See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/228644743

Modelling the thermal coupling between internal power semiconductor dies

of a water-cooled 3300V/1200A HiPak IGBT module

Article · January 2007

CITATIONS READS

31 2,593

5 authors, including:

Uwe Drofenik Andreas Müsing

ABB ETH Zurich
106 PUBLICATIONS   3,227 CITATIONS    29 PUBLICATIONS   701 CITATIONS   

SEE PROFILE SEE PROFILE

Johann Walter Kolar

ETH Zurich
936 PUBLICATIONS   42,681 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Rotary-Linear Actuators with Magnetic Bearings View project

Multi Gap Core Losses View project

All content following this page was uploaded by Johann Walter Kolar on 25 February 2014.

The user has requested enhancement of the downloaded file.

Modelling the Thermal Coupling between Internal Power Semiconductor Dies of
a Water-Cooled 3300V/1200A HiPak IGBT Module
Uwe DROFENIK*, Didier COTTET**, Andreas MÜSING*, Jean-Marc MEYER*** and Johann W. KOLAR*
*
Power Electronic Systems Laboratory, ETH Zurich, ETH-Zentrum / ETL H13, CH-8092 Zurich, Switzerland
Phone: +41-1-632-4267, Fax: +41-1-632-1212, E-mail: [email protected]
**
ABB Switzerland Ltd, Corporate Research, CH-5405 Baden-Dättwil, Switzerland
***
ABB Switzerland Ltd, CH-5300 Turgi, Switzerland

Abstract – Calculating the transient junction temperature of material properties is needed. Single dies inside the power
power semiconductors is important for analysing converter module that are connected in parallel via wire-bonding,
reliability or investigating short-term overload conditions. A cannot be analyzed separately. For reliability analysis it is
dynamic thermal model, which includes the mutual thermal often desirable to know internal temperatures at layer
coupling of neighbouring dies and permits easy integration
into a circuit simulator, is essential to perform such a task.
interfaces in order to evaluate the thermo-mechanical
In this paper, the thermal modelling procedure for a stresses at these critical points ([1], [2], [3]). If the
3300V/1200A IGBT module based on numerical simulations internal geometry and the material properties of the power
and infrared temperature measurements is presented. module are known, the thermal step responses of the
individual dies can be calculated numerically via 3D-
I. INTRODUCTION FEM based on a three-dimensional model of the power
In order to analyze the reliability of a converter, module. One big advantage of the numerical method is
knowledge of the transient junction temperature of the the accessibility of transient temperatures of all internal
power semiconductors during operation is essential, points of interest, especially layer interfaces and junctions.
especially maximum temperature values and temperature In this paper a dynamic thermal equivalent circuit model
amplitudes. Transient temperature rise during short term of a power module suitable for embedding into a circuit
converter overloads is also an important issue that simulator is presented. The equivalent circuit model is
requires dynamic thermal models for the semiconductor based on thermal step responses simulated via 3D-FEM.
devices. The ability to simulate transient semiconductor Therefore, it is very important to define a reliable 3D-
temperatures with a circuit simulator will become FEM model of the power module which is the main scope
increasingly important in a future virtual design of this paper. In section (II) the experimental setup for
environment, where as much of the converter as possible stationary measuring junction temperatures of dies inside
is designed, simulated and optimized on a computer a 3.3kV/1.2kA ABB HiPak IGBT module using infrared
before building a prototype. is described. The measured temperatures are compared to
Directly coupling of a thermal three-dimensional finite a 3D-FEM model that not only includes the power
element (3D-FEM) simulation with a circuit simulation module but also the water-cooled heat sink. The
will result in inadmissible long simulation times because measurements confirm the proposed 3D-FEM model with
of the huge matrix equation that has to be solved for very good accuracy. Details of the setup of the 3D-FEM
three-dimensional field problems at repeating time steps model are discussed in section (III). In section (IV) a
of the circuit simulation. Another problem that contributes general procedure for parameterizing a dynamic thermal
to the huge computational effort is that the time constants equivalent circuit model that describes the thermal
in the circuit simulation are typically defined by switching coupling between dies inside the power module is
frequencies and are therefore in the microsecond-range, discussed. The procedure is based on the simulated
while thermal time constants range from seconds for transient thermal step responses.
power semiconductors up to many minutes for heat sinks.
A solution of this well-known problem is to extract a
thermal equivalent circuit from the three-dimensional
thermal model of the power module mounted onto the
heat sink, which can be directly embedded into the circuit
simulator.
A dynamic thermal equivalent circuit model suited for Fig.1. 3300V/1200A ABB HiPak IGBT Module [4].
transient simulations is typically derived from the thermal
step response of the power semiconductor which can be II. TEMPERATURE MEASUREMENTS
measured or numerically simulated. If the thermal step
response is measured by employing a test current and A. Experimental Setup
measuring the semiconductor voltage drop as a function In this section temperature measurements are presented
of the junction temperature, it is not necessary to open the and compared to results from a 3D-FEM simulation that
power module or manipulate it otherwise. No exact will be discussed in detail in section (III). As case study in
knowledge of layer thickness, internal module design and this paper a water-cooled 3300V/1200A HiPak IGBT
module [4] has been chosen. The ABB HiPak IGBT image temperature. The reference procedure was repeated
modules have been presented in [5] showing excellent in temperature steps of 10°C steps over the range
electrical performances, capable of withstanding extreme [20°C…150°C]. Finally, the thermal images of the power
conditions during turn-off and short circuit operation. module were obtained under operation. Adjusting the
These modules aim at wide SOA applications operating module measurements with an image processing software
under hard switching conditions such as traction and employing the reference pictures, an estimated absolute
industrial drives. The effort undertaken to obtain such accuracy of 2°C is possible. Due to the reference image
performance at semiconductor level has as a consequence technique, the relative spatial accuracy is estimated to be
also increased the demands to the package design. even more accurate than 1°C. The experimental setup is
Extensive electrical and thermal simulations have been shown in Fig.2.
performed to avoid having the package as performance
bottleneck. In order to guarantee high converter system B. Heating of Selected IGBTs and Diodes
reliability, transient thermal simulations are planned to be
employed in future design phases. The power module,
shown in Fig.1 employs 36 internal dies, with 24 IGBTs
in parallel forming the switch and 12 diodes in parallel s11 s21 m1 s31 s41 m9 s51 s61

AlN-plate
AlN-plate
AlN-plate
forming the anti-parallel diode.
s12 s22 s32 s42 s52 s62
m2 m8
d11 d21 d31 d41 d51 d61
m5
m3 m4 m7
m6
d12 d22 d32 d42 d52 d62

AlN-plate

AlN-plate
AlN-plate
s13 s23 s33 s43 s53 s63

s14 s24 s34 s44 s54 s64

AlSiC-basplate

Fig.3. Internal layout of the 3300V/1200A ABB HiPak IGBT module.

Here, 24 dies (labeled ‘s’) are connected in parallel forming one switch,
and the remaining 12 dies (labeled ‘d’) connected in parallel form the
anti-parallel diode. All measurements referring to ‘s’- or ‘d’-points in
this paper are taken at the geometric center of the according die. Nine
Fig.2. Experimental Setup: The housing of the power module has been points (labeled ‘m’) located directly on the AlSiC-base plate were
removed. All surfaces are painted with a black coating for the infrared chosen for additional temperature measurements.
temperature measurements.

For the thermal measurements an open power module was

used. The power module was attached via a thermal
interface (thermal grease) to an aluminum plate
employing water cooling. The water temperature at the
inlet of the pipes was 20°C, the water flow through the
plate was set to 23.3 liter/minute.
The camera system used to obtain the temperature
pictures was a "Thermosensorik CMT 384M" infrared
camera with a 284x288 pixel detector matrix and a
dynamic range of 14 bits, which is sensitive in the
wavelength range of λ=3-5μm. The detector pixel value is
a function of temperature, spectral camera sensitivity and
object emissivity. In order to improve the absolute
accuracy of the thermal images, the measurements were Fig.4. Stationary temperature distribution for heating all 24 IGBTs. The
obtained in the following way: The module surface was IGBTs s11, s12, s13, s14, s21, s22, s23, s24 connected in parallel
painted with a black coating (3M Nextel Black Paint equally share a thermal power of 1430W (179W each), the IGBTs s31,
Velvet) which has - especially in the infrared range - an s32, s33, s34, s41, s42, s43, s44 connected in parallel equally share a
thermal power of 1290W (161W each), the IGBTs s51, s52, s53, s54,
emissivity close to one. This reduces measurement errors s61, s62, s63, s64 connected in parallel equally share a thermal power of
due to (metal) reflexions and varying surface properties of 1760W (220W each). The total thermal power dissipated by the module
the power module surface. For a calibration of the is 4480W in this experiment.
absolute temperatures, the power module was heated to a
spatial constant, homogeneous temperature. This results Temperatures are measured at geometric centers of dies as
in nearly uniform reference images, containing as defined in Fig.3. Additional points of interest are defined
information only the pixel to pixel sensitivity variations, and labeled ‘m1’, ‘m2’, … ‘m9’as shown in Fig.3. By
imperfections of the black coating and the reference removing certain wire bond connections it was possible to
heat selected semiconductors of the power module. By Pts. T[°C] pts. T[°C] pts. T[°C] pts. T[°C]
measuring voltage and current, the thermal power of s11 23 s21 24 s31 24 s41 24
s12 23 s22 25 s32 26 s42 25
groups of semiconductors connected in parallel could be d11 24 d21 26 d31 28 d41 28
easily calculated. All losses are conduction losses. The d12 25 d22 32 d32 41 d42 40
measurements were taken after a stationary temperature s13 23 s23 36 s33 100 s43 99
has been reached. Stationary temperature distributions s14 27 s24 38 s34 105 s44 102
measured with the infrared camera for different pts. T[°C] pts. T[°C] pts. T[°C] pts. T[°C]
configurations are given in Fig.4, Fig.5, Fig.6 and Fig.7. s51 24 s61 23 m1 24 m6 30
The temperatures measured at points defined in Fig.3 are s52 24 s62 22 m2 27 m7 28
compared to 3D-FEM simulations (see section (III)) in d51 26 d61 24 m3 31 m8 27
d52 30 d62 25 m4 32 m9 24
Tab.1 – Tab.8. In the tables, the heated semiconductors s53 34 s63 25 m5 33
are highlighted. s54 35 s64 26
Tab.3. Stationary temperatures measured via infrared as given in Fig.5.
Pts. T[°C] pts. T[°C] pts. T[°C] pts. T[°C]
s11 114 s21 126 s31 128 s41 127 pts. T[°C] pts. T[°C] pts. T[°C] pts. T[°C]
s12 - s22 124 s32 129 s42 124 s11 21 s21 22 s31 22 s41 22
d11 64 d21 68 d31 66 d41 66 s12 21 s22 23 s32 24 s42 24
d12 63 d22 68 d32 66 d42 65 d11 22 d21 25 d31 27 d41 27
s13 - s23 129 s33 129 s43 129 d12 23 d22 31 d32 44 d42 43
s14 122 s24 131 s34 134 s44 133 s13 23 s23 35 s33 105 s43 105
s14 23 s24 35 s34 107 s44 107
pts. T[°C] pts. T[°C] pts. T[°C] pts. T[°C]
s51 138 s61 133 m1 92 m6 60 pts. T[°C] pts. T[°C] pts. T[°C] pts. T[°C]
s52 141 s62 135 m2 66 m7 62 s51 22 s61 21 m1 23 m6 30
d51 71 d61 78 m3 60 m8 66 s52 23 s62 21 m2 26 m7 27
d52 68 d62 72 m4 60 m9 94 d51 25 d61 22 m3 30 m8 26
s53 137 s63 133 m5 60 d52 31 d62 23 m4 32 m9 23
s54 138 s64 132 s53 35 s63 23 m5 33
Tab.1. Stationary temperatures measured via infrared as given in Fig.4. s54 35 s64 23
Tab.4. Stationary temperatures derived via 3D-FEM simulation for the
Pts. T[°C] pts. T[°C] pts. T[°C] pts. T[°C] heat distribution as given in Fig.5.
s11 115 s21 122 s31 117 s41 120
s12 116 s22 124 s32 118 s42 121
d11 58 d21 66 d31 66 d41 69
d12 59 d22 66 d32 66 d42 69
s13 117 s23 124 s33 118 s43 121
s14 116 s24 123 s34 117 s44 120
pts. T[°C] pts. T[°C] pts. T[°C] pts. T[°C]
s51 141 s61 136 m1 86 m6 64
s52 143 s62 136 m2 65 m7 64
d51 72 d61 66 m3 60 m8 70
d52 71 d62 66 m4 60 m9 93
s53 143 s63 136 m5 59
s54 142 s64 135
Tab.2. Stationary temperatures derived via 3D-FEM simulation for the
heat distribution as given in Fig.4.

Fig.6. Stationary temperature distribution for heating one selected IGBT.

The IGBT s32 dissipates a thermal power of 168W.

pts. T[°C] pts. T[°C] pts. T[°C] pts. T[°C]

s11 23 s21 27 s31 32 s41 24
s12 22 s22 28 s32 73 s42 24
d11 23 d21 26 d31 31 d41 24
d12 22 d22 24 d32 24 d42 23
s13 - s23 23 s33 23 s43 23
s14 22 s24 23 s34 23 s44 22
pts. T[°C] pts. T[°C] pts. T[°C] pts. T[°C]
s51 23 s61 22 m1 33 m6 23
s52 23 s62 - m2 29 m7 23
d51 23 d61 22 m3 25 m8 23
d52 22 d62 22 m4 26 m9 23
Fig.5. Stationary temperature distribution for heating four selected s53 22 s63 22 m5 25
IGBTs. The IGBTs s33, s34, s43, s44 connected in parallel equally share s54 22 s64 22
a thermal power of 750W (188W each). Tab.5. Stationary temperatures measured via infrared as given in Fig.6.
 pts. T[°C] pts. T[°C] pts. T[°C] pts. T[°C] III. 3D-FEM THERMAL SIMULATION
s11 21 s21 25 s31 31 s41 22
s12 21 s22 26 s32 77 s42 23 A. General Considerations
d11 21 d21 25 d31 31 d41 22
d12 21 d22 22 d32 23 d42 21 If detailed stationary temperature distributions are
s13 20 s23 21 s33 21 s43 21 available as presented in section (II), it would be not too
s14 20 s24 21 s34 21 s44 21
difficult to tune material properties and/or geometries like
pts. T[°C] pts. T[°C] pts. T[°C] pts. T[°C] layer thickness in a way that the 3D-FEM simulations
s51 21 s61 20 m1 32 m6 21 match the measured temperatures in excellent agreement.
s52 21 s62 20 m2 28 m7 21 But typically, such temperature distributions are not
d51 21 d61 20 m3 24 m8 21
d52 21 d62 20 m4 25 m9 22 available due to the high experimental effort. Therefore,
s53 20 s63 20 m5 24 when building the 3D-FEM model some material
s54 20 s64 20 properties might not be exactly known, or geometries
Tab.6. Stationary temperatures derived via 3D-FEM simulation for the might be simplified in order to reduce the computational
heat distribution as given in Fig.6.
effort. In this section we investigate how to generally set
up a reliable 3D-FEM model for power modules, and we
discuss its sensitivity concerning changes of certain
parameters and/or simplifications. In this study the model
always includes the water-cooled heat sink plate. For a
different heat sink the model would be different, but could
be easily calculated with the procedure discussed in the
following. All thermal simulations are performed with the
fluid-thermal 3D-FEM simulation software ICEPAK [6].

Fig.7. Stationary temperature distribution for heating eight selected

diodes. The diodes d31, d32, d41, d42 connected in parallel equally
share a thermal power of 763W (191W each), the diodes d51, d52, d61,
d62 connected in parallel equally share a thermal power of 985W (246
each). The total thermal power dissipated by the module is 1748W in
this experiment.
pts. T[°C] pts. T[°C] pts. T[°C] pts. T[°C]
s11 25 s21 30 s31 33 s41 38
s12 25 s22 32 s32 42 s42 50 (a)
d11 26 d21 35 d31 89 d41 102
d12 26 d22 36 d32 89 d42 109
s13 30 s23 33 s33 43 s43 51
s14 26 s24 31 s34 35 s44 40
pts. T[°C] pts. T[°C] pts. T[°C] pts. T[°C]
s51 40 s61 39 m1 33 m6 69
s52 55 s62 - m2 47 m7 80 (b) (c)
d51 128 d61 130 m3 45 m8 71
Fig.8. (a) 3D-FEM simulation of the temperature distribution at the
d52 122 d62 130 m4 55 m9 44
surface of the Al-cooling plate directly below the power module for heat
s53 54 s63 55 m5 52
dissipation as defined in Fig.5. The positions of the IGBTs and diodes
s54 40 s64 38
are shown (see Fig.3). (b) 3D-view of the Al-cooling plate with a
Tab.7. Stationary temperatures measured via infrared as given in Fig.7. temperature distribution according to a heat dissipation as defined in
pts. T[°C] pts. T[°C] pts. T[°C] pts. T[°C] Fig.5. View from top, which is the side directly below the power module,
s11 22 s21 27 s31 32 s41 38 and (c) view from bottom, which is the side in contact with the water
s12 23 s22 31 s32 44 s42 54 (convective cooling).
d11 23 d21 35 d31 103 d41 117
d12 23 d22 34 d32 103 d42 117 B. Modeling the Water-Cooled Heat Sink
s13 23 s23 31 s33 44 s43 54
s14 22 s24 27 s34 33 s44 38 The heat is removed from the water-cooled aluminum
plate via convection. Generally, for calculating
pts. T[°C] pts. T[°C] pts. T[°C] pts. T[°C]
s51 40 s61 37 m1 32 m6 81
convection, computational fluid dynamics (CFD) software
s52 58 s62 54 m2 48 m7 88 is necessary. This is a kind of 3D-FEM software, where in
d51 137 d61 132 m3 49 m8 81 addition to the single heat-conduction equation another
d52 137 d62 132 m4 65 m9 46 four scalar equations (Navier-Stokes) describing fluid
s53 58 s63 54 m5 58
s54 40 s64 37
flow have to be linearized within each element. In
Tab.8. Stationary temperatures derived via 3D-FEM simulation for the addition to the detailed geometric layer structures within
heat distribution as given in Fig.7. the power module resulting in a very large mesh, the
matrix equation to be solved by the CFD software 0.1s the cooling plate temperature underneath s31 and d31
becomes very large, which often causes convergence (which are closest to s32) will start to rise as shown in
problems and numerical instabilities for such kind of Fig.9. As the hot spot grows larger, more neighbor dies
problems. Another problem is that the pipes inside the will experience temperature rise. There are additional heat
cooling-plate employ very fine three-dimensional flow paths, e.g. horizontally through the AlN-plate
structures acting as turbulence promoters which are not (λ=180W/Km-1, d=1mm) with s32, s31, d31 and s42
defined geometrically. Therefore, the flow through these being attached to it. If this path would be dominating,
pipes cannot be simulated with acceptable accuracy. there would be no significant time delay between the start
To avoid these problems arising from convective cooling, of the temperature rise between, e.g., s31 and s42. But as
the water-cooled plate was modeled for all 3D-FEM shown in Fig.9, the actual time delay between s31 and s42
simulations as an aluminum plate with a defined thickness. is very large (0.1s versus 1s). Besides the aluminum
The bottom side of this plate is defined as boundary of the cooling plate, the AlSiC base plate (see following section)
3D-FEM model by setting a homogenous heat transfer is a major contributor to heat spreading and, therefore,
coefficient h [W/K m-2] which describes with good thermal coupling of the internal semiconductors.
accuracy convective heat transfer. Now, only the heat
transfer equation has to be considered within each C. Parameters of the 3D-FEM Model
element (no CFD necessary), and the resulting matrix The three-dimensional structure of the power module
equation is significantly reduced, which results in modeled with the 3D-FEM software (ICEPAK) is shown
comparable small simulation times, avoidance of in Fig.10. The interface layer between the Al-cooling
numerical instabilities, and higher accuracy of the plate and the AlSiC-base plate is modeled as a
simulation results. The effect of convective cooling via homogenous layer of thermal grease with thermal
water is still described with high accuracy. conductivity 1.0W/Km-1 and thickness 50μm. In order to
It is important to model the cooling plate as a plate with a simplify the 3D-model, a solder layer (λ=50W/Km-1 [6],
certain thickness [7], because the cooling plates acts as a d=200μm) and a copper layer (λ=380W/Km-1, d=300μm)
heat spreader and contributes significantly to thermal underneath the AlN-plate (λ=175W/Km-1, d=1mm) are
coupling between the semiconductors inside the power represented by a single simplified plate (“solder & copper
module. This can be clearly seen in Fig.8(a) which shows & AlN-plate” in Fig.10) with a thickness dAlN* = 1.5mm
the simulated temperature distribution on the surface of and an equivalent thermal conductivity
the cooling plate directly below the power module. If one d AlN * d + dCu + d AlN 200 + 300 + 1000
λAlN * = = Solder = ≈ 140 Km
W

would set the boundary condition to be of constant ∑i i i

d λ d Solder
+
d Cu
+
d AlN 200
+
300 1000
+
temperature, the simulation results would show errors, λSolder λCu λ AlN 50 380 175
especially significant underestimation of thermal coupling. (1)
Also from the viewpoint of dynamic modeling, it is Also the Si-layer (λ=180W/Km-1, dIGBT =500μm / dDiode
essential to define the cooling plate with the correct =450μm) and the solder (λ=380W/Km-1, d=70μm)
thickness because it represents the largest mass and, underneath are represented by a simplified equivalent
therefore, has a major effect on the thermal time constants plate (“solder & silicon” in Fig.10) with dIGBT* =570μm /
([7]). dDiode* =520μm and λIGBT ≈ λDiode ≈ 130W/Km-1.
100
90
solder & silicon
80
s32 (λ=130W/K.m-1, dIGBT=570μm / dIGBT=520μm)
70 2D - heat source
60 copper (λ=388W/K.m-1, d=300μm)
50
T [°C]

s21,d21 solder & copper & AlN-plate

40 (λ=140W/K.m-1, d=1.5mm)
AlSiC-plate
30 s31,d31 (λ=165W/K.m-1, d=5mm)
s22
grease (λ=1.0W/K.m-1, d=50μm)
s42
20 Al - cooling plate
(λ=205W/K.m-1, d=10mm)
0.001 0.01 0.1 1 10 100 1000 cooling plate bottom:
t [s] heat transfer coefficient (h = 4400W/K.m2)
Fig.9. Thermal step responses for heating IGBT s32. Fig.10. Structure of the layers showing the layer thicknesses and
material properties as employed in the 3D-FEM simulation.

Thermal step responses for heating one single IGBT s32

with 168W (see Fig.6) are shown in Fig.9. The If the thermal resistance of the heat sink Rth,HS would be
temperature of the heated IGBT s32 starts to rise after given, the heat transfer coefficient h used to define the
boundary condition could be simply calculated as
approximately 100μs. Here, the smallest time constants
h = ( Rth , HS ⋅ AHS )
−1
are defined by the masses of the silicon die and the (2)
attached solder layer. A hot spot grows underneath s32 at with heat sink surface AHS (which is covered by the test-
the surface of the aluminum cooling plate, and after about heat source for Rth-measurement). In this study Rth,HS was
 150
not known. The heat transfer coefficient was derived from s52
s51 150 s52
s51
s62
temperature measurement at the power module center 140 140
point m5 in the scenario where all IGBTs are heated 130 130

T [°C]
s22

T [°C]
s22
(Fig.4). The reason for the selection of this point is that it 120
s61
120
s62 s61
is located directly on the surface of the AlSiC-plate. As 110 110
one can see from Fig.10, below the AlSiC-plate there is s11 s12 s11 s12
100 100
the thermal interface layer (grease) and then the Al- 100 120 140 160 180 20 40 60 80
cooling plate. Heat flow through these three layers is λAlN* [K/W m-1] λSolder [K/W m-1]
much better defined than, e.g. heat flow through the (a) (c)
semiconductor center with six additional layers involved 76
d51 76 d51
(solder – Cu – AlN – Cu – solder – Si), where each layer 72 72
d41
might contributed small errors and/or uncertainity 68
d41
68

T [°C]
concerning material properties and layer geometry

T [°C]
64 d21
(thickness) which might distort and/or obfuscate the d31,d61 64 d31,d61
d21
d11 d11
measurements. From the temperature at the AlSiC-plate 60 60
center point m5 as given in Tab.1 the heat transfer 56 56
coefficient was calculated according to (2) as 100 120 140 160 180 20 40 60 80
−1 −1
λAlN* [K/W m-1] λSolder [K/W m-1]
⎛ ΔT ⎞ ⎛ 60 − 20 ⎞ (b) (d)
h = ⎜ m 5 ⋅ AHS , AlSiC − plate ⎟ = ⎜ ⋅ (0.186 ⋅ 0.138) ⎟ = 4400 Wm
K
⎜P ⎟ ⎝ 4480 ⎠
⎝ V ,total ⎠ Fig.12. All IGBTs are heated as defined in Fig.4. Curves show the
(3) dependency of the temperature at certain measurement points (see Fig.3)
on a variation of the thermal conductivity (a), (b) of the AlN*-plate, and
One key assumption for building a thermal equivalent (c), (d) of the solder employed between layers of the power module.
circuit model for embedding into a circuit simulator is
that the thermal material properties are temperature- If the thermal conductivity of the solder is varying, the
independent. Strictly speaking, this assumption is violated IGBT-temperatures change according to Fig.12(c). For an
in most cases, but for many important materials like Cu, increase of λSolder from 40K/Wm-1 to 60W/Km-1, the
Al or Sn the thermal conductivity can be assumed to be
temperature decrease will be quite small (ΔT ≈ 3..4°C).
approximately constant over a wide temperature range as
The temperatures measured at the diode center points will
shown in Fig.11(a). The thermal conductivities of Si, AlN
not change in this case (Fig.12(d)). The reason is, that
and AlSiC are shown in Fig.11(b). Over a temperature
because only the IGBTs are heated, the temperature rise
range [-50°C … +250°C] they cannot be assumed to be
measured at the diode center points is caused not by heat
temperature-independent.
flow through the diode layers (including the solder layers),
500 350 but by the general rise of the temperature level at the
400
Cu
300
AlN cooling plate surface and/or AlSiC-plate due to the hot
spot caused by the total thermal power of 4480W
λ [K/Wm-1]

λ [K/Wm-1]

300 250
Al
Si dissipated by all 24 IGBTs.
200 200

100 Sn 150
AlSiC
D. Modeling the Active Area of a Semiconductor
0 100
-50 0 50 100 150 200 250 -50 0 50 100 150 200 250 In Fig.13 the detailed temperature distribution of IGBT
T [°C] T [°C] s33 is shown for heat dissipation as defined in Fig.5 (four
Fig.11. (a) Temperature dependency of the thermal conductivity of Cu IGBTs heated with 188W each). The mesh is defined to
[8], Al [8] and Sn [8]. (b) Temperature dependency of the thermal show at least 64 2D-elements over each two-dimensional
conductivity of Si [9], AlN [10] and AlSiC [11]. The ellipsoid highlights
a typical thermal operating area of the power module.
heat source representing the active area of the chip.
Therefore, the resulting mesh of the whole power module
Fig.11(b) shows that the thermal conductivity of AlN is very fine (Fig.13(a)), resulting in a total of
would be λ=200W/Km-1 at 75°C, but would decrease to approximately 1,000,000 elements for the whole 3D-FEM
model (power module plus cooling-plate). This large
λ=165W/Km-1 at 125°C. Fig.12(a) shows the effect of a
number of elements would result in extremely long
change of the equivalent thermal conductivity of the
simulation times and a strong tendency to numerical
AlN*-plate on the IGBT-temperatures for heating all
instability in case of calculating the Navier-Stokes
IGBTs (see Fig.4). In case of a decrease of the thermal
equations additionally to the heat conduction equation
conductivity of AlN from 200W/Km-1 (75°C) to 165/Km-1
(see discussion in section (III.B)). This again shows the
(125°C), the equivalent thermal conductivity of AlN*
merits of modeling the cooling plate with a convective
would decrease according to (1) from 153W/Km-1 (75°C)
boundary condition instead of directly simulating the
to 138W/Km-1 (125°C). According to Fig.12(a) this
water flow.
would result in a very small temperature increase of about
ΔT ≈ 2..3°C. Because AlN shows stronger temperature- The active area of the semiconductor is modelled by a
dependency than AlSiC and Si, it represents a worst-case 2D-heat source providing homogenously distributed
scenario, and temperature dependencies of the material thermal power. No wire bonds and no gate drive area are
properties can be neglected in good approximation. modelled. As one can see when comparing the simulation
in Fig.13(b) to the experimental measurement in Fig.13(c), temperature T [K]. To create a simplified thermal model
the differences in the absolute temperature values are describing the dependency between the chip junction
about 5°C or smaller at the selected measurement points temperature and the thermal losses, two different
which verifies the simple thermal semiconductor model approaches are possible. Both modeling approaches have
employed in all 3D-FEM simulations. It is also interesting been described in literature, and also the integration into a
to note that the temperature gradient on the top surface of circuit simulation is well known (e.g. [12] - [18]).
the silicon chip is about 25°C…30°C which is in the One method is based on the Finite Difference Method
range of 33% with regard to the temperature difference of (FDM) where the 3D-geometry is divided into many
semiconductor to ambient ΔΤ ≈100-20 =80°C. small volume elements, with the heat conduction equation
(4) linearized within each volume element [19], [20]. The
alternative modeling approach is based on an impedance
matrix [21]. Here, equation (4) is assumed to be a linear
differential equation (cP, ρ, and λ not dependent on
temperature) over the whole volume which is in good
approximation true for many applications (see discussion
in section (III.C)). Applying superposition, all thermal
contributions are modeled by thermal impedance circuits
that show approximately the same signal-behavior as the
3D-structure, but do not have physical meaning.

⎛ T junc ,(1) (t ) ⎞ ⎛ z1,1 z1,2 " z1,35 z1,36 ⎞ ⎛ pV ,(1) (t ) ⎞

⎜ ⎟ ⎜ ⎟ ⎜ ⎟
(a) (b) ⎜ T junc ,(2) (t ) ⎟ ⎜ z2,1 z2,2 " z2,35 z2,36 ⎟ ⎜ pV ,(2) (t ) ⎟
⎜ # ⎟=⎜ # # % # # ⎟⋅⎜ # ⎟ + Tambient
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
T (t )
⎜ junc ,(35) ⎟ ⎜ z35,1 z35,2 " z35,35 z35,36 ⎟ ⎜ pV ,(35) (t ) ⎟
⎜T ⎟ ⎜ z36,36 ⎟⎠ ⎜⎝ pV ,(36) (t ) ⎟⎠
⎝ junc ,(36) (t ) ⎠ ⎝ z36,1 z36,2 " z35,36
(5)
Equation (5) describes the general structure of the thermal
impedance matrix for a power module (plus cooling-plate)
with 36 internal dies, where the transient junction
temperature of each individual die is dependent on the
transient thermal power dissipated by each die. The very
complex and highly coupled thermal equivalent circuit
resulting from this equation is shown graphically in [22],
where the embedding into a circuit simulation is also
discussed in more detail. The coupling occurs via
elements zj,k that can be realized by simple single- or
multistage circuits as shown in Fig.14.

Rth,(13),i
Tjunc,(13) (t) Fig.14. Thermal equivalent
circuit (Foster-type) that
(c) represents one single entry z1,3
pV,(1) Cth,(13),i in the matrix in (5).
Fig.13. (a) Surface-mesh of six dies attached to an AlN-plate as
employed in all 3D-FEM simulations performed in this paper. (b) 3D-
FEM simulation of the temperature distribution of IGBT s33 for heating In order to find the parameters Rth,(jk),i and Cth,(jk),i of all zj,k
according to Fig.5. (c) Infrared-measurement of IGBT s33. sub-circuits in (5), one has to measure or calculate the
transient thermal step responses of all semiconductors of
the power module. In case of the ABB HiPak IGBT
IV. THERMAL MODELLING OF A POWER MODULE FOR module, there is the need to calculate 36 step responses
EMBEDDING INTO A CIRCUIT SIMULATOR and measure always the temperature rise at each of the 36
dies resulting in a total of 36×36=1296 curves. Each
A. Thermal Impedance Matrix curve has to be represented by the according matrix entry
The thermal conduction inside a solid structure is zj,k in (5). Now, based on each curve the according values
described by the heat conduction equation of Rth,(jk),i and Cth,(jk),i can be extracted. With higher order
of the sub-circuit the number of Rth- and Cth-components
∂T G
cP (T ) ⋅ ρ = ∇ [ λ (T ) ⋅ ∇T ] + w( x , t ) (4) increases which results in better curve fitting and higher
∂t accuracy, but also increases the computational effort
with thermal capacitance cP [Ws/(K.kg)], material density when embedding the model into the circuit simulation
ρ [kg/m3], temperature dependent thermal conductivity λ ([22]). As an example, Fig.15 shows all 36 thermal step
[W/Km-1], thermal power density w [W/m3], and responses for heating IGBT s32 with 168W.
 27 80 Electro-Thermal Simulation of IGBT Modules in Hybrid Electric
75
26 70
Vehicles”, in Microelectronics Reliability 45, pp. 1694-1699, 2005.
s22 s32
65 [3] K. Oila, M. Ciappa, N. Seliger, W. Fichtner, “Thermal Modelling,
25
s21 60 Simulation and Characterization fo a High-Temperature Converter
24 55 for Automotive Applications”, in Proc. of the 10th European
T [°C]

T [°C]
50
23 45
Conference on Power Electronics and Applications (EPE’03),
s13,s14,s24 Toulouse, France, Sept. 2 - 4, 2003.
40
22 s11,s12,s23 s31
35 [4] ABB Ltd, “3300V/1200A IGBT Module 5SNA 1200G330100”,
21 30 datasheet published at http://www.abb.com/semiconductors
s33,s34,s4i
25
20 20
[5] M. Rahimo, A. Kopta, R. Schnell, U. Schlapbach, R. Zehringer, S.
Linder, “2.5kV-6.5kV Industry Standard IGBT Modules Setting a
0.1 1 10 100 1000 0.001 0.01 0.1 1 10 100 1000 New Benchmark in SOA Capability”, Proc. of the 10th European
t [s] t [s]
34 Power Quality Conference (PCIM’04), Nuremberg, Germany,
21
s52 May 25 - 27, 2004.
s51 32
20.8 d31 [6] http://www.icepak.com/ (March 2007).
30 [7] U. Drofenik, J. W. Kolar, “A Thermal Model of a Forced-Cooled
20.6 28 Heat Sink for Transient Temperature Calculations Employing a
T [°C]

s61,s62
T [°C]

d32,d41,d22,d42 Circuit Simulator”, in the IEEJ Trans. of the Institute of Electrical

s53 26
20.4 Engineers of Japan, Volume 126-D, no. 7, pp. 841-851, July 2006.
s54 d21
24 [8] J. F. Shackelford, W. Alexander, “The CRC Materials Science and
20.2
22 Engineering Handbook”, ISBN 0-8493-4276-7, CRC Press, 1992.
s63,s64
[9] C. J. Glassbrenner, G. A. Slack, Phys. Rev. 134, 4A, A1058-A1069,
20 20 1964.
1 10 100 1000 0.1 1 10 100 1000 [10] G. A. Slack, R. A. Tanzilli, R. O. Pohl, J. W. Vandersande, J. Phys.
t [s] t [s] Chem. Solids 48, 7 (1987), 641-647.
Fig.15. Transient thermal step responses from a 3D-FEM simulation. [11] M. A. Occhionero, R. W. Adams, K. P. Fennessy, "A New
IGBT s32 is heated by a thermal power of 168W (see Fig.6) giving Substrate for Electronic Packaging: Aluminum Silicon Carbide
transient temperatures at the centers of all 36 dies labeled sij and dik. (AlSiC) Composites," in Proc. of the 4th Annual Portable by
Design Conference, March 24 - 27, pp 398 - 403, 1997.
[12] U. Drofenik, J. W. Kolar, “Thermal Analysis of a Multi-Chip
B. Embedding the Thermal Model into Circuit Simulator Si/SiC-Power Module for Realization of a Bridge Leg of a 10kW
A virtual design platform for power electronic systems is Vienna Rectifier”, in Proc. of the 25th IEEE International
Telecommunications Energy Conference (INTELEC’03),
under development at PES (ETH Zurich), where the Yokohama, Japan, pp. 826 - 833, Oct. 19 - 23, 2003.
procedure of generating a thermal equivalent circuit for [13] U. Drofenik, J. W. Kolar, “A General Scheme for Calculating
power modules mounted onto heat sinks as described in Switching- and Conduction-Losses of Power Semiconductors in
section (IV.A) has been implemented. One feature of this Numerical Circuit Simulations of Power Electronic Systems”, in
Proc. of the 2005 International Power Electronics Conf. (IPEC'05),
simulation platform is that complex thermally coupled Niigata, Japan, April 4 - 8, CD-ROM, ISBN: 4-88686-065-6, 2005.
models as given in (5) are built automatically from 3D- [14] C.-S. Yun, “Static and Dynamic Thermal Behavior of IGBT Power
CAD models of the power module with minimum user- Modules,” Series in Microelectronics, Vol. 124, Dissertation ETH
input necessary. This saves the user a lot of time and No. 13784, 2001.
[15] J. J. Rodriguez, Z. Parilla, M. Velez-Reyes, A. Hefner, D. Berning,
guarantees reliability of the model. The transient power J. Reichl, J. Lai, “Thermal Component Models for Electro Thermal
losses, consisting of temperature-dependent conduction Analysis of Multichip Power Modules,” in Proc. of the 2003 CPES
and switching losses, are calculated as described in [14], Annual Seminar, Blacksburg (VA), USA, April 27 - 29, 2003.
[19]. Junction temperatures of all 36 chips can be [16] Z. Jakopovic, V. Sunde, Z. Bencic, “From Transient Thermal
Impedance Measurement to Successful Electrothermal
measured directly in the circuit simulation. For a first Simulation,” in Proc. of the 10th International Power Electronics
description of the design platform see [22]. and Motion Control Conference (PEMC’02), Dubrovnik, Croatia,
Sept. 9 - 11, CD-ROM, ISBN: 953-184-047-4, 2002.
V. CONCLUSION [17] J. Chen, S. Downer, “MOSFET Loss and Junction Temperature
Calculation Model in MATLAB,” in Proc. of the 10th European
A 3D-FEM model of a 3300V/1200A HiPak IGBT power Power Quality Conf., Nuremberg, Germany, May 23 - 27, 2004.
module has been presented. Possible simplifications and [18] J. Z. Chen, Y. Wu, D. Borojevich, J. H. Bohn, “Integrated
Electrical and Thermal Modelling and Analysis of IPEMs,” in
numerical stability issues of the FEM simulation have Proc. of the 16th IEEE Applied Power Electronics Conference
been discussed in detail, and the sensitivity of the 3D (APEC’01), Anaheim (California), USA, March 4 - 8, 2001.
model against parameter variations was tested. A [19] U. Drofenik, “Embedding Thermal Models in Power Electronic
comparison of the simulated temperature distribution with Circuit Simulation,” in Proc. of the ECPE Power Electronics
Packaging Seminar, Baden-Dättwil, Switzerland, June 7 - 8, 2004.
the experimental measurements obtained with an infrared [20] O. Schepp, M. Lenz, “A Versatile Electrothermal Model of an
camera showed a very good agreement. Therefore, the Integrated Full Bridge Device Taking into Account Various
3D-FEM model is well-suited for calculating transient Boundary Conditions,” in Proc. of the 12th IEEE Applied Power
thermal step responses that can be used to derive an Electronics Conference (APEC’97), Atlanta, USA, Feb. 23 - 27,
Vol. 1, pp. 185 - 190, 1997.
equivalent thermal network model for embedding into a [21] T. Franke, G. Zaiser, J. Otto, M. Honsberg-Riedl, R. Sommer,
circuit simulation. “Current and Temperature Distribution in Multi-Chip Modules
REFERENCES under Inverter Operation“, in Proc. of the 8th European
Conference on Power Electronics and Applications (EPE’99),
[1] A. Castellazzi, M. Ciappa, W. Fichtner, G. Lourdel, M. Mermet- Lausanne, Switzerland, 1999.
Guyennet, “Compact Modelling and Analysis of Power-Sharing [22] U. Drofenik, D. Cottet, A. Müsing, J.-M. Meyer, J. W. Kolar,
Unbalances in IGBT-Modules Used in Traction Applications”, in “Computationally Efficient Integration of Complex Thermal
Microelectronics Reliability 46, pp. 1754 - 1759, 2006. Multi-Chip Power Module Models into Circuit Simulators”, Proc.
[2] M. Ciappa, W. Fichtner, T. Kojima, Y. Yamada, Y. Nishibe, of 4th Power Conversion Conf. (PCC’07), Nagoya, Japan, April 2
“Extraction of Accurate Thermal Compact Models for Fast - 5, 2007.

View publication stats