Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Small-signal model parameter extraction for microwave SiGe HBTs based on Y- and Z-
parameter characterization
This content has been downloaded from IOPscience. Please scroll down to see the full text.
(http://iopscience.iop.org/1674-4926/30/8/084005)
View the table of contents for this issue, or go to the journal homepage for more
Download details:
IP Address: 207.162.240.147
This content was downloaded on 15/07/2017 at 06:51
Small-signal modeling of GaN HEMT switch with a new intrinsic elements extraction method
Miao Geng, Pei-Xian Li, Wei-Jun Luo et al.
Novel Extraction Method for Source and Drain Series Resistances in Silicon Nanowire
Metal–Oxide–Semiconductor Field-Effect-Transistors Based on Radio-Frequency Analysis
Kyung Rok Kim, Sunhae Shin, Seongjae Cho et al.
Apparent base resistance decomposition by means of small-signal and high-frequency noise analyses
of submicron InP/InGaAs HBTs
E Ramirez-Garcia, F P Aniel, M A Enciso-Aguilar et al.
(Tsinghua National Laboratory for Information Science and Technology (TNList), Institute of Microelectronics, Tsinghua University,
Beijing 100084, China)
Abstract: High frequency intrinsic small-signal model parameter extraction for microwave SiGe heterojunction
bipolar transistors is studied, with a focus on the main feedback elements including the emitter series resistor, internal
and external base–collector capacitors as well as the base series resistor, all of which are important in determining
the behavior of the device equivalent circuit. In accordance with the respective features of definition of the Y- and Z-
parameters, a novel combined use of them succeeds in reasonably simplifying the device equivalent circuit and thus
decoupling the extraction of base–collector capacitances from other model parameters. As a result, a very simple
direct extraction method is proposed. The proposed method is applied for determining the SiGe HBT small-signal
model parameters by taking numerically simulated Y- and Z-parameters as nominal “measurement data” with the
help of a Taurus-device simulator. The validity of the method is preliminarily confirmed by the observation of certain
linear relations of device frequency behavior as predicted by the corresponding theoretical analysis. Furthermore, the
extraction results can be used to reasonably account for the dependence of the extracted model parameters on device
geometry and process parameters, reflecting the explicit physical meanings of parameters, and especially revealing
the distributed nature of the base series resistor and its complex interactions with base–collector capacitors. Finally,
the accuracy of our model parameter extraction method is further validated by comparing the modeled and simulated
S -parameters as a function of frequency.
Key words: SiGe; heterojunction bipolar transistors; small-signal; model; parameter extraction
DOI: 10.1088/1674-4926/30/8/084005 EEACC: 1350F; 2560B; 2560J
084005-1
J. Semicond. 30(8) Fu Jun
qIC
gm = gm0 exp(-jωτd ) ≈ (1 − jωτd ), (1)
kB T
Fig. 1. Common-emitter hybrid π-mode high frequency small-signal where ω is the angular frequency, and τd stands for the transit
equivalent circuit of the intrinsic part of an SiGe HBT. time phase delay of the transconductance.
The examination of the circuit in Fig.1 indicates that
errors, but also certain extrinsic parasitics e.g. collector-to- the presence of the feedback capacitors Cµ and Cbc span-
substrate capacitances cannot be deembedded at all through ning across input and output makes it difficult to formulate
the use of industry standard dummy test structures. On the the small-signal model analytically. On the other hand, even if
other hand, although most of the extrinsic parasitic elements normally the values for Cµ and Cbc may be much less than Cπ ,
can normally be determined by biasing the device in “hot” neither of them can be neglected arbitrarily due to the well-
(overdriven) or/and “cold” (cut-off) operation[4, 6, 8−12] , this known Miller effect. In other words, seen from the input point
kind of technique is unsuitable for extracting bias dependent of view, the feedback capacitances will be approximately mag-
extrinsic parameters. Therefore, in Ref. [7], this kind of tech- nified by a large factor of 1 + |Av | if e.g. the output is left open-
nique was applied in conjunction with the use of dummy struc- circuited, resulting in a high value of the small-signal voltage
tures for deembedding correction. No doubt, anyway, the im- magnification Av . It is worth noting that, nevertheless, if the
perfect deembedding and extrinsic parameter determination output (collector terminal) is shorted to ground (emitter ter-
will in turn interfere with the intrinsic small-signal modeling minal), the Miller effect will be greatly alleviated because of
parameter extraction, and even may cover up the physical na- the nearly zero value of Av in this case. Consequently, when
ture of device modeling to a great degree. In this sense, the characterizing the device using Y-parameters which are de-
separation of intrinsic parameter extraction from extrinsic par- fined under a short-circuited port condition, the small-signal
asitics determination is necessary. equivalent circuit can be reasonably simplified by removing
The purpose of this work is to explore physical insights the base–collector capacitors. Furthermore, as the output re-
into the device modeling of intrinsic small-signal equivalent sistor ro is normally large enough (in view of the very large
circuits for SiGe HBTs, while leaving parasitic extrinsic el- values of Early voltages for SiGe HBTs) to be neglected from
ements and associated deembedding techniques beyond the the above simplified circuit, one can obtain the following rela-
scope of this paper. As we know, a device simulator is a good tions based on Y-parameter characterization.
choice for creating an ideal device architecture excluding any !
extrinsic elements which may otherwise be difficult or even y11 1 τdCπ 2
Re = − ω , (2)
impossible to filter from an actual measurement set-up. So, in y21 gm0 rπ gm0
order to eliminate interference from extrinsic elements with
τ
the intrinsic part of a device equivalent circuit, numerical sim- ! Cπ + d
y11 rπ
ulation results instead of real measurement data are used for Im = ω, (3)
y21 gm0
parameter extraction in this paper. Based on the simulation
data, through the combination of Y- and Z-parameter charac- rb + re
! 1+
1 rπ (rb + re )Cπ τd 2
terization and reasonable approximation, a simple direct ex- Re = re + − ω . (4)
traction method is proposed to extract small-signal model pa- y21 gm0 gm0
rameters including base and emitter series resistances, and According to the above formulae, the linear regression of
external and internal base–collector capacitances, for SiGe Re(y11 /y21 ) and Re(1/y21 ) with respect to ω2 and the linear re-
HBTs. gression of Im(y11 /y21 ) versus ω, respectively, will determine
both slopes and intercepts of these linear relations, leading to
2. Theory altogether five equations, which can just be combined to solve
for the five unknown small-signal model parameters rb , re , rπ ,
After leaving out all of the extrinsic parasitic elements, as Cπ and τd .
illustrated in Fig. 1, the common-emitter (CE) hybrid π-mode We now turn to the Z-parameters which are defined with
high frequency small-signal equivalent circuit of the intrinsic input and output kept open-circuited, respectively. Because of
part of a SiGe HBT is investigated, with a focus on base series the severe Miller effect in this case, neither Cµ nor Cbc can be
084005-2
Fu Jun August 2009
1 rb2Cµ2Cbc
2
= − Cµ + Cbc ω − ω3 . (6)
Im (z22 − z21 ) Cµ + Cbc
As will be demonstrated in Section 4, the second term is nor-
mally negligibly small compared to the first one at frequencies
of interest in the right-hand side of the above expression, so it
can be approximated as below. Fig. 2. Device architecture of an SiGe HBT for simulation with only
right half taken considering the device symmetry.
1
≈ − Cµ + Cbc ω. (7)
Im (z22 − z21 ) emitter are also shown in the inset of the figure. Regarding
Obviously from Eq. (7), the linear dependence of 1/Im(z22 – lateral dimensions, as labeled inside Fig. 2, the widths of the
z21 ) on ω can be used to determine the value of the slope i.e. emitter and base electrodes are denoted as ew and xb, respec-
–(Cµ + Cbc ). In combination with Eq. (5) where rb has already tively, with the link-up base width lb defined as the distance
been extracted, the internal and external base–collector capac- between them. It should be noted that, unless otherwise stated,
itances can be obtained in the end. ew = 0.55 µm, lb = 0.3 µm, xb = 1.35 µm, n- epi doping con-
To sum up, owing to the combination of Y- and Z- centration Nepi = 1 × 1016 cm−3 and top plateau base doping
parameter characterization, a very simple direct extraction concentration NB = 5 × 1019 cm−3 are used for simulation with
method is proposed here while taking all the main feedback the total device width (along the direction perpendicular to the
elements (Cµ , Cbc and re ) as well as base series resistor rb into cross-section) fixed at 1 µm.
account. An AC small-signal simulation on the above defined de-
vice structure was carried out under bias conditions of Vbe =
3. Simulation 0.85 V and Vce = 5 V at frequencies ranging from 0.125 to 2
GHz. During the simulation, physical models of bandgap nar-
In this work, the simulation of a SiGe HBT with a nu- rowing induced by Ge content as well as the heavy doping
merical device simulator Taurus-device[15] is used to provide effect, Hansch field dependent carrier mobility and impurity
the relevant small-signal Y- and Z-parameters required for the concentration dependent SRH and Auger recombination are
model parameter extraction. employed by the Taurus-device simulator[15] . Finally, the sim-
The cross-section of the SiGe HBT device architecture ulated Y- and Z-parameters are collected for the subsequent
for simulation is schematically shown in Fig. 2, where only small-signal parameter extraction.
the right half is taken in view of the device symmetry. The
geometry and process details are summarized below. On top 4. Results and discussion
of a 2 µm thick lightly doped n-type epitaxial layer, a p-type-
doped base Si/SiGe/Si stack epitaxial layer with a total thick- Based on these Y- and Z-parameters obtained from nu-
ness of 55 nm is placed. In addition, heavily doped n+ and p+ merical high frequency small-signal simulation of the SiGe
ohmic regions are introduced at the surface of the device as HBT as illustrated in Fig. 2, the left hand side of the expres-
emitter and external base contacts, respectively. Meanwhile, sions of Eqs. (2)–(4), Eq. (5) and Eq. (7) are calculated and
another n+ division at the backside of the device is used to used to plot the corresponding curves as a function of ω2 or ω,
construct the collector contact. The emitter, base and collec- respectively, in Figs. 3, 4, and 5. As shown in Fig. 3, it is clear
tor electrodes are tied to the corresponding contacts. All of that both Re(y11 /y21 ) and Re(1/y21 ) exhibit descending linear
the doping concentration profiles in the contact regions and dependence on ω2 , just as predicted by Eq. (2) and Eq. (4),
inside the Si/SiGe/Si epitaxial layer are created according to respectively. Meanwhile, Figure 4 clearly shows an ascending
the Gauss distribution function. Also, trapezoidal Ge distribu- relation of Im(y11 /y21 ) in direct proportion to ω, in accordance
tion is employed with top plateau Ge percentage being 20%. with Eq. (3). On the other hand, as can be seen from Fig. 5,
Correspondingly, the vertical profiles of Ge mole fraction and Re(1/(z11 –z12 )) manifests itself nearly as a constant indepen-
impurity density along the cutline through the center of the dent of ω while 1/Im(z22 –z21 ) decreases directly proportional
084005-3
J. Semicond. 30(8) Fu Jun
Fig. 3. Re(y11 /y21 ) and Re(1/y21 ) versus ω2 calculated from high fre- Fig. 6. Variation of extracted base-collector capacitances (internal Cµ ,
quency small-signal simulation results simulated by a Taurus-device external Cbc and total Ct ) with doping concentration of n- epitaxial
simulator for the SiGe HBT as illustrated in Fig. 2. layer Nepi for the SiGe HBT as illustrated in Fig. 2.
Fig. 4. Im(y11 /y21 ) versus ω calculated from high frequency small- Fig. 7. Variation of extracted base-collector capacitances (internal Cµ ,
signal simulation results simulated by a Taurus-device simulator for external Cbc and total Ct ) with emitter width ew for the SiGe HBT as
the SiGe HBT as illustrated in Fig. 2. illustrated in Fig. 2.
Fig. 5. Re(1/(z11 − z12 )) and 1/Im(z22 − z21 ) versus ω calculated from Fig. 8. Variation of extracted base-collector capacitances (internal Cµ ,
high frequency small-signal simulation results simulated by a Taurus- external Cbc and total Ct ) with external base width xb for the SiGe
device simulator for the SiGe HBT as illustrated in Fig. 2. HBT as illustrated in Fig. 2.
to ω with a negative slope, in good agreement with prediction tration Nepi , emitter width ew, and external base width xb in
given by Eq. (5) and Eq. (7), respectively. As a consequence, Figs. 6, 7 and 8, respectively. Beginning with Fig. 6, the ex-
the theoretical analysis given in Section 2, as a basis of the tracted values of Cµ , Cbc and their sum Ct = (Cµ + Cbc ) re-
method proposed for extracting the small-signal equivalent cir- flect a reasonable increase in base–collector p–n junction ca-
cuit model parameters of the SiGe HBT, has been justified, at pacitances with Nepi . Apparently it would be beneficial in the
least to some extent, by the device simulation. reduction of base–collector capacitances to adopt as lightly
From the plots of Figs. 3, 4, and 5, the linear regres- doped a collector epi as possible. Then, as shown in Fig. 7,
sion lines are derived and drawn together with the simulation Cµ decreases in direct proportion to the shrinking emitter
data symbols as shown in these figures, from which the lin- width and at the same time Cbc is found to remain almost
ear regression slopes and intercepts can be readily determined. unchanged with ew shrinkage. In contrast, Figure 8 shows
Therefore, the small-signal equivalent circuit model parame- that Cµ is hardly affected by variation in xb while Cbc fol-
ters have been extracted in this simple manner, with a focus lows a decreasing linear tendency as xb narrows. The ob-
upon analysis of rb , re , Cµ , and Cbc below in this section. served dependence of Cµ and Cbc agrees well with the split-
The extracted values for the base–collector capacitors ting of the base–collector capacitor defined in the equivalent
are plotted as a function of n- collector epi doping concen- circuit, reflecting an explicit and reasonable physical meaning
084005-4
Fu Jun August 2009
084005-5
J. Semicond. 30(8) Fu Jun
Fig. 12. Magnitudes of S 11 and S 22 versus frequencies calculated with Fig. 15. Angles of S 21 and S 12 versus frequencies calculated with an
an ADS circuit simulator based on our model compared with the cor- ADS circuit simulator based on our model compared with the corre-
responding simulation results given by a Taurus-device simulator for sponding simulation results given by a Taurus-device simulator for
the SiGe HBT as illustrated in Fig. 2. the SiGe HBT as illustrated in Fig. 2.
5. Conclusions
In this paper, a simple method is proposed to directly ex-
tract high frequency CE hybrid-π mode small-signal model pa-
rameters for microwave SiGe HBTs, with a focus on the main
feedback elements including the emitter series resistor, inter-
nal and external base–collector capacitors as well as the base
series resistor in the device equivalent circuit. A novel com-
Fig. 13. Angles of S 11 and S 22 versus frequencies calculated with an
bined use of Y- and Z-parameters enables reasonable simplifi-
ADS circuit simulator based on our model compared with the corre-
cation of the device equivalent circuit and thus decoupling the
sponding simulation results given by a Taurus-device simulator for
extraction of base–collector capacitances from other model pa-
the SiGe HBT as illustrated in Fig. 2.
rameters. As a result, the parameter extraction is greatly sim-
plified through two-step Y- and Z-parameter characterization.
The validity of the method is preliminarily confirmed by us-
ing numerical device simulation data for extracting the cor-
responding model parameters. Furthermore, the extraction re-
sults can be used to reasonably account for the dependence
of the extracted model parameters on device geometry and
process parameters, reflecting the explicit physical meanings
of parameters, and especially revealing the distributed nature
of the base series resistor and its complex interactions with
base–collector capacitors. Finally, the accuracy of our model
parameter extraction method is further validated by compar-
Fig. 14. Magnitudes of S 21 and S 12 versus frequencies calculated with
ing the modeled and simulated S -parameters as a function of
an ADS circuit simulator based on our model compared with the cor-
frequency.
responding simulation results given by a Taurus-device simulator for
the SiGe HBT as illustrated in Fig. 2. References
device model parameter extraction is based. It is worth men- [1] Nakadai T, Hashimoto K. Measuring the base resistance of
tioning here that the value of collector series resistance rc , bipolar transistors. IEEE Bipolar Circuits and Technology
which is necessary for ADS calculation, is estimated following Meeting, 1991: 200
Eq. (6) in Ref. [3], i.e. [2] Pehlke D R, Pavlidis D. Direct calculation of the HBT equiva-
lent circuit from measured S -parameters. IEEE MTT-S Digest,
rc = Re(z22 − z21 ). (10) 1992: 735
[3] Lee S, Ryum B R, Kang S W. A new parameter extraction tech-
As shown in Figs. 12–15, the calculated curves are found nique for small-signal equivalent circuit of polysilicon emitter
to be in good agreement with the corresponding simulated bipolar transistors. IEEE Trans Electron Devices, 1994, 41(2):
084005-6
Fu Jun August 2009
084005-7