introduction to large-signal modeling of compound...
TRANSCRIPT
W. R. Curtice 2008
Introduction to Large-Signal Modeling of Compound
Semiconductor TransistorsWalter R. Curtice, Ph. D.W. R. Curtice Consulting
53 Commanders DriveWashington Crossing, PA 18977
W. R. Curtice 2008
Introduction
• Philosophy of Modeling
• Modeling of HEMT Devices
• Modeling of Bipolar Devices
W. R. Curtice 2008
Modeling Considerations• Device application must be considered when
choosing a model and when characterizing a device.
• Pulsed modulated applications require more attention to the time domain behavior than do CW applications.
• Large-signal, nonlinear equivalent circuit models are more useful than behavior models(for trimming model, for scaling model, etc.).
W. R. Curtice 2008
Modeling Considerations
• Modeling accuracy requirements are related to the device application
• The model’s accuracy need not be better than the variation in the device data
• Any given model can always be improved upon, at a price
• All models can be shown to behave non-physically under some operating condition
W. R. Curtice 2008
Modeling Choices
• Range of usefulness?• Number of model parameters?• Include self-heating effects?• Include dispersion effects?• Include breakdown effects?• Model useful for statistical analysis?
W. R. Curtice 2008
Types of Large-Signal Transistor Models
• Physical or “Physics-Based” Device ModelsCalibrated Using Data
• Measurement-Based Models: Analytical Models Black Box ModelsTable-Based
• Artificial Neural Network (ANN) Models
This presentation willonly deal with this type!
(Also called SPICE Models)
W. R. Curtice 2008
Angelov Model: an Analytical Example
Plot Angelov_ADS/dc_iv_2008/id_vds __1/id_vd
Angelov Model
vd [E+0] i
d.s
[E-3
] 0 2 4 6 8 10
0
100
200
300
400
500
Plot Angelov_ADS/dc_iv_2008/id_vds __1/id_vd
GaAs HEMT Data
vd [E+0]
id.
m [
E-3
]
0 2 4 6 8 100
100
200
300
400
500
Vgs =-1V
-2V
-3V
-4V
Ids = IPK0*(1 + Tanh(P1*(Vgs-Vpk))*Tanh(A*Vds)
The model parameters for current are IPK0, P1, Vpk and A plus Rth and temperature coefficients
W. R. Curtice 2008
W. R. Curtice 2008
Part 1: HEMT Modeling
Characterization and Modeling of GaAs, GaN, SiC and Si LDMOS RF
Power Devices
W. R. Curtice 2008
Modeling Pluses
• GaAs, GaN, SiC and Si LDMOS devices are all three terminal devices
• Many large-signal circuit models available for use with these devices
• Many conventional models can be enhanced to be more accurate for a specific device type.
W. R. Curtice 2008
ADS* FET Models, 2006
Angelov is the onlymodel here that iselectro thermal!
This model has acrude thermal correctionterm applied
19801985
*Agilent Technologies, Inc.
W. R. Curtice 2008
FET Models in MWO(AWR Design Environment 2006)
Angelov2 is the only
FET
electro thermal!
model here that is
MET_LDMOSis electro thermal
EEHEMTalso here
W. R. Curtice 2008
Other HEMT Model Choices• EE_HEMT, Materka, Tajima, etc. models can
be enhanced to be electro thermal • A new model for large devices is reported by
Tijama in Microwave Journal, May 2007• C_FET model (Curtice Consulting) is electro
thermal• Cabral, Pedro et al. model looks promising
(MTT Trans. Nov 2004)• Modified models can be entered into ADS or
MWO • Verilog-A models may be used with ADS
W. R. Curtice 2008
Example of Self HeatingGaN Id vs. Vd for Vg=0 V
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0 5 10 15
Vg (V)
Id (A
)
DC
Pulsed
Vds
W. R. Curtice 2008
C_FET GaN Chip Model(W. R. Curtice Consulting)
Typical GaN Modelusing C_FET8D
PortP1Num=1
LL2
R=L=.09 nH
Temp2
C_FET8DC_FET8D1
CGD_S=-5CGD_P=300RDS_S=8GGS=1.e-9 SGM_RF=1.AREA=1
X_EXP=.55FFE=1.AF=1.KF=0.R_TH=10T_TH=1.e-6A3T=-2.796mA2T=-27.47mA1T=.3383A0T=1.168TSNK=25TMPH=125TMP=25LAMB=3.368mP12=0
P11=2.7P10=-1P09=0P08=-1P07=500P06=700P05=2800P04=9.5P03=-657.8pP02=2.8P01=430VBR=500.NF=1.0IS=5.2e-13R2=10.
R1=1000.GAMMA=10.24BETA=617.6uA3=-4mA2=-45.42mA1=.3676A0=1.564VDSO=10TAU=2pCDS=.4pRIN=.1RDSO=150RS=.6RG=1RD=1.5
LL1
R=L=.07 nH
PortP2Num=2
NewRds and Cgd Parameters
Vds VoltageFor Characterization(10V)
Temperature monitored here
W. R. Curtice 2008
C_FET Model
*GM dispersion circuit not shown
*Thermal analogCircuit is critical forLarge periphery devices
W. R. Curtice 2008
Charge-based Capacitance FunctionsNeeded for Cap(Vgs, Vds)
Example of capacitance modeling fora Triquent HEMT
Cgs(Vgs)
Vds
Cgd(Vgs)
Vds
W. R. Curtice 2008
local
Q(Vi, Vj)
• Qgs(Vgs,Vds) gives rise to two capacitive current terms
• Likewise, Qgd(Vgs,Vds) gives rise to two current terms
• One term is the same as the small-signal cap term. The other is a transcapacitance term
remote
W. R. Curtice 2008
Jansen et al. Approach MTT Transactions, January 1995
• Assume Qgd = 0
• Then Qg = Qgs(Vgs,Vds)
• The transcapacitance then Accounts for the conventional drain to gate capacitance
W. R. Curtice 2008
Cgs (Vgs) Function needs toAccommodate Capacitance “Overshoot”
Behavior of HEMTs
-4.5 -4.0 -3.5 -3.0 -2.5 -2.0 -1.5 -1.0-5.0 -0.5
1.0
1.5
2.0
2.5
0.5
3.0
VGS
data
_8x2
40_1
0v_C
GS.
.CG
S_10
Vcg
s VDS = 10V
-4 -3 -2 -1-5 0
1.0E-12
1.5E-12
2.0E-12
2.5E-12
3.0E-12
5.0E-13
3.5E-12
VGS
Cgs
EEHEMT
C_FET
Note decreasingValues after “turn-on”
Red is data
Both models usea charge functionto compute Cgs and Cgd
RF Transconductance must be modeled accurately to get agreement with IMD3 and IMD5
GaN MOSHFET Example
W. R. Curtice 2008
0
10
20
30
-10 -5 0Gate Source Voltage (V)
Tran
scon
duct
ance
(mS)
Transconductance at VD=10V extracted respectively from CW-IV (xxx), pulsed I-V (ooo), CW s-parameter (∆∆∆), pulsed s-parameter (���), compared with simulated result(—) Quiescent biasing: VGQ=-3V, VDQ=25V.
Fundamental and 2-tone outputPower at 1.9GHz for 25V, -4V operation
See Also:The COBRA Model Cojocaru and Brazil; MTT Trans. 1997
W. R. Curtice 2008
Characterization Problems
• It is difficult to test large periphery devices, so many important characterization tests are done on small devices and the results scaled
• DC and pulsed I/V data are influenced by many test conditions
• RF I/V and reactance data are also affected by measurement and environmental conditions
W. R. Curtice 2008
Parameter Extraction
• Pulsed I/V and pulsed S-parameters for various ambient temperatures enables separation of temperature effects from voltage effects
• Compound semiconductor HEMTs exhibit “surface gating” effects, due to traps
• Must characterize Gm and Gds dispersion• For switching or logic applications, lag effects due
to traps are important to characterize
W. R. Curtice 2008
Parameter Extraction
• Need short-pulse testing capability as well as longer pulses. See Teyssier et al., [Ref. 7]
• IC-CAP (Agilent Tech.) provides testing capability and parameter extraction for “SPICE” models
W. R. Curtice 2008
Some Factors Influencing DC I/V Data
• Ambient temperature effects• Self heating effects- Importance increases
with device peripheryDevice current scales directly with area, but thermal resistance does not scale inversely with area.
• Trapping effects- Charge traps cause I/V “kinks”
W. R. Curtice 2008
Importance of Modeling Self-Heating in Large HEMTs
Thermal Effect in GaN/Si for 50mA/mm
0
50
100
150
200
250
0 10 20 30 40
Periphery, mm
Tem
pera
ture
Ris
e, C
28V56V
Self-Heating is more importantin larger devices because thethermal resistance does not decrease strongly with periphery increase
W. R. Curtice 2008
Example of I/V Kink
Notice kinkhere
W. R. Curtice 2008
I/V Kink Due to Aval. & Traps
-1V
-1V
-.75V
-.75V
-1.25V
-1.25V
Kink
Gate currentoccurring afterkink
W. R. Curtice 2008
Short Pulse Testing Permits Transistor Characterization for
Temperature Effects and Trapping Effects
W. R. Curtice 2008
The following viewgraphs show typical data for a 1-mm
AlGaN/GaN device on SiC substrate and with low dispersion
(Courtesy, WPAFB)
W. R. Curtice 2008
Evaluate the Dispersion in GaN HEMTs by comparing DC and
Pulsed I/VDC vs. Pulsed I/V
0.00100.00200.00300.00400.00500.00600.00700.00800.00900.00
0.00 1.00 2.00 3.00 4.00 5.00
Vds, V
Id, m
A
DC
.2us Pulsed
Large Dispersion Case(Other Vendor)
DC
PULSED
Note that Ids is about the samefor VDS=5V
Low Dispersion Case (WPAFB)
W. R. Curtice 2008
Final DC I/V Family Data and the C_FET Model
Red = DataBlue = C_FET Model
5 W heating power(R_TH = 8)
W. R. Curtice 2008
DC Sub-threshold Characteristic
Note that modelfollows the sub-thresholdcharacteristic very well!
W. R. Curtice 2008
Transconductance Nonlinearity
• gm(V) is the most important nonlinearity in any transistor
• Fundamental frequency output is directly related to gm
• Second harmonic output is directly related to first derivative of gm
• Third harmonic output (and IMD3) is directly related to second derivative of gm
• Nonlinear capacitance becomes more important at higher RF frequencies
W. R. Curtice 2008
External Transconductance, gmx
gmx 1st Deriv
2nd Deriv
VDS = 5V
Model and Data Are in good agreement
W. R. Curtice 2008
RF Modeling Procedure
• Use IC-CAP to extract capacitance parameters using capacitance data
• Optimize model parameters using “hot-FET” S- parameter data
• Evaluate the temperature dependence of the capacitance parameters
• Use load-pull data and power-sweep data to optimize the model parameter set
W. R. Curtice 2008
Typical S-Parameter Comparison
-8 -6 -4 -2 0 2 4 6 8-10 10
freq (500.0MHz to 26.00GHz)
Smea
s(2,
1)Sm
odel
(2,1
)
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8-1.0 1.0
freq (500.0MHz to 26.00GHz)
Smea
s(1,
1)Sm
odel
(1,1
)Sm
eas(
2,2)
Smod
el(2
,2)
-0.10 -0.05 0.00 0.05 0.10-0.15 0.15
freq (500.0MHz to 26.00GHz)
Smea
s(1,
2)Sm
odel
(1,2
) 5 10 15 20 250 30
-5
0
5
10
15
-10
20
freq, GHz
dB(S
mea
s(2,
1))
dB(S
mod
el(2
,1))
1E101E9 3E10
10
20
30
0
40
freq, Hz
dB(h
_val
(2,1
))dB
(h_v
al_m
od(2
,1))
c fet8 VDS=10 V, VGS = 0 V, 6-2-04
S11, S22S21
S12
S21_dB
H21_dB
0.5-26GHz
W. R. Curtice 2008
Tuned Power Sweep at 10GHz
10 15 20 255 30
10
20
30
40
0
50
RFpower
net_
pae
Pin
data
_20v
..PA
E*1
00
10 15 20 255 30
20
25
30
35
15
40
RFpower
p1_d
b
Pin
data
_20v
..Pou
t
10 15 20 255 30
0.30
0.35
0.40
0.25
0.45
0.01
0.02
0.03
0.04
0.00
0.05
RFpower
real
(I_P
robe
1.i[:
:,0])
Pin
data
_20v
..Ic*
.001
real(I_Probe3.i[::,0])
10 15 20 255 30
9
10
11
12
8
13
RFpower
gain
_db
Pin
data
_20v
..Pou
t-dat
a_20
v..P
in
I_GATE_SIM
C_FET8 Simulation vs. 441 Data, VDS=20V, Iq=300 mA, 6-02-04
DRAIN_CURRENT
Pout_dBm
PAE
GAINRed is modelBlue is data
4W/mm outputT_rise=36CT_rise=47C
W. R. Curtice 2008
Modeling the GaN on Silicon Substrate HEMT(Courtesy, Nitronex Corp.)
W. R. Curtice 2008
Over 100W Pulsed Power Obtained 36mm Packaged Device
25 30 35 4020 45
6
8
10
12
4
14
Pin
data
_36m
m_2
8V_P
ULS
ED..G
ain
RFpower
gain
_db
VDS = 28VFreq = 3.5GHzInput,output tuned
RED = data, Blue=model
25 30 35 4020 45
20
40
60
80
100
120
0
140
Pin
pow
(10,
data
_36m
m_2
8V_P
ULS
ED
..Pou
t/10)
*.00
RFpower
p1_w
W. R. Curtice 2008
Modeling Large Periphery Devices
• I/V data can often be scaled accurately from smaller devices
• S-parameter data may or may not be useful due to the low impedance values
• Load-pull and power sweep data can help with capacitance modeling
• Accurate packaging models with losses are extremely important for the large devices
• More data needed for modeling larger devices operated at frequencies at or above 10GHz
W. R. Curtice 2008
Conclusion: Part 1- HEMT Models
• Thermal analog circuit for self-heating effects• Rds, Cgd function of Vds• Charge function for modeling capacitance• Cgs(Vgs) permits fitting of “overshoot” • RF gm(Vgs) must be modeled accurately to
several derivatives• Rds and gm dispersion should be accommodated• Trapping effects may need to be included
W. R. Curtice 2008
Part 2: Bipolar Modeling
A thermal analog circuit is critical for accurately simulate temperature effects in
power bipolar devices.
W. R. Curtice 2008
Simulator Models Availablewith Self-Heating Effects
• ADSVBIC, HICUM, AgilentHBT
• MWOVBIC, HICUM, UCSD HBT
One Problem with VBIC:It is hard coded for Silicon
W. R. Curtice 2008
Compact BJT/HBT Models
• Gummel-Poon Model –All SPICE versions hard coded for Silicon only (TF Equation), No self-heating effects
• VBIC - VBIC95 BCTM Meeting 1995, Weak Aval., Self-heating effects, Some NQS effects
• MEXTRAM -1995, Weak Aval., NQS effects• HICUM –Weak Aval., Self-Heating, NQS
W. R. Curtice 2008
Cut-off Frequency, Ft
• Ft is frequency of unity current gain• Ft is a measure of the frequency response
Slope=-2.242E+001 [LIN/DEC] Yo=2.378E+002 Xo=4.068E+010freq [LOG]
10*
log(
s2h.
m.2
1) [
E+0
]
1E+8 1E+9 1E+10 1E+110
10
20
30
40
50
Ft = 40.68 GHz
(Data = Red points)
HEMTExample
H21 vs. freq
W. R. Curtice 2008
Bipolar Model must Accommodatethe Behavior of the Material
Ft goes up With VcIn Silicon
Ft goes down With VcIn GaAs
GaAsHBT Data
SiGeHBT Data
This behavior is related to the electron velocity-fieldrelationship.
In the modelsFt behavior is controlledby the parameters of theTF function.
W. R. Curtice 2008
GP_M Model (Curtice Consulting)
• Uses GP current and capacitance expressions• The thermal analog circuit was added to model
self-heating and elevated heat sink temperatures• B-C breakdown current, punch-through
capacitance and other effects were added• TF function changed for GaAs to get correct
dependence of FT upon VC
W. R. Curtice 2008
Thermal Analog Circuit
RRR =R _th
CCC =C _th
ItUs e rDe fS R C 1I_Tra n=I(t)*V(t) V_th
I_Tran= Instantaneous power dissipation
R_th*C_th = Thermal time constant
V_th is numerically equal to the temperature rise
Note: Some models useTwo thermal time constants
W. R. Curtice 2008
Thermal Modeling Example
• AlGaAs/GaAs HBT with emitter ballast resistance
• 4-finger(2x20) device on a thinned wafer• R_th = 200 C/W• Modified Gummel-Poon model (GP_M)
used to model device • GP has most model parameters as function
of device temperature
W. R. Curtice 2008
Model and Data forIC-VC Family for IB = .2, .4, .6, .8 ma
Heat Sink = 25 C
VB vs VCIC vs VC
Data is red
W. R. Curtice 2008
RF ParametersVC=1.5VIB=.6 mA
H21 S21
S22
S11
W. R. Curtice 2008
Thermal run-a-way simulation and dataHeat Sink = 25 C
IC vs VCVB=1.4 V
1.375 V
1.350 V
1.325 V1.3 V
Data is red
W. R. Curtice 2008
Previous data and thermal run-a-way simulationwith ballast resistance removed
Heat Sink = 25 C
IC vs VCVB=1.4 V
1.375 V
1.350 V
1.325 V1.3 V
Data is red
W. R. Curtice 2008
Thermal run-a-way for SiGe HBTusing HICUM model
Heat Sink = 25 C
IC vs VCVB=0.85 V
0.825V
0.80 V0.775 V0.75V
Data is red
W. R. Curtice 2008
Conclusion: Bipolar Modeling
• The thermal analog circuit is needed to accurately simulate temperature effects in GaAs-based HBTs
• The TF function must be adaptable to characteristics like GaAs, which is opposite to Si BJT behavior
• Breakdown behavior is important to model