Authors:
- Thomas Schmitt, Modelon GmbH, Munich
- Markus Andres, Modelon GmbH, Munich
- Patrick Denz, Vorarlberg University of Applied Sciences
This paper introduces behavioral (macro) models of power semiconductors, i.e. diodes, MOSFETs and IGBTs, being part of a library for simulating power electronics utilized, e.g. in electrified powertrains of either hybrid electric vehicles (HEV) or purely battery electric vehicles (BEV).
The models consider static, dynamic (switching mode) and thermal effects and in most cases can be fully parameterized solely on the basis of characteristic curves and parameters specified in datasheets. The main purpose of behavioral models is an accurate representation of the semiconductor signals to, e.g. calculate the overall losses.
The MOSFET models are verified in simulations with
various test circuits and are validated with measurement
data provided by a company developing electric drive systems. Furthermore, the arising numerical problems are discussed and possible solutions are provided on how to modify the models in order to use them in e.g. system simulation.
Full text at: https://www.modelica.org/events/modelica2014/proceedings/html/submissions/ECP14096343_DenzSchmittAndres.pdf
4. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Our Customers Need...
• ... models of electric powertrains that answer questions
regarding
1 lifetime
2 maximum driving range
3 temperature development
4 overall efficiency
• One major demand of our customers working in the
automotive field is to have models that can be
parameterized easily, i.e. via available parameters.
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 4 / 29
5. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Our Customers Need...
• ... models of electric powertrains that answer questions
regarding
1 lifetime
2 maximum driving range
3 temperature development
4 overall efficiency
• One major demand of our customers working in the
automotive field is to have models that can be
parameterized easily, i.e. via available parameters.
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 4 / 29
7. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Modeling Approach
To answer this question correctly, an appropriate modeling
approach was demanded...
• Modelica Standard Library (MSL)
• Package: Modelica.Electrical.Analog.Ideal
• Package: Modelica.Electrical.Analog.Semiconductors
• Semiconductor Physics
• modeling the motion and distribution of charge carriers
• geometrical data and doping profile needed
• Behavioral Approach
• modeling the behavior observed at the semiconductor's pin
• real physical structure is lost but model can be
parameterized using datasheets (in most cases)
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 6 / 29
8. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Modeling Approach
To answer this question correctly, an appropriate modeling
approach was demanded...
• Modelica Standard Library (MSL)
• Package: Modelica.Electrical.Analog.Ideal
• Package: Modelica.Electrical.Analog.Semiconductors
• Semiconductor Physics
• modeling the motion and distribution of charge carriers
• geometrical data and doping profile needed
• Behavioral Approach
• modeling the behavior observed at the semiconductor's pin
• real physical structure is lost but model can be
parameterized using datasheets (in most cases)
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 6 / 29
9. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Modeling Approach
To answer this question correctly, an appropriate modeling
approach was demanded...
• Modelica Standard Library (MSL)
• Package: Modelica.Electrical.Analog.Ideal
• Package: Modelica.Electrical.Analog.Semiconductors
• Semiconductor Physics
• modeling the motion and distribution of charge carriers
• geometrical data and doping profile needed
• Behavioral Approach
• modeling the behavior observed at the semiconductor's pin
• real physical structure is lost but model can be
parameterized using datasheets (in most cases)
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 6 / 29
10. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Modeling Approach
To answer this question correctly, an appropriate modeling
approach was demanded...
• Modelica Standard Library (MSL)
• Package: Modelica.Electrical.Analog.Ideal
• Package: Modelica.Electrical.Analog.Semiconductors
• Semiconductor Physics
• modeling the motion and distribution of charge carriers
• geometrical data and doping profile needed
• Behavioral Approach
• modeling the behavior observed at the semiconductor's pin
• real physical structure is lost but model can be
parameterized using datasheets (in most cases)
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 6 / 29
12. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Fundamentals
A behavioral model of a power semiconductor is usually divided
into a static- and a dynamic model.
Static Model:
• describes the state (behavior) of the model while it is either
conducting or blocking
Dynamic Model:
• adds information about the behavior if the model switches
from static mode to dynamic mode, i.e. describes the
switch-on, switch-off behavior of the semiconductor.
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 8 / 29
13. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Fundamentals
A behavioral model of a power semiconductor is usually divided
into a static- and a dynamic model.
Static Model:
• describes the state (behavior) of the model while it is either
conducting or blocking
Dynamic Model:
• adds information about the behavior if the model switches
from static mode to dynamic mode, i.e. describes the
switch-on, switch-off behavior of the semiconductor.
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 8 / 29
14. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Fundamentals
A behavioral model of a power semiconductor is usually divided
into a static- and a dynamic model.
Static Model:
• describes the state (behavior) of the model while it is either
conducting or blocking
Dynamic Model:
• adds information about the behavior if the model switches
from static mode to dynamic mode, i.e. describes the
switch-on, switch-off behavior of the semiconductor.
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 8 / 29
15. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Behavioral Modeling of MOSFETs
The aim is to develop a model that mimics the real behavior of
the MOSFET, i.e. generating the following simulation result.
+-
+-
Tj
T=50
R=200e-3
R1
dynamicMOSFET
ground
pulseVoltage
constVoltage=20
2.0E-7 4.0E-7 6.0E-7 8.0E-7 1.0E-6 1.2E-6 1.4E-6 1.6E-6 1.8E-6
0
40
80
120
1.005E-6 1.010E-6 1.015E-6 1.020E-6
0
40
80
120
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 9 / 29
16. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Deriving a Static MOSFET Model
UGS > Uth,V generates a
conducting channel.
Idb=bf(Vgs)
transferCharacteristic
V
voltageSensor
R=RonFW
RonFw
IdFw
freeWheelingDiodeFw
reverseBlocking
Diode
R=RonBW
RonBw
IdBwforwardBlocking
Diode
freeWheelingDiodeBw
D
S
G
bodyDiode
B
S G D
n n
p
UGS>0
1 on-state forward cond.
2 on-state reverse cond.
3 on-state reverse cond. with
body diode forward biased
4 off-state reverse cond.
5 off-state forward blocking
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 10 / 29
17. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Deriving a Static MOSFET Model
UDS > 0 generates a current
flow ID.
Id1=1f(Vgs)
transferCharacteristic
V
voltageSensor
R=RonFW
RonFw
IdFw
freeWheelingDiodeFw
reverseBlocking
Diode
R=RonBW
RonBw
IdBwforwardBlocking
Diode
freeWheelingDiodeBw
D
S
G
1
bodyDiode
B
S G D
n n
p
UDS>0
1 on-state forward cond.
2 on-state reverse cond.
3 on-state reverse cond. with
body diode forward biased
4 off-state reverse cond.
5 off-state forward blocking
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 11 / 29
18. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Deriving a Static MOSFET Model
UDS < 0 changes the direction of
ID.
Id2=2f(Vgs)
transferCharacteristic
V
voltageSensor
R=RonFW
RonFw
IdFw
freeWheelingDiodeFw
reverseBlocking
Diode
R=RonBW
RonBw
IdBwforwardBlocking
Diode
freeWheelingDiodeBw
D
S
G
2
bodyDiode
B
S G D
n n
p
UDS<0
1 on-state forward cond.
2 on-state reverse cond.
3 on-state reverse cond. with
body diode forward biased
4 off-state reverse cond.
5 off-state forward blocking
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 12 / 29
19. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Deriving a Static MOSFET Model
If ID is increased and UDS > Uth,D
body-diode will conduct too...
Id3=3f(Vgs)
transferCharacteristic
V
voltageSensor
R=RonFW
RonFw
IdFw
freeWheelingDiodeFw
reverseBlocking
Diode
R=RonBW
RonBw
IdBwforwardBlocking
Diode
freeWheelingDiodeBw
D
S
G
3
3
bodyDiode
B
S G D
n n
p
UDS<0
1 on-state forward cond.
2 on-state reverse cond.
3 on-state reverse cond. with
body diode forward biased
4 off-state reverse cond.
5 off-state forward blocking
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 13 / 29
20. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Deriving a Static MOSFET Model
If UGS = 0 and UDS < 0 solely the
body-diode conducts.
Id4=4f(Vgs)
transferCharacteristic
V
voltageSensor
R=RonFW
RonFw
IdFw
freeWheelingDiodeFw
reverseBlocking
Diode
R=RonBW
RonBw
IdBwforwardBlocking
Diode
freeWheelingDiodeBw
D
S
G
4
bodyDiode
B
S G D
n n
p
UDS<0
1 on-state forward cond.
2 on-state reverse cond.
3 on-state reverse cond. with
body diode forward biased
4 off-state reverse cond.
5 off-state forward blocking
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 14 / 29
21. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Deriving a Static MOSFET Model
If UGS = 0 and UDS > 0 no
current will flow.
Idb=bf(Vgs)
transferCharacteristic
V
voltageSensor
R=RonFW
RonFw
IdFw
freeWheelingDiodeFw
reverseBlocking
Diode
R=RonBW
RonBw
IdBwforwardBlocking
Diode
freeWheelingDiodeBw
D
S
G
bodyDiode
B
S G D
n n
p
UDS>0
1 on-state forward cond.
2 on-state reverse cond.
3 on-state reverse cond. with
body diode forward biased
4 off-state reverse cond.
5 off-state forward blocking
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 15 / 29
22. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Dynamic Model
• dynamics are mainly an effect of capacitors between the
MOSFET's pins
• in datasheets one can find characteristic curves for the
input capacitance Ciss, the output capacitance Coss and the
reverse transfer capacitance Crss
D
S
G
CDS
CGS
CDS
• Cds = Ciss − Crss
• Cgd = Crss
• Cgs = Coss − Crss
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 16 / 29
23. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Dynamic Model
• dynamics are mainly an effect of capacitors between the
MOSFET's pins
• in datasheets one can find characteristic curves for the
input capacitance Ciss, the output capacitance Coss and the
reverse transfer capacitance Crss
D
S
G
CDS
CGS
CDS
• Cds = Ciss − Crss
• Cgd = Crss
• Cgs = Coss − Crss
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 16 / 29
24. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Dynamic Model
• dynamics are mainly an effect of capacitors between the
MOSFET's pins
• in datasheets one can find characteristic curves for the
input capacitance Ciss, the output capacitance Coss and the
reverse transfer capacitance Crss
D
S
G
CDS
CGS
CDS
• Cds = Ciss − Crss
• Cgd = Crss
• Cgs = Coss − Crss
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 16 / 29
43. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Summary: IGBT
• PT-IGBTs and IGBTs with field-stop layer cannot be
implemented directly since neither information about the
tail current nor the additional output capacitor is available in
the datasheet.
• In order to model the behavior correctly one has to measure
the switch-off signals thoroughly and fed the results into a
look-up table.
• If it is not possible - for whatever reason - to measure the
switch-off behavior we can utilize the static models and
whenever a discontinuity appears, i.e. switch-on or
switch-off, add the Eon, Eoff = f(IC, T) losses specified in the
datasheet.
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 26 / 29
44. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Summary: IGBT
• PT-IGBTs and IGBTs with field-stop layer cannot be
implemented directly since neither information about the
tail current nor the additional output capacitor is available in
the datasheet.
• In order to model the behavior correctly one has to measure
the switch-off signals thoroughly and fed the results into a
look-up table.
• If it is not possible - for whatever reason - to measure the
switch-off behavior we can utilize the static models and
whenever a discontinuity appears, i.e. switch-on or
switch-off, add the Eon, Eoff = f(IC, T) losses specified in the
datasheet.
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 26 / 29
45. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Summary: IGBT
• PT-IGBTs and IGBTs with field-stop layer cannot be
implemented directly since neither information about the
tail current nor the additional output capacitor is available in
the datasheet.
• In order to model the behavior correctly one has to measure
the switch-off signals thoroughly and fed the results into a
look-up table.
• If it is not possible - for whatever reason - to measure the
switch-off behavior we can utilize the static models and
whenever a discontinuity appears, i.e. switch-on or
switch-off, add the Eon, Eoff = f(IC, T) losses specified in the
datasheet.
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 26 / 29
46. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Summary: IGBT
• This can be done using state graphs. Such models are
already integrated in the EPTL developed at Modelon GmbH.
turn-on
gate = true
conduction
delay = ton
turn-off
gate = false
blocking
delay = toff
IGBT: Pon IGBT: Pcond
Diode: Pcond
IGBT: Poff
Diode: Prec
• This in turn increases the simulation performance and
ensures that the losses are modeled correctly.
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 27 / 29
47. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Summary: IGBT
• This can be done using state graphs. Such models are
already integrated in the EPTL developed at Modelon GmbH.
turn-on
gate = true
conduction
delay = ton
turn-off
gate = false
blocking
delay = toff
IGBT: Pon IGBT: Pcond
Diode: Pcond
IGBT: Poff
Diode: Prec
• This in turn increases the simulation performance and
ensures that the losses are modeled correctly.
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 27 / 29
48. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Simulation Performance
• Since behavioral models of power electronic components
are stiff systems that also generate many events, the
simulation becomes slow or does not converge.
• When integrating the model into a simulation of the entire
vehicle, it must be transformed into an efficiency map model
which stores the power losses of different operating points
in a table.
• The EPTL already provide the means to generate efficiency
maps for the elcectric machines and those for the inverters
are developed too but not fully tested yet.
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 28 / 29
49. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Simulation Performance
• Since behavioral models of power electronic components
are stiff systems that also generate many events, the
simulation becomes slow or does not converge.
• When integrating the model into a simulation of the entire
vehicle, it must be transformed into an efficiency map model
which stores the power losses of different operating points
in a table.
• The EPTL already provide the means to generate efficiency
maps for the elcectric machines and those for the inverters
are developed too but not fully tested yet.
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 28 / 29
50. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Simulation Performance
• Since behavioral models of power electronic components
are stiff systems that also generate many events, the
simulation becomes slow or does not converge.
• When integrating the model into a simulation of the entire
vehicle, it must be transformed into an efficiency map model
which stores the power losses of different operating points
in a table.
• The EPTL already provide the means to generate efficiency
maps for the elcectric machines and those for the inverters
are developed too but not fully tested yet.
Schmitt, Denz, Andres Behavioral Modeling Modelica 2014 28 / 29