Transactions of the Indian National Academy of Engineering (2022) 7:47–60

https://doi.org/10.1007/s41403-021-00268-1

REVIEW ARTICLE

A Critical Review and Perspective on Thermal Management of Power

Electronics Modules for Inverters and Converters
Gautam Ghaisas1,2 · Shankar Krishnan1

Received: 2 June 2021 / Accepted: 13 September 2021 / Published online: 24 September 2021
© Indian National Academy of Engineering 2021

Abstract
A cooling technology roadmap is defined for power electronics modules such as insulated gate bipolar transistors (IGBT)
and wideband gap semiconductor modules (SiC/GaN). These power modules are widely used for inverter and converter
applications. Cooling technologies studied include natural air cooling, forced air cooling, forced liquid cooling, and thermal
management technologies employing phase change. This work reviews the standard modes of failures of power electronics
modules. The work helps to understand the thermal resistances at various interfaces in the module stack. The study high-
lights the importance of using electrothermal models for power loss estimation. Effective thermal management must ensure
the power electronics components' junction (die) temperature is within the desired limits. The developed cooling technol-
ogy roadmap suggests using heat flux at the module base and heat generation per temperature rise in the die as a metric for
selecting cooling technology. Further, the review also stresses the importance of junction temperature swings experienced
by the die during transient operation.

Keywords Cooling technology roadmap · Insulated gate bipolar transistors · Wide-bandgap semiconductors · Power
electronics modules · Thermal management
Abbreviations k Thermal conductivity
A Area L Length of copper tubes
b Thickness of thermal interface material m Modulation index
c Specific heat capacity ṁ Mass flow rate of the coolant
d Diameter of copper tubes N Number of IGBT modules
E Energy n Number of transistor/diodes die
f Frequency n′ Exponent of Prandtl number
g Acceleration due to gravity Nu Nusselt number
h Heat transfer coefficient P Power
hfg Latent heat of vaporisation p Power density
I Current Pr Prandtl number
KI Exponents for current dependency of switching Q Heat loss
losses q′′s Surface heat flux
KI Exponents for voltage dependency of switching R Thermal resistance
losses Re Reynolds number
r Electrical resistance
rCE Temperature-dependent bulk resistance of the on-
* Shankar Krishnan state characteristic for transistor
[email protected]
rF Temperature-dependent bulk resistance of the on-
Gautam Ghaisas state characteristic for diode
[email protected]
s Number of transistor/diode switches active in the
1
Department of Mechanical Engineering, Indian Institute circuit
of Technology Bombay, Mumbai 400076, India T Temperature
2
Cummins Technologies India Private Limited, Pune 411038, TC Temperature coefficients for the switching losses
India ΔT Temperature difference

123
Vol.:(0123456789)
48 Transactions of the Indian National Academy of Engineering (2022) 7:47–60

V Voltage Introduction
Vceo Temperature-dependent threshold voltage of the
on-state characteristic for transistor Power electronics have gained widespread importance in
VFo Temperature-dependent threshold voltage of the domestic and industrial applications (Qian et al. 2018),
on-state characteristic for diode leading to energy-efficient solutions (Okumura 2018). Power
Greek symbols electronics devices use semiconductor switches to control
𝜌 Density the flow of electrical energy through circuits. These devices
𝜇 Viscosity often use a module-based construction in which one or more
𝜎 Surface tension switches are enclosed within a plastic case and mounted on
𝜀 Effectiveness of cooling plate a base plate. The base plate often is made of copper, which
𝜆 Parameter used in calculation of spreading helps dissipate the heat from the die (also referred to as junc-
resistance tion) to the coolant or the external ambient. Each switch
cos Φ Power factor consists of an insulated gate bipolar transistor (IGBT) and its
ΔTe Degree of superheat antiparallel diode. Recently, metal oxide semiconductor field
effect transistors (MOSFET) with wideband gap materials
Subscripts such as silicon carbide (SiC) and gallium nitride (GaN) can
cond Conduction also be used instead of the IGBT. There are myriad applica-
cp Cooling plate tions of IGBTs in electric vehicles (EVs), hybrid electric
D Diode vehicles (HEVs), wind turbines, UPS, photovoltaics, and
E Energy railways (Qian et al. 2018). About 50% of applications are
j Junction expected to be related to EVs and HEVs by 2020. Figure 1
l Liquid shows a semiconductor technology development roadmap
p Plate based on Okumura (2018), who has studied the evolution of
peak Peak wideband gap semiconductor devices and their applications.
ref Reference The EVs and HEVs correspond to the 2nd generation
rms Root mean square of semiconductor technology development, which corre-
s Source sponds to desirable features such as high voltage blocking
sp Spreading capability, ability to sustain high temperatures, operating
sw Switching over a wide range of frequencies, and reducing conduction
th Thermal and switching losses (Okumura 2018). These devices were
T Transistor expected to work with a maximum power density demand
v Vapor up to 25 W/cm3 at the module level by 2020.

Fig. 1  Semiconductor technol- 60

Medical accelarator
ogy development roadmap Smart grids
55
based on (Okumura 2018) Power transmission and distribution infrasturcture
50
45
Power density (W/cm3)

40
Hybrid and electric hybrid
35 vehicles
Electric trains
30 Heavy electric machinery
Industrial machinery
25 Home appliances
20 Lighting
General purpose inverter
15 Power units for IT
equipment
10
5
0
1st generation 2nd generation 3rd generation
2010-2015 2015-2020 >2020

123
Transactions of the Indian National Academy of Engineering (2022) 7:47–60 49

The authors have tried to understand how the cool- Power Electronics Module Stack, Thermal
ing technology roadmap should develop in line with the Resistances, Failure Modes, and Prevalent
semiconductor technology roadmap by considering below Cooling Technologies
aspects:
A typical power electronics module stack is shown in Fig. 2.
1. Study of a typical power electronics module stack. The IGBT/MOSFET and diode chips are soldered to the
2. Understanding the typical failure mechanisms for a substrate (copper foil + ceramic + copper foil). The complete
power electronics module. stack is soldered to the copper base. The module is covered
3. Understanding the thermal resistances offered by the with a plastic case. A heat sink or cooling plate conducts and
module stack, interface between the baseplate and the convects the heat out from the die to the coolant or ambient.
cooling plate, and the cooling plate itself. Soft, conforming materials are used as gap fillers between
4. Study of existing cooling technologies for IGBT mod- the copper base and cooling plate or heat sink. These materi-
ules and literature related to promising technologies to als help to reduce the contact resistance at the interface and
meet future heat flux demands. are called thermal interface materials (TIM).
5. Development of electrothermal models for estimation Following thermal resistances need to be considered
of heat loss and junction temperatures within a power while designing the cooling solution, considering the mod-
electronics module—to understand the inter-dependence ule construction.
of the power loss on junction temperature.
1. Junction to case: Thermal resistance offered due to the
All these aspects are explained in detail in the subse- construction of the stack. It can be reduced by reducing
quent sections. the number of layers in the stack. The manufacturer in
the component datasheet mentions the value of junction
to case thermal resistance.
2. Resistance due to thermal interface material at the mod-
ule base—heat sink interface.

Fig. 2  Power electronics module stack

123
50 Transactions of the Indian National Academy of Engineering (2022) 7:47–60

3. Thermal spreading resistance: This is the resistance couple the thermal parameters of the IGBT with the circuit
to heat spreading from a smaller heat source area to a simulator. Many researchers have adopted different meth-
larger heat sink area. odologies to estimate the power electronics modules' power
4. Cooling plate or heat sink thermal resistance in convec- losses and junction temperatures (Hafezi and Faranda 2021).
tion. These include the analytical methods, physics-based simula-
tion approach, experimental approach, and behavior-based
The contribution of thermal resistances is shown in modeling.
Fig. 3. Kang (2012) for forced air-cooling; while designing Cougo et al. (2020) built a 15 kVA inverter and measured
a cooling solution, thermal resistances other than the cooling the data in experiments to confirm their loss calculations for
plate can also be significant. Therefore, not considering the aircraft applications. The focus of the work was on the junc-
other thermal resistances may lead to the ineffective design tion temperature swing of SiC semiconductors. An analysis
of the cooling solution. of the inverter, load, and PWM methods on thermal cycling
Ineffective thermal management strategies tend to neglect was predicted. Gu et al. (2019) proposed an improved space
or underpredict the effect of one or many of the above- vector pulse width modulation strategy to reduce the switch-
mentioned thermal resistances. It results in higher junction ing losses in the inverter. Similar optimization to reduce the
temperature during the actual operation of the power elec- switching losses was proposed by Albatran et al. (2017).
tronics module, causing temperature-related failures. The Albatran et al. proposed an algorithm for obtaining opti-
standard failure modes of the modules include bond wire mum switching frequencies. As the power output demand
cracking and bond wire lift-off. Also, because the power from the inverter increased, the switching frequency was
loss within the die is temperature-dependent, if the thermal reduced to reduce switching losses. Zhu et al. calculated
management is improper, the power losses and tempera- power losses and junction temperatures in modular multi-
ture both increase until a thermal runaway occurs, leading level converters. The average and RMS currents in the sub-
to malfunctioning or even burning the module (Valentine modules were obtained analytically, and the conduction and
et al. 2015; Hu et al. 2019). Electrothermal models proposed switching losses were evaluated. It was concluded that the
could be updated after solder fatigue; updated parameters analytical method could efficiently predict the reliability of
were based on the case temperature. The updated electro- a cooling system (Zhu et al. 2007). Pou et al. calculated
thermal parameters could predict junction temperatures bet- the switch level and total power losses from inverter with
ter and influence reliability calculations. Even if the thermal applications in electric vehicles (Pou et al. 2011). They vali-
management ensures maintaining the junction temperature dated the power loss calculations done using Matlab-Sim-
within operating limits, reliability considerations can war- ulink with SemiSel tool from Semikron and PSIM, and the
rant additional thermal margin (Yun et al. 2001; Bennion results obtained agreed well with both Semisel and PSIM.
and Thornton 2010). Yun et al. developed 3-D finite element The work helps select modulation strategies, heat sink, and
method-based RC network models and used them to estimate power devices suitable particularly for EV applications.
the dynamic response of the RC component model quickly. The power losses depend on the junction temperature, and
The approach developed is intended to help the designer hence the total power losses vary non-linearly with junction
to ambient thermal resistance.
The method of estimation of average junction temperature
needs to be self-consistent by considering the electrothermal
1.30% Silicon
0.40% Silver
effects (Lin et al. 2007). Electrothermal modeling considers
0.50% Copper the effect of temperature-dependent electrical parameters
9.70% Al-oxide
0.50% Copper and thermal resistances when calculating power loss and
1.30% Solder junction temperatures. Xu et al. developed such electrother-
1.60% Copper baseplate
10.00% Thermal paste mal modeling calculations for power semiconductors and
28.90% HS- Base spreading compared them with their experiments. The essential con-
45.80% HS- Fins convective
clusions from their work were that the aging of the module
has an effect of an increase in the thermal resistance, IGBT
collector–emitter voltage, and switching loss. In turn, the
rise in switching loss increases the module average junction
temperature and the temperature variation.
On the other hand, the rise in the thermal resistance after
thermal cycling had a dominant effect on only the average
Fig. 3  Contribution of stack thermal resistances to the total thermal junction temperature and not so much on the temperature
resistance from junction to ambient, based on (Kang 2012) variation because the thermal capacity remained reasonably

123
Transactions of the Indian National Academy of Engineering (2022) 7:47–60 51

constant (Xu et al. 2002). Zhihong et al. studied the internal the literature has focused on TIM characterization that
structure of the IGBT module and used a 1D physical model deals with idealized flat surfaces. In practice, the IGBT
to calculate the thermal resistance and thermal capacitance module baseplate is a wavy surface. Therefore, treating
of each layer (Zhihong et al. 2018). They further developed the TIMs as ideal flat surfaces results in improper predic-
a Cauer model to estimate the thermal time constant for tions of thermal performance and reliability aspects. They
their model. They also estimated and compared the impact tested three non-silicone oil thermal greases from three
of coolant temperature in their experiment and their model. different suppliers. They concluded that the best thermal
Bahman et al. (2018) developed a 3D lumped modeling performance was obtained from a thermal grease with the
methodology, including the heat source and heat sink. They smallest particle size (Morris et al. 2015). According to
compared their calculations with experimental results and Semenic et al. large temperature swings are experienced
found a good match. They also found that their models can by the power modules, which results in TIM degradation
be used accurately in the estimation of real-time operations. under harsh environments and can ultimately cause fail-
The work shows that the solder related to the junction to chip ures in the modules; hence, while the selection of TIMs,
region are affected by the power cycling because of thermal degradation with time should be as minimum as possible
impedance variations. Also it is proved by Bahman et al. (Semenic et al. 2017).
that making changes to the cooling system has more effect Wang et al. proposed analytical modeling and system-
on the lower layers of the power electronics module stack. level simulations for single-phase and two-phase cooling
A significant amount of research has been conducted on plates (Wang et al. 2013). They found that the IGBT junc-
the thermal management of power electronics modules such tion temperatures were reduced with R134a as the cooling
as IGBTs. These include air-cooling, liquid-cooling, micro- fluid compared to single-phase liquid cooling. Also, the
channel heat sinks, spray cooling, two-phase forced convec- pressure drop and coolant flow rate required reduced. Two-
tion cooling (Qian et al. 2018). In this paper, the authors phase cooling plates also help reduce the temperature vari-
have also attempted to show how the electrothermal models ations between the die that result from their locations on the
can derive the limits of a cooling technology based on heat baseplate.
flux at the power electronics module base plate and die level. Qian et al. (2018) conducted an exhaustive study of the
Air cooling provides lower effectiveness of heat exchange prevalent cooling technologies and came up with basic rules
between heat sink and ambient compared to liquid cooling. for selecting cooling technology for a particular application.
Kandlikar and Hayner have documented the thermal design These rules are,
and manufacturing considerations for liquid cooling plates
suitable for high-power industrial electronics (Kandlikar 1. The junction-to-case thermal resistance (Rthjc, K/W) of
and Hayner 2009). They classified the cooling plates into IGBT on a heat spread is on or below the line of the fol-
different types. These include the formed tube cold plate lowing equation.
(FTCP) and machined channel cold plate (MCCP). Accord-
ing to Kandlikar and Hayner, power electronics systems that Rthjc = 160P−1 , (1)
employ liquid cooling have to resort to unique solutions.
 where P is the total thermal power (W).
These solutions are dictated by the target boundary condi-
2. The equivalent heat transfer coefficient (h, W/m2 K) of
tions of temperature, pressure drop, materials used for cool-
the integrated metal plate of IGBT is on or above the
ing plate, and choice of cooling fluid. As per the authors,
line of the following equation.
early involvement of thermal engineer in the project would
result in best design and cost reduction. h = 290p0.9 , (2)
The thermal interface materials (TIM), which are soft
and conforming materials at the baseplate- heat sink  where p is the power density or heat flux (W/cm2).
interface- have also attracted the researcher's attention.
These materials reduce the thermal contact resistance at Although the work done by Qian et al. (2018) is very
the module base plate and heat sink or cooling plate inter- comprehensive and the rules appear to be quite simple, some
face. Types of TIM include tapes, pads, gels, grease, and difficulties may present to engineers looking to select a cool-
phase change materials. Jorda et al. characterized TIMs ing technology. The difficulty would arise in the applica-
for IGBT inverter applications. According to their study, tion of the second rule. While the second rule estimates the
datasheets often provide incomplete information, and the minimum heat transfer coefficient required, it would be chal-
thermal performance of the TIMs depends on mounting lenging to know in advance whether the selection of cooling
pressure and surface roughness. Therefore, direct meas- technology can yield heat transfer coefficients greater than
urement of thermal performance through tests should be or lesser than predicted by rule 2. In this perspective, the
carried out (Jordà et al. 2016). Morris et al. mention that present electrothermal modeling work done by the authors

123
52 Transactions of the Indian National Academy of Engineering (2022) 7:47–60

might bridge the gap. It is done by defining the heat transfer • Thermal Interface resistance calculation does not con-
technology roadmap in terms of the below metrics: sider degradation with time.
Heat flux at the module baseplate in W/cm2 and heat flux
at the die level in W/cm2 and.
Heat generation per temperature rise in die (W/cm3)/K. Electrothermal Modeling
These metrics are comparatively much easier to obtain
because, As the power loss within the power electronics module
depends on junction temperature (Fig. 4a), an electrothermal
1. The information about baseplate size is given in the modeling approach is used to simultaneously estimate power
datasheet or can be readily available from the manufac- loss and junction temperature in the present work. The need
turer. for electrothermal modeling can be seen in Fig. 4b. The
2. The die size can be obtained from the power electronics junction temperature limits are reached at a lower junction
component manufacturer. to ambient thermal resistance than the ideal case (Ghaisas
3. The electrothermal modeling can be used to obtain and Krishnan 2021).
power loss at any desired junction temperature. The methodology for estimating the conduction and
switching losses is based on the Application Note from
Once the above information is obtained, the defined met- Semikron (Wintrich et al. 2015). A DC-3 phase AC inverter
rics can be easily calculated based on the power loss and application is considered, as depicted in Fig. 5.
area. Thus, the selection of cooling technologies upfront can An IGBT module with 2 IGBTs and 2 diodes is consid-
become more straightforward. Also, in the present work, a ered in the present case. But the methodology mentioned in
clear differentiation is made between the heat flux limits at the subsequent section can be applied to any other configura-
the die level vs. the heat flux limits at the module base plate tion. The combination of IGBT and diode is referred to as an
level. arm. T1 and D1 refer to the IGBT and the diode. The heat
Although the datasheet of the power electronics compo- losses can be classified as conduction and switching losses.
nent mentions the limit of temperature that the device can IGBT conduction and switching losses are given in Eqs. 3
sustain during continuous operation, the design is seldom and 4 (Wintrich et al. 2015), respectively:
done according to the limits. Considering the reliability of Conduction loss,
the electronics, operating the device at a lower junction tem-
perature is desirable. The difference between the maximum
P = f (Tj)
allowed junction temperature and desired junction tempera- (a) 360
ture is the thermal margin desired for the particular design. 320
The maximum allowable temperature on the heat sink is P = constant
calculated from the power loss and the junction to case and 280
Tj (°C)

thermal spreading resistance from the desired junction tem- 240

perature. So the effectiveness of the heat sink design and 200
selection depends on its ability to maintain the desired junc-
tion temperature for given power loss, coolant flow rate, and 160

material properties. 120

Below are the assumptions for the results of the current 0.4 0.6 0.8 1.0 1.2 1.4
study. Rth (j-a)

(b) 200
• Power loss in all the transistor die is equal, power loss in
all the diode die is equal, resulting in equal distribution 180
of power loss in all the power electronics modules.
Tj (°C)

• The coolant is distributed evenly below all the IGBT 160

modules, and the electrical performance and conversion
efficiencies from all the IGBT modules are constant. 140 Rth(j-a) actual Rth(j-a) ideal
• The focus of the study is mainly on the module type con-
struction of power electronics and not so much on the 120
discrete components. 0.4 0.5 0.6 0.7 0.8
• The increase in temperature of the cooling fluid from Rth (j-a)
inlet to outlet causes negligible variation in thermophysi-
cal properties. Fig. 4  Electrothermal modeling

123
Transactions of the Indian National Academy of Engineering (2022) 7:47–60 53

Fig. 5  Power electronics

modules used in DC- 3 ph AC
inverter

b
( )
1 m cos 𝜙
Qcond(T) = + VCEO (Tj )Ipeak Rth(TIM) = . (7)
2𝜋 8 kA
(3)
For air cooling, the results from previous work done by
( )
1 m cos 𝜙 2
+ + rCE (Tj )Ipeak .
8 3𝜋 the authors is used (Gautam Ghaisas and Krishnan 2019).
For forced liquid cooling, the below methodology is used
Switching loss, (Kang 2012; Ghaisas and Krishnan 2021; Wintrich et al.
√ � � Kv 2015).
2 Iout Vin
Cooling plate,
� � ��
Qsw(T) = fsw EON+OFF 1 + TCEsw Tj − Tref .
𝜋 Iref Vref
(4) 1
Rth(cp) = , (8)
Diode conduction and switching losses are given in mC
̇ p𝜀
Eqs. 5 and 6 (Wintrich et al. 2015), respectively:
Conduction loss, −hAs
.
𝜀 = 1 − e mCp . (9)
( )
1 m cos 𝜙
Qcond(D) =
2𝜋
−
8
VFO (Tj )Ipeak The Nusselt number is obtained from the Dittus–Boelter
( ) (5) correlation as,
1 m cos 𝜙 2
+ − rF (Tj )Ipeak . 0.4
8 3𝜋 (10)
NuD = 0.023Re0.8
D
Pr .
Switching loss,
Spreading thermal resistance ((Rth(sp))),
√ � �K � �Kv
2 Iout I Vin � � �� √ √
Qsw(D) = fsw Err 1 + TCEsw Tj − Tref . Ap − As 𝜆kAp R0 + tan h(𝜆t)
𝜋 Iref Vref Rth(sp) = √ × , (11)
(6) k 𝜋Ap As 1 + 𝜆kAp R0 tan h(𝜆t)
Thermal resistances are estimated as below:
where,
TIM,

123
54 Transactions of the Indian National Academy of Engineering (2022) 7:47–60

100000
𝜋 3∕2 1 TP1
TP2 TP3
𝜆= √ +√ . (12) 10000 h=290(q'')0.9
Ap As
FTCP MCCP

h (W/m2K)
1000 Natural air-cooling
The heat sink, copper base plate and junction tempera- 300 cfm airflow Forced air cooling
tures of the IGBT and diode die are given by the Eqs. 100
Forced liquid cooling (WEG 50%-50%)

(13)–(15), respectively: 10 Two-phase

Qian et al.(Minimum required 'h')

1
( ( ))
Theat sink = Tcoolant,in + Qtotal Rth(hs−coolant) + Rth(spreading) ,
0 10 20 30 40 50 60
(13) q''(W/cm2)

(14)
( )
Tcopperbase = Theatsiunk + Qmodule Rth(TIM) , TP1 : 2 propanol-water (0.05M), Tsat = 84.9 °C, Theatsink= 95°C
TP2 : 2 propanol-water (0.2M), Tsat= 81.2 °C, Theatsink= 95°C
TP3: HFE7100 , Tsat= 60.4°C, Theatsink= 78°C

(15)
( )
Tj = Tcopperbase + Qtotal Rth(j−c) .
Fig. 6  Heat transfer coefficient as a function of heat flux at the mod-
ule base, for a choice of cooling technology
The module-level heat loss is calculated as

(16)
( )
Qmodule = s Qcond,T + Qsw,T + Qcond,D + Qsw,D .
q
h= ( ). (19)
and, total heat loss on the heat sink is given by Abase Theat sink − Tambient∕coolant,in

Qtotal = NQmodule . (17)

A comparison is made between the heat transfer technolo-
The below methodology estimates heat fluxes at the die
gies predicted by the present study and the correlation devel-
level and baseplate level using two-phase technologies.
oped by Qian et al. Natural air cooling, forced air cooling,
For a given choice of working fluid that changes its phase
forced liquid cooling, and two-phase technologies are con-
from liquid to vapor:
sidered in the present study. Figure 7 shows the maximum
allowed heat flux at the module base plate and maximum
1. Saturation temperature (Tsat) is determined at the given
allowed heat generation per temperature rise of the die for a
operating pressure.
given choice of cooling technology. Figures 6 and 7 empha-
2. Heat sink temperature is assumed in the electrothermal
size why natural convection with air can rarely be feasible
model.
for cooling the power modules (~ 0.5 W/ c­ m2 at the module
3. The value of total heat loss is obtained at the desired
base and ~ 4.5 W/ ­cm2 at the die level). Forced air cooling
maximum junction temperature.
can only dissipate around 3 W/cm2 at the module base and
4. Pool boiling correlation by Rohsenow is used, and the
around 22 W/ ­cm2 at the die level.
value of the total heat loss is obtained from above is
Forced air-cooled heat sinks are often combined with heat
plugged to estimate the temperature rise of the heat sink
pipes to dissipate higher fluxes and reduce local hotspots
or cooling plate surface.
by over 20% (Wang et al. 2021). However, liquid cooling
  Rohsenow correlation (Rohsenow 1952),
becomes the frequent choice for cooling power modules.
[ ( ) ]1∕2 ( )3 Liquid cooling plates with simple construction such as
g 𝜌l − 𝜌v Cp,l ΔTe
��
qs = 𝜇l hfg �
. (18) formed tube cooling plates (FTCP) can dissipate around
𝜎 Cs,f hfg Prnl dissipate approximately 25 W/cm2 at the module base and
around 170 W/cm2 at the die level.
5. The process is repeated until the assumed value of heat The performance of liquid cooling plates can be signifi-
sink temperature matches that obtained from the Rohse- cantly optimized using a machined channel cooling plate
now correlation. (MCCP) to 32 W/cm2 at the module base and around 250 W/
cm2 at the die level. However, to achieve the required perfor-
mance and the cooling plate design, the selection of thermal
interface material and an IGBT/SiC stack with minimum
Results and Discussion junction to case thermal resistance becomes essential. For
further enhancement in heat transfer, direct integration of
Figure 6 shows the heat transfer coefficient based on mod- copper base plate with the cooling solution and double-
ule area (consistent with Qian et al. 2018) as a function of sided cooling are a few options. Wang et al. (2015) studied
heat flux, as shown in Eq. 19, the effect of removing the thermal grease layer between the
IGBT and heat sink on the die temperature. As per their

123
Transactions of the Indian National Academy of Engineering (2022) 7:47–60 55

Fig. 7  Maximum allowed heat A- Natural air-cooling

flux at the module base plate B- Forced air-cooling
and maximum allowed heat C- Liquid-cooling (Formed tube cold plate (FTCP)
50 D- Liquid-cooling (Machined channel cold plate (MCCP) or 600
generation per temperature rise Pocketed Folded Fin Technology (PFCP))
of the die for a given choice of E- Two-phase : 2 propanol-water (0.05M), Tsat = 84.9 °C, Theatsink= 95°C 550
cooling technology 45
F- Two-phase : 2 propanol-water (0.2M), Tsat= 81.2 °C, Theatsink= 95°C
G- Two-phase : HFE7100 , Tsat= 60.4°C, Theatsink= 78°C 500
40
450
35

Q'''die/ΔT ((W/cm3)/°C)
400
30 350

q'' (W/cm2)
25 300

20 250
200
15
150
10
100
5 50
0 0
A B C D E F G

study, more than 50% reduction in thermal resistance from dissipating die level heat fluxes around 150–200 W/cm2
junction to the heat sink can be achieved by directly integrat- while still maintaining the die temperatures less than
ing the copper baseplate with the cooling solution, resulting 125 °C. They identified HFE-7100 as the optimum coolant
in enhanced reliability and active power cycling. for this spray cooling. The authors used their previously
An example of a directly integrated cooling plate is shown developed electrothermal model to calculate the maximum
in Fig. 8. However, it should be noted that for a directly inte- heat flux at IGBT die level. The thermophysical properties
grated cooling plate, along with the reduced thermal resist- of HFE-7100 are shown in Table 1, and the IGBT die tem-
ance, sealing the module becomes a challenge as there is perature is maintained at 125 °C. The maximum heat flux
more chance of cooling fluid coming into contact with the predicted at the die level from the electrothermal model is
electronics. around 200.3 W/cm2. If the die temperatures are allowed
Figure 9 shows a double-sided cooling technology sche- to reach typical operating temperature limits of 150 °C,
matic representation (Gillot et al. 2001) for the IGBT mod- the corresponding maximum allowable heat flux at the die
ule. In their work, the microchannel heat sinks were brazed level is estimated to increase to about 296 W/cm2.
onto the DBC substrate. Also, they used the flip-chip tech- The heat fluxes at the IGBT chip level for IGBTs used
nology, thus replacing wire bonds with solder bumps. for hybrid vehicles are close to 100–150 W/cm 2 (Wang,
An important practical aspect that needs to be dealt with Qian). For heat fluxes in this range, liquid cooling and
while designing thermal management solutions for power two-phase cooling promise to be the prime path. In the
electronics for EVs/HEVs is the size of the heat sink or future, the heat fluxes may reach as high as 500 W/cm 2.
cooling plate. As the inverters are mounted on the vehi- For these heat flux levels, two-phase cooling with addi-
cles, the space available to fit the inverter may be limited. tional technological advances needs to be explored. As
Hence, efforts are directed to make the heat sinks compact. per the present study and calculations, natural air -cool-
Figure 10 shows the percentage heat sink area required as ing (heat flux at die level ~ 4.5 W/cm2) may be suitable
a function of cooling technology. The difference between for discrete IGBT or MOSFET components or for com-
heat sink temperature and coolant or ambient temperature ponents that have lower power loss, such as diode mod-
is assumed to be 30 °C for the calculations. Forced air cool- ules). On the other hand, forced air cooling (heat flux at
ing can make the heat sinks around five times compact than die level ~ 22 W/cm2) would be more suitable for inverters
natural air cooling, while forced liquid cooling can make and converters where the output power demands are lesser
the cooling plate around 100 times compact compared with than electric and hybrid electric vehicles. Examples would
natural air cooling. Compact designs help to increase the be solar photovoltaic inverters or inverters used in wind
power densities from inverters or converters. turbines with a power output of about 5 kW.
For hybrid vehicle electronics, Mudawar et al. (2009)
explored using two-phase spray cooling technology for

123
56 Transactions of the Indian National Academy of Engineering (2022) 7:47–60

Fig. 8  Direct integration of

module base with the cooling
solution

Fig. 9  Double-sided cooling

From the study, it is clear that two-phase technologies 1. The saturation temperature of the working fluid should
with the below combinations must be explored to dissipate be below 60 °C—This opens up the possibility of using
heat fluxes at the die level exceeding 300 W/cm2. refrigeration for cooling power modules.

123
Transactions of the Indian National Academy of Engineering (2022) 7:47–60 57

A - Natural air cooling potential solutions for the thermal management of power
B- Forced air cooling electronics modules.
C - Forced liquid cooling - FTCP The cooling system should also be designed to overcome
D - Forced liquid cooling - MCCP the fan or pump's pressure drop across the heat sink or cool-
ing plate. For air-cooled systems, the operating point on the
100 fan curve is determined. For liquid cooling or a two-phase
Percentage of heat sink area

90 system, a pump must maintain the desired coolant volume

Percentage of heat sink area

5
80 4 flow rate. Therefore, the pressure drop across the cooling
70 3 system is vital as it determines the pumping power require-
60 2
ment. For small pumping power requirements, the size and
50 1
space required for the pump can be optimized, and the sys-
0
40 A B C D
tem can be made more compact. Wang et al. (2013) demon-
Cooling technology
30 strated the effectiveness of using a two-phase cooling plate
20 over a single-phase cooling plate. Using R134a as the cool-
10 ing fluid, they maintained the maximum IGBT temperature
0 lower than 68.3 °C, for the same pumping power require-
A B C D ment of 0.3 W, the ethylene–glycol/ water-cooled system
Cooling technology maintained the IGBT maximum temperature at 115.3 °C.
At the calculated pumping power of 0.3 W, the temperature
Fig. 10  Percentage heat sink area required as a function of cooling non-uniformity between the IGBT was reduced from 32 °C
technology for single-phase cooling to 3.9 °C for two-phase cooling.
Temperature uniformity ensures balanced electrical perfor-
mance between different IGBTs.
The main limitation of two-phase cooling is the boiling
Table 1  Thermophysical properties of HFE-7100 at atmospheric
pressure instabilities that occur during operation. As a result, many
researchers have attempted to improve the boiling stabil-
Quantity Value Unit ity using geometric and fluid modifications (Mathew et al.
Saturation temperature (Tsat) 60.4 °C 2021). When the base plate temperature due to the heat flux
Liquid viscosity (μl) 0.000355 N s/m2 from power electronics is less than the saturation tempera-
Latent heat of vaporization (hfg) 112,100 J/kg ture of the cooling fluid, single-phase liquid cooling may be
Gravitational acceleration (g) 9.81 m/s2 more beneficial, provided the cooling solution can maintain
Liquid density (ρl) 1372 kg/m3 the desired junction temperature limits. A heat flux value
Vapor density (ρv) 9.66 kg/m3 will exist beyond which the transition from a single phase
Surface tension (σ) 0.0117 N/m to two-phase flow technology should be adopted. Also, the
Specific heat of liquid (Cp) 1254 J/kg K amount of cooling fluid in the loop should be carefully con-
Thermal conductivity of liquid (kl) 0.0619 W/m K trolled by designing a condenser for two-phase cooling sys-
tems such as a pumpless loop. Also, an accumulator type of
arrangement should be considered to compensate for any
loss in the working fluid due to evaporation.
2. The die material should sustain an elevated operat- Since their introduction, microchannel heat sinks have
ing temperature greater than 150 °C—This opens up stirred much interest in the research community (Tuck-
research related to wideband gap material such as SiC/ erman and Pease 1981). In their work, 790 W/cm2 could
GaN instead of the conventional IGBT modules. be achieved at a temperature rise of 71°C above the inlet
3. In addition, the manufacturers need to create power elec- water temperature. The ceramic substrate on which the sili-
tronics modules with reduced junction to case tempera- con chips are mounted is integrated with the microchannel
tures, switches that require less energy to turn on and heat sink. Microchannel heat sinks offer the advantages of
off. being compact, lightweight, and have a surface-volume ratio
4. Better performing thermal interface materials need to be (Biswal et al. 2009).
developed to reduce the contact resistance further. However, there is a need to consider the heat sink and
heat source sizes used for these studies and compare them
Recently, two-phase pumpless loops with surfactants with the practical sizes of IGBT modules used for elec-
(Rama Raju and Krishnan 2020) and 2-propanol/water tric/ hybrid electric vehicle applications as considered
mixtures (Mathew et al. 2021) are also proposed as in the present study. For example Biswal et al. (2009),

123
58 Transactions of the Indian National Academy of Engineering (2022) 7:47–60

in their study of microchannel heat sinks, heat sink size

considered is 10 mm × 10 mm, microchannel dimen-
sions 100-micron × 300-micron × 10 mm length). For this
size and at a volume flow rate of coolant is 0.18 L/min,
the pressure drop across the microchannel cooling plate
is 12.5 kPa. For the present study, we consider IGBT
modules with dimensions around 50–60 mm in width
and 100–150 mm in length. If there are 3 such modules
mounted on the heat sink, the minimum heat sink length
would be of the order of a few hundred mm. For a micro-
channel heat sink of this size, the pressure drop across the
cooling plate will be at least tenfold (of the order of a few
hundred of kPa), resulting in an increase in the pump size (ΔTmax/P), tcycle=1s σ, tcycle=1s
for even minimal volume flow rates. (ΔTmax/P), tcycle=0.2s σ, tcycle=0.2s
Wang et al. (2013) instead used channel dimensions 0.18 10
(0.5 mm width × 0.5 mm to 8 mm height), length of the

(ΔTdie,max / P) °C/W
0.17 8
cooling plate is 216.4 mm, and the width of the cooling
plate is 81.3 mm. Using R134a as the cooling fluid, they 0.16 6

σ , °C
maintained the maximum IGBT temperature lower than
0.15 4
68.3 °C, for the same pumping power requirement of
0.3 W, the ethylene–glycol/ water-cooled system main- 0.14 2
tained the IGBT maximum temperature at 115.3 °C. An
0.13 0
alternate way to look at this would be if the die tempera- 0 50 100 150 200 250
ture is maintained at 115.3 °C in two-phase cooling, much Power(W)
lower pumping power would be needed.
From the studies mentioned above, it can be summa- Fig. 11  Thermal resistance and temperature swing as a function of
rized that microchannel cooling would be more suitable power for different cycle times
for applications involving cooling high heat flux discrete
components (where the maximum heat sink size is of the
order of a few mm). On the other hand, for cooling power C 10C
1.5
electronics modules (especially for applications such as
(ΔTdie,max / Ptpulse) °C/J

EVs/HEVs, the heat sink size is of the order of a few hun-

dred mm, and heat sources are distributed across the heat
sink). In such situations, single-phase cooling with opti- 1.0
mally designed channels or two-phase cooling systems
would be preferred.
Power electronics systems are often subjected to pulsat- 0.5
ing loads in which the peak power requirements in a short
interval of time may be much higher than the average power
requirements (Jankowski and Mccluskey 2009). The mass
0.0
of the die is much less compared to the mass of the copper
0 10 20 30 40 50
base plate. As a result, the baseplate does not react thermally
to the pulsed load profile while the die immediately starts t/tpulse
responding (the die reaches a steady temperature within
seconds). Due to this difference in the response between Fig. 12  Effect of thermal capacitance on temperature swing
the baseplate and die to the pulsed profile, a temperature
difference or temperature swing develops between the die
and the baseplate. This temperature swing depends upon remains the same as the cycle time changes, the standard
the magnitude of the applied pulse, duration of the applied deviation of temperature swing changes with the cycle time.
pulse, material properties of the module stack (thermal con- Figure 12 shows the effect of thermal capacitance on
ductivity, thermal capacitance), and frequency of the applied temperature swing. The authors considered two stacks in
pulse pattern. Thermal simulations performed by the authors the present work, one with thermal capacitance 'C' and the
show (Fig. 11)that though the thermal resistance of the die other with thermal capacitance ‘10C’. The higher thermal

123
Transactions of the Indian National Academy of Engineering (2022) 7:47–60 59

capacitance clearly can reduce the temperature swings estimated. Notably, the temperature swings can become
exhibited by the die. Similar observations for thermal capac- important for cases where the peak power demands from
itance were recorded by Mathew and Krishnan (2021). It is the inverter or converter for some time durations are signifi-
concluded that power electronics modules such as IGBTs cantly higher than the average power demands. Therefore,
should have stack materials that can offer low thermal resist- power electronics modules should use stack materials that
ance but high thermal capacitance. offer low thermal resistance but high thermal capacitance
Andresen and Liserre (2014) employed active thermal to dampen the temperature swings from a thermal perspec-
management and controlled the maximum die tempera- tive. If possible, active thermal management techniques can
ture and temperature swings. The principle is to reduce be adopted to reduce the device's switching losses during a
the switching losses of the devices in peak power condi- short time interval.
tions. Alternately, the switching losses can be shifted to the
devices that are not subjected to high stresses during that
short time interval. Albatran et al. (2017) showed that it is Funding This work was partially funded by Cummins India.
possible to obtain higher efficiencies from the inverter and
lower case temperatures on the modules by reducing the
switching frequencies at higher power output demands.
References
Albatran S, Smadi IA, Ahmad HJ, Koran A (2017) Online optimal
Conclusions switching frequency selection for grid-connected voltage source
inverters. Electronics (switzerland) 6(4):1–22. https://​doi.​org/​10.​
A cooling technology roadmap is prepared for power elec- 3390/​elect​ronic​s6040​110
tronics modules with specific applications to inverters and Andresen M, Liserre M (2014) Impact of active thermal manage-
ment on power electronics design. Microelectron Reliab 54(9–
converters. The authors' previous work on electrothermal 10):1935–1939. https://​doi.​org/​10.​1016/j.​micro​rel.​2014.​07.​069
modeling is used to develop the roadmap. It is shown that Bahman AS, Ma K, Blaabjerg F (2018) Coupling and thermal bound-
natural air cooling does not work for cooling power modules ary conditions for. IEEE Trans Power Electron 33(3):2518–2530
due to very low heat flux dissipation capability; forced air Bennion K, Thornton M (2010) Integrated vehicle thermal manage-
ment for advanced vehicle propulsion technologies. SAE Techni-
cooling has limited use cases at lower heat fluxes, while cal Papers, no. February. https://​doi.​org/​10.​4271/​2010-​01-​0836
forced liquid cooling is the most preferred technology pres- Biswal L, Chakraborty S, Som SK (2009) Design and optimization of
ently. The thermal resistances due to the thermal interface single-phase liquid cooled microchannel heat sink. IEEE Trans
material and the module stack (junction to case) are also Compon Packag Technol 32(4):876–886
Cougo B, Morais LM, Segond G, Riva R, Tran Duc H (2020) Influence
significant in addition to the cooling plate/ heat sink ther- of PWM methods on semiconductor losses and thermal cycling of
mal resistance. The power loss from the die is temperature- 15-KVA three-phase SiC inverter for aircraft applications. Elec-
dependent and electrothermal models are required to esti- tronics (switzerland). https://​doi.​org/​10.​3390/​elect​ronic​s9040​620
mate the nonlinear behavior of power loss as a function of Ghaisas G, Krishnan S (2019) IHMTC2019-POS-141 analysis of ther-
mal interactions among multiple discrete sources in air-cooled
thermal resistance. For dissipating heat fluxes greater than power electronics for vehicles. In: Proceedings of the 25th
200 W/cm2 at the die level, all the thermal resistances must National and 3rd International ISHMT-ASTFE Heat and Mass
be optimized in the component design and selection stages. Transfer Conference (IHMTC-2019), December 28–21, 2019. IIT
Compact designs help to increase the power densities from Roorkee, Roorkee
Ghaisas G, Krishnan S (2021) Thermal influence coefficients-based
inverters or converters, and cooling technology should also electrothermal modeling approach for power electronics. IEEE
be selected considering the available space to package the Trans Compon Packag Manuf Technol 11(8):1187–1196. https://​
heat sink within the inverter. Also, integrated baseplate and doi.​org/​10.​1109/​tcpmt.​2021.​30960​73
double-sided cooling are additional options for the design at Gillot C, Schaeffer C, Massit C, Meysenc L (2001) Double-sided
cooling for high power IGBT modules using flip chip technology.
these heat flux levels. Adoption of two-phase technologies IEEE Trans Compon Packag Technol 24(4):698–704
can become necessary for heat fluxes close to 250 W/cm2. Gu X, Wei B, Zhang G, Wang Z, Chen W (2019) Improved synchro-
For die level fluxes greater than 300 W/cm2, further research nized space vector PWM strategy for three-level inverter at low
may be needed to be carried out to find die materials that modulation Index. Electronics (switzerland). https://​doi.​org/​10.​
3390/​elect​ronic​s8121​400
sustain elevated temperatures. The use of refrigerants as Hafezi H, Faranda R (2021) A new approach for power losses evalua-
cooling fluids can also be an option. The manufacturers need tion of IGBT/diode module. Electronics (switzerland) 10(3):1–21.
to create power electronics modules with reduced junction https://​doi.​org/​10.​3390/​elect​ronic​s1003​0280
to case temperatures and switches that require less energy Hu Z, Zhang W, Wu J (2019) An improved electrothermal model to
estimate the junction temperature of IGBT module. Electronics
to turn on and off. (switzerland). https://​doi.​org/​10.​3390/​elect​ronic​s8101​066
In addition to the steady-state performance of the power Jankowski NR, Mccluskey FP (2009) Modeling transient thermal
modules, the temperature swings in the die also need to be response of pulsed power electronic packages. In: 2009 IEEE

123
60 Transactions of the Indian National Academy of Engineering (2022) 7:47–60

Pulsed Power Conference, pp 820–825. IEEE. https://​doi.​org/​10.​ Semenic T, Bhunia A, Gollhardt N, Morris G, Lukaszewski R (2017)
1109/​PPC.​2009.​53863​68 Low thermal resistance thermal pad for power converter modules.
Jordà X, Perpiñà X, Vellvehi M, Fernández M, Llorente S, Aranda S In: 2017 IEEE 12th International Conference on power electronics
(2016) Characterization of thermal interface materials for IGBT and drive systems (PEDS), pp 453–55. https://​doi.​org/​10.​1109/​
inverter applications. In: THERMINIC 2016—22nd International PEDS.​2017.​82891​21
Workshop on Thermal Investigations of ICs and Systems, pp 136– Tuckerman DB, Pease RFW (1981) High-performance heat sinking for
140. https://​doi.​org/​10.​1109/​THERM​INIC.​2016.​77490​41 VLSI. IEEE Electron Device Lett 2(5):126–129
Kandlikar SG, Hayner CN (2009) Liquid cooled cold plates for indus- Valentine N, Das D, Pecht M (2015) Failure mechanisms of insulated
trial high- power electronic devices—thermal design and manu- gate bipolar transistors ( IGBTs ). In: 2015 NREL Photovoltaic
facturing considerations. Heat Transfer Eng 30(12):918–930. Reliability Workshop, CALCE, Univeristy of Maryland, pp 1–30.
https://​doi.​org/​10.​1080/​01457​63090​28373​43 http://​ w ww.​n rel.​g ov/​p v/​p erfo ​r mance_​r elia​ b ility/​p dfs/​2 015_​
Lin SC, Chrysler G, Mahajan R, De VK, Banerjee K (2007) A self-con- pvmrw_​131_​das.​pdf
sistent substrate thermal profile estimation technique for nanoscale Wang P, Mccluskey P, Bar-Cohen A (2013) Two-phase liquid cool-
ICs—Part I: electrothermal couplings and full-chip package ther- ing for thermal management of IGBT power electronic module. J
mal model. IEEE Trans Electron Dev 54(12):3342–3350. https://​ Electron Packag Trans ASME. https://d​ oi.o​ rg/1​ 0.1​ 115/1.4​ 02321​ 5
doi.​org/​10.​1109/​TED.​2007.​909039 Wang B, Wang L, Yang F, Mu W, Qin M et al (2021) Air-cooling
Mathew J, Bhagat A, Ghaisas G, Krishnan S (2021) Control of boiling system optimization for IGBT modules in MMC using embed-
instabilities in a two-phase pumpless loop using water-alcohol ded o-shaped heat pipes”. IEEE J Emerg Sel Top Power Electron.
mixtures. J Thermal Sci Eng Appl 13(April):1–7. https://​doi.​org/​ https://​doi.​org/​10.​1109/​JESTPE.​2021.​30549​18
10.​1115/1.​40497​55 Wang Y, Dai X, Liu G, Wu Y, Li D, Jones S (2015) Integrated liquid
Mathew J, Krishnan S (2021) A review on transient thermal manage- cooling automotive igbt module for high temperatures coolant
ment of electronic devices. J Electron Packag 144(1) application abstract. In: Proceedings of PCIM Europe 2015; Inter-
Morris GK, Polakowski M, Wei L, Ball MD, Mark G Phillips, Mosey national Exhibition and Conference for power electronics, intelli-
C, Lukaszewski RA (2015) Thermal interface material evalua- gent motion, renewable energy and energy management, pp 19–21
tion for igbt modules under realistic power cycling conditions. In: Wintrich A, Ulrich N, Werner T, Reimann T (2015) Application man-
2015 IEEE International Workshop on integrated power packag- ual power semiconductors. Semikron. https://​doi.​org/​10.​1109/​50.​
ing (IWIPP), pp 111–114. https://​doi.​org/​10.​1109/​IWIPP.​2015.​ 762905
72959​91 Xu D, Lu H, Huang L, Azuma S, Kimata M, Uchida R (2002) Power
Mudawar I, Bharathan D, Kelly K, Narumanchi S (2009) Two-phase loss and junction temperature analysis of power semiconductor
spray cooling of hybrid vehicle electronics. IEEE Trans Compon devices. IEEE Trans Ind Appl 38(5):1426–1431. https://​doi.​org/​
Packag Technol 32(2):501–512. https://​doi.​org/​10.​1109/​TCAPT.​ 10.​1109/​TIA.​2002.​802995
2008.​20069​07 Yun CS, Malberti P, Ciappa M, Fichtner W (2001) Thermal compo-
Okumura H (2018) A roadmap for future wide bandgap semiconductor nent model for electrothermal analysis of IGBT module systems.
power electronics. MRS Bull 40(5):439–444. https://​doi.​org/​10.​ IEEE Trans Adv Packag 24(3):401–406. https://​doi.​org/​10.​1109/​
1557/​mrs.​2015.​97 6040.​938309
Pou J, Osorno D, Zaragoza J, Ceballos S, Jaen C (2011) Power losses Zhihong W, Xiezu S, Yuan Z (2018) IGBT junction and coolant tem-
calculation methodology to evaluate inverter efficiency in elec- perature estimation by thermal model. Microelectron Reliab
trical vehicles. In: 2011 7th International Conference-Workshop 87(June):168–182. https://d​ oi.o​ rg/1​ 0.1​ 016/j.m​ icror​ el.2​ 018.0​ 6.0​ 18
compatibility and power electronics, CPE 2011—conference pro- Zhu Y, Xiao M, Su X, Yang G, Lu K, Wu Z, Hu Z et al (2007) Power
ceedings, pp 404–409. https://d​ oi.o​ rg/1​ 0.1​ 109/C
​ PE.2​ 011.5​ 94226​ 9 loss and junction temperature analysis in the modular multilevel
Qian C, Gheitaghy AM, Fan J, Tang H, Sun B, Ye H, Zhang G (2018) converters for HVDC transmission systems. J Power Electron
Thermal management on IGBT power electronic devices and 10(3):1097–1108. https://​doi.​org/​10.​1109/​CPE.​2011.​59422​69
modules. IEEE Access 6:12868–12884. https://​doi.​org/​10.​1109/​
ACCESS.​2018.​27933​00 Publisher's Note Springer Nature remains neutral with regard to
Rama Raju VSV, Krishnan S (2020) Experimental investigation of two- jurisdictional claims in published maps and institutional affiliations.
phase pumpless loop with aqueous anionic surfactant as working
fluid. Int J Therm Sci 154(March):106400. https://​doi.​org/​10.​
1016/j.​ijthe​rmals​ci.​2020.​106400
Rohsenow WM (1952) MITLibraries. Trans ASME 74:969

123