Realistic benchmarking of IGBT-modules with the help of a fast and

easy to use simulation-tool

R. Schnell, U. Schlapbach
ABB Switzerland Ltd, Semiconductors, Fabrikstrasse 3, CH 5600 Lenzburg, Switzerland
e-mail: raffael.schnell@ch.abb.com

Abstract:
IGBT modules normally contain the rated current and voltage in their type designation.
Therefore the rated current is often taken as the first value to compare modules from different
competitors. This is a misleading practice since the definition of the rated current differs
significantly between the modules and manufacturers. In order to allow the user of IGBT
modules to compare various types of IGBTs in a realistic application environment we have
developed a fast and simple to use simulation-tool. With this tool it is possible to compute
static and transient losses and temperature rises in IGBT modules as well as optionally
specified heatsinks in a two-level voltage source inverter topology.
The paper shows examples of simulations with different conditions and module types and
describes the calculation methods employed in the simulation-tool.

the static and switching characteristics are

The definition of the nominal current, dc-collector temperature dependent.
current, rated current, etc. varies significantly The power dissipation calculation of the IGBT and
between different manufacturers as well as diode are done by an average calculation for the
conduction and switching losses over one period
between modules from even the same producer:
T0 of the output frequency [1].
This approach yields in accurate results for twoTJ TC
(1) level voltage source inverters with naturally
IC =
Rth ( J C ) VCEsat
sampled PWM and sinusoidal output currents.
Whereas VCEsat is a function of IC.
IGBT loss calculation: Since the IGBT of one
switch
conducts only over one half period the
Equation 1 shows the general definition of the
conduction
losses are given by the integration of
nominal current. Depending on the module type
also the heatsink temperature can be the the forward losses (VCE0, rCE including RCC'+REE')
reference. Unfortunately different values of TJ or up to T0/2:
T /2
TC are used for different modules and
1 0
=
P
(VCE0 sin(t) + rCE ( sin(t))2 ) (t))dt
manufacturers. Therefore a comparison does cond IGBT
T0 0
hardly make sense. Additionally the definition of
the rated current does not include any switching
(2)
losses but at least within a given chip-technology
a lower VCEsat means higher switching losses or
vice versa. Therefore IGBTs with low VCEsat with (t) as the function of pulse pattern with IGBT
offer a higher nominal current but might fairly well turned-on=1 and IGBT turned-off=0. (t) can be
offer less output current if operated with higher substituted by a function of modulation (m) and
switching frequency.
phase angle ().
With an infinite switching frequency we get the
duty cycle variation over time (PWM pattern).
Loss Calculation:
The simulation-tool offers a relatively exact and
fast method for loss calculation.
1
(2a)
The model data for the IGBT modules are based (t ) = (1 + m sin(t + ))
2
on their respective data-sheet values.
The calculations are done with a linear
approximation
of
the
devices
forward Inserting (t) into the formula and solving the
characteristics and with a polynomial function of integral we get the conduction losses:
the IGBT and diode switching energy. Additionally

Introduction:

PCIM'04 Power Electronics Conference, Nuremberg

revised 01.07.05

Page 1 of 6


2
1

2
1
Pcond = 12 (VCE0 + rCE ) + m cos (VCE0 + rCE 2 ) Pcond = 1 (VF0 + rT ) m cos (VT 0 + rT 2 )
2

4
8 3

4
8 3
(2b)
(5)
The simulation-tool restricts the modulation index In case of the diode the turn-on energy can be
to m1, which is the linear mode of the PWM.
neglected. Therefore only the recovery energy
counts. The recovery energy given in the dataThe switching losses are the sum of all turn-on sheet diagram can be described as a polynomial
and turn-off energies at the switching events.
function:
The measured turn-on and turn-off energies given
in the data-sheet can be described as a
E rec = (a + b I + c I 2 )
(6)
polynomial function (Esw=f(I)):

E sw = Eon + Eoff = (a + b I + c I 2 )

(3)

Since the DC-link voltage can vary in different

applications the dependence of the switching
energy on the DC-voltage needs to be
considered. Within certain limits this dependence
can be assumed linear:

E sw = (a + b + c 2 )

VDC
Vnom

(3a)

To calculate the switching losses the switching

energies have to be added up.

1
Psw = E sw ( )
T0 n

(3b)

Therefore we can calculate the switching losses

as a function of phase-current and switching
frequency.
(3c)

The total IGBT losses are the sum of the

conduction- and switching losses:

PIGBT = Pcond + Psw

(4)

Diode loss calculation:

The diode losses can be nearly calculated the
same way as the IGBT. Since the freewheeling
diode conducts when the IGBT is turned-off the
function of the pulse pattern has to be negated:

PCIM'04 Power Electronics Conference, Nuremberg

a b c 2 V DC
Prec = f sw ( +
+
)
2
4
Vnom

(6a)

The total Diode losses are the sum of the

conduction- and switching losses:

PDiode = Pcond + Psw

(7)

Thermal Calculation:

whereas n depends on the switching frequency.

a b c 2 V DC
Psw = f sw ( +
+
)
2

4
Vnom

The recovery losses as a function of phasecurrent and switching frequency and VDC can be
written as:

The loss calculation yields in average losses over

one period of the output frequency. In fact the
losses per switch only occurs during one halfperiod whereas during the other half period the
antagonist switch is producing losses.
Figure 1 shows in the upper graph a PWM
chopped current as it appears in case of a two
level VSI. In addition the resulting junction
temperature is shown (as calculated by
convolution of momentary power loss and thermal
impedance). It is obvious that the junction
temperature oscillates with the frequency of the
output current. In the lower graph the
corresponding losses Psmoothed are shown. As a
comparison the calculated average losses (Pav)
from simulation-tool are given (dashed line).
If the junction temperature is calculated with
Tjav = Pav * Rth evidently the peak value of junction
temperature exceeds the result of Tj calculated
with the average losses Pav. Therefore the
calculation of Tj with average losses yields in a to
optimistic value.
In order to minimise this inaccuracy the
simulation-tool calculates with Pmod which is two
times the average losses Pav dissipated during
one half period of the phase current. The resulting
junction temperature Tjmod is shown in the lower

revised 01.07.05

Page 2 of 6

400

Tj

Ic

graph of figure 1 and matches in terms of the Thermal equivalent block diagram:
peak and bottom values quite well with the real The simplification done in the simulation-tool is
value of Tj.
the assumption of common reference temperature
point where the temperature is assumed to be
800
125
homogenous over the full area. In case of base120
less devices this reference is the heatsink
600
temperature (figure 2):
115
Ic [A]
Tj [C]
200

110

Base-less Modules

105

0
0.060

0.065

0.070

0.075

0.080

0.085

0.090

0.095

Switch 1

Switch n

100
0.100
PIGBT

t [s]

PDiode
ZthJH

Psmoothed [W]
Pav [W]
Pmod [W]
Tjmod [C]
Tjav [C]

1600
1400
1200

PDiode
ZthJH

ZthJH

125
Reference: Th

120
ZthHA

1000

115
Tj

PIGBT
ZthJH

800
110

600
400

Figure 2: thermal equivalent block diagram for base-less

devices

105

200
0
0.060

0.065

0.070

0.075

0.080

0.085

0.090

0.095

100
0.100

t [s]

Figure 1: junction temperature as a function of fO

The static calculation of the temperature rise in

the heatsink can be calculate with the thermal
resistance of the cooler and the number of
dissipating heat sources mounted on it:

The maximum junction temperature Tvj max in

(9)
function of the phase output current frequency fO TH = T A + ( PIGBT + PDiode ) n S RthHA
can be calculated if the transient thermal
resistance is known:
With nS as the number of switches mounted on
1
the cooler.

Tvj max = 2 PAV Ri

1 e

2 f O i

+ Tref

For devices with a base-plate the reference is the

case temperature. The thermal characteristic of
1 e
devices with a base-plate is normally specified
with a ZthJC (thermal impedance junction case)
With Tref as the reference heatsink temperature Th for the IGBT and Diode part. Additionally the
for base-less modules or case temperature Tc for interface resistance case-heatsink RthCH is given
based modules.
for the whole module. Figure 3 shows the thermal
equivalent block diagram for modules with baseplate:
Heatsink and Transient Calculations:
In addition to the module calculation the
simulation-tool allows the user to calculate the
Based Modules
temperature rises and module losses of the
IGBT-module together with a specified cooler.
The simulation-tool uses a simplified approach to
calculate temperature rises and cross-talk effects.
This has the advantage that the calculations can
be done with the given values in the module- and
heatsink data-sheets.
More accurate methods based on finite element
calculations require specific knowledge about
module and cooler construction and need much
more time for parameter extraction as well as for
the calculations. On the other hand the simplified
method of the simulation-tool allows quick and Figure 3: thermal equivalent block diagram for devices with
accurate simulation without detailed knowledge base-plate
on cooler and module construction.
i =1

1
f O i

(8)

Switch 1

PIGBT

Switch n

PDiode

ZthJC

PIGBT

ZthJC

PDiode

ZthJC

ZthJC

Reference: Tc

RthCH

ZthHA

PCIM'04 Power Electronics Conference, Nuremberg

revised 01.07.05

Page 3 of 6

As in case of base-less modules the static

Tstart
Z th (t ) + P(T ) Z th (t )
(12)
calculation of the temperature rise in the heatsink T (t ) =
Rth
can be calculate with the thermal resistance of the
cooler and the number of dissipating heat sources
mounted on it:
In order to include the starting conditions for T
the first term is introduced. Tstart/Rth describes
TH = T A + ( PIGBT + PDiode ) nS RthHA
(9b) the initial constant power that resulted in Tstart.
This deposited power influences the thermal
Additionally the temperature rise in the interface behaviour until Zth(t)=Rth(t).
TCH needs to be calculated. Since it is possible The second term describes the heating with the
more than one module is mounted on the heatsink temperature dependant power P(T). Since T is as
and that the heat sources are distributed in well a function of the dissipated power and the
several modules, as for example in a three phase thermal impedance iterations are necessary to get
inverter built with three halve-bridge modules an accurate result.
mounted on a single cooler, it becomes necessary For the transient heatsink temperature we can
write:
to scale the interface resistance accordingly:

TCH = ( PIGBT

R
+ PDiode ) n S thCH
nM

TH (t ) = TA
(10)

With nM as the number of modules mounted on a

single heatsink.
In order to calculate the temperature dependent
semiconductor losses the simulation-tool adjusts
the junction temperature and the corresponding
losses in several iterations. Depending on the
module type and the calculation (with / without
heatsink) the calculation starts with the initial
conditions Tvj = TA, Tvj = TH or Tvj = TC .
Transient Overload Calculation:
The simulation-tool offers the possibility to
additionally calculate the transient thermal
behaviour of the IGBT module and the heatsink.
For the transient calculation overload conditions
can be specified. In order to have full flexibility the
start conditions of the temperature excursion
calculation need to be specified. This includes the
overload duration, the start junction temperature
and the start case or heatsink temperature.
Therefore it is possible to study the transient
thermal behaviour from any temperature level and
even the thermal recovery behaviour from an
overload can be computed.
The transient temperature rise can be calculated
with the thermal impedance.

THstart TA
Z thHA (t ) + P(T ) Z thHA (t )
RthHA
(12a)

whereas P(T) depends on the number of switches

mounted on the heatsink

P(T ) = ( PIGBT (T ) + PDiode (T )) n s

(12b)

For the transient average junction temperature we

can write:

Tvj av(t) = TH (t)

Tvj start TH start

RthJH

ZthJH(t) + P(T) ZthJH(t)

(13)

In case of modules with a base plate TH has to be

replaced with TC and RthJH and ZthJH have to be
replaced with RthJC and ZthJC respectively. In
addition the temperature drop across the interface
TCH has to be calculated. Since the interface has
no heat capacity this can be done similar to the
static calculations (10).
Therefore we are able to calculate the transient
average junction temperature. As already
mentioned the average junction temperature
yields in too optimistic value if the temperature
ripple in function of the output current needs to be
considered.
To avoid a too high complexity the simulation-tool
n
uses a simplified approach that is valid in most
t / i
(11) relevant application cases.
Z th (t ) = Ri (1 e
)
i =1
The approach is to calculate the "overshoot
temperature" (Tov) due to the output frequency
The temperature rise T(t) is a function of Zth(t) ripple at the end of the overload pulse and add it
and the temperature dependent power dissipation to the calculated average temperature:
in the semiconductor P(T).

Tvj max (t ) = Tvj av (t ) + Tov

PCIM'04 Power Electronics Conference, Nuremberg

revised 01.07.05

(14)

Page 4 of 6

1
The same modules have been simulated as well

2 fOi
n

with forced air-cooled heatsink with a thermal

1 e
(14a) resistance heatsink to ambient RthHA=24K/kW.
(
)
Tov = 2 P(T) Ri
+
T

P
T

ref
th
1
i=1

This represents the typical cooling condition for

fOi
1

modules of that size. For this case the simulation

Iout, rms [A]

was carried out with a fixed ambient temperature

This is valid as long as the Zth(t) of the module is of TA=40C (figure 6).
close to its Rth value. For most modules this is
Tvj=125C, Tc=100C, VDC=900V, cos=0.85, m=1,
the case between 0.5 and 1s. Below this time
fout=50Hz
3000
span the values for Tvj max(t) are slightly optimistic.
Inom=3600A@Tc=80C
Therefore the simulation-tool does not allow
2500
Inom=2400A@Tc=80C
transient calculations with duration below 1s.
Inom=1800A@Tc=80C
2000
In figure 4 an example output of a transient
simulation is shown.
Temperature Excursion During Overload 5SNA 1200E330100
120
100

1500
1000
500

T [C]

80

60

40
Tj IGBT [C]
Tj Diode [C]
Tc [C]
Th [C]

20
0
0

10

15

20

25

30

35

1000

2000

3000
fsw [Hz]

4000

5000

6000

Figure 5: Output current versus switching frequency, fixed

case-temperature Tc=100C

t overload [s]

Such transient/overload simulations are especially

useful for calculating temperature cycles and
predicting the useful life of the module.

Benchmarking:
The Simulation-tool allows for two types of
calculations. One of them is with a fixed
heatsink/case temperature and the other with a
fixed ambient temperature, which allows the user
to include the heatsink into the calculation.
The calculation with a fixed heatsink/case
temperature is a quick method that does not
require specifying or selecting a heatsink. It is
suitable for comparisons of IGBTs with exactly the
same package. Nevertheless if devices with
different packages have to be compared or if the
results have to be more application relevant the
method including the heatsink in the calculation
should be used.
Figure 5 shows the simulation results for three
modules in the same package all with the nominal
current rated at Tc=80C. The diagram shows the
maximum rms-output current versus the switching
frequency if the junction temperature is limited to
125C with a given case-temperature of
Tc=100C.

If we compare figure 5 and figure 6 we can

observe quite a difference. The assumption of
Tc=100C is too optimistic for air-cooling,
therefore the output-currents are lower for the
modules simulated with the cooler. In addition one
can observe that the difference between the
modules will be significantly minimised if
calculated with a cooler. As a matter of fact the
heatsink dominates the thermal behaviour since
its Rth is significantly higher than the one of the
modules (table 1). Additionally the power
dissipated by the diode has to flow as well
through the cooler and contributes to the
temperature rise THA in the heatsink. The
calculation with a fixed case-temperature does not
consider this and neglects any cross-talk effects
between diode and IGBT.
Tvj=125C, Ta=40C, VDC=900V, cos=0.85, m=1,
fout=50Hz, RthHA=24K/kW
2500
Inom=3600A@Tc=80C
Inom=2400A@Tc=80C

2000

Inom=1800A@Tc=80C
Iout, rms [A]

Figure 4: Output current versus switching frequency, fixed

case-temperature Tc=100C

1500

1000

500

0
0

1000

2000

3000
fsw [Hz]

4000

5000

6000

Figure 6: Output current versus switching frequency. Heatsink

RthHA=24K/kW, fixed ambient temperature TA=40C.

PCIM'04 Power Electronics Conference, Nuremberg

revised 01.07.05

Page 5 of 6

Module

IGBT

Diode

RthJC [K/kW]

RthJC [K/kW]

9.0
7.0
7.0

17
12
16

Inom=1800A
Inom=2400A
Inom=3600A

Table 1: RthJC values for the modules used in the simulation

As a third simulation study we calculated the

same modules with a high-performance watercooled heatsink with RthHA=10K/kW. Figure 7
Shows the result of this simulation. Obviously the
different modules are more distinguishable if the
Rth of the cooler is lower.
Tvj=125C, Ta=40C, VDC=900V, cos=0.85, m=1,
fout=50Hz, RthHA=10K/kW
3000.0
Inom=3600A@Tc=80C
Inom=2400A@Tc=80C

2500.0
Iout, rms [A]

Inom=1800A@Tc=80C
2000.0

In order to offer a simple and realistic method to

benchmark IGBT modules we have created the
simulation-tool that allows the calculation of
temperature rises in the module including the
conduction and switching losses with their
temperature dependence. In addition the
simulation-tool is capable to compute static and
transient temperature rises of the module together
with a free definable heatsink.
We have presented the calculation methods
employed in the simulation-tool and showed
calculation examples for different modules.
This example study shows that for proper
application relevant benchmarking the right
heatsink has to be included in the simulation and
that a simulation only with a fixed case/heatsink
temperature leads to inaccurate results.
Thanks to the simulation-tool such calculations
can be done quickly and accurate.

1500.0

References:
1000.0

1.

D.Srajber, W.Lukasch, "The calculation of the power

dissipation for the IGBT and the inverse diode in circuits
with sinusoidal output voltage", electronica'92, Proc, pp.
51-58

2.

R.Schnell, U.Schlapbach, K.Haas, G.Debled, "Parallel

Operation of LoPak Modules", Proc. PCIM'03, Nrnberg
2003.

3.

M.Rahimo, A.Kopta, R.Schnell, U.Schlapbach,

R.Zehringer, S.Linder, "High-Power Modules with SoftPunch Through IGBT Technology for Traction
Application", Proc. PCIM'03, Nrnberg 2003.

500.0
0.0
0

1000

2000

3000
fsw [Hz]

4000

5000

6000

Figure 7: Output current versus switching frequency. Heatsink

RthHA=10K/kW, fixed ambient temperature TA=40C.

Comparing all three figures we additionally

observe that even with the best cooling the output
current hardly reaches the current rating of the
module even at very low switching frequency. 4.
Especially for modules with higher current rating
the deviation of rms output-current and module
nominal current becomes significant.
Analysing the three cases in more detail we can 5.
observe that the achievable output-current of the
3600A rated module becomes lower at higher
switching frequencies than the output current of
the 2400A module. This crossing point is because
the 3600A module is optimised for low on-state 1.
losses but features higher switching losses than
the 2400A module. That allows the module to be
rated at 3600A. The 2400A and 1800A rated
modules are from the same technology.
As a further observation from the three studies we
can conclude that for optimum utilisation of higher
rated modules a good cooling is necessary.

M.Rahimo, A.Kopta, R.Schnell, G.Debled, J.de Oate,

S.Linder, "New Industry Standard LoPak Modules with
1700V SPT-IGBT and Diode chip set", Proc. PCIM'02,
Nrnberg 2002.
M.Rahimo, W.Lukasch, C. von Arx, A.Kopta, R.Schnell,
S.Dewar, S.Linder Novel Soft-Punch-Through (SPT)
1700V IGBT Sets Benchmark on Technology Curve,
Proc. PCIM01, Nrnberg 2001.

Acknowledges:
Dr. J.Waldmeyer for providing simulation results using the
convolution method (fig. 1)

Conclusions:
The comparison of IGBTs with the nominal
current is inaccurate and can lead to wrong
results. Due to thermal limitations it is in most
cases not possible to reach an inverter outputcurrent as high as the module nominal current in a
realistic application environment.

PCIM'04 Power Electronics Conference, Nuremberg

revised 01.07.05

Page 6 of 6