Sentaurus Technology Template:

DC and RF Characterization of HEMTs

Abstract
The purpose of this Sentaurus TCAD simulation project is to provide a template
setup for the DC and RF characterization of high electron mobility transistors
(HEMTs).

For HEMTs, the following simulations are performed: IdVgs, a family of IdVds
curves, on-state and off-state breakdown, and RF analysis. The template is based on
an InGaAs HEMT; however, the project structure and input files can be used for any
HEMT with only minor modifications.

For each of the simulated IV curves, relevant electrical parameters such as pinch-off
voltage, transconductance, drain saturation current, and breakdown voltage are
calculated. Furthermore, RF parameters such as ft and fmax, as well as other two-
port network and RF gain parameters, are computed.
Synopsys and the Synopsys logo are registered trademarks of Synopsys, Inc.

Copyright © 2007 Synopsys, Inc. All rights reserved.

Introduction Project setup: HEMT_DC
This project provides standard templates for Sentaurus The first Sentaurus Workbench template project performs
Device, which can be used to perform the most common device simulations relevant for the DC characterization of
types of simulation for the characterization and performance HEMT devices.
assessment of HEMTs. There are three different Sentaurus
Workbench projects: Sentaurus Device
■ A DC characterization project HEMT_DC, which
performs an IdVgs simulation and computes a family of Sentaurus Device performs IdVgs and IdVds sweeps. The
IdVds curves. bias conditions and the number of sweeps are controlled
■ A breakdown simulation project HEMT_BV, which using the following Sentaurus Workbench parameters:
computes the on-state and off-state breakdown. ■ Vgmin [V] defines the minimum gate voltage for the
■ A small-signal simulation project HEMT_RF, which IdVgs or IdVds sweeps. Here, it is set to -1.5 for the
performs a two-port network analysis. IdVgs sweeps and to -0.8 for the IdVds sweeps.
■ Vgmax [V] defines the maximum gate voltage for the
All three projects are based on an analytic InGaAs HEMT IdVgs or IdVds sweeps. Here, it is set to 0.8.
device structure, which is created using Sentaurus Structure
■ Vd [V] defines the drain bias for the IdVgs sweeps and
Editor.
the final drain voltage for the IdVds sweeps. It is set to
0.05 for the low bias IdVgs sweep and to 1.5 for the
It is assumed that the user is familiar with the Sentaurus tool
high bias IdVgs sweep, as well as the IdVds sweeps.
suite, in particular, Sentaurus Workbench, Sentaurus
Structure Editor, Sentaurus Device, and Inspect. For an ■ IdVd = 0,1,2,3,... For IdVd = 0, an IdVgs
introduction and tutorials, refer to the Sentaurus training sweep is performed. For IdVd = N, a family of N
material. IdVds sweeps is simulated at gate biases equally spaced
between Vgmin and Vgmax. Here, the parameter is set
The focus of this project is to provide setups that can be to 0 (IdVgs) and 5 (IdVds).
used as they are or adapted to specific needs. The
documentation focuses on aspects of the project setup. For Inspect
details about tool uses and specific tool syntax, refer to the
respective manuals. Inspect plots the respective IV characteristics and extracts:
■ Vtgm [V] is the threshold voltage, defined as the
General simulation setup intersection of the tangent at the maximum
transconductance gm with the Vgs axis.
The simulations are organized in three Sentaurus
■ Vti [V] is the threshold voltage, defined as Vgs at
Workbench projects. The tool flow of the projects are
which Id = 1 mA/mm.
discussed here. The tool flow consists of Sentaurus
Structure Editor, which creates the analytic HEMT ■ Idmax [mA/mm] is the maximum value of the IdVgs
structure; Sentaurus Device, which calculates the device curve.
characteristics; and the visualization tool Inspect. For each ■ Ron [kΩ mm] is the on-state resistance, extracted for
tool, the associated Sentaurus Workbench input parameters, Vgs = 0 V and Vds = 1 V.
as well as the extracted parameters, are discussed.

Project setup: HEMT_BV

All projects
The second Sentaurus Workbench template project
Sentaurus Structure Editor performs on-state and off-state breakdown simulations of
the HEMT device.
The analytic HEMT structure is defined with Sentaurus
Structure Editor. The Sentaurus Structure Editor setup is Sentaurus Device
identical for all three HEMT template projects.
Sentaurus Device performs breakdown simulations using
The generated structure, mesh, and doping information are two different methods. In the direct method, a large resistor
stored in a TDR file, which is then passed to Sentaurus is attached to the drain (load line) and the drain is ramped to
Device. a very high voltage. Depending on the gate bias, this yields
the on-state or off-state breakdown. Alternatively, the drain-
current injection method [1] is used to perform an off-state

Copyright © 2007 Synopsys, Inc. All rights reserved. 3

breakdown simulation. In this method, the drain current
level is held constant while the gate is closed. To maintain
Project setup: HEMT_RF
the drain current, the drain voltage must rise sharply when
the gate closes. When breakdown sets in, the drain voltage The third Sentaurus Workbench project is designed to
saturates. investigate the RF characteristics of the HEMT device.

The bias condition and type of breakdown simulation are Sentaurus Device
controlled using the Sentaurus Workbench parameters:
Sentaurus Device performs a small-signal analysis during
■ Vg [V]: Gate bias. For the on-state breakdown IdVds
an IdVgs sweep. The bias conditions and feedback circuit
sweep, the bias is set to 0. For the off-state breakdown
are controlled using the Sentaurus Workbench parameters:
IdVds sweep, the bias is set to -1.5. For the drain-
current injection method, this parameter sets the end ■ Rfb [Ω μm]: For Rfb = 0, no external resistive
point of the gate voltage sweep. Here, the value –5 is feedback from drain to gate is considered. Otherwise, a
used. resistor of the given magnitude between the drain
electrode and the gate electrode is included in the circuit
■ Vd [V]: For the direct method, this parameter sets the
setup. Here, the values 0 and 1e5 are used.
end point of the drain bias sweep. Values of 200 and
400 are used for the on-state and off-state breakdown ■ Vgmin [V] defines the minimum gate voltage for the
sweeps, respectively. For the drain-current injection IdVgs sweep. Here, it is set to -0.5.
method, this parameter is used pre-bias the drain, before ■ Vgmax [V] defines the maximum gate voltage for the
switching to the current boundary condition. Here, a IdVgs sweep. Here, it is set to 0.5.
value of 0.1 is used.
■ Vd [V] defines the drain bias for the IdVgs sweep. Here,
■ Rd [Ω μm]: This parameter sets the value of the drain it is set to 1.0.
contact resistance. In the direct method, the resistance is
set to a high value in order to emulate a load line. Values Inspect
of 1e4 and 1e5 are used for the on-state and off-state
breakdown sweeps, respectively. For the drain-current
The Sentaurus Workbench variable Extract controls
injection method, this parameter is set to the regular
which RF parameter Inspect plots and extracts:
drain contact resistance of 150.
■ IdVg: IdVgs and the DC transconductance are plotted.
■ Id [mA/mm]: If this parameter is set to 0, the direct
method is used. For a nonzero value, the drain-current ■ fmax: The maximum frequency of oscillation (fmax) is
injection method is used and the value gives the current plotted as a function of bias for three different extraction
level. Here, three different current levels are used (200, methods: unit-gain-point, extract-at-dBPoint, and
300, and 500). extract-at-frequency. The maximum fmax value over all
bias points for the unit-gain-point method and the
Inspect extract-at-dBPoint method is written to the Sentaurus
Workbench Family Tree together with the bias point at
Inspect plots the respective IV characteristics and extracts: which the maximum is reached.

■ BVv [V]: Breakdown voltage defined as the maximum ■ MUG: A family of Mason’s unilateral gain curves is
inner voltage on the drain. The inner voltage is plotted as a function of frequency for all bias points.
measured at the device contact directly and does not ■ MUG_b: A family of Mason’s unilateral gain curves is
include the voltage drop across the contact resistance. plotted as a function of bias for all frequencies.
■ BVi [V]: Breakdown voltage defined as the inner drain ■ ft: The cut-off frequency (ft) is plotted as a function of
voltage at which the drain current reaches a certain bias for three different extraction methods: unit-gain-
value. Here, a value of 1000 mA/mm is used. point, extract-at-dBPoint, and extract-at-frequency. The
■ BVc [V]: Breakdown voltage determined using the maximum ft value over all bias points for the unit-gain-
drain-current injection method. Here, the breakdown point method and the extract-at-dBPoint method is
voltage is defined as the drain voltage reached after the written to the Sentaurus Workbench Family Tree
gate has been biased well below the pinch-off voltage. together with the bias point at which the maximum is
Here, a value of Vgs = –1.5 V is used. reached.
■ h21: The RF parameter h21 is plotted as a function of
frequency for a gate bias of 0 V. The plot shows the real
and imaginary parts of h21 as well as the magnitude
and phase.

4 Copyright © 2007 Synopsys, Inc. All rights reserved.

■ h21_b: The RF parameter h21 is plotted as a function The HEMT layer structure is as follows (the actual names of
of bias for a frequency of 1 GHz. The plot shows the real the Scheme variables used in the parameterization of the
and imaginary parts of h21 as well as the magnitude HEMT structure are given in parentheses):
and phase.
■ Substrate: 0.8 μm GaAs (HGaAsSub)
■ fK1: A plot of the boundaries of the region of
■ Channel: 10 nm InGaAs (HChannel)
unconditional stability is shown. This boundary is
defined as the isolines at which the Rollett stability ■ Spacer: 34.5 nm AlGaAs (HSpacer)
factor (K) is one. If two boundaries exist for a given bias ■ Cap layer: 30 nm GaAs (HGaAsCap)
point, the low frequency line is red and the high
frequency line is blue. To highlight in which regions the ■ Passivation layer: 50 nm nitride (HPass)
device is unconditionally stable, the isolines at which ■ Location of delta-doping layer: 31 nm (YDelta)
K=1.1 are also shown as green dashes.
■ Thickness of delta-doping layer: 2 nm (DeltaThick)
■ MSG: The maximum stable gain (MSG), the maximum
available gain (MAG), and the Rollett stability factor The gate Schottky contact is etched into the top spacer layer.
(K) are plotted as a function of frequency for a gate bias This gate recess is 15 nm deep (GRecess). The gate length
of 0 V. is 0.25 μm (Lg). At each side of the gate, a 40 nm wide
■ MSG_b: MSG, MAG, and the Rollett stability factor (K) oxide layer isolates the gate from the cap layer (Wsp). The
are plotted as a function of bias for a frequency of sheet doping concentration is 5.4 x 1012 cm–2
1 GHz. (SheetCharge). It is assumed that the dopants are in a
tight Gaussian distribution, with a diffusion length given by
■ Smith: The RF parameters s11 and s22 are plotted on a the thickness of the delta-doping layer. Mole fractions and
Smith chart for all frequencies and a gate bias of 0 V. trap concentrations are defined in the input file of Sentaurus
Note that the frequency associated with a given point on Device.
the Smith chart can be identified by clicking the
respective frequency label on the legend. To adjust details of the devices, users can modify the top
■ Smith_b: The RF parameters s11 and s22 are plotted section of the Sentaurus Structure Editor input file
on a Smith chart for all bias points and a frequency of sde_dvs.cmd. For example, the width and sheet doping
1 GHz. Note that the bias associated with a given point concentration of the delta-doping layer are defined by the
on the Smith chart can be identified by clicking the Scheme variables:
respective bias label on the legend. (define DeltaThick 0.002) ; [um]
(define SheetCharge 5.4e12) ; [1/cm2]
■ S21S12: The RF parameters s21 and s12 are plotted on
a polar plot for all frequencies and a gate bias of 0 V.
Note that the frequency associated with a given point on Other geometric, doping, and meshing parameters are
the polar plot can be identified by clicking the respective accessible in a similar manner.
frequency label on the legend.
The meshing strategy is designed to result in a high-quality
■ Export2CSV: The RF parameters Y, H, Z, and S are mesh without excessive node counts for a large range of
written to a text file using a comma-separated values geometric parameters.
(CSV) format, which can be read by any spreadsheet
program. The file contains a header, which lists the Most of the meshing strategy is defined using regular
number of bias and frequency points as well as the refinement boxes, which specify the allowed minimum and
minimum and maximum bias points and frequencies. maximum mesh spacing within a specified area. It is the
The header is followed by a table, which contains the meshing algorithm that places the actual lines. For the delta-
frequency, the bias, and the real and imaginary parts of doping layer, a new feature is applied that allows users to
the respective components of the RF parameters matrix place lines explicitly, by defined a list of x- and y-values at
(for example y11, y12, y21, and y22). Two versions of which a line is to be added. To use this feature, a stanza is
the CSV file are written. One is sorted by frequencies added to the Sentaurus Mesh command file, which has the
and the other is sorted by bias points. form:
AxisAligned {
Tool-specific setups xCuts = ( <list of coordinates> )
yCuts = ( <list of coordinates> ) }

Sentaurus Structure Editor and Sentaurus Sentaurus Structure Editor calls a meshing engine to
Mesh generate the structure files for Sentaurus Device. Sentaurus
Mesh is called from within Sentaurus Structure Editor with:
Sentaurus Structure Editor is used to define the HEMT (sde:build-mesh "snmesh" "n@node@_half_msh")
devices in a fully parameterized manner.

Copyright © 2007 Synopsys, Inc. All rights reserved. 5

This command generates a device structure in TDR format, The mole fraction in the AlxGa(1–x)As spacer layer is set to
containing doping and grid data. Note that only half of the x = 0.3 with:
HEMT structure is created by Sentaurus Structure Editor Physics(Material="AlGaAs"){
and meshed with Sentaurus Mesh. It is subsequently MoleFraction(xFraction=0.30 Grading=0)
reflected about the vertical axis to obtain the full device. }
The reflection is performed in Sentaurus Structure Editor by
a system call to Sentaurus Data Explorer (tdx): In the substrate layer, a donor trap concentration of
(system:command "tdx -mtt -x -ren drain=source 1016 cm–3 is defined with:
n@node@_half_msh n@node@_msh") Physics( Region="R.GaAsSub" ){
Traps(
The option -x instructs Sentaurus Data Explorer to reflect (Donor Conc=1e16 EnergyMid=0.61 fromCondBand
the device along an axis defined by x = x min . The given eXsection=2.5e-18 hXsection=2.5e-16)
half-structure has three contacts: drain, gate, and )
}
substrate, which are defined in sde_dvs.cmd. Of
these, the gate and substrate contacts touch the axis of
Hydrodynamic transport is used for the electrons. High-
reflection and, upon reflection, are extended and thereby
field mobility saturation is selected. Furthermore,
preserve their names. However, the drain contact in the
Shockley–Read–Hall (SRH), Auger, and radiative
reflected half is named drainmirrored by default. This
recombination models are activated. For the breakdown
contact is explicitly renamed to source with the Sentaurus
simulations, the avalanche generation model is used as well.
Data Explorer command-line option -ren. The device
obtained using Sentaurus Structure Editor is shown in
At the Ohmic source and drain contact, a contact resistance
Figure 1.
of 150 Ω μm is set, and the gate is defined as a Schottky
contact with a Schottky barrier height of 0.9 eV and electron
and hole recombination velocities of 107 cm/s:
Electrode{
{ Name="source" Voltage=0.0 Resistor= 150 }
{ Name="drain" Voltage=0.0 Resistor= 150 }
{ Name="gate" Voltage=0.0 Schottky Barrier=0.9
eRecVelocity=1.e7 hRecVelocity=1.e7 }
...
}

Figure 1 HEMT device generated by Sentaurus Structure Editor. For better

viewing, different scales are used for the x-axis and y-axis.

Sentaurus Device

Sentaurus Device is used to simulate the various DC

current-voltage as well as the RF characteristics.

The mole fraction in the In(1–x)GaxAs channel is set to

x = 0.75 with:
Figure 2 Electron concentration under the gate at Vgs = –1.5 V and
Physics(Material="InGaAs"){ Vds = 1.5 V. The mesh in this critical area is shown. For better
viewing, different scales are used for the x-axis and y-axis.
MoleFraction(xFraction=0.75 Grading=0)
}

6 Copyright © 2007 Synopsys, Inc. All rights reserved.

To track accurately the electron and hole concentrations, the For the IdVds sweeps (see Figure 4), first the gate is ramped
error control parameter ErrRef(Electron|Hole) is to Vgmin and then to Vgmax. During the sweep from
reduced from the default value of 1010 cm–3 to 107 cm–3 for Vgmin to Vgmax, the solution is saved at a number of
the electrons and 104 cm–3 for the holes. In regions with equidistant bias points. This number is set by the Sentaurus
very low concentrations, such as the depletion region under Workbench parameter IdVd. The solutions are reloaded
the Schottky contact, the carrier temperature is not a well- one by one and a drain bias sweep is performed.
defined quantity. It is possible that the numeric solution of
the partial differential equation (PDE) results in 700
nonphysically high temperatures in such regions, which can
have a negative effect on convergence. 600

To suppress such artifacts, the energy relaxation time is 500

Drain Current [mA/mm]

effectively reduced in regions where the carrier density is
less than RelTermMinDensity: 400
Math{...
ErrRef(Electron) = 1e7 300
ErrRef(Hole) = 1e4 Vgs = 0.8 V
Vgs = 0.4 V
RelTermMinDensity = 1e4
200 Vgs = 0.0 V
RelTermMinDensityZero = 1e7 Vgs = -0.4 V
} Vgs = -0.8 V
100
(RelTermMinDensityZero controls the reduction of
the energy relaxation time in low density regions in which 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
artificial cooling occurs.) Drain Voltage [V]

Figure 4 Drain current as function of drain voltage for a gate bias of

DC characterization Vgs = –0.8, –0.4, 0.0, 0.4, and 0.8 V

For the IdVgs sweeps (see Figure 3), the gate and drain are Breakdown simulations
first ramped together to the bias conditions defined by the
Sentaurus Workbench parameters Vgmax and Vd. Then, the For the breakdown simulations, two different methods are
gate is ramped to the final bias point defined by the used. In the direct method (see Figure 5 on page 8), a large
Sentaurus Workbench parameter Vgmin. (Note that, in load-line resistor is attached to the drain contact and the
HEMT devices, often faster convergence can be obtained by drain is ramped to a very large value. After the onset of
sweeping an open device into pinch-off, instead of starting impact ionization, most of the voltage drop occurs across
at pinch-off.) this resistor if the value chosen is sufficiently large.
Therefore, this simple technique achieves an automatic
switching from a voltage-controlled prebreakdown regime
Vds = 1.5 V
to a current-controlled post-breakdown regime.
600 Vds = 50 mV

The breakdown IV characteristics can be seen by plotting

Drain Current [mA/mm]

the terminal current versus the inner contact voltage instead

of the outer contact voltage. The appropriate value for the
400 attached resistor is of the order of R = V/I at the onset of
impact ionization. This technique is used here for the on-
state as well as the off-state breakdown simulations. The
values of the external resistor are defined using the
200 Sentaurus Workbench parameter Rd. The external resistor is
declared in the Sentaurus Device Electrode section with:
Electrode{...
{ Name="drain" Voltage=0.0 Resistor= @Rd@ }
}
0
-1.5 -1 -0.5 0 0.5
Gate Voltage [V]
Figure 3 Drain current as function of gate voltage for a drain bias of
Vds = 50 mV (blue) and 1.5V (red)

Copyright © 2007 Synopsys, Inc. All rights reserved. 7

 function of the gate bias originates from the resistance of
the substrate layer, which now carries most of the current.)
2500

RF characterization
Drain Current [mA/mm]

2000
For the simulation of the RF characteristics, the same
biasing and sweeping scheme as for the IdVgs simulation is
1500
used. At equidistant bias points, Sentaurus Device performs
a small-signal analysis (AC) for various frequencies and it
1000 stores the small-signal admittances and capacitances for all
contact-to-contact combinations (this data is equivalent to
the Y-matrix). During postprocessing with Inspect, this data
500 is used to determine the various RF parameters.
On-state BV: Vgs = 0 V
Off-state BV: Vgs = -1.5 V
A mixed-mode environment is required for AC simulation
0
0 5 10 15 20 in Sentaurus Device, that is, instead of simulating an
Drain Voltage [V]
isolated HEMT, the HEMT is embedded in an external
Figure 5 On-state and off-state breakdown calculated using the direct
method: blue curve is IdVds at Vgs = 0 V (on-state breakdown) and
circuit. Here, a two-port network circuit configuration (see
red curve is IdVds at Vgs = –1.5 V (off-state breakdown) Figure 7 on page 10) is used, in which voltage sources are
attached to the gate (port 1) and drain (port 2) terminals. All
With the drain-current injection method [1] (see Figure 6), other terminals are grounded. The circuit is defined in the
the HEMT is first biased to a nonzero drain bias, given by System section:
the Sentaurus Workbench parameter Vd, in order to have a
System {
nonzero current flow. Then, the contact boundary condition HEMT hemt (gate=1 drain=2 source=0 substrate=0)
is switched from voltage controlled to current controlled Vsource_pset vg ( 1 0 ) { dc = 0 }
with: Vsource_pset vd ( 2 0 ) { dc = 0 }
Set ( "drain" mode Current ) #if @Rfb@ != 0
Resistor_pset Rfb (1 2) { resistance = @Rfb@ }
#endif
The drain current is ramped to a value given by the }
Sentaurus Workbench parameter Id. Finally, the gate is
closed by ramping it to a large negative value, given by the If the Sentaurus Workbench parameter Rfb is set to a
Sentaurus Workbench parameter Vg. nonzero value, an external feedback resistor is also included
in the circuit.

The AC analysis is activated with the ACCoupled

20
statement:
Quasistationary( ...
Drain Voltage [V]

15 Goal{ Parameter=vg.dc Voltage=@Vgmax@ }

){ ACCoupled(
StartFrequency= 1e8 EndFrequency= 1e12
10
NumberOfPoints= 13 Decade
Node(1 2) Exclude(vg vd)
Id = 500 mA/mm
ACCompute( Time=(Range=(0 1) Intervals=20) )
Id = 300 mA/mm
Id = 200 mA/mm ){ Poisson Electron Hole eTemperature }
5
}

The ACCompute statement defines that the AC analysis

-5 -4 -3 -2 -1
Gate Voltage [V]
will be performed at 21 evenly spaced bias points between
the starting bias point Vgmin and the bias end point
Figure 6 Off-state breakdown calculated using the drain-current injection
method: drain voltage as function of gate voltage for a constant Vgmax. The frequency range is set from 100 MHz (1e8) to
drain current level of Id = 500 (blue), 300 (green), and 200 (red) 1 THz (1e12). This range is sampled by 13 logarithmically
mA/mm
distributed points (that is, 3 points/decade).
Near pinch-off, the channel resistivity rises sharply. To
maintain the current level, the drain voltage must rise DC extractions with Inspect
sharply. At some point, impact ionization sets in and the
device resistivity stabilizes. Then, the drain voltage In the Sentaurus Workbench tool flow, Sentaurus Device is
saturates. (The residual slope of the drain voltage as a followed by the Sentaurus visualization tool Inspect. This

8 Copyright © 2007 Synopsys, Inc. All rights reserved.

tool plots the corresponding IV characteristics and extracts ExtractVti <Name> <Curve> <Ilevel>
relevant electrical parameters, as discussed in General
simulation setup on page 3. where Name defines the name of the extracted parameter,
Curve refers to the name of the Inspect IdVgs curve, and
The extractions are performed with the help of the Inspect Ilevel defines the drain current level at which the gate
library EXTRACT. The extraction routines are activated voltage is extracted. For example, the call:
with: set Io 1 ; # [mA/mm]
load_library EXTRACT set Vti [ExtractVti Vti Id $Io]

(Refer to the Inspect User Guide for more information results in output such as:
about this library.) DOE: Vti -1.076
Vti (Vg at Io=1.000e+00): -1.076 V
Scaling IV data
Extracting maximum transconductance
By default, for 2D simulations, Sentaurus Device uses the
units of A/μm if no explicit AreaFactor is specified in The routine ExtractGm extracts the maximum
the Physics section. For HEMT devices, it is common to transconductance. The routine is called with:
use the units mA/mm. Rescaling to these units is performed ExtractGm <Name> <Curve> <Type>
with the utility routine cv_scaleCurve, which is part of
the EXTRACT library. where Name defines the name of the extracted parameter,
cv_scaleCurve Id 1 1e6 y Curve refers to the name of the Inspect IdVgs curve, and
Type can be either "nMOS" or "pMOS". For example, the
The first argument gives the name of the curve to be scaled, call:
the second and third arguments give the scaling factors for set gm [ExtractGm gm Id "nMOS"]
the x-axis and the y-axis, and the last argument defines on
which y-axis the scaled curve should be displayed (y or results in output such as:
y2).
DOE: gm 1.955e+02
gmb: 1.955e+02 mS/mm
Extracting threshold voltage using Max gmb is at Vg= -0.918 V
transconductance method
Extracting maximum y-value
The routine ExtractVtgm extracts the threshold voltage
using the maximum transconductance method. The routine The routine ExtractMax extracts the maximum y-value
is called with: of a given curve. The routine is called with:
ExtractVtgm <Name> <Curve> <Type> ExtractMax <Name> <Curve>

where Name defines the name of the extracted parameter as where Name defines the name of the extracted parameter
it appears in the Variable Values column of Sentaurus and Curve refers to the name of the Inspect IdVgs curve.
Workbench, Curve refers to the name of the Inspect IdVgs For example, the call:
curve, and Type can be either "nMOS" or "pMOS". The
set IdMax [ExtractMax IdMax Id]
routine passes the extracted value to Sentaurus Workbench
and prints it to the log file. It also returns the value to results in output such as:
Inspect. For example, the call:
DOE: IdMax 9.383e+01
set Vt [ExtractVtgm Vtgm IdVg nMOS] Max: 9.383e+01

results in output such as: Extracting breakdown voltage

DOE: Vtgm -1.029
Vt (Max gm-B method): -1.029 V The breakdown voltage can be defined as the maximum
voltage that can be applied to a contact. The routine
Extracting threshold voltage using current ExtractBVv extracts this value. The routine is called
level method with:
ExtractBVv <Name> <Curve> <Sign>
The routine ExtractVti extracts the gate voltage at
which the drain current exceeds a given current level. The where Name defines the name of the extracted parameter,
routine is called with: Curve refers to the name of the Inspect curve, and the

Copyright © 2007 Synopsys, Inc. All rights reserved. 9

parameter Sign can take the values +1 (NMOS) or –1 Loading Sentaurus Device small-signal data
(PMOS) and is used to distinguish between the different file
types of transistor. (In general, set Sign to –1 if the
breakdown occurs at a negative bias.) The Sentaurus Device small-signal output file is loaded with
the routine rfx_load:
For example, the call:
rfx_load "ACFileName=AC_@acplot@ Port1Name=1 \
set BVv [ExtractBVv BVv Id 1] Port2Name=2 BiasPortName=v(1) DeviceWidth=25"

results in output such as: The port names do not have to be numbers but they must
DOE: BVv 1.09e+01 agree with the node names defined in the Sentaurus Device
BVv: 1.09e+01 V System section. The parameter BiasPortName defines
which port is biased and whether the biasing is a voltage or
The breakdown curve sometimes exhibits a pronounced a current condition. The parameter DeviceWidth is
snapback. In this case, another relevant definition is the bias optional. If present, the Y matrix is multiplied by this
voltage at which the current reaches a certain level. This number. This allows users to specify the actual width of the
type of extraction is performed with the routine device if no AreaFactor has been defined in Sentaurus
ExtractBVi, which is called with: Device. The device width is not very important when the
ExtractBVi <Name> <Curve> <Ilevel> main interest is the Y, Z, or H matrices and the derived gain
parameters, since these parameters scale trivially with the
where Name defines the name of the extracted parameter, device width. However, for the S matrix, the device width is
Curve refers to the name of the Inspect curve, and very important because the reference impedance is 50 Ω.
Ilevel refers to the mentioned current level. For example,
the call: NOTE For a current condition, the syntax for
BiasPortName is more complex. For example, if the
set BVi [ExtractBVi BVi Id 1000]
name of the current source is vc and the name of the port is
2, the declaration would be:
results in output such as:
BiasPortName=i(vc,2)
DOE: BVi 4.094e+00
BVi: 4.094e+00 V
In this case, however, Sentaurus Device must also be
instructed to include the current at this node through this
RF extractions with Inspect device in the small-signal data file. This is performed in the
System section of the Sentaurus Device input file with, for
The RF parameter extractions are performed using the example:
Inspect library RFX. The extraction routines are activated System {
with: HBT hbt (base=1 collector=2 emitter=0)
load_library RFX Vsource_pset vb ( 1 0 ){ dc = 0 }
Vsource_pset vc ( 2 0 ){ dc = 0 }
ACPlot(v(1) v(2) i(vb 1) i(vc 2))
(Refer to the Inspect User Guide for more information }
about this library.)
Cutoff frequency (ft)
The following routines of this library are used in this
project. (Refer to the Inspect User Guide, Chapter 3, for a
complete command reference and list of formulas used to The routine rfx_GetFt returns a list of ft values. For
compute the RF parameters. These formulas have been example, the following command returns the ft values using
taken from the literature [2]–[5].) the unit-gain-point method:
set ft0 [rfx_GetFt "Method=unit-gain-point \
The RFX library assumes a two-port network–like Scale=1e9"]
configuration such as shown in Figure 7.
The optional Scale argument allows users to change the
Port1 Port2 unit of ft. Here, the unit is set to GHz.
Two-Port
Network
The extraction routine supports three different methods to
extract ft from the RF parameter h21:
Figure 7 Schematic of two-port network
■ Method=unit-gain-point
With this method, the frequency at which h21 = 1 is
returned.

10 Copyright © 2007 Synopsys, Inc. All rights reserved.

■ Method=extract-at-dBPoint NOTE Unfortunately, there is no single definition of ft and
fmax that is appropriate under all circumstances. You may
With this method, the frequency at which h21 drops by
a predefined amount of decibels (dB) is located. Then, it think that the direct search for the unit-gain-point should be
the most appropriate definition. However, at high
is assumed that the gain curve exhibits a 20 dB/decade
frequencies, additional parasitic elements may become
decay at this point and, using this assumption, the unit-
dominant and may alter the 20 dB/decade slope near the
gain-point is computed by extrapolation.
unit-gain-point. Furthermore, in experiments, it is
The dB point at which the extrapolation is to be sometimes difficult to trace the gain curve to high enough
calculated is set, for example, to 10 dB using: frequencies to see the unit-gain-point directly. Therefore,
set dBPoint 10 extrapolating the experimental data to the unit-gain-point is
set ft10 [rfx_GetFt "Method=extract-at-dBPoint \ common. In this case, the experiment may not ‘see’ the
Parameter=$dBPoint Scale=1e9"] altered slope due to the parasitics (also some parasitics may
■ Method=extract-at-frequency not be included in the simulation). Therefore, for a
comparison with experimental results, the extrapolation
With this method, the extrapolation of the gain to the methods may be better.
unit-gain-point is performed by assuming that the 20 dB/
decade decay is established at a given frequency. The transition between the flat low-frequency region of the
To set the frequency used for the extrapolation, for gain curves to the 20 dB/decade slope at higher frequencies
example to 10 GHz, use: can be drawn out. In fact, sometimes, a clear 20 dB/decade
slope is never reached. In this case, the extract-at-dBPoint
set frequency 1e10
set ft1e10 [rfx_GetFt \ method may give misleading results. (Often, the results can
"Method=extract-at-frequency \ be improved by adjusting the dB point used for the
Parameter=$frequency Scale=1e9"] extrapolation.)

The rfx_load command creates automatically a list of all Sometimes, it is known beforehand that at a given frequency
bias points, called rfx_BiasPoints, as well as a list of the gain curves fall off at a 20 dB/decade slope. In this case,
all frequencies, called rfx_Frequencies. To plot the ft the unit-gain-point can be determined by the extract-at-
values as a function of bias, create an Inspect curve as frequency method. However, the band of frequencies for
follows: which the 20 dB/decade slope assumption holds true can be
cv_createFromScript ft0 $rfx_BiasPoints $ft0 y
very narrow and can depend on the bias conditions.
cv_display ft0 y
This discussion shows that using solely one method may
give inappropriate results. Therefore, it is recommended to
always use all three methods concurrently. If the fmax or ft
60
curves for all three methods agree well, the results can be
trusted with a high level of confidence. If they give differing
results, it is suggested to examine the underlying gain
50 curves (that is, Mason’s unilateral gain or h21). In most
ft [GHz]

cases, it is clear that the assumptions on which the

extractions are based are not fulfilled. That is, the gain curve
40 does not fall off with a clean 20 dB/decade slope at high
frequencies.
unit-gain-point
extract-at-frequency
30 extract-at-dBPoint
The problem is illustrated in Figure 9 on page 12 where the
fmax curves for the HEMT device are shown. The unit-gain-
-0.4 -0.2 0 0.2 0.4 point method gives fmax values between 100 and 200 GHz.
Gate Voltage [V] The other two methods predict substantially higher fmax
Figure 8 Cutoff frequency ft as a function of gate bias, extracted with three values (400–600 GHz). Moreover, the extract-at-dBPoint
different methods: unit-gain-point (blue), extract-at-frequency
(green), and extract-at-dBPoint (red). This data is computed with a method gives very noise results.
feedback resistor of 100 kΩ μm.

Maximum frequency of oscillation (fmax)

The routine rfx_GetFmax returns a list of fmax values. It

uses the same methods and control parameters as the routine
rfx_GetFt. The main difference is that rfx_GetFmax
operates on Mason’s unilateral gain instead of h21.

Copyright © 2007 Synopsys, Inc. All rights reserved. 11

 to-drain capacitance result in a second-order pole in the gain
(that is, a 40 dB/decade slope). Refer to the literature [6].
extract-at-dBPoint
600
extract-at-frequency
-
unit-gain-point For higher gate voltages, the gain curve exhibit a spike-like
500 gain enhancement between 10 and 100 GHz. (This
resonance is actually much narrower than the spacing
frequency sample points used in the simulation.)
fmax [GHz]

400

This feature is caused by a process in the channel under the

300 gate: As the gate restricts the current flow in the channel,
the parallel electric field rises. This raises the electron
200 temperature and, due to high-field saturation, the effective
mobility degrades, which in turn changes the charge and
field distribution in the channel area under the gate. In
100
HEMTs, the channel and the gate actually form a parasitic
-0.4 -0.2 0 0.2 0.4 capacitor, which provides a RF signal feedback to the gate.
Gate Voltage [V]
The temperature response in the channel area is delayed
Figure 9 Maximum frequency of oscillation fmax as function of gate bias,
extracted with three different methods: unit-gain-point (blue),
with respect to changes in the parallel field. Consequently,
extract-at-frequency (green), and extract-at-dBPoint (red). This the charge and field distribution response to an AC change
data is computed without a feedback resistor.
in the gate voltage is phase shifted. Under certain
conditions, this phase shift results in an increased power
The reason for this unusual behavior is better understood by
transfer from the gate to the drain. This causes the
looking at Figure 10, where the underlying Mason’s
‘resonance’ in MUG. (In the literature [6], it is shown that
unilateral gain (MUG) curves are shown as a function of
such resonances occur also in HEMT compact models if it
frequency. Clearly, these curves have a more complex shape
is assumed that the transconductance element reacts with a
than assumed (flat low-frequency region, followed by a
delay to gate bias changes.)
20 dB/decade slope).
60 Vgs = -0.5 Identifying bias point or frequency index
Vgs = -0.4
Vgs = -0.3
Vgs = -0.2
Vgs = -0.1
Internally, the RF extraction script stores the Y matrix as a
Vgs = 0.0 function of bias and frequency in the form of a two-
Mason’s Unilateral Gain [dB]

Vgs = 0.1
Vgs = 0.2 dimensional array. To access an RF parameter for a given
40 Vgs = 0.3
bias or frequency, the appropriate array index must be given.
Vgs = 0.4
Vgs = 0.5 Use the routine rfx_GetNearestIndex to obtain the
20 dB/decade
relevant index.

The routine operates on the list of bias points

20 (rfx_BiasPoints) or the list of frequencies
(rfx_Frequencies), which is created when loading the
Sentaurus Device small-signal output file.

For example, to find the index of the bias point closest to

0.025 V, use:
0
8 9 10 11 12
10 10 10 10 10 set BiasPointIndex [rfx_GetNearestIndex \
Frequency [Hz]
Target=-0.025 $rfx_BiasPoints]
Figure 10 Mason’s unilateral gain as function of frequency for various gate
voltages between Vgs = –0.5 and 0.5 V at a drain bias of Vds = 1 V.
This data is computed without a feedback resistor. The black The Inspect log file contains information about the bias
dashed line shows an ideal 20 dB/decade slope. associated with the returned index:
!rfx_GetNearestIndex: Nearest Index is 11.
The topmost curves (–0.5 V < Vgs < 0 V) exhibit a 20 dB/
! Corresponding value is 5.0E-02.
decade slope up to about 100 GHz, but then they fall off at ! Target value is 0.025.
40 dB/decade. For this reason, the actual unit-gain-point is
found at much lower frequencies than an extrapolation The actual bias point can be accessed in the script with:
based on a 20 dB/decade slope predicts.
set BiasPoint [lindex $rfx_BiasPoints \
$BiasPointIndex]
Such behavior is typical for HEMTs where parasitic
elements such as the gate-to-source resistance and the gate-

12 Copyright © 2007 Synopsys, Inc. All rights reserved.

The identification of a frequency index works the same way. For example, to plot the real and imaginary parts of h21 as a
For example, to find the index for a frequency close to function of frequency at a bias of 0 V, use:
1 GHz, use: set BiasPointIndex [rfx_GetNearestIndex \
set FrequencyIndex [rfx_GetNearestIndex \ Target=0.0 $rfx_BiasPoints]
Target=1e9 $rfx_Frequencies]
set Frequency [lindex $rfx_Frequencies \ set ReIm [rfx_GetRFCList "ParP1P2=h21 \
$FrequencyIndex] XAxis=frequency BiasOrFqIndex=$BiasPointIndex"]
set Re [lindex $ReIm 0]
set Im [lindex $ReIm 1]
Mason’s unilateral gain
cv_createFromScript Re_h21 $rfx_Frequencies $Re y
The routine rfx_GetMUG returns a list of Mason’s cv_display Re_h21 y
unilateral gain (MUG) values. For example, to compute the
gain as a function of frequency at the fifth bias point index, cv_createFromScript Im_h21 $rfx_Frequencies $Im y
use: cv_display Im_h21 y

set MUG [rfx_GetMUG "XAxis=frequency \ Any parameter can be accessed by selecting the proper
BiasOrFqIndex=5"]
value of the ParP1P2 argument. Valid choices are, for
example, y11, z12, h21, and s22.
To plot the gain curve as a function of frequency, create an
Inspect curve as follows:
Plotting a parameter as a function of bias or plotting a
cv_createFromScript MUG $rfx_Frequencies $MUG y family of curves works is a similar way as for MUG (see
cv_display MUG y Mason’s unilateral gain on page 13).

To plot the gain curve as a function of bias for a given

Plotting magnitude and phase of an RF
frequency, for example 1 GHz, use:
parameter
set FrequencyIndex [rfx_GetNearestIndex \
Target=1.e9 $rfx_Frequencies]
When the real and imaginary parts of an RF parameter as a
set MUG [rfx_GetMUG "XAxis=bias \ function of bias or frequency are available (see Plotting a
BiasOrFqIndex=$FrequencyIndex"] complex RF parameter), they can be converted to magnitude
and phase using the utility routines rfx_ReIm2Abs and
cv_createFromScript MUG $rfx_BiasPoints $MUG y rfx_ReIm2Phase:
cv_display MUG y
set dB 0
set Abs [rfx_ReIm2Abs $ReIm $dB]
To plot a family of gain curves, for example between the set Phase [rfx_ReIm2Phase $ReIm]
bias points on –0.5 V and 0.5 V, use a Tcl loop:
set StartIndex [rfx_GetNearestIndex \ cv_createFromScript Abs_h21 $rfx_Frequencies $Abs y
Target=-0.5 $rfx_BiasPoints] cv_display Abs_h21 y
set EndIndex [rfx_GetNearestIndex \
Target=0.5 $rfx_BiasPoints] cv_createFromScript Phase_h21 $rfx_Frequencies $Phase
cv_display Phase_h21 y2
for {set i $StartIndex} {$i <= $EndIndex} {incr i} {
set MUG [rfx_GetMUG "XAxis=frequency \ The second argument of rfx_ReIm2Abs can take the
BiasOrFqIndex=$i"] values 0, 10, or 20. For 10 or 20, the magnitude is
converted to decibel using 10*log(abs) or 20*log(abs),
set BiasPoint [lindex $rfx_BiasPoints $i]
respectively.
cv_createFromScript MUG($i) $rfx_Frequencies $MUG y
cv_display MUG($i) y
cv_setCurveAttr MUG($i) "MUG B=$BiasPoint" \
red solid 3 circle 5 defcolor 1 defcolor
}

Plotting a complex RF parameter

The routine rfx_GetRFCList is used to access a matrix

element of the complex Y, Z, H, or S parameters as a
function of either frequency or bias.

Copyright © 2007 Synopsys, Inc. All rights reserved. 13

 12
40 10
Abs_h21
Re_h21
30 Im_h21
Phase_h21
11
-50 10
20

fK1 [Hz]
h21

h21 Phase
10 10
10
-100
0

9
-10 10

-150
-20
108 109 1010 1011 1012 8
Frequency [Hz] 10
-0.4 -0.2 0 0.2 0.4
Figure 11 RF parameter h21 as function of frequency for a gate bias of Gate Voltage [V]
Vgs = 0 V: absolute value |h21| (red), real part (blue), imaginary
Figure 12 Region of unconditional stability: The K=1 isolines are red and
part (green), and phase (black dashes) blue. The K=1.1 isolines are green dashes. It follows that the
HEMT is unconditionally stable below about 5 GHz and above
100 GHz. This data is computed with a feedback resistor of
Region of unconditional stability 100 kΩ μm.

An amplifier is unconditionally stable if the Rollett stability Rollett stability factor, maximum stable gain,
factor (K) is greater than one. To plot the boundaries of the and maximum available gain
region of stability (that is, the isolines K=1), use the routine
rfx_GetFK1:
The routine rfx_GetK_MSG_MAG returns the Rollett
set fK1_1 [rfx_GetFK1 "XScale=log Scale=1 \ stability factor (K), the maximum stable gain (MSG), and
Occurrence=1"] the maximum available gain (MAG). The arguments for this
routine are the same as for Mason’s unilateral gain (see
cv_createFromScript fK1_1 $rfx_BiasPoints $fK1_1
cv_display fK1_1 y
Mason’s unilateral gain on page 13). However, the routine
returns three lists instead of one. The first list contains the
set fK1_2 [rfx_GetFK1 "XScale=log Scale=1 \ Rollett stability factor, the second list contains the MSG,
Occurrence=2"] and the last list contains the MAG. To plot these three
curves as a function of frequency for the first bias point, use
cv_createFromScript fK1_2 $rfx_BiasPoints $fK1_2 for example:
cv_display fK1_2 y
set K_MSG_MAG [rfx_GetK_MSG_MAG "XAxis=frequency \
BiasOrFqIndex=0"]
Use the optional keyword Occurrence=<n> if there is
set K [lindex $K_MSG_MAG 0]
more than one boundary of a region of stability for a given set MSG [lindex $K_MSG_MAG 1]
bias point. If, for example, Occurrence=2, the location set MAG [lindex $K_MSG_MAG 2]
of the second occurrence of K=1 is returned. If no second
K=1 occurrence is found, 0 is returned instead. cv_createFromScript K $rfx_Frequencies $K y
cv_display K y2
If only the K=1 boundaries are shown, it is unclear as to
cv_createFromScript MSG $rfx_Frequencies $MSG y
which side of the boundary the amplifier is stable. Use the
cv_display MSG y
optional keyword Target to draw another isoline. For
example, the isoline K=1.1 is drawn with: cv_createFromScript MAG $rfx_Frequencies $MAG y
set fK2_1 [rfx_GetFK1 "XScale=log Scale=1 \ cv_display MAG y
Target=1.1 Occurrence=1"]

cv_createFromScript fK2_1 $rfx_BiasPoints $fK2_1 y

cv_display fK2_1 y

14 Copyright © 2007 Synopsys, Inc. All rights reserved.

 You can now click the frequency label of interest in the
4 legend and the corresponding point in the Smith chart is
30 highlighted.
3.5
Maximum Stable/Available Gain [dB]

20 3

K 2.5
10 MSG

K(1)
MAG
2

0
1.5

-10 1

8 9 10 11 12 0.5
10 10 10 10 10
Frequency [Hz]
Figure 13 Rollett stability factor (red), MSG (blue), and MAG (green) as
function of frequency for a gate bias of Vgs = 0 V. In the region
where the amplifier is not unconditionally stable (K < 1), MAG is set
to zero. This data is computed with a feedback resistor of
100 kΩ μm. Figure 14 The RF parameters s11 and s22 are shown on a Smith chart for a
gate bias of 0 V and frequencies between 100 MHz and 1 THz.
Shown is a screen capture of the Inspect window. By clicking a
Smith charts frequency label in the legend, the corresponding data point in the
Smith chart is highlighted.

For a Smith chart, the imaginary part of an RF parameter is

To plot a Smith chart for a different RF parameter, such as
plotted against the real part on a backdrop of a family of
s22, replace s11 with s22 in the script segment shown
circles of constant resistive/conductive components (R-
above. To plot the parameter for a fixed frequency as a
circles) and family of constant capacitive/inductive
function of bias, change XAxis=frequency to
components (x-circles) [5]. The backdrop is generated with
XAxis=bias, and, in the foreach loop, change
the routine rfx_SmithBackdrop. It requires two lists as
$rfx_Frequencies to $rfx_BiasPoints.
arguments. The first list contains the R-values and the
second list contains the x-values. For example, the backdrop
shown in Figure 14 is generated with the command: Polar plots
set RCList [list 0 0.3333 1.0 3.0]
Polar plots can be generated in a similar way as Smith
set CIList [list 0.268 0.575 1 1.73 3.75]
rfx_SmithBackdrop $RCList $CIList
charts. The routine for creating a polar backdrop is
rfx_PolarBackdrop. It takes two lists as arguments.
For example, to plot the imaginary part of s11 against the The first list contains the radii of the concentric circles to be
real part as a function of frequencies at the first bias point, drawn and the second list contains the angles for which lines
use: are to be drawn. For example, the backdrop shown in
Figure 15 on page 16 is generated with:
set ReIm [rfx_GetRFCList "ParP1P2=s11 \
XAxis=frequency BiasOrFqIndex=0"] set RList [list 0.25 0.5 0.75 1.0]
set Re [lindex $ReIm 0] set PhiList [list 0 30 60 90 120 150]
set Im [lindex $ReIm 1] rfx_PolarBackdrop $RList $PhiList

cv_createFromScript s11 $Re $Im y Plotting the RF parameter and adding labels work in the
cv_display s11 y same way as for the Smith charts (see Smith charts).

The following script segment enables you to see which

frequency is associated with a given point on the Smith
chart:
foreach X $Re Y $Im Fq $rfx_Frequencies {
set LBL [format %.3e $Fq]
cv_createFromScript s11($LBL) $X $Y y
cv_display s11($LBL) y
cv_setCurveAttr s11($LBL) "s11 Fq=$LBL" \
blue solid 0 circle 5 defcolor 1 defcolor
}

Copyright © 2007 Synopsys, Inc. All rights reserved. 15

 [5] K. F. Sander, Microwave Components and Systems,
1 Wokingham, England: Addison-Wesley, 1987.
0.8 [6] R. J. Trew, “Equivalent Circuits for High Frequency
Transistors,” in Proceedings of the IEEE/Cornell Conference
0.6 on Advanced Concepts in High Speed Semiconductor
0.4
Devices and Circuits, pp. 199–208, August 1987.

0.2
Im S

-0.2

-0.4

-0.6

s21
-0.8
s12
-1
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
Re S
Figure 15 The RF parameters s21 and s12 are shown on a polar plot for a
gate bias of 0 V and frequencies between 100 MHz and 1 THz.

Exporting RF parameters

To export the Y, Z, H, or S matrix into a comma-separated

values (CSV) file, use:
rfx_Export2csv "FileName=Y.csv \
Parameter=Y XAxis=frequency

The file name is set with the keyword FileName. The

keyword Parameter defines the parameter to be exported.
The keyword XAxis specifies whether the parameters
should be sorted by frequency first (XAxis=frequency),
with bias being the secondary parameter, or by bias first
(XAxis=bias) with the frequency being the secondary
parameter. The CSV file can be loaded into any spreadsheet
application. It contains a header that starts with the number
of bias and frequency points as well as the value of the first
and last bias and frequency points. A table follows, which
contains, for the example given above, the columns:
bias, freq, R: Y.11, I: Y.11, R: Y.12, I: Y.12, \
R: Y.21, I: Y.21, R: Y.22, I: Y.22

References
[1] S. R. Bahl and J. A. del Alamo, “A New Drain-Current
Injection Technique for the Measurement of Off-State
Breakdown Voltage in FET’s,” IEEE Transactions on
Electron Devices, vol. 40, no. 8, pp. 1558–1560, 1993.
[2] R. S. Carson, High-Frequency Amplifiers, New York: John
Wiley & Sons, 2nd ed., 1982.
[3] W. Liu, Handbook of III-V Heterojunction Bipolar
Transistors, New York: John Wiley & Sons, 1998.
[4] S. M. Sze, Physics of Semiconductor Devices, New York:
John Wiley & Sons, 2nd ed., 1981.

16 Copyright © 2007 Synopsys, Inc. All rights reserved.