nonlinear characterization and modeling through pulsed iv ...small-signal iv model non-linear...
TRANSCRIPT
Nonlinear Characterization and Modeling Through Pulsed IV/S-Parameters
– Introduction
– Core device model extraction
– Model Enhancement
– Model Validation
OUTLINE
Types of Large-Signal Transistor Models
Compact
model
Behavioral
model
Physic
model
Physical insight
Operating rangeConvergence
Extrapolation
Accuracy
Easy modeling
processUsability for Circuit design
Commercial compact FET models
• Mostly used models for GaN HEMTs
FET modelsNumber of parameters
Electro-thermal effect
TrappingEffects
Original DeviceContext
Curtice3 [1] 59 No No GaAs FET
CFET [2] 53 Yes No HEMT
EEHEMT1 [3] 71 No No HEMT
Angelov [4] 80 Yes No HEMT/MESFET
AMCAD HEMT1 [5] 65 Yes Yes GaN HEMT
• AMCAD GaN HEMT1 is the only model here with a complete extraction flow
based on pulsed IV/RF measurements
Small-Signal IV ModelNon-linear
capacitancesThermal model
Trappingeffects
y = 0.0029x + 0.6375
y = 0.0049x + 0.6889
0.4
0.6
0.8
1
1.2
1.4
1.6
0 50 100 150 200
T°C
Rs, R
d
Rs
Rd
y = -0.0008x + 1.1543
1.02
1.04
1.06
1.08
1.1
1.12
1.14
1.16
1.18
0 50 100 150 200
T°C
Idss
Rg
Lg
Cpg
Ls
Cpd
Ld
Rs
Rd
Ri
Cds
τ
Gm
Gd
Cgs
Cgd
Rgd
Dgs=f(Vgs)
Dgd=f(Vgd)
Ids=f(Vgs,Vds)
Cgs=f(Vgs)
Cgd=f(Vgd)
Rs=f(T)
Rd=f(T)
Dgs=f(Vgs,T)
Dgd=f(Vgd,T)
Ids=f(Vgs,Vds,T)
Ids=f(Vgs_trap,Vds,T)
Various effects are successively added
Compact FET model extraction flow
Parameter extraction methodology
– Introduction
– Core device model extraction
– Model Enhancement
– Model Validation
OUTLINE
Parameter extraction methodology
Pulsed IV / S parameter
Multibias set of linear models
Nonlinear model
1st step: bias-dependant S
parameters
Core device model
2nd step: large
signal fitting
Measurements
Modeling
Parameter extraction methodology
Pulsed IV / S parameter
Multibias set of linear models
Nonlinear model
1st step: bias-dependant S
parameters
Core device model
2nd step: large
signal fitting
Measurements
Modeling
Pulsed IV measurements
Several quiescent bias point
Short pulse : Quasi-isothermal conditions
Low duty cycle : Constant mean temperature
Quiescent bias point : Thermal conditions fixed
Modelling Process
Small-Signal
Rg
Lg
Cpg
Ls
Cpd
Ld
Rs
Rd
Ri
Cds
τ
Gm
Gd
Cgs
Cgd
Rgd
Bias Bias
Pulsed S parameter measurements
Compact FET model extraction flow
Pulsed S parameter measurements
The first & most important point :
• Pulsed S parameter measurements must not
be noisy
• Small S2P measurement variation = strong
influence over the linear model extraction :
optimization algorithm
Dynamic range in pulsed mode > 90dB for Duty Cycle ~ 5%
IVCAD
Requirements :
Pulsed S-parameter Measurements
Pulsed S-parameter measurements must not be noisy at low duty cycle with narrow pulse width
Pulse detection methods
Wideband detection Narrowband detection
• No pulse desensitization
• Increased noise with narrow pulse width due
to wider IF bandwidth
• Limited pulse width by maximum available IF
bandwidth
• Narrower minimum pulse width than
wideband pulse
• Reduced dynamic range with low duty cycle
due to pulse desensitization by 20*log(duty
cycle)
IF filterIF filterReceiver samples Receiver samples
PNA/PNA-X Noise reduction techniques and performances
Peak-to-peak noise with wideband detection at 10% duty cycle
Averaging 20 times in calibration and measurements
10 us
(150 kHz)
Pulse
width
(IFBW)
5 us
(280 kHz)
1 us
(1.5 MHz)
500 ns
(3 MHz)
0.03 dB
0.04 dB
0.06 dB
0.09 dB
0.005 dB
0.006 dB
0.012 dB
0.013 dB
No averaging in calibration and measurements
Dynamic range with wideband detection at 10% duty cycle with 10 us, 5 us, 1 us,
500 ns pulse width
Averaging 20 times in calibration and measurements,
1% smoothing on
No averaging in calibration and measurements, 1%
smoothing on
PNA/PNA-X Noise reduction techniques and performances
10 us
Pulse
width
5 us
1 us
500 ns
0.009 dB
0.009 dB
0.012 dB
0.011 dB
0.005 dB
0.006 dB
0.012 dB
0.013 dB
Wideband detection with 20 times averaging in
calibration and measurements
Narrowband detection with no averaging in calibration
and measurements
Peak-to-peak noise at 10% duty cycle
PNA/PNA-X Noise reduction techniques and performances
Dynamic range with narrowband detection at 500 ns pulse width
No Averaging, 1% smoothing on, 500 Hz IF bandwidthHardware
gating
Crystal
filter
Software
gating
Spectral
nulling
>100 dB at 10%
>100 dB at 5%
90 dB at 1%
85 dB at 0.5%
PNA/PNA-X Noise reduction techniques and performances
E836x Legacy PNA N524xA PNA-X N522xA New PNA
Pulse generator External Internal/External Internal/External
Pulse modulator External Internal/External Internal/External
Wideband detection
Max BW/Min PW 35 kHz / 50 us 15 MHz / 100 ns 15 MHz / 100 ns
High level noise* 0.006 dBrms 0.002 dBrms 0.002 to 0.003 dBrms
Dynamic range** 114 to 123 dB 124 to 129 dB 127 dB
Narrowband detection
Min IF gate width 20 ns <20 ns <20 ns
Dynamic range*** 85 dB <105 dB <105 dB
* Specified as trace noise magnitude, at 20 GHz, at 1 kHz IF bandwidth
** Specified performance at 20 GHz, at 10 Hz IF bandwidth
*** Measured performance at 10 GHz at 10 Hz IF bandwidth, 1% duty cycle
Small-Signal IV Model
Rg
Lg
Cpg
Ls
Cpd
Ld
Rs
Rd
Ri
Cds
τ
Gm
Gd
Cgs
Cgd
Rgd
Dgs=f(Vgs)
Dgd=f(Vgd)
Ids=f(Vgs,Vds)
RF RF
DC DC
Compact FET model extraction flow
Pulsed IV parameter measurements• Pulsed IV measurements must be accurate
from low to high voltage/current values
• Accurate IV data =
Reliable current source
Transconductance
Leakage current
Ideality factor schottky diode
IVCAD
How to get accurate pulsed IV measurements ?
PIV system
Pulsed IV parameter measurements
How to get accurate pulsed IV measurements ?
15 bits + sign Drain250V
16 bits
16 bits25V
Gate
+15V
-15V
15 bits + sign
Pulse shape monitoring
20ns time resolution
Pulsed IV parameter measurements
How to get accurate pulsed IV measurements ?
Measurement Resolution
Voltage Absolute Accuracy
Voltage Range
0,6mV
65mV
15V
0V-15V
0mA
20mA
200mA
1µA
40µA
20mA
9µA
400µA
200mA
Pulsed IV parameter measurements
How to get accurate pulsed IV measurements ?
0V 25V 250V
Measurement Resolution
Voltage Absolute Accuracy
Voltage Range
0,53mV 4,9mV
50mV 500mV
0A
1A
10A
22µA
2mA
200µA
20mA
Pulsed IV parameter measurements
Pulsed IV & S parameter measurements
How to get accurate pulsed IV measurements ?
Synchronisation between Pulse IV and pulse S parameters
Pulsed S parameter measurements
How to get accurate pulsed IV measurements ?
Synchronisation between Pulse IV and pulse S parameters
Parameter extraction methodology
Pulsed IV / S parameter
Multibias set of linear models
Nonlinear model
1st step: bias-dependant S
parameters
Core device model
2nd step: large
signal fitting
Measurements
Modeling
• Extrinsic parameters
- pad capacitances Cpg, Cpd
- port metallisation inductances Lg, Ld, Ls
- port ohmic resistances Rg, Rd, Rs
• Intrinsic parameters
- channel capacitances Cgs, Cgd
- voltage-controlled current source with
transconductance gm and transit time delay
tau
- ohmic resistances Ri, Rgd
- output capacitance Cds and resistance Rds
Small signal FET modeling
Rs
RdRg
Grille Drain
SourceSource
Rgd
Ri
Cgs
Cgd
CdsGm Rds
Gm = Gm 0 e-jTransistor intrinsèque
Transistorintrinsèque
CpdCpg
Lg Ld
Ls
G
S
D
The small-signal model presents
two parts :
- an intrinsic circuit
- an extrinsic circuit related to the
parasitic elements
An algorithm developped at the
IRCOM lab allows to optimize the
extrinsic elements in order to get
intrinsic parameters that do not
depend on the frequency
Rs
RdRg
Grille Drain
SourceSource
Rgd
Ri
Cgs
Cgd
CdsGm Rds
Gm = Gm 0 e-jTransistor intrinsèque
Transistorintrinsèque
CpdCpg
Lg Ld
Ls
G
S
D
The small-signal model presents
two parts :
- an intrinsic circuit
- an extrinsic circuit related to the
parasitic elements
An algorithm developped at the
IRCOM lab allows to optimize the
extrinsic elements in order to get
intrinsic parameters that do not
depend on the frequency
intrinsic
transistor
intrinsic transistor
Gate Drain
Source
Small signal FET modeling
• Extraction of extrinsic and intrinsic
parameters:
There is only one set of extrinsic parameters
for which intrinsic parameters are independent
from the frequency
For a given set of extrinsic parameters, intrinsic
admittance matrix of the device is extracted
from measured [S] parameters
Multi-biasing extraction of the
linear model
Set min. and max. for each extrinsic parameter- user choice- initiated by cold FET meas.
Optimization algorithm: annealing, fast simulated diffusion(intrinsic parameters calculus)
Fit ?
No
Yes
– Introduction
– Core device model extraction
– Model Enhancement
– Model Validation
OUTLINE
Parameter extraction methodology
Pulsed IV / S parameter measurement results
Specific measurements
Multibias set of linear models
Nonlinear modelEnhanced
Nonlinear model
1st step: bias-dependant S
parameters
- diodes
- g-d breakdown
- thermal effects
- charge carrier trapping
3rd step:
setting of
additional
parameters
Core device model
2nd step: large
signal fitting
Model Enhancement
Small-Signal IV ModelNon-linear
capacitancesThermal model
Trappingeffects
y = 0.0029x + 0.6375
y = 0.0049x + 0.6889
0.4
0.6
0.8
1
1.2
1.4
1.6
0 50 100 150 200
T°C
Rs, R
d
Rs
Rd
y = -0.0008x + 1.1543
1.02
1.04
1.06
1.08
1.1
1.12
1.14
1.16
1.18
0 50 100 150 200
T°C
Idss
Rg
Lg
Cpg
Ls
Cpd
Ld
Rs
Rd
Ri
Cds
τ
Gm
Gd
Cgs
Cgd
Rgd
Dgs=f(Vgs)
Dgd=f(Vgd)
Ids=f(Vgs,Vds)
Cgs=f(Vgs)
Cgd=f(Vgd)
Rs=f(T)
Rd=f(T)
Dgs=f(Vgs,T)
Dgd=f(Vgd,T)
Ids=f(Vgs,Vds,T)
Ids=f(Vgs_trap,Vds,T)
Various effects are successively added
Compact FET model extraction flow
• Gate charge is partitioned into gate source and gate-drain charge. Each charge expression is
a function of both VDS and VGS.
• For power amplifier applications, 2D models do not bring a breakthrough in the precision, but
they are much more complex
• 1 dimension capacitances extracted along optimal load-line are preferred due to simplicity.
1D capacitances with equations based on hyperbolic tangents are naturally charge
conservatives
• Output Capacitance Cds is linear – no voltage dependence (weak anyway)
Cgs=f(Vgs)
+ Modeling simplicity. Very good convergence
- Validity of the 1D ?
Cgd=f(Vgd)
Non linear capacitances
33
Non linear capacitances
• Capacitances modeling: 1D vs 2D
Better fit of the 2D
model in wider domain
But the 1D model
has a good behavior
34
Cgd capacitance extracted
along optimal load-line for
power amplification
Cgd • Feedback capacitance Cgd is a strong function of drain voltage.
Non linear capacitances
Cgs • Input capacitance Cgs is a strong function of gate voltage.
The gate-voltage non-linearity also effects model’s harmonic generation
Cgs capacitance extracted
along optimal load-line for
power amplification
Non linear capacitances
Output current source
meas.
model
10 20 30 40 50 600 70
0.2
0.4
0.0
0.6
Vds (V)
Id (
A)
-9 -8 -7 -6 -5 -4 -3 -2 -1 0 1-10 2
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
0.45
0.00
0.50
Vgs (V)
Gm
(S
)
• AMCAD drain current model formulation allows to predict
very accurately the I-V curves, the partial derivatives gm
and gd, the knee voltage and the transconductance
decrease at high current.
Output current sourceIdss ↔ amplitude
Vp0 ↔ pinch-off
Vdsp, A ↔ slope
P ↔ gd
M, P ↔ fitting parameters
AlphaGm, Vgm, BetaGm, Vdm ↔ gm (derivative)
Diodes
• Gate-drain and gate-source diode
equations include forward conduction
of gate current
n)formulatio classical (a. . VdAlpha
eIsID
Breakdown generator
• Gate-drain Breakdown generator
50 100 150 2000 250
0
200
400
600
-200
800
Vds (V)
Ids (
mA
)
50 100 150 2000 250
-200
-100
0
100
-300
200
Vds (V)
Igs (
mA
)
The breakdown phenomena leads to a current from the drain to the gate when the
device is pinched-off and for high values of Vds voltage. In this case, the whole
negative current characterized on the gate is seen in positive on the drain
A polynomial expression with order 4 is necessary to model the cross of breakdown
curves, with varies depending on the process
Thermal effects
• Static and Dynamic self-heating effects
20 40 60 800 100
-0.0
0.2
0.4
0.6
0.8
-0.2
1.0
Vds (V)
Ids (
A)
Static
Dynamic
1
2
25°C
5 10 15 20 25 30 35 400 45
-0.0
0.2
0.4
0.6
0.8
-0.2
1.0
Vds (V)
Ids (
A)
-40°C
5 10 15 20 25 30 35 400 45
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
-0.1
0.9
Vds (V)
Ids (
A)
150°C
5 10 15 20 25 30 35 400 45
0.0
0.1
0.2
0.3
0.4
-0.1
0.5
Vds (V)
Ids (
A)
• Temperature dependence with ambient or chuck temperature
Thermal effects
AMCAD original current source
Diodes
Non linear Capacitances
Breakdown source
Thermal circuit
Dissipated Power
Extrinsics
Gate access
Drain access
Source access
• Drain current is only temperature dependent model element
Takes into account ambient temperature and self-heating effects
Thermal analog circuit to model self-heating and elevated heat sink temperatures
RC cells
Thermal effects
Trapping effects
AMCAD original current source
Diodes
Non linear Capacitances
Breakdown source
Thermal circuit
Dissipated Power
Extrinsics
Gate access
Drain access
Source access
• Charging and discharging of traps has influence on Ids and leads to current collapse. This is
described in the model by trapping effects modifying the gate command and separated into
gate and drain lag sub-circuits
Igs(T°) Vgs
_intGate- & Drain-lag
Vgs Vds
Cgs
Ri
Rs(T°)
Ls
Cds
Ids (Vgs_int(t-τ), Vds(t), T°)
Cpd
LdRd(T°)RgdCgd
Igd(T°)
RgLg
Cpg
Ibk
Transistor intrinsèque
Igs(T°) Vgs
_intGate- & Drain-lag
Vgs Vds
Cgs
Ri
Rs(T°)
Ls
Cds
Ids (Vgs_int(t-τ), Vds(t), T°)
Cpd
LdRd(T°)RgdCgd
Igd(T°)
RgLg
Cpg
Ibk
Transistor intrinsèque
Trapping effects
Trapping effects
• Charge of the capacitance = Ionized traps
Charge through Rcapture, Emission through Rémission
Diode = dissymmetry of the
capture and emission process
Tuning of the magnitude of the
trapping effects
Fundamental assumption : dissymetry of the capture and emission process
signal reshaping
circuit Port
Vout
Port
Vin
Diode
diode
R
Remission
R
Rcapture
C
C8
Trapping effects
• Bias dependant gate lag -> current reduction over the entire characteristic
Bias dependent drain lag -> current reduction and shifts the knee-voltage to a higher Vds
Model covers knee walkout to avoid
errors in calculation of output power.
5 10 15 20 25 30 35 40 45 50 550 60
0.0
0.1
0.2
0.3
0.4
0.5
0.6
-0.1
0.7
Vds (V)
Ids (
A)
(H
)
H
gate-lag : Id ↘ => Pout ↘
Id ↘Vknee↗
=> Pout ↘drain-lag :
Trapping effects
• Decreasing form of the mean output current only reproduced with traps modeled
-5 0 5 10 15 20 25-10 30
0.20
0.25
0.30
0.15
0.35
Pin( dBm)
Ids (
A)
(H
)Id
s (
A)
(H
) (
H)
H
0.05 0.10 0.15 0.20 0.25 0.300.00 0.35
1
2
3
0
4
Pin (W)
Pout
W
(H)
Pout
W
(H)
(H
)
H
-5 0 5 10 15 20 25-10 30
0.20
0.25
0.30
0.15
0.35
Pin( dBm)
Ids (
A)
(H
)Id
s (
A)
(H
) (
H)
H
0.05 0.10 0.15 0.20 0.25 0.300.00 0.35
1
2
3
0
4
Pin (W)
Pout
W
(H)
Pout
W
(H)
(H
)
H
Ids
(A)
Po
ut (
W)
Ids ↘
Pout ↘
meas
model without traps model
model with traps model
meas
model without traps model
model with traps model
– Introduction
– Core device model extraction
– Model Enhancement
– Model Validation
OUTLINE
Parameter extraction methodology
Pulsed IV / S parameter measurement results
Specific measurements
Power measurements
Multibias set of linear models
Nonlinear modelEnhanced
Nonlinear modelFinal Nonlinear
model
1st step: bias-dependant S
parameters
- diodes
- g-d breakdown
- thermal effects
- charge carrier trapping
-load-pull measurements
CW, pulsed
2-tones
time domain
3rd step:
setting of
additional
parameters
4th step:
implementation in
commercial simulator
5th step:
validation and
refinement
Core device model
2nd step: large
signal fitting
Model Enhancement Model Validation
meas.
model
-5 0 5 10 15 20 25-10 30
10
15
20
25
30
35
5
40
10
20
30
0
40
Pin dBm
Pout
(dB
m)
and G
ain
(dB
)
PA
E (%
)
PoutPAE
gain
0 5 10 15 20 25-5 30
0
20
40
-20
60
0
20
40
-20
60
Pin dBm
Po
ut (d
Bm
) a
nd
Ga
in (
dB
)
PA
E (%
)
meas.
model
Pout PAE
gain
Model validation of a 8x75 µm GaN HEMT with
load-pull measurements performed at 6 GHz
for optimum PAE load impedance in class-AB
Model validation of a 8x400 µm GaN HEMT with
load-pull measurements performed at 3 GHz for
the optimum Pout load impedance in class-B
• Large-signal
Model validation
Model validationVNA Based load pull system is preferred for model validation
Specific Architecture
PA
VNATuner f0, 2f0, 3f0
50
DUTT T
Gate
Drain
DC or pulse DC supplies
+ meas Units
CW or pulse RF signal
f0 or f1+f2
Low loss directional
couplers
Tuner f0
Model validation
VNA Based load pull system is preferred for model validation
Power meter based system are wideband measurement system
f0
2.f0 3.f0
VNA based system can be narrowband measurement system
More information
for model
validation or
efficient design
f0
f0
f0
2.f0 3.f0
2.f0 3.f0
2.f0 3.f0
VNA based Load Pull systems
Some Measurement definition
P_in: Power delivered to the DUT by the source
P_out: Power delivered to the load impedance
Power gain is the ratio of the power delivered to the load (Pout) to the
power delivered to the transistor by the source (Pin).
Model validationVNA Based load pull system is preferred for model validation
Specific Architecture
PA
VNATuner f0, 2f0, 3f0
50
DUTT T
Gate
Drain
DC or pulse DC supplies
+ meas Units
CW or pulse RF signal
f0 or f1+f2
Low loss directional
couplers
Tuner f0
Phase
reference
Time domain load pull
measurements
Deembedding in the intrinsic
reference plane
Parasitic extrinsic elements must
be accurately extracted by previous
S parameter measurements
Large signal impact - class AB, 25V, 10 GHz – Comparison with measurements
With non optimal loads :
Testing Model Validity with Pulsed IV Simulations(Using Agilent ADS)
• Is your FET model suitable for high-frequency design?
– Is dynamic behavior included?
– Is it well-fitted for high-frequency, large signals?
– What is the model’s valid range of use?
“Accounting for Dynamic Behavior in FET Device Models,” Microwave Journal, July, 2011: http://cp.literature.agilent.com/litweb/pdf/5990-8706EN.pdf
Download the Pulsed IV Curve Simulation DesignGuide for FETs: http://edocs.soco.agilent.com/display/eesofkcads/Pulsed+IV+Curve+Simulation+DesignGuide+for+FETs
Maury – Agilent – AMCAD Solution
Microwave Journal – March 2012
Technical Feature
“Compact Transistor Models: The Roadmap to First-Pass Amplifier Design Success”
Hiro Maehara
Applications Expert – Agilent Technologies
Tony Gasseling
General Manager– AMCAD Engineering
Steve Dudkiewicz
Director, Business Development – Maury Microwave