The 2010 International Power Electronics Conference

New Physical Model for Lifetime Estimation of

Power Modules
I. F. Kovaþeviü, U. Drofenik, and J. W. Kolar
ETH Zurich, Power Electronic Systems Laboratory, Physikstrasse 3, 8092 Zurich, Switzerland
[email protected]

Abstract--In this paper a physical model for lifetime The layers, having typically different thermal
estimation of standard power modules is proposed. The properties, have great influence on the heating and
lifetime prediction is based on the assumption that the cooling rates of the module and determine its overall
solder interconnections are the weakest part of the module performance. Consequently, the failures of power
assembly and that the failure cause is the inelastic modules are directly related to the power modules’
deformation energy accumulated within the solder material.
Unlike the well-known Coffin-Manson model, the proposed
assembly structure and packaging technology. During
model can be used to physically explain the dependency of thermal and/or termo-mechanical loadings, the module
lifetime on the various properties of a temperature profile assembly layers experience shear stresses due to the
i.e. frequency, dwell-ramp time, minimum/maximum differing coefficients of thermal expansion (CTE), i.e. the
temperature. The model is based on Clech’s algorithm for bimetallic effect. The failure modes observed so far in
simulation of stress-strain solder response under cyclical modern power modules are described in [1].
thermal loading and on the solder deformation mechanism
map used to define the dominant failure mechanism under
observed stress-temperature conditions. Either accelerated
cycling tests or existing field databases are needed to
parameterize the model. To verify the approach, the results
of power cycling tests for a high power IGBT module found
in literature are applied and the impacts of two mission
profiles on the module lifetime are examined.

Index Terms--Clech's algorithm, deformation mechanism

map, lifetime modeling, mission profile.

I. INTRODUCTION Fig. 1. Multilayered structure of a standard power module: power

Specifying, designing and maintaining power semiconductor chips are soldered to a direct bonded copper (DBC)
ceramic substrate [2]; at the bottom side, the substrate is soldered to the
electronics systems is a complex task since there is an
package (metal) base plate; the chip interconnections are performed by
ever increasing demand for higher reliable products in all aluminium (Al) wire bonds; the layer thickness are not shown to scale.
industries. Reliability is an important issue for power
electronics systems that require long-term lifetime To study dominant wear-out failure mechanisms of
guarantees like those used in cars, locomotives, airplanes power modules, active power and passive temperature
etc. Power electronics converter employed in these cycling tests are performed. Setting test conditions to
systems are exposed to rather severe operating induce a specific failure mechanism is the most important
conditions: high voltages and currents, temperature issue in such accelerated experiments. So far, the known
cycles of large amplitude, etc., and the reliability of a weak points are the interconnections inside the power
whole system is determined by the endurance of each module including wire bonds and solder joints i.e. die
element i.e. each power module comprising high power attachment and the joint between the base plate and the
IGBTs or MOSFETs and diodes. substrate. The most common failures occurring at these
A standard power module can be seen as a interconnections are mainly ascribed to unmatched
multilayered structure (see Fig. 1). The heat dissipated coefficients of thermal expansion (CTE) of adjacent
inside a power semiconductor chip is conducted through materials: wire bond and silicon, silicon and DBC
the multilayered structure into a heat sink, and it is then substrate, DBC substrate and base plate.
transferred by convection through the heat sink to the In recent years advanced wire bonding techniques
ambient. Due to their low cost and appropriate physical have been applied suggesting that the bond wire lift-off
properties high power modules with lead-based solder failure mode might be avoided with certainty [3].
interconnections (eutectic SnPb, or high lead solders Furthermore, effort has been made to analyze solder
PbSn with a small addition of silver) are still in use. layers in more detail i.e. to develop an accurate lifetime
However, since lead adversely impacts the environment model for solder joint failures. So far, extensive research
and health, other candidates like SnAg alloys attract more has been done to develop an accurate lifetime prediction
attention nowadays. for power modules but a generally applicable model is
still not developed.

978-1-4244-5393-1/10/$26.00 ©2010 IEEE 2106

 The 2010 International Power Electronics Conference

II. POWER MODULE LIFETIME PREDICTION Lifetime prediction of a power module using an
Several well-known lifetime models are summarized analytical approach is performed by means of Miner’s
in this section to gain a brief insight into the current state rule for damage accumulation. The correlation between
of research in the field. In general, two different lifetime temperature changes and the damage produced within the
modelling approaches can be distinguished: analytical module has to be defined and then lifetime is represented
and physical. The models are built based on the thermo- as inverse of the total damage accumulated within a
mechanical characteristics of the module assembly, on power module until the cessation of its correct
the knowledge of failure mechanisms under certain functioning. In a real operating environment power
conditions, and on the mission profile dedicated to the modules are exposed to rather non-uniform temperature
module i.e. the temperature profile T(t) appearing within variations and thus for the analytical lifetime modelling a
the solder layer of a module during its operation. proper definition of a temperature cycle within the given
mission profile is required. Accordingly, mission-profile-
A. Analytical Lifetime Models transformation into a sequence of temperature cycles is
Analytical models describe the dependence of the the main issue of the analytical approach.
number of cycles to failure (Nf) on the parameters of The performance of three observed analytical models
temperature cycles i.e. amplitude, duration, frequency, can be analyzed using the results of the power cycling
mean value, dwell time, maximum and minimum tests (Nf results) for high power IGBT traction modules
temperature, etc. Several analytical models published in (Table I) [7]. Table I summarizes Nf results for three
literature are analysed: the Coffin-Manson model [4], power cycling tests PC1-3 with equal temperature swings
Norris-Landzberg model [5], and Bayerer’s model [6]. but different heating and cooling times. The number of
The well-known Coffin-Manson model has been cycles to failure predicted by the analytical models and
widely used even though it has obvious shortcomings. the experimental Nf results in Table I are compared by (4)
and (5).
Nf a ˜ ('T j ) n ˜ e Ea /( k ˜Tm ) (1) Apparently, the results in Table I cannot be explained
by the Coffin-Manson model. On the other hand, if the
According to the Coffin-Manson formula (1), the power cycling tests PC1 and PC2 are employed to find the
number of cycles to failure depends only on the unknown parameter n2 in the Norris-Landzberg equation
temperature amplitude of the junction temperature ǻTj and the Nf for PC3 is calculated by (2), then the resulted Nf
and the medium temperature Tm. As it has been proven [6, would be 2.1 times higher than the experimentally
7] that the other parameters i.e. frequency of cycles, reported value in Table I. This implies that (2)
heating and cooling times also take significant influence underestimates the lifetime of the observed IGBT
on lifetime, the Coffin-Manson model is rather simple modules. Similarly, using (3) Nf ratios can be calculated
and not accurate enough. and then compared to the experimental ratios defined by
The second considered analytical model is the Norris- Table 1. The resulting Nf ratios are given by (4) and (5),
Landzberg model, which includes the frequency
parameter but neglects the influence of other parameters N f , PC1 N f , PC1
1.54, 2.36 (4)
such as heating and cooling time, N f , PC2 N f , PC3
(3) (3)
Nf A ˜ f  n2 ('T j ) n1 ˜ e Ea /( k ˜Tmax ) . (2)
N f , PC1 N f , PC1
The multi-parameter model of Bayerer (3) is the most 2.77, 11.91. (5)
comprehensive analytical model as it is built based on a N f , PC2 N f , PC3
(Table I ) (Table I )
large amount of power cycling data from different
module technologies. It includes the influence of the The difference between (4) and (5) points out that
various parameters of power cycling tests and also power Bayerer’s model does not comply with the experimental
module characteristics: temperature swing ǻTj, the results presented in Table I.
maximum junction temperature Tj, the heating time ton, TABLE I
the applied DC current I, the diameter D of the bond POWER CYCLING TESTS FOR THREE POWER TRACTION MODULES

wires and the blocking voltage V. Module ǻTj [°C] Tj,max[°C] IC[A] ton [s] toff[s] Nf [·1k]
PC1 50 110 166 3.1 2.3 305
E1 E 2 /(T j  273 K ) E3 E4 E5 E6 PC2 50 110 149 10 4 110
Nf K ('T j ) ˜e ˜ ton ˜I ˜V ˜D (3)
PC3 50 110 138 30 4.6 25.6
The constants K and ȕ are extracted from a large data
set collected in the long-term reliability testing
B. Physical Lifetime Models
experiments.
Assuming that the failure of solder interconnections is Physical modelling requires failure and deformation
the dominant failure mechanism, a more accurate mechanisms to be priorly known. It is based on the
approach would be to use the solder layer temperature [8] knowledge of stress/strain deformations within devices
in these Nf -models instead of the junction temperature Tj that can be gained either by experiments or simulations.
measured on the top of chip. Direct measurements of stress and strain in electronic
packages demand the usage of high resolution measuring

2107
 The 2010 International Power Electronics Conference

methods i.e. infrared microscopy, etc. Therefore, described by means of a hysteresis loop. As a result, the
computational mechanics, like Finite Element Analysis stress-strain plot can be employed as a tool for lifetime
(FEA), are often employed. estimation of solder joints in electronic devices.
The physical models for estimating the lifetime of Al Depending on the stress level and temperature, solder
bond wires are based on fracture mechanics and basically experiences different physical deformations that
rely on FE simulations ([9]). On the other hand, physical gradually contribute to damage accumulation and finally
models of solder behaviour under thermal cycling to the failure of solder interconnections. The stress-strain
operation have been developed for high-density response is modelled by solder constitutive equations that
electronics packages mostly used in telecommunication mathematically describe time-independent elastic and
and consumer electronics: flip chips, ball grid arrays plastic deformations and time-dependent plasticity called
(BGA), chip size packages (CSP), and fine-pitch surface creep.
mounted assemblies. These models are essentially based By their nature, elastic and plastic deformations and
on the experimental data collected in low-cycle tests creep can be defined respectively by strain and strain-rate
needed for model parameterization. Hence, the functions,
corresponding databases for frequently employed solder
materials can be found in literature.
J f (W , T )
(6)
Four classes for the lifetime prediction of solder joints d J / dt f (W , T , t )
can be distinguished: stress-based methods, strain-based
methods, energy-based methods, and damage-based where Ȗ is strain, IJ is stress, T is the temperature and t is
methods. Energy-based models are seen to be the most time. According to mechanical analysis [13], the
convenient, as they have ability to capture test conditions connection between shear stress and strain within a solder
with more accuracy [10]. joint can be described by the reduction lines (7), where K
An energy-based method specially intended for power is the reference assembly stiffness depending on
modules is introduced in [11]. The stress-strain response geometry (the height h and the cross section of solder
to an arbitrary temperature profile of a bimetallic joint A) and D1 is a constant reflecting both the geometry
interconnection is numerically calculated and the (the length of solder joint L) and CTE mismatch between
resulting stress-strain curve is further used to calculate layers connected by solder, Į1-Į2. The explanation of
the total deformation energy accumulated within the reduction lines is given in Fig. 2.
module. The lifetime model presented in [11] is based on W shear
the assumption that the end-of-life of a device is J shear  D1 ˜ (T  T0 )
K (7)
determined by the total deformation energy accumulated
during the operation of a device: a device fails when the h
K  , D1 L ˜ (D1  D 2 )
deformation work reaches the critical value ǻWtot. The A
main drawback of this model is that elastic and plastic W shear K (J th (T )  J shear )
deformations are neglected in the stress-strain relation
formulas and only time-dependent creep is taken into
K ˜ J th,(3)
account.
A new physical model for lifetime estimation of power K ˜Jth,(2)
modules relying fundamentally on the energy-based
approach is proposed in the next section. The physical K ˜Jth,(1)
model enables lifetime estimation of power modules
which should operate under arbitrary temperature
profiles. J shear
Jth(T) D1(T T0)
III. THE PROPOSED PHYSICAL LIFETIME MODEL
The proposed method for lifetime estimation of power
modules is derived from comprehensive equations Fig. 2. Isothermal stress reduction lines at different temperatures and
describing solder behaviour, i.e. simultaneously taking comparison of stress reduction to stress relaxation and creep.
into account elastic, plastic and creep deformation. The
stress-strain curves are obtained from modelling solder The assembly layers change their dimensions
response to different temperature profiles. Then, applying according to their CTEs. The strain created by thermal
the energy-based approach, these equations are used to mismatch of materials joined by solder is accommodated
calculate the total deformation energy producing the final within the solder layer by the strain change ǻȖ composed
device failure. The stress-strain response of a solder joint of a plastic, elastic and/or creep component. The
is generated by means of Clech’s algorithm [12]. The equations describing solder behavior are given by (8)-
brief summary of the theory behind the proposed model (11).
is given in the following subsections. Based on Hook’s law, elasticity of solder can be
defined by
A. Solder Behavior
Solder response to a thermal cycling load can be

2108
 The 2010 International Power Electronics Conference

W cyclic load. The resulting algorithm for generating a

J . (8) stress-strain response for eutectic solder is introduced in
G (T )
[12].
To describe time-independent plastic behavior Clech’s algorithm can be explained by Fig. 3.
Darveaux’s stress-strain (IJ, Ȗ) dependency [14] Knowing the stress-strain state (Ȗ1, IJ1) at time t, the stress-
strain state (Ȗ2, IJ2) at time t+ǻt can be calculated using the
W mp constitutive equations describing solder behaviour. It is
J Cp ˜( ) (9)
G (T ) assumed that: (a) ǻt is a relatively small time step, (b) the
stress is built up instantaneously for the temperature
is used, where G is the shear modulus constant dependent increase from T = T1 to T2 = T1+ǻT, assuming that
on temperature, G(T)=G0 – G1(T-273K). dominant strain components are time-independent plastic
Analyzing the creep nature of solder, two types of and elastic components ǻȖelastic and ǻȖplastic, and (c) at
creep deformation can be distinguished: dislocation temperature T2, the change of stress follows the T2-
controlled creep at higher and diffusion controlled creep reduction line building the strain component ǻȖcreep.
at lower stress levels, defined respectively by strain rate To calculate ǻȖelastic and ǻȖplastic, (8) and (9) are used
equations (10) and (11). and ǻȖcreep is calculated using either (10) or (11). The
implemented algorithm detects changing from one region
dJ G (T ) D ˜W nl Ql / RT
Cl ˜ sinh[( ) ]˜ e into a neighboring region of the deformation mechanism
dt T G (T ) map and selects the according creep differential equation.
(10)
G (T ) D ˜W nh Qh / RT This approach provides the explanation of solder
 Ch ˜ sinh[( ) ]˜ e
T G (T ) behaviour under different temperature profiles. The
influence of ramp time, hold time, and frequency of
dJ G (T ) W cycling can be captured by the information carried within
B1 ˜ ˜ ( ) ˜ e Qb / RT the corresponding hysteresis (stress-strain curve) returned
dt T G
(11) by the described simulation.
G (T ) W
 B2 ˜ ˜ ( ) ˜ e Qm / RT
T G 1
Each strain rate equation consists of two parts referring
10-1 Ideal Strength
to different deformation mechanisms at low and high Dislocation Glide (plastic flow)
temperature ranges. To cover the full stress range both
Normalized stress IJ/G

Equation 9
10-2 Yield Strength
creep equations are needed. The term G/T has been Dislocation Creep
introduced to take into account the temperature effect. 10-3 Low Temperature Creep

The constants and parameters for 63Sn37Pb solder can be

Left Term of Equation 10
Elastic Region High Temperature Creep
found in [15]. It was shown in [15] that these constitutive 10-4
Equation 8 Right Term of Equation 10

equations could be used to describe the creep deformation Diffusional Creep

10-5 Low Temp. High Temp.
of solder for different temperatures over a wide stress Creep
Left Term of
Creep
Right Term of

range. 10-6
Equation 11 Equation 11

0.0 0.2 0.4 0.6 0.8 1.0

B. Deformation Mechanism Map
Homologous Temperatur T / TM
The nature of deformations occurring within metals Fig. 3. The regions of the deformation mechanism map are described
and ceramics exposed to some external stresses, e.g. by corresponding solder constitutive equations.
thermal stresses, is explained in detail in [16]. The
deformation mechanism map is a stress-temperature
diagram presenting the dependency of normalized stress 'J plastic  'J elastic ' J creep
IJ/G on homologous temperature T/TM (TM is the melting
W
temperature). For a known stress and temperature range,
the dominant damage mechanism can be identified from
J12,W12
the map. An example of a deformation mechanism map is
presented in Fig. 3 with the axis scaled just as a guide.
J2,W2
The regions of the deformation mechanism map are
defined by the solder constitutive equations (8)-(11). The J1,W1
borders between low temperature and high temperature T2 T  'T
creep regions can be found by equalizing the left and the
right terms of (10) or (11). In the same manner, the stress- T1 T
temperature condition for transition from one creep
mechanism to another can be calculated by equalizing J
(10) and (11). Fig. 4. The solder stress-strain simulation step for a temperature
increase from T1 to T2= T1+ ǻT.
C. Clech’s Algorithm
Clech’s intention was to find a way to accurately Implementing Clech’s algorithm based on (8)-(11) d in
simulate the response of SMT solder joints exposed to MATLAB allows to numerically solve the solder

2109
 The 2010 International Power Electronics Conference

constitutive equations and to calculate hysteresis loops The parameterization of the proposed physical model
corresponding to temperature cycling tests or an input is defined as an optimization routine implemented in
temperature profile (mission profile) with an arbitrary MATLAB. The optimal parameters should minimize the
shape. difference of two calculated deformation energies Wtot,k1
and Wtot,k2 corresponding to two different input
IV. RELATIVE LIFETIME ESTIMATION temperature profiles Tk1 and Tk2. The ratio of the
In combination with the deformation mechanism map, parameterization defined as (8) should be ideally equal to
the constitutive solder equations are used to accurately 1,
determine strain components in the entire stress and r Wtot ,k1 / Wtot ,k 2 . (8)
temperature range. The proposed lifetime model needs 17
material/geometry parameters to be known. The material The search procedure for the optimal unknown
parameters of 60Sn40Pb (i.e. 63Sn37Pb) solder typically parameters is based on a built-in MATLAB optimization
used in standard power modules are taken from [14] and routine that returns only local minimums and cannot
[15]. K and D1 are material/geometry dependent guarantee finding the global minimum of the objective
parameters that have to be additionally determined as function. Therefore, an approach to deal with local
described in the following. minimums is needed in order to ensure the choice of the
right parameters K and D1. An idea to find out which
A. Model Parameterization
parameter set most correctly describes theoretical solder
To fully parameterize the model, the results of at least behaviour is suggested in Section IV.C.
two accelerated power cycling tests are needed i.e. the
number of cycles to failure for different temperature B. Verification of the Proposed Lifetime Modeling
profiles, Nf1 and Nf2. The total deformation energy Wtot,k Approach
accumulated in the solder layer that leads to the module’s To verify the proposed lifetime prediction approach,
end of life can be calculated by the presented physical model is applied to a high power
IGBT traction module already discussed in Section II.A.
Wtot ,k N f ,k ˜ Whys,k (7)
The model parameterization is performed using the
results of three power cycling tests (see Table I) with
where Nf,k is the number of cycles to failure for the k-th temperature profiles presented in Fig. 6. The power
temperature profile and Whys.k is the deformation energy cycling tests were originally performed to identify the
equal to the area of one corresponding simulated influence of cycling time on the lifetime behavior of
hysteresis loop. For power modules of same type, Wtot,k IGBT modules. Therefore, in all three power cycling
should theoretically be constant. An optimization routine tests, the temperature swings were set to 50°C with the
is used to find the best set of parameters (K, D1) so that maximum temperature of 110°C. The final failure driving
the ratio, Wtot1/Wtot2, is close to one. The power cycling force was ascribed to solder interconnections.
tests have to be specified such that all deformation Temperature Profile for Power Cycling 1

mechanisms are invoked. The parameterization procedure 120

Temperature [oC]

can be described by the flow chart in Fig. 5.

100

80

60

40
20 25 30 35 40 45 50 55 60
time [s]

Temperature Profile for Power Cycling 2

120
Temperature [oC]

100

80

60

40
20 40 60 80 100 120 140
time [s]
Temperature Profile for Power Cycling 3
120
Temperature [oC]

100

80

60

40
50 100 150 200 250 300
time [s]

Fig. 6. The temperature profiles during three power cycling tests of the
IGBT traction modules defined in Table I.

Following the procedure described by the flow chart in

Fig. 5, several optimal parameter sets (K, D1) are found.
Two parameter sets, s1, 2 = (K, D1) (see Table II) returning
a similar ratio of parameterisation are chosen to analyze
Fig. 5. Flow chart describing the parameterization of the proposed the simulation results and to describe an idea how the
lifetime model. stress-strain curves for different parameter sets can be
distinguished among each other.

2110
 The 2010 International Power Electronics Conference

TABLE II results of power cycling tests used to parameterize the

TWO OPTIMAL PARAMETER SETS RETURNED BY THE
PARAMETERIZATION PROCEDURE
model have significant influence on the determination of
Parameter Set K D1 Ratio
the unknown material/geometry parameters.
s1 2454 2.7e-4 1.614 TABLE III
s2 1034 7.3e-4 1.586 THE POWER CYCLING RESULTS FOR TWO DIFFERENT LEVELS OF FAILURE
CRITERION: 20% AND 50% INCREASE OF RTHJ

The simulations of solder stress-strain response using Module 20% increase of Rthj 50% increase of Rthj
s1 and s2 are depicted respectively in Fig. 7 and Fig. 8. PC1 305 000 480 000
PC2 110 000 186 000
The y-axis represents stress in MPa, while the x-axis
PC3 25 600 59 400
relative strain i.e. strain scaled by the height of the solder
joint, in percent. From the figures, it can be seen that the
hystereses shapes and the stress levels are similar, but the C. Selection of Optimal Parameters K and D1
strain magnitudes belong to different ranges so that the The total strain is the sum of the creep and time-
hystereses are shifted from each other along the strain independent plastic and elastic strain components. The
axis. numerical implementation makes possible to separately
15
Hystereses: s1
evaluate the strain components and to find out which
component becomes more dominant in which part of
10
temperature cycle. Analyzing how strain develops during
5
a temperature cycle, gives deeper theoretical insight into
the solder behaviour and enables to investigate the
0
influence of the unknown material/geometry parameters
stress [MPa]

-5
on stress-strain response.
Based on theory, creep cannot be developed at fast
-10 temperature rates and time-independent plasticity and
-15
elasticity produce the most of strain deformations in that
PC1: Nf1 = 305000
case. On the other hand, at lower temperature rates, creep
-20 PC2: Nf2 = 110000
PC3: Nf3 = 25600
becomes gradually dominant and determines almost alone
the total strain. Consequently, for a comprehensive
analysis, it is important to choose the temperature profiles
-25
1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
strain [%]
such that the impact of both creep and time-independent
Fig. 7. The stress-strain simulation applying the parameters set s1 = (K, plastic and elastic deformations can be observed.
D1) for the IGBT traction modules exposed to three power cycling
profiles PC1-3.
Accordingly, the creep behavior should be best observed
Hystereses: s2
in PC3 with the longest turn-on time, while for PC1 with
15
fast temperature changes the time-independent plastic and
10
elastic components should be dominant over the creep
strain component. The strain components for the
5 temperature profiles PC1 and PC3 are presented
0
respectively in Appendix Fig. A2 and Fig. A3.
For the parameter set s1, plasticity and elasticity almost
stress [MPa]

-5 solely determine the total strain during the whole

-10
temperature cycle PC1 (Fig. A2a), but in the simulation
with the parameter set s2 the creep component takes some
-15 higher share in the total strain at the end of both down-
-20
PC1: Nf1 = 305000
PC2: Nf2 = 110000
and up-ramp times of the temperature cycle PC1 (Fig.
PC3: Nf3 = 25600 A2b).
-25
4.6 4.7 4.8 4.9 5 5.1 5.2 5.3 In the case of temperature profile PC3 with slower
strain [%]
temperature changes, it can be seen that the creep
Fig. 8. The stress-strain simulation applying the parameters set s2 = (K, component becomes dominant at the dwell period of
D1) for the IGBT traction modules exposed to three power cycling temperature cycle (Fig. A3), building almost solely the
profiles PC1-3.
total strain in both simulations with s1 and s2. These
Regarding the ratio of parameterization, it was results comply with theoretical behaviour of solder
observed that better results would be achieved if the material. However, an apparent difference between the
failure criterion of IGBT modules was differently chosen simulations with the parameter sets s1 and s2, is the
i.e. 50% increase of junction thermal resistance Rthj magnitude of creep component at the up- and down-ramp
instead of 20%. The Nf -results for these two criterions are times of PC3 i.e. for faster changes of PC3. In the
compared in Table III. Using the experimental Nf -results simulation with s1, the creep is always smaller than the
returned by the 50% failure criterion as input of the time-independent plasticity, while in the simulation with
parameterization procedure, the calculated ratio of s2, the value of creep component becomes almost equal to
parameterization is on average 1.1. This implies that the the value of time-independent strain.

2111
 The 2010 International Power Electronics Conference

Accordingly, it can be concluded that by taking the mechanical engineering. Therefore, the approach
parameters set s2 as input for the simulation, the creep proposed in this paper enables engineers to estimate the
component is higher for both temperature profiles PC1 lifetime of power modules without detailed specifications
and PC3 so that the stress-strain simulation with the and without the knowledge of geometric properties which
parameter set s1 better agrees with the theory of solder are hard to measure. At the same time, it is less
behaviour. demanding than FE analyses from the aspects of
computational complexity and run-time.
D. Lifetime Prediction Approach The verification of the model is demonstrated on the
Having parameterized the solder model, different example of a power cycling test taken from literature [7].
mission profiles can be compared to estimate the worst The parameterization procedure is described through a
operation conditions of the power module. For two flow chart diagram and the simulated stress-strain curves
mission profiles, T1(t) and T2(t) the stress-strain for two different parameter sets producing similar error of
simulation returns two hystereses carrying the parameterization are analyzed. In a next step at the PES
information about the solder deformation losses (Wtot1, Laboratory of ETH Zurich, power cycling equipment will
and Wtot2). Equation (8) defines the relative lifetime of the be built which allows to define temperature cycles of
power module. arbitrary shape in order to further evaluated the proposed
physical model [18].
Wtot1
r ³ W dJ / ³ W dJ (8) 20
Wtot 2 T1 (t ) T2 (t )
15

The ratio (8) means that the first profile results in r

10
times more deformation damage than the second profile
implying that an increase of the lifetime of
stress [MPa]
5

semiconductors by a factor r can be expected if T2(t) is

0
applied instead of T1(t). The lifetime prediction can be
demonstrated on an example of two mission profiles A -5

and M from [17]. The resulting stress-strain simulations

-10
applying the parameter set s1 are presented in Fig. 9 and
10. The relative lifetime is found to be r = 2.5 (9). -15
0 0.5 1 1.5 2 2.5 3
strain [%]
WtotA Fig. 9. Stress-strain response to the temperature profile A of [16].
r 2.5 (9)
WtotM
15

It means that the small oscillations of temperature 10

within the temperature profile A would produce more
damage to an IGBT traction module than would be 5
produced by the constant temperature levels within the
stress [MPa]

mission profile M. 0

V. CONCLUSIONS -5

The proposed physical model for lifetime prediction of -10

power modules is developed based on Clech’s algorithm

and the energy-based lifetime modelling approach. The -15
0 0.5 1 1.5 2 2.5
used solder constitutive equations simultaneously take strain [%]

into account different deformation mechanisms: Fig. 10. Stress-strain response to the temperature profile M of [16].
elasticity, plasticity and creep. The implemented
numerical simulation returns the hysteresis-loops of the APPENDIX
stress-strain behaviour taking as input an arbitrary The stress and strain variations for third temperature
temperature profile T(t). The model has the ability to cycling profile are separately shown in Fig. A1. It can be
handle all information comprised within the mission seen that stress is proportional to temperature and strain
profile. changes nonlinearly with temperature. The temperature
The transformation of a mission profile into the profiles and the corresponding strain components, split
frequency distribution of temperature changes, into the time independent plastic/elastic ǻȖpl&el (blue) and
characteristic to analytical methods, is avoided. This time-dependent creep parts ǻȖcreep (green), are shown in
increases the accuracy of the proposed approach Fig. A2 and Fig. A3. The fast slopes of strain components
significantly, but still power cycling test results are actually correspond to the derivative of strain as the
needed to accurately parameterize the model. Typically, constant time step is used in the simulation, ǻȖ = ǻt·dȖ/dt.
power electronic engineers do not know the full internal
physical structure of power modules which would be
required in FE analysis and also they are not specialized
in characterizing materials what is a much closer topic to

2112
 The 2010 International Power Electronics Conference

Temperature Profile PC3

temperature [oC]

150
120

100 110

100
50
0 20 40 60 80 100 120 140 90
time [s]
20 80
stress [MPa]

70
0
60
-20 40 60 80 100 120
time [s]
-40
0 20 40 60 80 100 120 140 (a)
time [s] -3 PC3@ s1: Strain Components
x 10
0.02 1.5 creep
plastic & elastic
0.019 1 total strain
strain

0.5
0.018

relative strain [%]

0
0.017
0 20 40 60 80 100 120 140 -0.5
time [s]
-1
Fig. A1. Strain and stress variations for the temperature profile PC3
-1.5
applying the parameters set s1.
-2
Temperature Profile PC1
120 -2.5
40 60 80 100 120
time [s]
110
(b)
100
temperature [oC]

90
-3 PC3@ s2: Strain Components
x 10
4 creep
plastic & elastic
80 total strain

70 2

60
relative strain [%]

0
8 10 12 14 16 18 20
time [s]
(a) -2
-4
x 10 PC1@ s1: Strain Components
6 creep
plastic & elastic -4
4 total strain

2 -6
relative strain [%]

40 60 80 100 120
time [s]
0 (c)
Fig. A3. Analyses of strain components: (a) temperature profile PC3,
-2
(b) strain components applying the parameters set s1, (c) strain
-4 components applying the parameters set s2.

-6
8 10 12 14
time [s]
16 18 20 REFERENCES
(b) [1] M. Ciappa, “Selected failure mechanisms of modern power
x 10
-3 PC1@ s2: Strain Components modules”, Microelectronics Reliability, vol. 42, no. 4, pp.
1.5 creep
plastic & elastic 653-667(15), April 2002.
total strain [2] U. Drofenik, D. Cottet, A. Müsing, J. M. Meyer, and J. W.
1
Kolar, “Modelling the thermal coupling between internal
power semiconductor dies of a water-cooled 3300V/1200A
relative strain [%]

0.5 HiPak IGBT module”, in Proc. of the Conf. on Power

Electronics, Intelligent Motion, and Power Quality
0 (PCIM), Nuremberg, Germany, May 22 - 24, 2007.
[3] H. Berg and E. Wolfgang, “Advanced IGBT modules for
-0.5 railway traction applications: reliability testing”,
Microelectronics Reliability, vol.38, no. 6, pp. 1319-1323,
-1 June 1998.
8 10 12 14 16 18 20
time [s] [4] S. S. Manson, Thermal stress and low cycle fatigue, New
(c) York: McGraw-Hill, 1966
Fig. A2. Analysis of strain components: (a) temperature profile PC1, (b) [5] K. C. Norris and A. H. Landzberg, “Reliability of
strain components applying the parameters set s1, (c) strain components controlled collapse interconnections,” IBM Journal of
applying the parameters set s2. Research and Development, vol. 13, no. 3, pp. 266–271,
May 1969.

2113
 The 2010 International Power Electronics Conference

[6] R. Bayerer,T. Hermann, T. Licht, J. Lutz, and M. Feller,

“Model for power cycling lifetime of IGBT modules –
various factors influencing lifetime”, in Proc. of the 5th
International Conference on Integrated Power Electronics
Systems (CIPS), Nuremberg, Germany, March 11-13,
2008.
[7] A. Hamidi, A. Stuck, N. Beck, and R. Zehringer, “Time
dependent thermal fatigue of HV-IGBT-modules”, in Proc.
of the 27th Kolloquium Halbleiter-Leistungsbauelemente
und Materialgüte von Silizium, Freiburg/Breisgau, Oct. 26
- 27, 1998.
[8] M. P. Rodriguez, N. Y. A. Shammas, A. T. Plumpton, D.
Newcombe, and D.E. Crees, “Static and dynamic finite
element modelling of thermal fatigue effects in insulated
gate bipolar transistor modules”, Microelectronics
Reliability, vol. 40, no. 3, pp. 455-463, March, 2000.
[9] K. Sasaki, N. Iwasa, T. Kurosu, K. Saito, Y. Koike, Y.
Kamita, and Y. Toyoda, “Thermal and structural
simulation techniques for estimating fatigue of an IGBT
module”, in Proc. of the 20th International Symposium on
Power Semiconductor Devices and IC’s (ISPSD), Orlando,
Florida, USA, May 18-22, 2008.
[10] W.W. Lee, L.T. Nguyen, and G.S. Selvaduray, “Solder
joint fatigue models: review and applicability to chip scale
packages”, Microelectronics Reliability, vol. 40, no. 2, pp.
231-244, Feb. 2000.
[11] M. Ciappa, F. Carhognani, P. Cow, and W. Fichtner,
“Lifetime prediction and design of reliability tests for high-
power devices in automotive applications”, IEEE
Transactions on Device and Materials Reliability, vol.3,
no.4, pp. 191-196, Dec. 2003.
[12] J.-P. Clech, “Review and analysis of lead-free solder
material properties”, Online available at:
http://www.metallurgy.nist.gov/solder/clech/Introduction.h
tm
[13] C. H. Raeder, L.E. Felton, R.W. Messler, and L.F.
Coffine,”Thermomechanical stress-strain hysteresis of Sn-
Bi eutectic solder alloy”, in Proc. of the IEEE/CPMT Int.
Electronics Manufacturing Technology Symposium, Las
Vegas, May 21-14, 1995.
[14] G. Z. Wang, Z.N. Cheng, K. Becker and J. Wilde,
“Applying ANAND model to represent the viscoplastic
deformation behaviour of solder alloys”, Journal of
Electronic Packaging, vol. 123, no. 3, pp. 247-254, Sept.
2001.
[15] X.Q. Shi, Z. P. Wang, Q.J. Yang, and H.L.J. Pang, “Creep
behaviour and deformation mechanism map of Sn-Pb
eutectic solder alloy”, Journal of Engineering Materials
and Technology, vol. 125, pp. 81-87, Jan. 2003
[16] N.E. Dowling, Mechanical behaviour of materials:
engineering methods for deformation, fracture, and fatigue,
Prentice-Hall, Inc., 1993.
[17] U. Drofenik, I. Kovacevic, R. Schmidt, and J.W. Kolar,
“Multi-domain simulation of transient junction
temperatures and resulting stress-strain behaviour of power
switches for long-term mission profile”, in Proc. of the 11th
IEEE Workshop on Control and Modeling for Power
Electronics (COMPEL), Zurich, Switzerland, Aug. 18-20,
2008.
[18] A. Stupar, D. Bortis, U. Drofenik, and J. W. Kolar,
“Advanced setup for thermal cycling of power modules
following definable junction temperature profiles”, in
Proc. of International Power Electronics Conference -
ECCE Asia- (IPEC), Sapporo, Japan, Jun 21-24, 2010.

2114