 in
power, Watts
lat
12 250
ircu
10 200 ntc
na
8 so
6
150 Re
4 100
2 50
0 0
0 100 200 0 5 10 15 20
Time, ps Incident power, Watts
c) d) π
π
ion
Effective phase shift, rad
t
dic
pre
red
π π
ua
sq
2 2
e
ess
Fin
0 0
0 100 200 0 5 10 15 20
Time, ps Incident power, Watts
5.2.6 Multistability
80 400
60 300
40 200
20 100
0 0
0 50 100 150 0 50 100 150
Time, ps Time, ps
500
4 400
300
2 200
100
0 0
0 50 100 150 0 20 40 60
Time, ps Incident power, Watts
a) c)
E3 E4 EB2
E1 E2 EA1 EB1
E5
b) d)
E3 E4 EB2
EA1 EB1
E1 E2
Fig. 5.12. (a) An allpass 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
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)
2
+ O(3), (5.26)
140 5. Nonlinear Optics with Microresonators
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)
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)
2.0
1.5
Normalized nonlinear
1.0 enhancement
0.5
Net transmission
0
0 0.02 0.04 0.06 0.08 0.10
Singlepass loss (1a2)
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.
(a − r )2
T (a) = → 0, (5.32)
(1 − r a)2 r =a
1
T (α2π R) ≈ 1 − B0 (α2π R) + 2 B20 (α2π R)2 + O(3). (5.33)
1
0.592 in the highﬁnesse limit.
142 5. Nonlinear Optics with Microresonators
e−αL
T = . (5.36)
1 + α2 ILα
Because twophoton absorption increases in proportion to the nonlinear
refractive index, the achievable nonlinear phase shift saturates logarith
mically with increasing intensity:
M2 M2 M1
L
100
10 L
Resonator
101 FL
2
10
103
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.)
a) 0.10 b) 30
25
0.08
Incident
Transmission
Power, Watts
20 pulse
0.06
15
0.04
10
Transmitted
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)
0
π
Effective phase shift, rad
0
π
4
π
4
π π
20 5 10 15 20 25 20 20 40 60 80 100 120 140
Incident power, Watts Time, ps
4
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 pdoped core, among others [203].
If the pump pulse is longer than the cavity lifetime, a quasisteady
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
eE
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)
m∗
which gives
ω2p
4π χf = − 2 , (5.46)
ω
so
ω2p
n2 = (ω) = b − 2 . (5.47)
ω
Since there is no imaginary term above, the refractive index is real. 5 Then,
ω2p
n2 = b − (5.48)
ω2
dn ωp
⇒ 2n = −2 2 . (5.49)
dωp ω
5
Realistic, noninﬁnite carrier mobilities lead to freecarrierinduced changes in
the absorption as well.
5.4 Enhanced FourWave Mixing Eﬃciency in Microring Resonators 147
Since we are concerned with the intensity, we consider the squared mag
nitude of the righthandside; 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 fourwave mixing is enhanced as B4 .
5.5 Summary
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. AllOptical Switching and Logic using Microresonators
the ring path. Assuming Br  Ar , the propagation of the pump and
the probe beams in the presence of TPAinduced 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 freecarrier density, Nfc , via ∆αfc = σa Nfc
and ∆nfc = σr Nfc , with σa and σr being the absorption cross section
and refraction volume, respectively. The freecarrier 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)
iφ
Ar (t) = r Ar (t − TR )e − i 1 − r Ai (t),
2 (6.5)
where r is the lumped selfcoupling coeﬃcient, TR is the cavity transit
time, and φ is the roundtrip 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 AllOptical Switching 151
0.8
Transmittance
0.6
0.4
0.2 measaured
model fit
0
0.4 0.2 0 0.2 0.4
Wavelength detuning, nm
Fig. 6.2. Measured spectral response and its theoretical ﬁt of a 10µm
radius GaAs/AlGaAs microring resonator at the 1562nm 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. AllOptical 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 300ps pump pulse through the
microring. The wavelength of the pulse was set at 1561.97 nm, that
is, 0.1nm 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
1
Input pump
Circulating pump
0.8
Normalized Intensity
0.6
0.2
0
0 200 400 600 800
Time, ps
Fig. 6.3. Simulated nonlinear propagation of a 300ps pump pulse
through a 10µmradius GaAsAlGaAs 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 AllOptical 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 lowintensity 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 threedimensional
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
0.2
0.2
800
600
400 1551.4
1551.2
Time, ps 200 1551
1550.8
0 1550.6 Probe wavelength, nm
Fig. 6.4. Threedimensional 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. AllOptical Switching and Logic using Microresonators
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
(a)
0.8
Normalized Intensity
0.6
0.4
(b)
0.2
0
0 200 400 600 800
Time, ps
Fig. 6.5. Simulated pumpprobe 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
resonance.
6.1 AllOptical Switching 155
a 300ps pump pulse signal with a 20MHz 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 bluedetuned
from the resonance. A 3mW 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 modemismatch 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 bandpass
ﬁlter, and detected using a 40GHz detector and a 50Ghz oscilloscope. In
Fig. 6.7, we plot a threedimensional graph for the measured throughput
probe intensity as a function of time and probe wavelength. It is in very
good agreement with the simulated 3dimensional graph shown earlier
156 6. AllOptical Switching and Logic using Microresonators
Normalized Intensity
0.2
0.2
800
600
1551.4
Time, ps 400
1551.2
200 1551
1550.8
0 Probe wavelength, nm
1550.6
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 switchon 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 surfacestate recombination at the microring sidewalls.
1
Input pump
0.8
Output probe
Normalized Intensity
0.6
(a)
Output pump
0.4
0.2
0
0 200 400 600 800
1
Input pump
0.8
Output probe
Normalized Intensity
0.6
(b)
0.4
Output pump
0.2
0
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. AllOptical Switching and Logic using Microresonators
1
Output
Input
Normalized Intensity
0.75
0.5
0.25
0
100 50 0 50 100
Time, ps
Fig. 6.9. Time traces of the input and output pulses of a GaAsAlGaAs
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.
Drop
20 micron
Add
Input Through
Fig. 6.10. Optical micrograph of a GaAs vertically coupled OCDF micror
ing resonator.
160 6. AllOptical 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.
Fig. 6.12. Time traces for (a) the input pump pulse at 1545 nm and (b) the
output dropped signal initially bluetuned to the 1555nm resonance.
Fig. 6.13. Time traces for (a) the input 5GHZ 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 AllOptical Logic 163
6.4.1 AND/NAND
In this section, alloptical AND and NAND logic gate operation via nonlin
ear absorption is demonstrated in single semiconductor microracetrack
resonators [211]. Operation near critical coupling allows a high switch
ing contrast [57]. The device is pumped with two counterpropagating
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 1550nm pump
pulses are absorbed through TPA inside the resonator, and free carriers
164 6. AllOptical Switching and Logic using Microresonators
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 AllOptical Logic 165
Fig. 6.15. Experimental setup for optical logic AND gate using the micro
racetrack resonator. PG: Pattern generator. PC: Polarization controller.
BPF: Bandpass ﬁlter.
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 modelocked 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 AllOptical Logic 167
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 bluetuned out
of resonance at 1558.6 nm.
carrier lifetime, e.g., by sweepout [202], or doping the guiding layer with
ptype 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
1
This results in a reduced ntype minority carrier lifetime, which is much
shorter than the ambipolar lifetime.
6.4 AllOptical Logic 169
6.4.2 NOR
Fig. 6.18. Optical micrograph of the NOR logic gate. The two microring
resonators are symmetric and of 9.6µm 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.
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 bluetuned to
the ﬁrst peak, whereas the probe beam (C) is initially bluedetuned 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 refractiveindex 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 reddetuned. Note that the pump wavelength is initially only
slightly bluedetuned [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. AllOptical 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.
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.
6.5 Summary
In this chapter we gave an overview of the theoretical framework along
with experimental demonstrations of alloptical switches and logic de
vices in III–V microring resonators by TPA using the pumpprobe method.
174 6. AllOptical Switching and Logic using Microresonators
2
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
E3 E4 R
E1 E2 L
Fig. 7.1. A microresonatorstructured, 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
L
kR
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 selfcoupling 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
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
0
i + δz
Ej+1 (ω) = e c L ×
⎧ ⎡ ⎤n ⎫
⎨ ∞ n
i ⎣ n0 ∞
1 δz dm Φ ⎬
m⎦
1+ ∆ωδz + (φ − φ0 ) E (ω) .
⎩ n! c m
m! L dφ φ0 ⎭ j
n=1 m=1
(7.3)
Taking the Fourier transform of this equation results in a diﬀerence equa
tion relating the pulse envelopes at the two points:
The diﬀerent terms in this equation are isolated and examined in what
follows.
0.3
Power, mW
0.2
0.1
0
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 10ps FWHM hyperbolic secant pulse tuned
for maximum anomalous GVD (B = 0.13) enters the system consisting
of 100 resonators each with a 5µm 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.)
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 thirdorder dispersion coeﬃcient
is
+ ,
1 d3 Φ TR3 −2r 1 − r 2 1 + r 2 cos φ0 − 3r + r cos 2φ0
k
eﬀ ≡ = 3
L dω3 L (1 − 2r cos φ0 + r 2 )
4 F 3 TR3
→ − . (7.8)
φ0 =0,r ≈1 π3 L
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 thirdorder 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 timedomain, 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 RBA + iTR A.
dz ⎩ c ∂t m! L dφm φ ∂t ⎭
m=1 0
(7.10)
For two nonlinearly interacting resonant pulses, the results here derived
for the selfphase modulation (SPM) eﬀect similarly apply to the eﬀect of
crossphase modulation (XPM) although with an extra degeneracy factor
of two [178].
2
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
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 − keﬀ 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 A2 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)
2
where
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
1.0
Soliton condition
(GVD) (SPM)
0.5 γeff
''
keff
(2FT/π)2/L γ (2F/π)2 (2πR/L)
0.0
0.5
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 resonatorinduced 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 :
π
Γ = √
√ B2 ≈ B2. (7.13)
2 3 arcsech2 1/ 2
0.3
Power, W
0.2
0.1
0
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 glassbased system. (After Ref. [129], ©2002, Optical
Society of America.)
3
The values of keﬀ and γeﬀ are, respectively, lowered by factors of 3/4 and 9/16
from their given maximum values when operating at dispersion extremum
points.
4
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
roundtrip within the resonator followed by interference at the coupler. Tra
ditional beam or pulse propagation splitstep 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 timedomain.
7.3 Nonlinear Propagation in Distributed Microresonators 183
Fig. 7.6 shows the propagation of the fundamental dark SCISSOR soliton
tuned to the normal dispersion peak.
Power, mW
30
20
10
0
0
20
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.)
5
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
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 secondorder SCISSOR soliton
1.2
Power, W
0.8
0.4
0
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 resonatorinduced intensitydependent group velocity
(selfsteepening). Here a secondorder 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.)
6
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 pedestalfree optical pulse compression. The
eﬀects of induced selfsteepening in a sequence of resonators can take
place for picosecond and even nanosecond pulses because unlike in the
case of intrinsic selfsteepening, 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 selfsteepening parame
ter (s/TP ) is 0.173. To observe the same eﬀect with traditional intrinsic
selfsteepening, a sixcycle or 30femtosecond pulse would be required.
90
Input Output
80
70
60
Power, W
50 Output
40 (1ps)
30
20 Input
10 (5ps)
0
0 10 20 30 40 50
Time, ps
Fig. 7.8. A highly compact microresonatorbased 5× pulse compressor. A
25 W transformlimited input pulse (NS = 5) of 5ps 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 intensitydependent group velocity. These extra pulses might be
eliminated through the use of a saturable absorbing material that may
even be microresonatorbased. Here, 10 AlGaAs or chalcogenidebased
microresonators of 10µmdiameter 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
crossphase modulation in a resonant SCISSOR.
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
∂ ∂
A + keﬀ A =
∂z ∂t
1 ∂2 1 ∂ 3 ∂ γeﬀ A2
− i k A + k A + i 1 + is A. (7.18)
2 eﬀ ∂t 2 6 eﬀ ∂t 3 ∂t 1 + A2
2
AS 
2.0
Power, W
1.5
1.0
0.5
0
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.)
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, fourwave 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
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 ultrahigh 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 ultrahigh 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 powerdetuned 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
powerdetuning 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 reddetuned initially by
π /F and allowing the resonator to be powerdetuned though resonance
for a total value of π radians [227].
190 7. Distributed Microresonator Systems
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
Assuming that the phase error of 1 radian is the maximum tolerable error,
1/3
3 1 1
the usable bandwidth is restricted to ∆ν = 2 F TR ≈ F TR . A SCISSOR,
composed of a sequence of N allpass 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 singledevice delay. Fortunately, the net accumulated phase
errors do not scale in the same fashion. That is, the phaseerror limitation
on usable bandwidth (governed on resonance by thirdorder √ dispersion)
3
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 thirdorder dispersion. This leaves fourthorder dispersion
as the limiting phase error. The dominant phase error in this regime is
thus
√
4
1 d4
Φ 1 27 3 2
Φerror  = φ =4
F φ 4
(7.25)
24 dφ 4 24 128 π
φD
4 1/8
2 1 1
with a similar bandwidth restriction of ∆ν = 35 F TR ≈ F TR for a
single
√ resonator. It is easily shown that a less restricting scaling law of
4
∆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 100ps pulse tuned to resonance
is delayed greatly, but broadens and acquires ripples associated with neg
ative thirdorder 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 thirdorder 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 nonvanishing secondorder (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, 10µm diameter resonators spaced by
10π µm with r = 0.98 corresponding to a group velocity reduction of
100×. (a) A weak, 100ps resonant pulse propagates at a group velocity of
(c/n)/100 through the SCISSOR and is corrupted by resonator induced
thirdorder 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 6.4mW peak power, 100ps
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 chalcogenideglass based
waveguiding structure, (γ/Aeﬀ = 60m− 1W− 1) were employed.
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 36ps pulses through a SCISSOR structure
composed of twenty lossy undercoupled resonators with 10µm 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 allpass ﬁlter, feedback is present
7.7 Generalized Periodic Resonator Systems 199
8
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 secondorder
Chebyschev reﬂection response, because a single notched microring with one
guide is equivalent to two coupled rings between two guides [150, 251].
9
Strictly speaking, the ﬁrst conﬁguration is constructed from twoport res
onators and is a special case.
200 7. Distributed Microresonator Systems
Bj Bj+1 Bj Bj+1
a)
Aj Aj+1 Aj Aj+1
b)
Aj Aj+1 Aj Aj+1
Bj Bj+1 Bj Bj+1
c)
Aj Aj+1 Aj Aj+1
Bj Bj+1 Bj Bj+1
d)
Aj Aj+1 Aj Aj+1
Table 7.1. Porttoport relations for the building blocks in Fig. 7.14.
M11
M12
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
E3 = r E1 + itE2 ,
E4 = itE1 + r E2 .
Table 7.1 shows the components Mij that represent the porttoport 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
R1 R2
1.546
R2 band 47
1.571 R1 band 31
R2 band 46
1.597
10 8 6 4 2 0  Lπ π
2L 0 π
2L
π
L 10 5 0 5 10
Norm. group index (ng/n) Bloch vector (keff) GVD (ps2/mm)
b) High finesse
1.498
Wavelength (µm)
1.546 R2 band 47
R1 band 31
1.571
R2 band 46
1.597
50 40 30 20 10 0  Lπ π
2L 0 π
2L
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 lowﬁnesse and (b) a highﬁnesse 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 doublechannel SCISSOR. (After Ref. [252], ©2004, Op
tical Society of America.)
204 7. Distributed Microresonator Systems
Next we revisit the singlechannel 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 feedforward 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
Singlechannel SCISSOR
a) Low finesse
1.498
Wavelength (µm)
1.522
1.546
1.571
1.597
10 8 6 4 2 0  Lπ π
2L 0 π
2L
π
L 10 5 0 5 10
Norm. group index (ng/n) Bloch vector (keff) GVD (ps2/mm)
b) High finesse
1.498
Wavelength (µm)
1.522
1.546
1.571
1.597
50 40 30 20 10 0  Lπ π
2L 0 π
2L
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 lowﬁnesse and (b) a highﬁnesse singlechannel 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 forwardtraveling wave is shown.
(After Ref. [252], ©2004, Optical Society of America.)
206 7. Distributed Microresonator Systems
Doublechannel SCISSOR
a) Low finesse
1.498
Wavelength (µm)
1.546
Bragg gap 47
Bragg gap 46
1.597
10 8 6 4 2 0  Lπ π
2L 0 π
2L
π
L 10 5 0 5 10
Norm. group index (ng/n) Bloch vector (keff) GVD (ps2/mm)
b) High finesse
1.498
Wavelength (µm)
Fig. 7.17. The dispersion relation, normalized group index, and GVD
for (a) a lowﬁnesse and (b) highﬁnesse doublechannel SCISSOR. Note
that unlike the dispersion relation for the singlechannel SCISSOR, the
doublechannel 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
resonatormediated backcoupling. 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
(interresonator 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 lowﬁnesse resonators. The interpretation is simple: In the high
ﬁnesse case, the band over which the individual resonators are reﬂecting
is narrow whereas in the lowﬁnesse 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 stopgaps, this structure is ideal for explor
ing nonlinear eﬀects [261] particularly within the bandgap, such as gap
solitons [243, 260].
1.522
1.546
1.571
1.597
10 8 6 4 2 0  Lπ π
2L 0 π
2L
π
L 10 5 0 5 10
Norm. group index (ng/n) Bloch vector (keff) GVD (ps2/mm)
b) High finesse
1.498
Wavelength (µm)
1.522
1.546
1.571
1.597
50 40 30 20 10 0  Lπ π
2L 0 π
2L
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 lowﬁnesse and (b) a highﬁnesse twisted doublechannel SCISSOR.
Bandgaps are absent in the dispersion relation, which resembles that of
the singlechannel SCISSOR, but with the presence of a second branch.
The two branches correspond to the two decoupled forwardtraveling
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 widebandwidth channeltochannel
coupling whereas in (b), a low coupling strength, t 2 = 0.1814, results
in narrowbandwidth channeltochannel coupling. To avoid redundancy,
and because the two forward and two backwardtraveling waves do not
couple, only the two forwardtraveling dispersion relation branches are
shown. (After Ref. [252], ©2004, Optical Society of America.)
210 7. Distributed Microresonator Systems
T T T
T T T
R R R
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 lowﬁnesse one, ﬁve, and
inﬁnite unitcelled structures. Figure 7.21 displays the transmission spec
tra for equivalent highﬁnesse 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 doublechannel 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
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 highreﬂectivity 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
−1
(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 lowﬁnesse cases whereas are strong
only in the highﬁnesse case for the CROW.
7.8 Summary
1.0
'
keff
(a)
(2FT/π)/L
γeff
0.5 (d)
(b)
''
keff (2F/π)2 γ 2πR/L
(2FT/π)2/L
0.0
'''
keff γeff
'
(c) (e)
(2FT/π)3/L (2F/π) γ 2πRT/L
3
0.5
4π 2π 0 2π 4π
F F F F
Normalized detuning, φ0
Fig. 7.22. Functional dependence of (a) the group velocity reduction, (b)
group velocity dispersion, (c) thirdorder dispersion, (d) selfphase mod
ulation coeﬃcient, and (e) selfsteepening 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
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 InPbased material systems.
8.1 Materials
Material systems being researched for microresonator fabrication include
various glasses and polymers (on a silicaonsilicon substrate), silicon
onsilica (using readily available silicononinsulator 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 massproduction 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, singlemode 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
8.3.1 Polymers
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. Lowtemperature 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].
Silica, silicon oxynitride (SiON), and silicon nitride (SiN) are typically de
posited with CVD (LowPressure CVD, PlasmaEnhanced 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.)
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 doublebonded device.
(c) Bonded interface. (Continued on next page.)
224 8. Fabrication Techniques for Microresonators
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 bondarea 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 sidecladding 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 lowindex 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 sidecladding (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 airclad 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
two.
Growth substrate
Upper/lower cladding
Core
(a) Epitaxially grown wafer.
Coupling gap
Bus Resonator Bus
(a)
(b)
Fig. 8.3. Scanning electron micrographs of waveguide crosssection. (a)
0.5µmwide waveguide in the ﬁlter section, (b) 3µmside waveguide at
the input and output(s). The dark bands are the quaternary layers that
comprise the core of the waveguide.
8.7 PolymerBonded, III–V Vertically Coupled Passive Microresonators 229
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 etchstop layer is removed with
8.7 PolymerBonded, III–V Vertically Coupled Passive Microresonators 231
Growth substrate
Etch stop
Upper/lower cladding
Core
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 200nm midlayer
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
Bus Bus
Polymer glue
Transfer substrate
(d) Flip patterned chip and bond to transfer
substrate.
Fig. 8.6. (Continued from previous page.) Process ﬂow for the fabrication
of polymerbonded, vertically coupled microresonators. (Continued on
next page.)
Polymer glue
Transfer substrate
(e) Remove growth substrate and etch stop to
expose “other”side of epilayer for processing.
Polymer encapsulation
Resonator
Bus Bus
Polymer glue
Transfer substrate
(f) Pattern etch mask for resonator, etch res
onator, and spinon 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 polymerbonded, vertically coupled microresonators.
conﬁned waveguides possible on both device layers. Also, the BCB en
capsulation provides the symmetric refractiveindex 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 layertolayer 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 etchedthrough alignment keys, provides a highquality
layertolayer alignment with errors smaller than 0.25 µm (measured
from onchip verniers; Fig. 8.8). The fabricated singlemicroring optical
channeldropping ﬁlter is shown in Fig. 8.9. Residuals from removal of
234 8. Fabrication Techniques for Microresonators
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.
Here is an example of the process ﬂow for active laterally coupled mi
croresonators, adapted from [174]. The waveguide stack is MBEgrown
pInP/GaInAsPMQW/nInP, with the central MQW region composed of a
8.8 Active III–V Laterally Coupled Microresonators 235
Coupling gap
Bus
pInP
Resonator
nInP
Contact
Planarization
InP cladding
Growth substrate
∼4.5 µm. The tall, grasslike 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 CrSiO2 mask
and the particles that caused micromasking 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 TiPtAu 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
shortcircuited.
Grover et al. chose to use the lateral coupling approach for the device de
scribed in this chapter because of the simplicity of fabrication. Realworld
applications typically will require the use of vertical coupling so that the
bus and the ring can be of diﬀerent bandgaps allowing for activepassive
8.9 Other Conﬁgurations 239
(a)
(b)
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 electrooptic modulators.
240 8. Fabrication Techniques for Microresonators
(a)
Fig. 8.15. (a) Rabiei’s process for the fabrication of large polymer micror
ings. (Continued on next page.)
Teflon
Si
SU8 SU8
Spin coating, UV 15 Protection Layer
Baking, RIE Teflon
Teflon Si
Si
Patterning RIE
Etching
Spin coating, UV Teflon
Cure, Baking, RIE
SU8 SU8
UV 15 Protection Layer
UV 15 Protection Layer
Teflon
Teflon
Si Si
SU8
UFC 170
UV 15 Protection Layer
Teflon
Teflon
SU8 SU8
Si
UV 15 Protection Layer
Teflon
Pattering, UV Exposure
Si
SU8 SU8
RIE Etching
UV 15 Protection Layer
Teflon
Teflon UFC170
Si
SU8 SU8
UV 15 Protection Layer
Teflon
Spin coating,
Si
Baking, RIE
(b)
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.
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, UrbanaChampaign [74], in collab
oration with researchers from Sarnoﬀ, uses MMIcoupling for laterally
coupled microresonators.
2. The group at Twente University [294] uses SiON waveguides with a
combination of lowpressure and plasmaenhanced chemical vapor
deposition, as does the group at Politecnico di Milano [295].
3. Little Optics (now Inﬁnera) uses glassbased 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.
Bibliography
64. K. Djordjev, S.J. Choi, S.J. Choi, and P. D. Dapkus, “HighQ 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,
“Lowthreshold InGaAsP ring lasers fabricated using bilevel dry etching,”
IEEE Photonics Technology Letters, vol. 12, pp. 146–148, 2000.
71. D. G. Rabus and M. Hamacher, “MMIcoupled 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, “HighQ 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 singlemode GaInAsPInP 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 InPbased
singlemode 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 GaInAsPInP 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 sodiumﬁlled 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
155. R. Kronig, “On the theory of the dispersion of Xrays,” 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, “KramersKronig rela
tions for an interferometer,” Optics Communications, vol. 138, pp. 118–126,
1997.
158. L. J. Wang, “Causal ‘allpass’ ﬁlters and KramersKronig 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 doublering opticalguidedwave
Vernier resonator,” IEE Proceedings (Optoelectronics), vol. 144, pp. 203–208,
1997.
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 highorder microringresonator
adddrop 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 parallelcoupled ringresonator ﬁlter,” Optics
Letters, vol. 26, pp. 917–919, 2001.
166. F. Niklaus, P. Enoksson, E. Kälvesten, and G. Stemme, “Lowtemperature 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,
“Lowtemperature 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 PdIn metal bonding and laser liftoﬀ,” 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 PdIn metal bonding and laser liftoﬀ,” 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, “3D
photonic circuit technology,” IEEE Journal of Selected Topics in Quantum
Electronics, vol. 8, pp. 935–942, 2002.
Bibliography 253
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
– electrooptic, 110 CROW, 202
– need for, 107 – bandgap engineering, 210
– thermal tuning, 110
add–drop rings, 84 delay lines, tunable, 195
– alloptical 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
alloptical switching, 6, 149
– add–drop, see add–drop rings eﬀective index method, 23
– allpass, see allpass rings electrooptic eﬀect, 110
– Fabry–Perot, 137
– pump reshaping, 156 fabrication
– pumpprobe experiments, 154, 159 – feature deﬁnition, 221
– pumpprobe simulations, 152 – lateral active III–V, 234
– response time measurement, 160 – lateral passive III–V, 225
– ringenhanced MZ, 137 – multilayer, 222
– routing, 161 – polymer rings, 242
– theory, 149 – vertical passive III–V, 227
– thresholding, 156 – wafer bonding, 222
– timedivision mux/demux, 158 fast light, 191, 197
allpass rings, 74, 124 ﬁgures of merit, 142
– alloptical switching, 137 ﬁlters, 97
– attenuation in, 83 – parallelcascade of rings, 99
– ﬁnesse, 77 – practical limitations to cascading,
– group delay, 79 103
– group delay dispersion, 79 – serialcascade of rings, 99
– intensity buildup, 75 – vernier eﬀect, 103
– phase shift in, 78 ﬁnesse
– enhanced nonlinearities, 124
261
262 Index
Molto più che documenti.
Scopri tutto ciò che Scribd ha da offrire, inclusi libri e audiolibri dei maggiori editori.
Annulla in qualsiasi momento.