Version of Record: https://www.sciencedirect.

com/science/article/pii/S0038110120303828
Manuscript_97a4b071c1bc33f1a14c67ec9baa6398

Solid State
Electronics
00 (2020) 1–8

TCAD and EM Co-Simulation Method to Verify SiGe HBT

Measurements up to 500 GHz
Soumya Ranjan Panda1,2 , Sebastien Fregonese1 , Marina Deng1 , Anjan Chakravorty2 and Thomas Zimmer1
1 IMS Laboratory, University of Bordeaux, Talence, France

2 Department of Electrical Engineering, IIT Madras, Chennai, India

Abstract
A systematic method for the verification of high frequency measurement (up to 500 GHz) of silicon germanium heterojunction
bipolar transistor (SiGe HBT) is proposed. First of all, the method involves an accurate estimation of the effects of passive envi-
ronment on the overall measurement by a detailed electro-magnetic (EM) simulation. This ensures that the complete measurement
environment like probes, pads and access lines along with the appropriate layouts are precisely included in the EM simulation
framework. In order to additionally include the active device like S iGe HBTs, technology computer aided design (TCAD) tool
is used to simulate the device S -parameters. TCAD simulation results are fed into an EM-plus-SPICE simulation framework to
emulate a complete on-wafer measurement environment. The final simulation results show appreciable correlation with the on-
wafer measurement data up to 500 GHz. Further the need of proper calibration, de-embedding in high frequency characterization
are emphasized through the investigation of the S-parameters corresponding to a narrow-band amplifier at 170 GHz suitable for
G-band RADAR applications.

Keywords: SiGe HBT, HF S -parameter measurement, on-wafer, TRL calibration, de-embedding, TCAD-EM co-simulation,
amplifier.

1. Introduction models to be deployed in VLSI tools. The physics-

Moderate cost and compatibility of SiGe based de-
vices with CMOS technology motivated and enabled
continuous horizontal and vertical scaling of these de-
vices over last two decades in order to achieve high-
speed performances [1]. Further for radiometers or
instruments for millimeter wave (mmW) remote sens-
ing on small satellites, traditionally used InP and GaAs
technologies do not come out to be high yielding and
100 mm wafer size makes the production expensive.
Here SiGe based technology stands strong with up to
300 mm wafers and other advantages like excellent 1/f
noise, total ionizing radiation tolerance and a potential
for single chip integration that make it better suited for
the above applications.
A reliable circuit design requires accurate compact
1
based models that are used in VLSI tools are highly de-
pendent on accurate DC and RF characterization. Hence
it is important to understand the S -parameter character-
ization in high frequency range and verify the models in
VLSI tools in that range. But the characterization above
110 GHz is still a challenge and there is no established
de-embedding method above this range. In fact, a very
few research groups have published S -parameter mea-
surement results above 110 GHz [2] [3] [4] [5] [6].
High frequency measurements require careful cali-
bration procedures. Off-wafer calibration is more pop-
ular in the production environment [6]. However, since
the calibration substrate is different from that of the de-
sired wafer to be characterized, the off-wafer calibration
leads to systematic errors, which is correlated to differ-

© 2020 published by Elsevier. This manuscript is made available under the Elsevier user license
https://www.elsevier.com/open-access/userlicense/1.0/

0 VBE= 0.7V
-5
mag(S 22) (dB)
-10
-15 1V with step size of 50mV
-20 Symbols : Measurement
Line : TCAD
-25
0 100 200 300 400
Frequency (GHz)
Figure 1: Frequency-dependent magnitude of S 22 for the SiGe HBT
corresponding to STMicroelectronics B55 process at different VBE :
comparison between actual measurement and TCAD simulation.
Figure 2: TEM picture (left) of SiGe HBT and corresponding simu-
lated TCAD structure (right).
ent probe-to-substrate electromagnetic coupling [7][8].
Similarly, as different probes are used to measure differ-
ent frequency bands, discontinuity appears in the mea-
sured data due to change in RF probes if the probe’s
cross-talk is not fully corrected by the calibration pro-
cedure. A detailed explanation of the work has been
reported in [8].
On-wafer calibration is a better choice for high fre-
quency (HF) measurement because it avoids some of
the above mentioned limitations observed in off-wafer
calibration. Thru-Reflect-Line (TRL) calibration is the
method of choice [2] when carrying out an on-wafer cal-
ibration. Using TRL, the reference plane is well defined
which helps to partially mitigate the error due to probe-
to-substrate coupling [2]. But while probing the Thru
and Line, it is necessary to shift one probe which may
Soumya et al. / 00 (2020) 1–8 2
)
As SIMS
-3
20
10 B SIMS
Doping Concentration (cm
TCAD
19
10
18
10
30 SIMS
TCAD
25
Germanium (%)
20
15
10
17
5
10 0
0.015 0.02 0.025 0.03
Distance
Distance
Figure 3: Vertical doping profiles: SIMS profiles of As and B doping
(shown in different symbols) and corresponding analytic profile (line)
incorporated in TCAD. Inset shows the Ge mole-fraction measured by
500
EDX (symbol) and one used in TCAD (line).
add errors in the measurement particularly in case of
manual probe stations [4][9]. The drawback of the on-
wafer calibration is the need for a specific design of test
structure which can be time-consuming. This design re-
quires engineering effort and expertise. For example,
the line lengths need to be chosen properly to cover the
required frequency band; the geometry of the line needs
to minimize the loss and allow only one mode to prop-
agate; the design of the pad itself needs to be optimized
to limit the probe-to-substrate coupling; the pad-to-pad
distance needs to be sufficiently long in order to ensure
a reduced cross-talk and finally one has to keep enough
space between the structures to avoid coupling [8][10]
[11].
In order to check the trend, consistency and accuracy
of the high frequency measurement results, one has to
compare the data with those obtained from calibrated
simulation. One can think of different options such as
ab-initio based simulation, TCAD simulation and elec-
trical compact model simulation for this purpose. In this
work, we have chosen TCAD tool as a good compro-
mise between the complexity, accuracy and simulation
time [12].
In this paper, using Sentaurus TCAD and Ansys-
HFSS electro-magnetic (EM) simulations, we explore
and verify the high frequency measurements of SiGe
HBTs. As a motivation for such an investigation,
Fig. 1 compares the measurement and calibrated TCAD
simulation of frequency-dependent S -parameters corre-
sponding to a SiGe HBT from STMicroelectronics B55
technology. It is observed that in the lower frequency
range (below 110 GHz), the measurement and simu-
lation are in good agreement. However, some discon-
tinuities are observed in the measurement data at the
cross-over points (110 GHz, 220 GHz and 325 GHz)
2
 Soumya et al. / 00 (2020) 1–8 3

of different frequency bands. Also the characteristics

in the upper frequency range tend to deflect from the 10
1
Symbols : Measurement

TCAD simulations. This motivates us to emulate a Line : TCAD

mm
0
10 AE=90nm x 4.8
more realistic measurement environment. To do so,
-1
we combine our TCAD simulation results with EM 10

IB, I C (mA)
simulation that includes the back-end-of-line (BOEL) 10
-2

along with the probes. This TCAD plus EM simula- -3

10
tion data together will be henceforth called as Virtual VCB=0V
-4
measurement throughout the paper. After that, we fol- 10 T=294K

low all the on-wafer calibration and de-embedding pro- 10

-5

cedures on this virtual measurement data. Finally for -6

10
verification, both TCAD and de-embedded virtual mea-
-7
surement data are compared with actual measurement 10

results. Also a circuit-level verification is carried out 10

-8

by designing a G-band single-stage amplifier based on 0.4 0.6 0.8 1.0

TCAD as well as de-embedded virtual measurement VBE (V)

data.
Figure 5: Measured and TCAD simulated Gummel characteristics for
Section 2 of this paper provides a comprehensive dis- 90 nm × 4.8 µm SiGe HBT at VCB =0V.
cussion on the calibration of TCAD simulation. In sec-
tion 3, a brief description of the measurement setup and
procedures are provided. The methodology of combin- The doping profiles have been reproduced using ana-
ing TCAD and EM simulation to produce virtual mea- lytic mathematical models as shown in Fig. 3 [13][12].
surement results is explained in section 4. In section All necessary physical effects are taken into account in
5, the assessment of virtual measurement data is car- the simulation using appropriate models and parameter
ried out in terms of some figures-of-merit (FoMs) of sets are appropriately tuned from Monte Carlo simula-
a narrow-band 170 GHz amplifier circuit. Finally, we tion as discussed below [14][15].
conclude in section 6. For the carrier transport, we have used hydrodynamic
model that solves drift-diffusion equations along with
25 the additional energy balance equations for minority
Symbols : Measurement

Line : TCAD
20 Symbols : Measurement

Line : TCAD

AE=90nm x 4.8 mm
carriers. The Slootboom bandgap narrowing model with
20
AE=90nm x 4.8 mm
15
appropriately tuned Nre f value is used to account for the
Cbe (fF)

Cbc (fF)

15
effect of carrier concentration at heavy doping [16]. Pa-
10
10
rameter in the Caughey-Thomas model has also been
5
tuned as suggested in [16]. Shockley-Read-Hall and
5
VC=0V, V E=0V

f = 5 GHz
VC=0V, V E=0V

0
-0.8 -0.4 0.0 0.4
0
-0.8
VB (V)
(a)
Figure 4: Base-emitter voltage dependent (a) base-emitter capacitance
(C BE ) and (b) base-collector capacitance (C BC ): comparison between
actual measurement and TCAD simulation.
2. TCAD Simulation Setup and Calibration
In commercially available Sentaurus TCAD simula-
tor an SiGe-HBT structure corresponding to ST Mi-
croelectronics’s B55 process is developed as shown in
Fig. 2. All the dimensions including the shallow and
deep trench isolation are ensured to be closely identical
with those of the B55 process during the device design.
f = 5 GHz
Auger recombination models are used to take care of
-0.4 0.0 0.4
the recombination effects and Lackner’s model has been
VB (V) considered to include the effect of breakdown at high
(b) electric field.
The base-emitter and base-collector junction capac-
itances, shown in Figs. 4(a) and (b) respectively, are
obtained from simulation at low-frequency and are
matched with the corresponding measured data to vali-
date the approximation used to emulate the doping pro-
file in TCAD. A maximum permissible error of 0.5 fF is
allowed while fitting the capacitances. The recombina-
tion parameter in poly-silicon emitter has been tuned to
match the base current. A more ideal germanium pro-
file is considered keeping the maximum mole fraction
within the permissible range. This effectively helped us
in capturing the collector current in the Gummel plot as
shown in Fig. 5.
Further, the TCAD simulated transit ( fT ) and maxi-
3
 Soumya et al. / 00 (2020) 1–8 4

350
Symbols : Measurement 350 Symbols : Measurement

Line : TCAD Line : TCAD

300
AE=90nm x 4.8 mm 300 AE=90nm x 4.8 mm
250

fmax (GHz)
250
fT (GHz)

200
200

150 V CB=0V 150 V CB=0V

T = 294K
T = 294K
100 100

50 50

0
Figure 8: Probe geometry used for different frequency bands: (a) 1-
0
0.01 0.1 1 10 0.01 0.1 1 10 110 GHz (b) 140-220 GHz (c) 220-325 GHz.
IC (mA) IC (mA)

(a) (b) EM
TCAD
Figure 6: Collector current dependent (a) transit frequency ( fT ) and
simulation
2-port
(b) maximum oscillation frequency ( fmax ) at VCB =0 V: comparison 4-port
between actual measurement and TCAD simulation.

Thru,
Reflect,
TRL Lines,
calibration Load,
procedure Trans.-
open,
Trans.-short
De-
embedding

Figure 9: TCAD-EM co-simulation flow leading to a virtual measure-

ment.

Figure 7: Side view of the EM simulation setup for probes enclosed

within an air-box.
This is also validated after looking into the profiles
received from ST Microelectronics (not disclosed here
due to intellectual properties agreements).
mum oscillation frequencies ( fmax ) are compared with
the measured data at VCB =0 V and T=294 K in
Figs. 6(a) and (b), respectively, prior to carry out the 3. Actual Measurement Procedure
S -parameter comparison [17]. It is to be noted that for
the fT and fmax extraction, a spot frequency at 63 GHz We have carried out the S -parameter measurements
has been chosen [18]. up to 500 GHz for VBE varying from 0.7 V to 1 V and
Since the Gummel (Fig. 4) and fT characteristics VBC =0V. In order to measure from 1 GHz to 500 GHz,
(Fig. 6(a)) show high level of correlation between the four different measurement benches such as (i) Agi-
TCAD and measured data and since these two charac- lent’s E8361A VNA up to 110 GHz using extenders
teristics are highly dependent on the intrinsic part of the (N5260-60003) above 67 GHz (ii) 140-220 GHz (iii)
HBT, one can conclude that a reasonably accurate cal- 220-330 GHz and (iv) 325-500 GHz bands with a four-
ibration of the 1D profile of the device has been ob- port Rohde & Schwarz ZVA24 VNA coupled with ex-
tained in TCAD. Since fmax is dependent on fT and tenders (ZC220-ZC330-ZC500) have been used. The
other external parameters such as base resistance and power level is set to below -32 dBm in the four bands for
base-collector and collector-substrate capacitances [19], measurement of both active and passive elements (the
the high level of agreement between the TCAD and latter are used for de-embedding purposes). Picoprobe
measured fmax characteristics ensures that the profile pa- RF probes with 50 µm pitch are used for covering the
rameters related to the extrinsic base, buried layer in the bands above 110 GHz and 100 µm pitch probes are used
collector region and the substrate doping at its contact below 110 GHz. On-wafer TRL calibration and stan-
and around the deep trenches are closely identical with dard Short-Open de-embedding has been performed.
those of the B55 process. The reference plane of the TRL is set at the top metal
4
 Soumya et al. / 00 (2020) 1–8 5

Symbols : Measurement 0 Symbols : Measurement

100
Symbols : Measurement
0 Short Dash : TCAD Short Dash : TCAD -10 VBE=0.8V Short Dash : TCAD

Line : Virtual measurement -20 Line : Virtual measurement Line : Virtual measurement

80
-2 -40

mag(S 12) (dB)

phase (S 11) ( )

phase (S 12) ( )
mag(S 11) (dB)

-20 VBE=0.9V

0
60
VBE=0.8V -60
-4
VBE=0.9V
-80 40
-30
VBE=0.8V
-6 -100
20
Symbols : Measurement
-120 -40 Short Dash : TCAD

-8 Line : Virtual measurement

0
VBE=0.8V
-140
VBE=0.9V -20
VBE=0.9V
-160 -50
-10

-180 -40
0 100 200 300 400 500 0 100 200 300 400 500 0 100 200 300 400 500 0 100 200 300 400 500

Frequency (GHz) Frequency (GHz) Frequency (GHz) Frequency (GHz)

(a) (b) (a) (b)

Figure 10: Frequency dependent (a) magnitude and (b) phase of S 11 Figure 11: Frequency-dependent (a) magnitude and (b) phase of S 12
at VCB =0 V at VBE = 0.8 V and 0.9 V for the SiGe HBT up to metal- at VCB =0 V VBE =0.8 V and 0.9 V for the SiGe HBT up to metal-
1: comparison of de-embedded actual and virtual measurements with 1: comparison of de-embedded actual and virtual measurements with
TCAD simulation. TCAD simulation.

and at sufficiently close proximity of the vias. This ref- 25

Symbols : Measurement

Short Dash : TCAD

180
Symbols : Measurement

Short Dash : TCAD

erence plane removes the distributed nature of the as- 20

Line : Virtual measurement

150
Line : Virtual measurement

Phase (S 21) ( )
sembly formed by the access line, vias, and transistor-
Mag(S 21) (dB)

0
15 120

open or transistor-short. 10
VBE=0.9V
90

60
5 VBE=0.9V

30
0 VBE=0.8V

4. Virtual Measurements, Results and Discussion -5

VBE=0.8V
0

-30

-10

As discussed in section 1, after observing a mismatch 0 100 200 300

Frequency (GHz)
400 500 0 100 200 300

Frequency (GHz)
400 500

between the actual measurement data for S -parameters

with calibrated TCAD simulation, we decided to in- (a) (b)

clude the non-ideal environment factor related to BEOL, Figure 12: Frequency dependent (a) magnitude and (b) phase of S 21
pad-parasitics, probes etc. into the TCAD simulation at VCB =0 V VBE =0.8 V and 0.9 V for the SiGe HBT up to metal-
results via additional EM simulation. For this purpose, 1: comparison of de-embedded actual and virtual measurements with
we performed three-dimensional (3D) EM simulations TCAD simulation.
using commercially available simulator, Ansys-HFSS.
The simulation layout is imported on a silicon substrate
and the probes are placed exactly the way they appear down to metal 1 where the transistor is connected, are
during the measurement. Then, the whole structure simulated with HFSS (4-port EM simulation). Next,
along with the probes is covered within an air box as the TCAD (2-port TCAD) and EM simulation results
shown in Fig. 7. During the setup, proper care has been are concatenated. This data set contains both the ac-
taken in choosing the meshing and assigning the appro- cess (probes, pads, BEOL) as well as the actual tran-
priate boundary conditions for absorbing the radiation sistor itself. The next step is to perform a calibration
at the faces of the air box in order to emulate an infinite and de-embedding as done in real measurement. For
free space environment. Three separate probe models that purpose, all TRL calibration standards are simu-
for each frequency band as shown in Fig. 8 are cho- lated with HFSS and the calibration is performed as
sen and finally simulations are performed by giving in- shown in Fig. 9. Finally, the Short-Open de-embedding
put excitation at each wave-port. Together the TCAD structures are simulated as well and the de-embedding
simulation of SiGe HBT up to metal-1 and EM simu- is completed leading to the virtual measurement data.
lation to consider the remaining non-ideal environment Had the calibration and de-embedding been ideal, the
factors as described above generate the so-called virtual de-embedded virtual measurement data would perfectly
measurement data for our further investigation. Now we match with the TCAD simulation of the actual transis-
compare actually measured S -parameter results with (i) tor. The results for the three data sets (actual measure-
TCAD simulation (of the actual device up to metal-1) ments, TCAD simulation and virtual measurements) are
and (ii) virtually measured data. Virtual measurement compared in Figs. 10 to 14 for the magnitude and phase
data are achieved following the flow graph described in of S 11 , S 12 , S 21 , S 22 and fT , fmax up to 500 GHz.
Fig. 9. First, the RF probes, the pads and the BEOL From the comparison plots following observations
5
 Soumya et al. / 00 (2020) 1–8 6

are made. Symbols : Measurement Symbols : Measurement

0 0 Short Dash : TCAD
Short Dash : TCAD

Line : Virtual measurement Line : Virtual measurement

-2
• Up to 100 GHz, a good correlation can be ob- VBE=0.8V
-20

phase (S 22) ( )
mag(S 22) (dB)

0
-4
served for both VBE bias points and for all four S - -40
VBE=0.8V

-6
parameters (magnitude as well as phase). VBE=0.9V
-60

-8
-80 VBE=0.9V

• When changing the frequency bands (especially at -10

-100

the transition into the 140 GHz to 220 GHz band), -12
-120

discontinuities are observed with the actual mea- 0 100 200 300

Frequency (GHz)
400 500 0 100 200 300

Frequency (GHz)
400 500

surements while comparing with TCAD and vir-

(a) (b)
tual measurements. Therefore, it can be concluded
that the TRL calibration along with Short-Open de- Figure 13: Frequency dependent (a) magnitude and (b) phase of S 22
embedding are not able to correct the coupling with at VCB =0 V VBE =0.8 V and 0.9 V for the SiGe HBT up to metal-
HF probes and the wafer surface accurately for the 1: comparison of de-embedded actual and virtual measurements with
TCAD simulation.
given test-structure and the HF probes used. Also,
the measurements are noisy and can be attributed
to some difficulties to have a reliable contact be- 400 450
Symbols : Measurement

tween the probes and pads. 350 VBE=0.9V

400
Short Dash : TCAD

Line : Virtual measurement

350 VBE=0.9V
300

fmax (GHz)
• The discontinuous transition of actual measure- 300
fT (GHz)

Symbols : Measurement

Short Dash : TCAD

250 Line : Virtual measurement

ments at 220 GHz is well predicted by the virtual 250

200 VBE=0.8V
200
measurement especially for the magnitude of S 12 - 150
150 VBE=0.8V

parameter. 100
100

50

• When going to higher frequencies beyond 0 100 200 300 400 500 0 100 200 300 400 500

Frequency (GHz) Frequency (GHz)

400 GHz, a deviation of the expected behavior

can be observed both in virtual as well as actual (a) (b)

measurements compared to TCAD data. Even
though no perfect agreement for the actual or
virtual measurement data is achieved with the ref-
erence TCAD data, at least the trend is reasonably
predicted. For some of the measured S -parameters
such as mag(S 12 ), mag(S 21 ) and mag(S 22 ), the
agreement is quite convincing.
This brings us to the conclusion that the calibration
and de-embedding procedures are not absolutely flaw-
less especially at the higher frequency bands and re-
quire further improvements in order to correct all the
demonstrated errors introduced by the measurement en-
vironment. The impact from these unavoidable errors
on circuit level will be explored in the next section.
5. G-band Amplifier Design
The preceding analysis makes it clear that there may
be a mismatch between the characteristics of the intrin-
sic device (up to metal-1) and the actual measurement
(from which the influences of the probes and pads in the
calibration process are not perfectly eliminated). How-
ever, parameter extraction is performed on actual mea-
surement data and the process design kit contains an
electrical compact model parameter set for the transistor
6
Figure 14: Frequency-dependent transit frequency ( fT ) and maximum
oscillation frequency ( fmax ) at VCB =0 V and VBE =0.8 V and 0.9 V for
the SiGe HBT up to metal-1: comparison of de-embedded actual and
virtual measurements with TCAD simulation.
based on these imperfectly de-embedded measurement
data.
In the following, we evaluate the effects of using
a slightly erroneous parameter set for circuit design
through a case study on a single-stage amplifier operat-
ing at 170 GHz. The same design has been re-simulated
with a perfect parameter set representing only the actual
device upto metal-1 that is obtained from TCAD simu-
lation.
First, as can be seen from Fig. 15, the S -parameter
box contains the virtual measurement data simulated at
VB =0.85 V and VCB =0.5 V, which stands for the opti-
mum gain bias conditions, below the breakdown volt-
age. Before going ahead with the design of matching
network, K-4 stability is checked for both virtual mea-
surement and TCAD data. The required conditions of K
> 1 and 4 < 1 are met for both data sets at 170 GHz.
Using the virtual measurement data for the transis-
tor, input and output matching networks for the G-band
single-stage amplifier are designed for the port match-
 Soumya et al. / 00 (2020) 1–8 7

Input Output

Z=50Ω
S-Parameter
Z=50Ω

6
matching matching
Box
network network TCAD

mag(S 21) (dB)

Figure 15: Block diagram representation of an single stage amplifier
4
circuit with input and output matching networks.

Virtual measurement
10
Symbols: TCAD 2
mag(S 11) & mag(S 22) (dB)

Line : Virtual measurement

0

-10 1

-20 S 11

S 22 0
-30 140 160 180 200 220

Frequency (GHz)
-40 -10

-20

-50 -30
Figure 17: Comparison of frequency-dependent gains (S 21 ) obtained
-40

-60 -50
from TCAD simulation (red line) and virtual measurement (black
-60
line) of the G-band single-stage amplifier
-70 -70

152 160 168 176 184 192

-80
120 140 160 180
Frequency (GHz)
Figure 16: Comparison of frequency-dependent S 11 , S 22 obtained
from two amplifier circuits designed at VBE =0.85 V and VCB =0.5 V:
one amplifier considers a transistor virtual measurement data while
the other corresponding to the TCAD simulation.
ing with 50 Ω termination at 170 GHz (see Fig. 15).
Now the virtual measurement data is replaced with the
TCAD data, keeping the matching networks and bias
point fixed. First, we check that the matching constraint
S 11 , S 22 < -10dB is met for the amplifier circuit using
both data sets for the transistor (as shown in Fig. 16).
Considering a -10 dB bandwith, a 30 GHz bandwith is
observed (160-190 GHz). In this 160-190 GHz range,
the gain S 21 in Fig. 17) shows a difference of around
1 dB between TCAD simulation and virtual measure-
ment, which is significant at such a high frequency. In
addition, the maximum gain value is shifted by around
10 GHz towards higher frequency for the amplifier’s vir-
tual measurement.
It is apparent from the results that if someone designs
an amplifier based on model simulation in millimeter-
wave range, the measured circuit performance may not
meet the predicted performance. Particularly, in the case
of a narrow band application, such a performance de-
viation of the measured circuit may potentially lead to
unsatisfactory specification requirements.
200 220
6. Conclusion
We expose the limitations involved in the state-of-
the-art calibration and de-embedding techniques when
applied to the high-frequency (up to 500 GHz) charac-
terized data. A comparison of carefully measured and
de-embedded S -parameters with a calibrated TCAD
simulation for the intrinsic transistor up to metal-
1 reveals certain measurement discontinuities above
110 GHz unlike TCAD data. A physics based method-
ology based on TCAD plus EM simulation leading to a
virtual measurement is applied to understand the source
of the actual measurement errors. The analysis reveals
that the parasitic components that are already present in
the actual measurement environment, are unintention-
ally transferred to the measured data even after care-
fully applying the state-of-the-art calibration and de-
embedding techniques. The measurement errors in the
upper band obstructs the extraction of accurate high fre-
quency compact model parameters. Using a G-band
single-stage amplifier, the impact of slightly erroneous
measurement and parameter extraction is exposed in the
circuit-level.
Acknowledgment
This work is partly funded by the French Nouvelle-
Aquitaine Authorities through the FAST project. The
authors also acknowledge financial support from the EU
under Project Taranto (No. 737454). The authors would
7
 Soumya et al. / 00 (2020) 1–8 8

like to thank ST-Microelectronics for supplying the Sil- and Compound semiconductor Integrated Circuits and Technol-
icon wafer. ogy Symposium (BCICTS), IEEE, 2019, pp. 1–4.
[16] V. Van-Tuan, D. Celi, T. Zimmer, S. Fregonese, P. Chevalier,
Tcad calibration of high-speed si/sige hbts in 55-nm bicmos,
ECS Transactions 75 (8) (2016) 113–119.
References
[17] J. Van Den Biesen, A simple regional analysis of transit times
[1] K. Yau, E. Dacquay, I. Sarkas, S. P. Voinigescu, Device and in bipolar transistors, Solid-state electronics 29 (5) (1986) 529–
ic characterization above 100 ghz, IEEE Microwave Magazine 534.
13 (1) (2012) 30–54. [18] B. M. Haugerud, M. M. Pratapgarhwala, J. P. Comeau, A. K.
[2] D. F. Williams, P. Corson, J. Sharma, H. Krishnaswamy, W. Tai, Sutton, A. G. Prakash, J. D. Cressler, P. W. Marshall, C. J. Mar-
Z. George, D. S. Ricketts, P. M. Watson, E. Dacquay, S. P. shall, R. L. Ladbury, M. El-Diwany, et al., Proton and gamma
Voinigescu, Calibrations for millimeter-wave silicon transistor radiation effects in a new first-generation sige hbt technology,
characterization, IEEE transactions on Microwave Theory and Solid-state electronics 50 (2) (2006) 181–190.
Techniques 62 (3) (2014) 658–668. [19] Y. Taur, T. H. Ning, Fundamentals of modern VLSI devices,
[3] S. P. Voinigescu, E. Dacquay, V. Adinolfi, I. Sarkas, A. Balteanu, Cambridge university press, 2013.
A. Tomkins, D. Celi, P. Chevalier, Characterization and model-
ing of an sige hbt technology for transceiver applications in the
100–300-ghz range, IEEE Transactions on Microwave Theory
and Techniques 60 (12) (2012) 4024–4034.
[4] S. Fregonese, M. Deng, M. Potereau, C. Ayela, K. Aufinger,
T. Zimmer, et al., On-wafer characterization of silicon transis-
tors up to 500 ghz and analysis of measurement discontinuities
between the frequency bands, IEEE Transactions on Microwave
Theory and Techniques 66 (7) (2018) 3332–3341.
[5] L. Galatro, A. Pawlak, M. Schroter, M. Spirito, Capacitively
loaded inverted cpws for distributed trl-based de-embedding at
(sub) mm-waves, IEEE Transactions on Microwave Theory and
Techniques 65 (12) (2017) 4914–4924.
[6] M. Deng, T. Quémerais, S. Bouvot, D. Gloria, P. Chevalier,
S. Lépilliet, F. Danneville, G. Dambrine, Small-signal charac-
terization and modelling of 55 nm sige bicmos hbt up to 325
ghz, Solid-State Electronics 129 (2017) 150–156.
[7] L. Galatro, M. Spirito, Millimeter-wave on-wafer trl calibration
employing 3-d em simulation-based characteristic impedance
extraction, IEEE Transactions on Microwave Theory and Tech-
niques 65 (4) (2017) 1315–1323.
[8] M. Seelmann-Eggebert, M. Ohlrogge, R. Weber, D. Peschel,
H. Maßler, M. Riessle, A. Tessmann, A. Leuther,
M. Schlechtweg, O. Ambacher, On the accurate measurement
and calibration of s-parameters for millimeter wavelengths
and beyond, IEEE Transactions on Microwave Theory and
Techniques 63 (7) (2015) 2335–2342.
[9] R. Sakamaki, M. Horibe, Uncertainty analysis method including
influence of probe alignment on on-wafer calibration process,
IEEE Transactions on Instrumentation and Measurement 68 (6)
(2019) 1748–1755.
[10] D. F. Williams, F.-J. Schmückle, R. Doerner, G. N. Phung,
U. Arz, W. Heinrich, Crosstalk corrections for coplanar-
waveguide scattering-parameter calibrations, IEEE transactions
on microwave theory and techniques 62 (8) (2014) 1748–1761.
[11] G. N. Phung, F. J. Schmückle, R. Doerner, B. Kähne, T. Fritzsch,
U. Arz, W. Heinrich, Influence of microwave probes on cal-
ibrated on-wafer measurements, IEEE Transactions on Mi-
crowave Theory and Techniques 67 (5) (2019) 1892–1900.
[12] T. Sentaurus, Manuals, synopsys inc, Mountain View, CA 94043
(2009).
[13] V. Vu, Exploration and evaluation of a novel si/sige heterojunc-
tion bipolar transistor architecture for next bicmos generation,
Ph.D. thesis, Ph. D. dissertation, School Phys. Sci. Eng., Univ.
Bordeaux, Bordeaux . . . (2016).
[14] F. M. Bufler, Full-band Monte Carlo simulation of electrons and
holes in strained Si and SiGe, Herbert Utz Verlag, 1998.
[15] S. R. Panda, S. Fregonese, A. Chakravorty, T. Zimmer, Tcad
simulation and assessment of anomalous deflection in measured
s-parameters of sige hbts in thz range, in: 2019 IEEE BiCMOS
8