4224 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 33, NO.

5, MAY 2018

Direct Single Impinging Jet Cooling of a MOSFET

Power Electronic Module
Johannes Jörg , Silvano Taraborrelli, Student Member, IEEE, Garikoitz Sarriegui, Rik W. De
Doncker, Member, IEEE, Reinhold Kneer, and Wilko Rohlfs

Abstract—This paper presents an approach of direct liquid jet D Nozzle diameter (mm).
impingement cooling of a typical MOSFET power module. A single H Nozzle height (mm).
micro water jet was installed for cooling the upper surface of a IDS Drain-to-source current (A).
MOSFET semiconductor package. In contrast to standard cooling
methods, this approach focuses on hot spot removal omitting any LN Nozzle length (mm).
kind of heat spreading device. The cooling chamber was directly P Pumping power (mW).
soldered to the MOSFET cover that represents a very efficient way Q̇el Electrical power (W).
of liquid cooling. Two different configurations with and without RAdapter Thermal resistance of the adapter plate
electrical insulation (TIM) were used to investigate the importance (cm2 · K/W).
of insulating material. In the range of 10–100 mL/min coolant flow
rates at an inlet temperature of 22.5 ◦ C a maximum power distribu- RJet Thermal resistance of the jet impingement
tion of 51 W (at 30 mL/min) next to a maximum measured MOSFET (cm2 · K/W).
temperature of 163 ◦ C could be realized. Heat transfer coefficients RJet+Adapter Thermal resistance of the jet impingement and
up to 12 000 W/m2 · K were achieved using only 10.8 cm3 of assem- adapter plate (cm2 · K/W).
bly space for the cooling device. With electrical insulation, the heat
t Time (s).
transfer coefficient exceeded 6000 W/m2 · K at a coolant flow rate of
30 mL/min and pumping power of 3 mW. The results illustrate the TI Inlet temperature (◦ ).
potential of direct liquid cooling using impinging microjets in com- TM MOSFET temperature (◦ ).
bination with a compact injection chamber. The individual cooling UDS Drain-to-source voltage (V).
of semiconductors offers new perspectives in the design of power UGS Gate-to-source voltage (V).
electronic modules. V̇ Coolant volume flow rate (mL/min).
Index Terms—Liquid cooling, power electronics, submerged jet ᾱ Spatially averaged heat transfer coefficient
impingement, thermal resistance. (W/m2 · K).
NOMENCLATURE ᾱJet Heat transfer coefficient of the jet impingement
(W/m2 · K).
DBC Direct bonded copper. δBrass Layer thickness of the adapter plate (brass) (mm).
CFD Computational fluid dynamics. δInsulation Layer thickness of the electrical insulation (mm).
GaN Gallium nitrite. Δp Pressure difference (mm).
IGBT Insulated-gate bipolar transistor. λBrass Heat conductivity of brass (W/m· K).
JBS Junction barrier Schottky. λC Heat conductivity of the coolant (water) (W/m· K).
MOSFET Metal–oxide–semiconductor field-effect λInsulation Heat conductivity of the electrical insulation
transistor. (W/m· K).
PCB Printed circuit board. ν Kinematic viscosity.
SiC Silicon carbide. Re Reynolds number (−).
TIM Thermal interface material. Nu Average Nusselt number (−).
AM Water covered surface area (mm2 ). Pr Prandtl number (−).

Manuscript received December 19, 2016; revised April 24, 2017; accepted I. INTRODUCTION
June 12, 2017. Date of publication June 29, 2017; date of current version Febru-
ary 1, 2018. This work was supported by the Deutsche Forschungsgemeinschaft OWER electronics have proven to fulfill the rising
(DFG, German Research Foundation) within the framework of the mobilEM
Post Graduate Program under Grant GRK 1856. Recommended for publication
by Associate Editor S. Ang. (Corresponding author: Johannes Jörg.)
P demands on control, energy efficiency, and reliability in an
increasing number of applications throughout the last decade.
J. Jörg, R. Kneer, and W. Rohlfs are with the Institute of Heat and Mass Trans- For future electronic application it is assumed that all electronic
fer, RWTH Aachen University, Aachen 52056, Germany (e-mail: joerg@wsa.
rwth-aachen.de; [email protected]; [email protected]). power will be controlled by at least one power electronic de-
S. Taraborrelli, G. Sarriegui, and R. W. De Doncker are with the Insti- vice [1]. In power electronic applications, power loss caused by
tute for Power Electronics and Electrical Drives, RWTH Aachen University, switching and conduction losses leads to extensive heat gener-
Aachen 52066, Germany (e-mail: [email protected];
[email protected]; [email protected]). ation confined to small areas. This heat needs to be removed
Digital Object Identifier 10.1109/TPEL.2017.2720963 by adequately designed heat sinks. Heat sinks have to provide a

0885-8993 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
JÖRG et al.: DIRECT SINGLE IMPINGING JET COOLING OF A MOSFET POWER ELECTRONIC MODULE 4225

sufficient surface area for convection, radiation, and conduction Deeper insights into these challenges of power modules’ ther-
as well as take into account steady state as well as transient mal management are presented in Leslie [8]. In addition to the
operating conditions. Heat sink designs focus on extensive heat high coolant temperature level the temperature difference be-
spreading. In common cooling applications, heat is dissipated tween the die (e.g., IGBT or MOSFET) and the coolant should
by conduction through base plate and heat spreader, followed not exceed 30 ◦ C. This ensures high-dissipating heat fluxes
by convective heat transfer via the coolant (air, water, or oil). whereas the IGBT or MOSFET is held below the maximum op-
Especially in transient operating conditions reducing the inter- erating temperature.
nal thermal resistance of a heat sink moderates the intensity of Besides temperature range, the required coolant pumping
peak heat fluxes [1]. power should be considered. Pumping power, represented by
Considering the increasing power density in contrast to the pressure drop of the heat sinks, is an important indicator
weight and space reduction of power electronic modules com- of the efficiency of the cooling device. Leslie [8] compared
pact and highly efficient cooling has become mandatory. As an two different cooling concepts: pin fin and microchannel cool-
example, automotive powertrain systems include power densi- ing. The author specifically points out the difference of two
ties of transistor packages with MOSFET, such as dc/dc convert- common flow arrangements (parallel or normal to the heat gen-
ers, that already approach heat fluxes up to 300 W/cm2 [2]. eration surface) regarding the subsequent pressure drop and
cooling performance. Contrary to parallel flow concepts, nor-
mal flow heat sinks that are associated with lower pressure drops
A. State-of-the-Art Systems and Future Challenges often necessitate short flow passages. This highlights that high-
Today’s power electronic and semiconductor devices as well performance cooling concepts also depend on a small pressure
as the future trends are reviewed by Benda [3] and Vobecky drop and low thermal resistance in the heat flow path.
[4]. By focusing on fundamental physical limits, these works A promising approach to a compact and highly efficient cool-
discuss the most important power semiconductor devices and ing system is represented by microchannel cold plates, such as,
possible future application. The authors point out that standard e.g., presented by Valenzuela and colleagues [9]. Microchannel
electronics, in general, and high-power electronics, in particu- cold plate designs focus on the integration of a normal flow
lar, will utilize silicon switching devices. While for the voltage configuration into an IGBT module and eliminate the necessity
range of 1–3 kV IGBT may remain the most important switch- of a heat spreader and TIM. Instead, the microchannel cooler is
ing devices the voltage range below 330 V may be dominated directly soldered to the direct copper bonded (DCB) device. It is
by power MOSFETs. Power electronic modules using IGBTs, thus characterized by very low thermal resistance and moderate
especially in combination with SiC or GaN Schottky or JBS pressure drop. Using a microchannel cold plate cooler results in
diodes, can handle higher junction temperatures. However, the very high heat dissipation rates (300 W/cm2 ) with a tempera-
operating temperature of power electronic modules and its semi- ture difference below 50 ◦ C. This illustrates the high potential
conductors should not exceed the critical temperature level for of normal flow configurations in combination with the reduction
save operations of 175 ◦ C [5]. of thermal resistance.
Today’s commonly used cooling solutions are listed as A thermohydraulic comparison of two liquid cooling devices
follows: under real application conditions is given by Robinson [1]. This
1) air-cooled heat sinks (natural and forced convection); work contrasted the performances of jet arrays and microchan-
2) liquid cold plates (single and double-sided cooling); nels. Furthermore, the required pumping power for a desired
3) microchannel coolers; heat removal rate was considered to illustrate the possible op-
a) liquid coolers built into the base plate; erating costs of each approach. Both cooling devices have to
b) liquid coolers integrated within the DBC substrate; dissipate a heat flux of 250 W/cm2 while the temperature of the
4) direct contact liquid cooling of the module base plates or power electronics (heated surface) is kept below a temperature
DBC substrate; of 85 ◦ C. The analysis focused on heat transfer coefficients, pres-
5) jet impingement cooling of the module base plates; sure drop, and pumping power of both cooling strategies. Pump-
6) two-phase liquid cold plates (phase change of a dielectric ing power was limited to 0.1 W, which restricted the maximum
refrigerant coolant at constant temperature). flow rate and, accordingly, the pressure drop. The flow rates of
Especially in the automotive sector, additional requirements both cooling strategies were adjusted to dissipate the mentioned
for power electronic cooling have to be taken into account. In heat flux and the hydraulic diameter of the microchannels, as
hybrid cars, e.g., compact packaging and cost limits, as well as well as the configuration of the jet arrays (number of jets and
the interconnection of peripheral components such as installed distances among them), were optimized.
cooling cycles have to be considered. This results from the fact Robinson [1] could show that microchannel coolers, which
that the temperature of the supplied coolant is in the order of are known for high pressure drops, have to be operated at
90 ◦ C. Here, the difference between inlet and outlet temperature small coolant flow rates to meet the restrictions of low pumping
of the power electronic cooling system has to be sufficiently low power. Performance levels of the microchannels indicate that the
to prevent from coolant boiling. Schulz-Harder [6] and Saums heat transfer coefficient is strongly influenced by channel width
[7] have provided a summary on state-of-the-art cooling solu- and that it decreases with the hydraulic diameter. Raising the
tions for power electronic modules in automotive applications flow rate through microchannels will increase heat transfer but
focused on thermal management. also pressure drop and, therefore, pumping power. In contrast,
4226 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 33, NO. 5, MAY 2018

impinging jet coolers can achieve high coolant flow rates due requiring a coolant flow rate of 2.5 L/min. The small dimen-
to their lower pressure drop. Here, the heat transfer coefficient sions of the treelike manifold and inlet nozzles cause a pres-
also increases with flow rate. In addition, for fixed nozzle di- sure drop of 0.35 bar. In summary, the authors demonstrate
ameters and a constant total flow rate the heat transfer coeffi- an impressive cooling performance, however, at a cost of high
cient decreases with a larger number of jets. In contrast to a pressure drop.
worse heat transfer coefficient, a large number of impinging A reliability and performance study on an impinging jet heat
jets decreases the pressure drop and, thus, the required pumping exchanger design using water-ethylene glycol (mixture by vol-
power. Concluding, under comparable operating conditions, the ume: 50%–50%) is presented by Narumanchi and colleagues
heat transfer coefficient of impinging jets is significantly higher [18]. Within the context of a commercially used inverter for
compared to microchannel coolers [1]. electric vehicle applications a channel flow-based cold plate is
Highlighting the challenges concerning reliability and main- compared to an impinging jet heat sink. Based on experimental
tenance in the automotive sector Robinson [1] further states that and numerical investigations, the results show that in compa-
lower operating pressure decreases the risk of leakage. At the rable coolant flow ranges (approximately 10 L/min) the ther-
same time, high flow rates lead to a more uniform temperature mal resistance of the impinging jet configuration is 34% lower
across the cooled surface of the electronic component. Valen- compared to the channel flow-based cold plate, the tempera-
zuela and colleagues [9] anticipated that power devices using ture uniformity significantly decreases, and the specific power
SiC or GaN semiconductor technologies, operated at higher increases by 82%. Further improvements are achieved using
switching frequencies and currents, will exceed heat generation a microfinned base plate for the impinging jet cooling (34%
rates of 500 W/cm2 and further improvements of the cooling in thermal resistance, 0.15 ◦ C of temperature uniformity, and
concepts are required. 118% in specific power). After 6 month contentious reliability
In order to overcome the theoretical maximum heat removal testing no significant regression of the thermal performance of
rate of liquid coolants phase changing concepts can provide a the impinging jet heat exchange, with and without microfins,
possible solution. One promising cooling concept is presented is reported. Apart from supporting the superiority of impinging
by Levett and colleagues [10]. The authors reviewed experi- jets’ cooling performance, the results also support the durability
mental results of IGBT modules equipped with a cold plate and of heat sinks using liquid impinging jets.
a vaporizable dielectric fluid. The results show that the main Common problems of jet arrays are the interaction of sur-
cooling effect is based on the liquid-to-gas phase change of rounding nozzles with the impingement as well as coolant
common refrigerant such as R134-A. As long as the two-phase cross flows. Depending on the position of the outlet nozzles
fluid runs liquid through the cold plate, the temperature level coolant cross flow perpendicular to the impingement direction
stays close to the refrigerant boiling temperature. At a flow rate can redirect the impinging jets in downstream direction. Both
of 350 mL/min a power distribution of 1 kW is achievable. effects, jet interaction and cross flow, can inhibit fully developed
Thus, low flow rates, small pumps, and reduced diameter tubes wall jets and, thus, decrease the heat removal rate of the imping-
can be used. However, as a disadvantage in contrast to typical ing jet heat sink. In their study, Han and colleagues [19] present
coolant concepts, such a phase change cooling concept includes a hybrid microcooler, which combines microjet array impinge-
the usage of a greenhouse gas (R134-A). ment with microtrenches for drainage. These microtrenches are
used to achieve fully developed impinging jets for all individual
nozzles. The hybrid microcooler is fabricated by bonding two
B. Submerged Impinging Jet Cooling Concepts silicone plates together and etching processes. The nozzles have
Over the last decades, impinging jet cooling has been ex- a diameter of 100 μ m and the microtrenches have a width of
tensively studied. Focusing on single-phase liquid jets, these 150 μm. Overall, more than 300 nozzles and 180 trenches are
studies have provided a wide range of experimental, theoretical, implemented into the heat sink for cooling an area of around
and numerical analyses of different impinging jet configurations 50 mm2 . For a coolant flow rate of 300 mL/min a pumping
[11]–[13]. The investigations range from single submerged jets power of 0.05 W is required to achieve heat transfer coefficients
[14], to multiple submerged jets [15], and impinging jet cooling in the range of 13 000 W/m2 · K. In conclusion, the authors can
of electronic modules [16]. demonstrate that the heat transfer performance of the impinging
The wide range of impinging jet configurations and their jets can be improved by the use of microtrenches, and by this,
cooling performance in real applications is presented by the the amount of nozzles can be reduced.
following studies. Finally, Gould and colleagues [20] present a high heat flux
A performance investigation of a submerged single-phase jet impingement cooled heat exchanger designed for harsh
direct liquid-jet-impingement cold plate using a large number operating environment (e.g., underhood installation in heavy
of impinging jet nozzles and water as a coolant is given by duty electrified vehicles). The heat exchanger was applied to a
Brunschwiler and colleagues [17]. For nozzle diameters be- 600 V/50A silicon (SiC) power module and optimized by experi-
tween 25 and 126-μm pressure drop, heat transfer coefficient, mental studies. The authors compared the impinging jet cooling
and heat removal rates were investigated feeding up to 50 000 heat sink with two commercial heat sinks (liquid-cooled heat
nozzles cooling a chip area of 400 mm2 . The results show sink with meandering copper tube and microchannel cooler).
that a maximum heat transfer coefficient of 87 000 W/m2 · Using a mixture of water-ethylene (50%–50%) as coolant, a
K could be achieved using a nozzle diameter of 43 μm and flow rate of 195 mL/min and device junction temperature limit
JÖRG et al.: DIRECT SINGLE IMPINGING JET COOLING OF A MOSFET POWER ELECTRONIC MODULE 4227

of 175 ◦ C the commercial heat sinks proofs to be insufficient. In Engelmann and colleagues [23] focus on the experimental in-
a worst case scenario with 151 W heat load the junction temper- vestigation of thermal resistances under real application condi-
ature increases up to 290 ◦ C for the copper tube heat sink and tions. This study includes solid materials (e.g., Al2 O3 ceramic),
up to 215 ◦ C for the microchannel heat sink. The optimized im- elastomeric materials (e.g., Gap PAD 1500) as well as thermal
pinging jet heat sink reduces the junction temperature to 169 ◦ C greases (e.g., Akasa Silver Compound 450). The used test setup
using the same coolant flow rate. The results in pressure drop consisted of a surface mounted MOSFET pressed on a heat sink.
and heat removal rate indicate that the best cooling performance The applied MOSFET was a DirectFET (L8 package).
can be achieved by one jet per heat source. With an operation The results of this study show that elastomeric TIMs lead
time of up to 50 h, this work highlights the long-term reliability to thermal resistances below 10 K/W, while solid TIMs (e.g.,
of impinging jet heat sinks. It can further be stated that heat Al-oxide) provide further reduction of the thermal resistance.
removal rates beyond 350 W/m2 can be realized, even in harsh The best results (below 5 K/W) are achieved with the combi-
environments. nation of solid TIMs with thermal greases and phase-changing
materials.
C. TIM and Direct Cooling Furthermore, the authors highlight the influence of the me-
Regardless of the employed cooling concept internal ther- chanical boundaries as a limiting factor to the heat removal rate
mal resistance of the semiconductor and the TIM becomes a of a heat sink. The measurements of the pressure distribution
dominating obstacle in improving overall cooling performance. on the DirectFETs of a dc/dc converter pressed against a planar
Narumanchi and colleagues [21] have stated TIMs to be even aluminum heat sink indicate nonhomogeneous heat conductivity
the main bottleneck to realize high heat flow rates from an IGBT between the DirectFET and the planar base plate of the heat sink.
or MOSFET into the coolant. In the automotive sector, a compact This poses an additional caveat that has to be considered when
design of a power electronic module relies on the each compo- all semiconductors of a power electronic module are cooled by
nent’s performance. Especially for the three main components, one base plate.
heat spreader, heat exchanger (microchannel or impinging jet In summary, pumping power, pressure drop, and total ther-
cooler), and TIM, the main performance measures consist of mal resistance are key characteristics when designing an ef-
their reliability, high heat removal performance, and cost effi- ficient and compact power electronic cooling device. Related
ciency. Therefore, the reduction of the TIM thermal resistance to the pumping power, small coolant flow rates in combination
is a promising approach to achieve these goals. with low pressure drops result in small power consumptions and
Taking into account numerous limiting conditions, e.g., a high efficiencies. Low thermal resistance can be realized by in-
maximum coolant temperature of 105 ◦ C and the temperature creasing corresponding heat transfer coefficients and removing
limitations of the device, Narumanchi and colleagues [21] inves- material layers in the heat flow path. Overall, a designer of a
tigated different electronic devices (conventional silicon-based, power module and its cooling device has to take into account
SiC-based, and IGBT) in combination with several TIM ma- several tradeoffs when selecting an adequate heat sink: the type
terials and thicknesses. The results of this study show that the of coolant and coolant flow rate, pressure drop and thermal re-
thermal resistance of the investigated materials achieves a mini- sistivity, compact size, and low weight, as well as the cost and
mal value of 0.33 cm2 · K/W. The thicknesses of the tested TIM reliability [8].
layers (bond-line thickness) are 100 μm or thinner. In contrast to
heat spreader, base plate, or heat sink, any of the thin TIM layers
D. Direct Single Impinging Jet Cooling
significantly increases the overall thermal resistance. Thus, for
extremely compact and high performance power electronics the Common impinging jet concepts, such as those presented in
usage of any kind of TIM should be avoided. Section I-B use a large number of small and short orifice-type
Direct cooling of a power electronic modules fulfills this nozzles to generate jet impingement. A high number of nozzles
requirement. The cooling fluid is guided directly inside the base- and small diameters come along with disadvantages, such as
plate of the module. Compared to the systems discussed above, high production efforts and challenging technologies for the
direct cooling reduces the distance between heat generation lo- manufacturing and maintenance. As new contribution to direct
cation and dissipating surface. Thus, thermal resistance can be liquid jet cooling, the present study uses one single nozzle with
reduced and efficiency is increased. This is supported by the larger diameter. Hereby, the design of the injection manifold is
results of Schulz-Harder and colleagues [22], who focus on the simplified and the size of the heat sink is reduced.
benefits of replacing a standard air-cooled heat sink with TIM Furthermore, the study applies one long tube nozzle, offering
by a liquid-cooled base plate and jet impingement cooling. The benefits regarding hot spot removal and heat transfer rates. As
results illustrate the negative influence of the TIM on overall shown before by our group, the velocity profile has a significant
thermal resistance, as well as the heat sink performances. The influence on local Nusselt number in the stagnation region of
authors conclude that changing from air cooling to direct liq- the impinging jet [24]. For the same Reynolds number (identical
uid cooling leads to significant reduction of thermal resistance flow rate), a parabolic velocity profile emerging from the cooling
between the semiconductor and the coolant. nozzle exhibits a much higher capability for heat removal than a
Despite the obvious disadvantages caused by TIMs, their us- uniform velocity profile. While long tube nozzles offer enough
age cannot be avoided if a water-based coolant is used and hydraulic length to form a fully developed velocity profile of
electrical insulation is required. However, new technologies and parabolic shape orifice-type nozzles induce uniform velocity
materials enable certain improvements of TIMs. profiles.
4228 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 33, NO. 5, MAY 2018

Fig. 1. Left picture shows the DirectFET Power MOSFET package mounted
on a PCB. The right panel displays a schematic cross section of the DirectFET
package equipped with a standard plate cooler and TIM.

Instead of mounting a PCB on the cooled base plate, heat

spreader, and heat sink, a single MOSFET semiconductor is cooled Fig. 2. Cross section of the two investigated cases, differing in the mounting
by a directly attached single cooling chamber. This sets way for to the MOSFET. Case A (left side): Thin wall chamber housing directly soldered
to the MOSFET top surface. Electrical conduction between MOSFET drain and
new possibilities in the design and layout of power electronic coolant. Case B (right side): Thick wall chamber housing (due to fixing screws)
modules due to the separate cooling of each MOSFET. Fig. 1 including an electrical insulation layer, mounted on top of a housing adapter.
illustrates the experimental setup. This adapter is soldered to the top surface of the MOSFET. In this case, there is
no electrical conduction between MOSFET drain and coolant.
The MOSFET, used as switching device of a dc–dc converter,
is mounted as DirectFET Power MOSFET package [25] on a PCB
the phenomena taking place on the MOSFET surface and inside
and is cooled by a common cold plate. Fig. 1 left panel illus-
the cooling chamber; the focus here is on fluid dynamics and heat
trates the aluminum metal cover of the DirectFET package as
transfer analysis. Section IV consists of six sections. First, in-
well as its dimensions. Fig. 1 right panel illustrates the cross
frared measurements of the MOSFET top cover during the heating
section of the DirectFET package equipped the plate cooler.
phase are discussed, illustrating the spatial heat generation of the
The DirectFET consists of bottom connections for the gate and
MOSFET. The second and third sections address the influences of
source supply, diode, and MOSFET in the center and drain via
nozzle length on pressure drop, pumping power, and heat trans-
the top cover. The package height is below 0.5 mm. The stan-
fer coefficient. In addition, the electrically noninsulated (Case
dard heat sink for this application consists of a liquid-cooled
A) and electrically insulated (Case B) setups are compared re-
heat sink with meandering copper tube, comparable to the heat
garding their heat transfer performance. Nusselt correlations for
sink investigated in [20]. It consists of a cooper base plate
both setups are presented in comparison to fundamental research
(heat spreader) and tube for the coolant flow. The heat sink
results on impinging jets. Furthermore, a detailed insight into
is pressed against the MOSFETs of the converter with TIM in
the thermal resistance of the two setups highlights the drawback
between.
of the additionally used electrical insulation layer. Section IV
The study presents first results on the heat dissipation perfor-
presents a failure scenario, in which a break-down of the coolant
mance of this approach. Taking into account the challenges for
supply is examined. It highlights the thermal response in case of
a compact, reliable, and highly efficient heat sink the presented
potential shutdown of the coolant flow. Section IV-F evaluates
cooling device is attached directly to the MOSFET top cover,
the efficiency of the impinging jet cooling approach and com-
thereby meeting the demands stated by Brunschwiler and col-
pares the required pumping power to the achieved heat transfer
leagues [17] who have suggested this low-temperature bonding
coefficient. Section V concludes the study.
process. The setup further uses single microjets, well-known for
high heat transfer coefficients. This setup offers the opportunity
to minimize the travel path length of the heat and, thus, ends up II. EXPERIMENTAL SETUP
in extremely compact chamber housing. The experimental setups used to examine the effect of di-
As illustrated in the cross sections of Fig. 2, the study includes rect liquid cooling of a MOSFET (DirectFET [25]) is shown in
two different test setups. The setup that is electrically noninsu- Figs. 3–6.
lated (Case A) is used to determine the upper limits of the pre- The core unit consists of a 12 mm × 10 mm cooling cham-
sented impinging jet cooling device. As the coolant is in direct ber, which is 9 mm in height (shown in Fig. 3). The chamber
contact with the top cover of the MOSFET, the thermal resistance itself is made of brass, and comprises a chamber housing and a
of the heat sink only depends on the heat transfer coefficient. separate top cover (both connected to each other by four remov-
The second setup (Case B) is equipped with an insulation layer able screws). The MOSFET and chamber housing were brazed
and mounting adapter to guarantee electrical insulation between jointly in one single step onto the PCB in a brazing furnace.
the cooling cycle and the power electronic module. The advantage of this manufacturing process is that the elec-
The remainder of this article is structured as follows: trical contacts will not be affected due to a second brazing
Section II outlines the concept of direct liquid jet cooling in the stage. Furthermore, the material connection is water-proof and
experimental setup. Section III provides a detailed insight into allows for small unit size. The separate top cover of the cooling
JÖRG et al.: DIRECT SINGLE IMPINGING JET COOLING OF A MOSFET POWER ELECTRONIC MODULE 4229

Fig. 4. Pictures of the investigated injection chamber housing, additionally

equipped with an adapter plate (Case B), electrical insulation (TIM), and slightly
changed housing outline. The left picture depicts the housing adapter soldered
onto the MOSFET, the right picture illustrates the completely mounted single jet
Fig. 3. Left picture depicts the electrically noninsulated cooling chamber cooling chamber.
(Case A), comprising injection and outlet nozzles, as well as electrical connec-
tion. The right picture shows the inside of the cooling chamber and the top cover
of the MOSFET. This setup was used for the IR-thermography measurements.
high thermal conductivity and insufficient thickness the thermal
resistance of the soldering layer is negligible.
chamber contains three holes with a diameter of 0.9 mm each. The pre-assembled MOSFET, PCB, and housing adapter are
The center hole is used for cooling supply through a LN = 12 shown in the left picture of Fig. 4. The completely mounted
mm or LN = 24 mm long cooling nozzle. The inner diameter of cooling device is presented in the right picture. In between the
all nozzles is D = 0.6 mm. The inlet height is H = 3 mm above housing adapter and chamber housing a thin layer of the elec-
the heat transferring surface (DirectFET top cover (CASE A) trical insulation pad (Gap PAD [26]) is visible. The thickness
or cooling chamber bottom surface (CASE B)). The shape and of the insulation pad consists of approximately 200 μm. Taking
length of the cooling nozzle influences the emerging velocity into account its thermal conductivity of 5 W/m· K the effective
profile of the cooling liquid and, hence, the local heat transfer. heat transfer coefficient is approximately 2.5 W/cm2 · K. Com-
To remove the cooling liquid from the chamber the two addi- pared to the thermal conductivity of the brass (around 120 W/m·
tional holes include short outlet pipes of same diameter, but at K) and the aspired goal of 300 W/cm2 dissipated heat fluxes the
a length of 10 mm. The prototype cooling nozzles were taken additional insulation pad with its low heat transfer coefficient
from dental medical supplies and have large connectors (yellow was expected to significantly contribute to the thermal resistance
plastic), allowing for a simple connection to a flexible hose sys- of the cooling device.
tem. The nozzles are glued to cooling chamber cover. In real-life The two different test cases (Case A and Case B) allow for
applications, these connectors can be of much smaller size. The a separation of the aspects attributed to the impinging jet heat
electrically noninsulated setup (Case A) was used to determine transfer from the aspects related to TIM and heat spreading.
the influence of the jet impingement parameters regarding the To compare the results of the electrically noninsulated setup
cooling performance of the device. To exclude convective and (Case A) to the electrically insulated setup (Case B) the in-
conductive heat dissipation to the surroundings, i.e., through the ner design of both cooling chambers is identical (top cover
PCB and chamber housing outer surface, the entire experimen- as well as nozzles positions are identical). Thus, the flow and
tal setup is thermally insulated. Thus, the cooling performance convective heat transfer characteristics of both test setups are
is determined only of the direct impinging jet cooling. This equivalent.
procedure is employed in both test cases. Fig. 5 provides an overview of the main elements of the cool-
The electrically insulated setup (Case B) is shown in Fig. 4. To ing system. Fed by the inlet manifold, the coolant leaves the
ensure a certain proximity to real-life applications, the second center nozzle perpendicular to the MOSFET surface and impinges
setup includes a layer of electrical insulation. Therefore, the at the center of the planar aluminum top cover. It then rebounds
chamber housing could not be directly soldered onto the MOSFET sideways and accelerates along the surface. The MOSFET gener-
top cover. Instead of the chamber housing, a housing adapter was ates heat, which is removed by the wall jet. Thus, the coolant
placed on top of the MOSFET and brazed jointly to each other in flow heats up along the bottom of the cooling chamber and is
the soldering furnace. This procedure enables preassembling of redirected upward by the chamber wall. Besides the convective
MOSFET + PCB + adapter. Also, the entire cooling cycle (piping, heat transfer on the lower surface an additional cooling effect
chamber housing, etc.) can be assembled separately from the could be represented by heat loss via the chamber walls and top
electrical devices. Because it holds additional plastic mounting cover. Also, mixing of the hot rebounded coolant and the re-
screws the size of the chamber housing slightly increased. Due to maining fluid inside the cooling chamber can occur. Finally, the
4230 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 33, NO. 5, MAY 2018

Fig. 7. Typical on-resistance versus gate voltage of the investigated MOSFET

[25] for a drain-to-source current of ID = 96 A and junction temperatures of
Fig. 5. Schematic display of the investigated injection chamber for direct
T J = 25–125 ◦ C. The scope of experiments is highlighted by the gray-shaded
impinging jet cooling as well as dimensions and electrical connections. The
area.
heat loss of the MOSFET is directly dissipated by the impinging jet.

connection of several cooling chambers the temperature differ-

ence across one cooling device needs to be determined. Im-
mediately prior and after the cooling chamber inlet and outlet
temperatures are measured using a Type-K thermocouple. The
temperature of the MOSFET is measured at the bottom surface of
the semiconductor. A probe hole was drilled through the PCB
base plate and the tip of thermocouple is pushed from below
against the MOSFET surface. The electrical circuit comprises of
an adjustable power source (current controlled) that imposes a
defined gate-to-source current. The electric power input of the
MOSFET is given by

Q̇el = UDS · IDS (2)

whereby the drain-to-source voltage UDS and the respective

current IDS are continuously monitored during the experimen-
tal runs. A second power source is applied to control the gate
Fig. 6. Schematic display of the cooling circuit, measurement, and electrical
setup. Both cooling chamber setups (with and without electrical insulation) were voltage of the MOSFET, which allows for a precise regulation
supplied by this experimental periphery. of the electrical loss of the MOSFET. The relation between the
electrical resistance and the gate-to-source voltage for different
coolant exits through the outlet nozzles. For both nozzle lengths temperatures is shown in Fig. 7.
(12 mm and 24 mm), the distance between the injection nozzle The applicable voltage level of the cooling approach is re-
and the heat dissipating surface (height) is 4 mm. stricted by the electrical insulation. Since in Case A the coolant
The entire cooling circuit is presented in Fig. 6 and con- (water) is in direct contact with the drain of the DirectFET
sists of a reservoir for liquid storage, from which the liquid package the applicable voltage level depends on the electrical
is pumped (adjustable gear pump Gather Industrie PM8060) conductivity of the coolant. In Case B, the applicable voltage
through a filter, a preheater (HAAKE AC 200), and a flow meter level depends on the electrical conductivity of the TIM. The used
(PKP DTH08) to the cooling chamber. Special attention was Gap PAD [26] has a dielectric breakdown voltage of 5000 V.
paid to highly accurate flow rate measurements, for which a For the experimental investigations, the drain-to-source voltage
calorimetric flow meter was employed. The coolant flow rate is UDS did not exceed 30 V.
adjusted by a high precision valve. Based on liquid flow rate V̇ ,
nozzle diameter D and kinematic viscosity of the fluid ν, the III. FLUID DYNAMICS AND HEAT TRANSFER CHARACTERISTICS
dimensionless Reynolds number is given by This study focuses on the investigation of a small-scale and
V̇ · D compact cooling device of a single MOSFET semiconductor. The
Re = . (1) heat generating area and, thus, the surface for the convective
ν · D2 · π4
heat transfer is extremely small. The water-covered surface area
For future applications, the temperature levels of the inlet AM is confined to the bottom surface of the cooling chamber.
and outlet coolants are of great interest. Especially for a series For the first test setup (Case A), it is equal to the water-covered
JÖRG et al.: DIRECT SINGLE IMPINGING JET COOLING OF A MOSFET POWER ELECTRONIC MODULE 4231

area of the MOSFET top surface (right picture of Fig. 3), with a
value of 40 mm2 . To provide a reasonable comparison of the two
test setups, the water covered surface area AM for the second
setup was defined the same way and, thus, features the same
value. The dissipated heat is defined by the power loss of the
MOSFET, induced by low gate-to-source voltages UGS . The gate-
to-source voltage ranges between 4.58 and 4.74 V, leading to
sensitive changes in resistance as given in Fig. 7. The heat loss of
the MOSFET is presented by the electrical power density Q̇el , as a
function of the drain-to-source current IDS and drain-to-source
voltage UDS .
Based on the temperature of the MOSFET, the water covered
surface AM , the electric power density Q̇el , and the inlet tem-
perature of the cooling liquid the spatially averaged heat transfer
coefficient is determined by Fig. 8. Distribution of the MOSFET surface temperature change after 0.01 s of
heating. The heat dissipation is caused by the power loss of the semiconductor,
Q̇el operated on low gate-to-source voltage V G S .
ᾱ = . (3)
(TI − TM ) · AM
This heat transfer coefficient can be translated to the dimen-
sionless Nusselt number
ᾱ · D
Nu = (4)
λC
where λC is the thermal conductivity of the cooling liquid. Based
on these equations the performance of the cooling system will
be determined in the following sections. Further analyses of the
fluid dynamics inside the cooling chamber are provided by the
results of the numerical investigations.

IV. RESULTS
In the first part of this chapter, the local heat generation of the
investigated type of MOSFET is analyzed using IR-thermography
Fig. 9. Influence of the nozzle length and coolant flow rate on the spatially
(see Section IV-A). In Section IV-B, the pressure drop and re- averaged heat transfer coefficient of the cooling chamber without electrical
quired pumping power are discussed based on numerical simu- insulation (Case A).
lation results of the respective cooling system. In Section IV-C,
the heat transfer characteristics of the respective chamber setup
documenting the measurement with an IR-camera. A frame rate
(Case A and B) are discussed and Nusselt number correlations
of 300 Hz yielded a temperature difference for the first 0.01 s.
are presented.
The temperature resolution of the IR-camera was in the order of
In Section IV-E, a shutdown scenario of the cooling cycle
10 × 10−3 K.
illustrates the failure behavior of the cooling device. Finally,
The surface temperature of the MOSFET only changes in the
Section IV-F evaluates the cooling efficiency, highlighting the
central area, represented by the gray to white spots in Fig. 8.
heat removal drawback of the TIM.
The central area representing heat generation of the MOSFET is
surrounded by a black colored ring, which indicates a constant
A. Local Heat Generation of the mosfet temperature over the observed time frame. Generally, the flow
In the design process of the cooling system heat generation directions of the generated heat can be vertical, from the hot spot
of the MOSFET (DirectFET) was analyzed by IR-thermography. inside the semiconductor to its surface. Due to the metal cover
The MOSFET package (see Fig. 1) including the chamber housing of the MOSFET and its high thermal conductivity the heat can
of the first test setup (electrically noninsulated, Case A) was also spread horizontally along the plane cover. The short time
placed in front of an IR-thermography camera (Infra Tec Image frame of the measurements reduces horizontal heat spreading
IR5325), as shown in Fig. 3 (right picture). Connected to a power and, thus, ensures the detection of hot spots. Based on these
supply and gate voltage the IR-images of the MOSFET top cover results, the injection nozzle was placed in the center position of
surface were recorded during the early heat up (time span of the chamber housing.
0.5 s after initiation). Results of the IR-measurements are shown Fig. 9 presents the experimental results of the spatially av-
in Fig. 8. The MOSFET was operated on a low gate-to-source eraged heat transfer coefficient for different nozzle lengths and
voltage of VGS = 4.58 V. A drain-to-source current of IDS = volume flow rates. For both nozzle lengths, the heat transfer
1 A was applied together with a nonpowered MOSFET while coefficient is nearly identical.
4232 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 33, NO. 5, MAY 2018

B. Pressure Drop and Pumping Power

The thermohydraulic comparison of Robinson [1] highlights
the tradeoff between pressure drop and volumetric flow rate
in impinging jet applications. High volume flow rates cause
high pressure drops and, thus, increase the required pumping
power. Increasing the outlet velocity leads to higher local and
spatially averaged heat transfer coefficients and, as a result, bet-
ter cooling performance. To determine these competing effects
on the cooling performance of the direct impinging jet cool-
ing the electrically noninsulated setup (Case A) was equipped
with two different nozzle lengths (LN = 12 mm and LN =
24 mm). For both nozzles, a flow rate variation was performed
and the dissipated heat was measured. All experiments and nu-
merical investigations were conducted under steady-state condi-
tions, using same coolant and ambiance temperatures. Further-
more, the test setup was thermally insulated to avoid heat loss
to the ambient.
Besides heat transfer characteristics and space requirements
of the cooling method pumping power is an important parameter.
The pumping power P depends on the pressure drop Δp and
the volumetric flow rate V̇

P = V̇ · Δp. (5)

To estimate the hydrodynamic characteristics numerical sim-

ulations were performed using the commercial CFD tool AN-
SYS Fluent. The domain of the numerical simulations based on
the geometric dimensions of the first test setup included the inlet
and outlet pipes as well as the entire cooling chamber. Not in- Fig. 10. Results of the numerical simulations. Upper panel: Influence of nozzle
cluded were hydrodynamic losses in the pumping system and of length L N and coolant flow rate V̇ on pressure drop Δp. Lower panel: Influence
the hoses (incoming and outgoing liquid). The thermophysical of nozzle length L N and coolant flow rate V̇ on pumping power P . Note: for
both investigated setups (with and without electrical insulation) pressure drop
properties of the fluid used were equal to the coolant property and pumping power are equal.
data at 22.5 ◦ C. The plots in Fig. 10 summarize the numerical
simulation results for a nozzle length of 12 mm and 24 mm and
various volume flow rates. C. Heat Transfer Characteristics
The evaluation of the results regarding pressure drop and Heat transfer characteristics of the two cooling system
pumping power highlights the influence of the volume flow (Case A and Case B) are experimentally obtained using the
rate. For both nozzle types the pressure drop increases. Pump- setup described in Section II.
ing power, the product of pressure drop and volume flow rate, Fig. 11 depicts the spatially averaged heat transfer coefficient
increases even more. Taking into account the moderate increase ᾱ in dependence of volume flow rate V̇ and MOSFET temperature
of the heat transfer coefficient in the same flow rate range (see TM . Furthermore, the influence of the electrical insulation on
Fig. 9), the cooling system has to be operated with relatively heat transfer is illustrated. For all experimental conditions, the
high pumping power to achieve only small improvements in electric current IDS was adjusted in order to provide a constant
cooling performance. Especially the additional pressure drop of MOSFET temperature during each flow rate variation. The drain-
the long nozzle does not lead to significant improvements of the to-source voltage VDS adjusts itself according to the MOSFET
heat transfer coefficient. resistance, set by the gate-to-source voltage with a value of
Fulfilling the requests for a compact, reliable, and cost- VGS = 4.58 V. Consequently, a nozzle length of LN = 12 mm
effective cooling system the short nozzle (LN = 12 mm) offers was chosen for all subsequent experiments and the test setup
the best performance and was chosen for the following investi- was thermally insulated. The coolant inlet temperature was TI =
gations. Also, the relatively low coolant flow rate of 30 mL/min 22.5 ◦ C.
proved adequate cooling performance. Thus, the liquid distri- Represented by the gray line with cross marker, the develop-
bution system can be designed with a very low pressure drop. ment of the heat transfer coefficient of the electrically noninsu-
For the flow rate of 30 mL/min the pumping power required is lated setup (Case A) starts at a flow rate of V̇ = 10 mL/min with
approximately 1 mW, which is low for a spatially averaged heat a value of around ᾱ = 4250 W/m2 · K, followed by a constant
transfer coefficient of 9000 W/m2 · K. increase up to 8660 W/m2 · K at 25 mL/min. In the range of
JÖRG et al.: DIRECT SINGLE IMPINGING JET COOLING OF A MOSFET POWER ELECTRONIC MODULE 4233

Fig. 11. Spatially averaged heat transfer coefficient of the electrically non-
insulated (Case A) and electrically insulated cooling chamber (Case B). The
MOSFET was operated with a gate-to-source voltage of V G S = 4.58 V.

Fig. 13. Results of the numerical investigations based on the x,z-symmetry

plane, highlighted green in Fig. 12. Upper panels: fluid and wall jet character-
istics at a coolant flow rate of V̇ = 15 mL/min. Bottom panels: fluid and wall
jet characteristics at a coolant flow rate of V̇ = 60 mL/min. The respective left
panels illustrate the velocity magnitude distributions inside the cooling cham-
ber. The respective right panels (upper and lower row) illustrate the stream lines
of the coolant, beginning at the inlet nozzle.

model. Taking advantage of the symmetric design of the cool-

ing chamber the domain represents one quarter of the liquid
phase. Also, the fluid inside the inlet and outlet nozzles are in-
cluded. The results of the fluid dynamics analyses are presented
in Fig. 13. The results are based on the numerical results at the
Fig. 12. Domain of the numerical investigations, as well as symmetry planes
of the coolant inside the cooling chamber and nozzles.
x, z-symmetry plane, highlighted by the cross-hatched area in
Fig. 12.
Fig. 13 presents the differences of the fluid dynamics inside
30–100 mL/min the dependency between heat transfer coeffi- the cooling chamber below and beyond V̇ = 25 mL/min. The
cient and volume flow rate exhibits a square root characteristic. upper plots are based on a coolant flow rate of V̇ = 15 mL/min,
With additional electrical insulation (Case B), the heat trans- the bottom plots on a coolant flow rate of V̇ = 60 mL/min.
fer coefficient is significantly lower, as illustrated by the black To evaluate the influence of the impinging jets and their corre-
curves (for MOSFET temperatures of TM = 50, 70, and 90 ◦ C). sponding wall jets on heat transfer the velocity magnitudes as
All three curves have the same exponential characteristic with well as stream lines are considered.
only minor variations in the absolute value, indicating that the At a coolant flow rate of V̇ = 15 mL/min the fluid enters
temperature level does not influence the heat transfer character- the cooling chamber through the inlet, gets redirected in the
istics. This suggests that forced convection is the dominant heat stagnation region (width  0 mm to 0.3 mm), and develops
transfer mechanism. a wall jet. As indicated by the separately observable stream-
Special emphasis should be drawn to the slopes of the curves lines the wall jet is of significant thickness. Along the bottom
in Fig. 11, which change abruptly around V̇ = 25 mL/min. surface the thickness of the wall jet increases and, finally, turns
This inflection point, especially prominent in Case A, indicates toward the outlet nozzle. Taking into account the low fluid veloc-
changing physical conditions regarding convective heat transfer ity in this area (below 1 m/s), the heat transfer is accordingly low.
and fluid dynamics in the cooling chamber. Increasing coolant flow rate (higher momentum of the imping-
To investigate this aspect further, numerical investigations ing jet) increases the redirected fluid velocity in the stagnation
were carried out. Fig. 12 illustrates the domain of the CFD region. The thickness of the wall jet decreases, but also, the wall
4234 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 33, NO. 5, MAY 2018

Fig. 14. Developed correlation of the Nusselt by Reynolds numbers compared Fig. 15. Break-down of the thermal resistances for the two test setups; differ-
to the experimental measurement data. The MOSFET temperature is adjusted to entiated by their heat transfer mechanism.
T M = 80 ◦ C.
For the given property data as well as the inlet nozzle diameter
jet area increases. Both effects, increasing the wall jet area as of D = 0.6 mm, the volume flow rate of V̇ = 30 mL/min leads
well as the wall jet velocity, result in a significant heat transfer to a Reynolds number of around Re = 940.
improvement and can be obtained up to a the flow rate of V̇ =
25 mL/min. D. Thermal Resistance
The bottom plots of Fig. 13 illustrate the wall jet characteris-
tics of coolant flow rates beyond 25 mL/min. The momentum of The reciprocal value of the heat transfer coefficient is the
the impinging jet is sufficient to establish the wall jet area along thermal resistance. This value is a good indicator to quantify
the entire bottom surface and decreases the wall jet thickness. the performance of the cooling device. The experimental inves-
The streamlines as well as the velocity distribution illustrate that tigations in this study comprise of two setups, with and without
the wall jet nearly reaches the cooling chamber wall (width = electrical insulation. This approach allows us to differentiate
3.5 mm). In contrast to coolant flow rates below 25 mL/min the thermal resistance of the impinging jet cooling and electri-
increasing the coolant flow rate leads to no further growth of cal insulation. For the test setup equipped with the electrical
the wall jet area. Thus, the development of the heat transfer insulation (Case B) the thermal resistance is given by
coefficient beyond 25 mL/min, as observed in Fig. 11, can only RJet+Adapter = RJet + RAdapter . (8)
be attributed to the influence of the increasing fluid velocities.
To enable scaling of the experimental results and a transfer In this case, the thermal resistance RJet+Adapter comprises the
to other cooling applications, Fig. 14 compares the dimension- jet impingement resistance RJet and the thermal resistance of
less Nusselt number with classical power-law dependencies of the adapter plate RAdapter . Both resistances can be determined
laminar jet impingement. The Nusselt number is proportional to by
the Reynolds number by Nu ∝ Re1/2 (for more details on scal-
1 δBrass δInsulation
ing law dependencies see [24]). The equations for the Nusselt RJet = , RAdapter = + . (9)
ᾱJet λBrass λInsulation
correlations of the first (Nu1 = f (Re)) and the second setup
(Nu2 = f (Re)), as well as the values of their corresponding The thermal resistance of the jet impingement cooling depends
coefficients are on the heat transfer coefficient ᾱJet and, due to this, on the
coolant flow rate V̇ . Thus, the thermal resistance of the electri-
Nu1 = a · Re1/2 + b a = 0.1052 b = 5.514 (6) cally noninsulated setup (Case A) can be divided from the heat
Nu2 = c · Re1/2 + d c = 0.0377 d = 4.803. (7) transfer measurements. The thermal resistance of the adapter
plate depends on the heat conductivity of the electrical insula-
Remarkably good agreement is found for Reynolds number tion λInsulation and brass λBrass , as well as their layer thicknesses
values above Re = 800. Below this Reynolds number, the Nus- (δInsulation and δBrass ).
selt number increases much stronger with Reynolds number, In order to highlight the adverse effects of the electrical insu-
which may indicate that the jet velocity and momentum are not lation on the heat transfer Fig. 15 illustrates the dependency of
sufficient to generate a fully developed wall jet on the surface thermal resistances on coolant flow rate of the investigated cool-
of the MOSFET. As such, the area covered by the wall jet in- ing chambers. The thermal resistance of the cooling chamber
creases with Reynolds number and the Nusselt number reveals equipped with electrical insulation (Case B) is represented by
a much stronger power-law dependency on Reynolds number. the black curve with square markers (Jet + Adapter). The ther-
Note, that critical hot spots may arise if the MOSFET is not suf- mal resistance characteristic of the cooling chamber without
ficiently cooled by the wall jet. Therefore, it is recommended electrical insulation (Case A) is represented by the gray curve
to operate the liquid jet impingement cooling above Re = 800. with diamond markers (Jet). The light gray curve with circle
JÖRG et al.: DIRECT SINGLE IMPINGING JET COOLING OF A MOSFET POWER ELECTRONIC MODULE 4235

markers (Adapter) of Fig. 15 represents their difference. Thus,

the thermal resistance of the adapter can be determined. The
thermal resistance of jet impingement decreases from RJet =
2.35 cm2 · K/W at a volume flow rate of V̇ = 10 mL/min, down
to RJet = 0.85 cm2 · K/W at a volume flow rate of V̇ = 100
mL/min. The curve of the thermal resistance representing the
setup including electrical insulation (Jet + Adapter) reveals the
same characteristic profile, but has higher thermal resistances.
At a volume flow rate of V̇ = 10 mL/min, the thermal resis-
tance decreases from RJet+Adapter = 2.65 cm2 K/W down to
RJet+Adapter = 1.64 cm2 · K/W. In contrast to the thermal re-
sistance curves of Cases A and B, the thermal resistance of the
adapter shows no apparent dependency on coolant flow rate.
Except for small volume flow rates (10–25 mL/min) the value
of the adapter thermal resistance is approximately RAdapter = Fig. 16. Failure scenario: Shut down of the coolant flow at t = 0 s. Tracking
0.58 cm2 · K/W. Compared to the jet impingement the addi- of the MOSFET temperature at a power distribution of approximately 11.0 W and
tionally installed electrical insulation (Case B) contributes to an inlet temperature of T I = 22.0 ◦ C.
more than 30% on the thermal resistance of the cooling device.
The layer thickness of the adapter plate δBrass = 0.6 mm and
electrical insulation pad δInsulation = 0.2 mm, as well as the The MOSFET temperature was continuously tracked (steady-
corresponding thermal heat conductivities (λBrass = 120 W/m· state value of TM = 67 ◦ C), while the semiconductor was oper-
K and λInsulation = 5 W/m· K) lead to a thermal resistance of ated with a power loss of approximately 11.0 W. The results in
the adapter of RAdapter = 0.45 cm2 · K/W. The difference be- Fig. 16 show that the coolant supply shutdown leads to an im-
tween the measured and the calculated value of the adapter’s mediate temperature rise with a gradient of around dTM /dt =
thermal resistance can be addressed to differences in insulation 3 k/s within the first seconds. After approximately t = 25 s,
thickness, but also to additional nonconsidered resistances (i.e., the MOSFET temperature reaches the maximum value of TM =
solder between the MOSFET surface and adapter lower surface). 97 ◦ C. To prevent any damage of the MOSFET and boiling of the
Taking into account the findings for pressure drop and pumping coolant the power supply was switched OFFat this point.
power (see Section IV-B), the maximum power distribution of The square root progress of the temperature curve can be ex-
the MOSFET is measured at a coolant flow rate of 30 mL/min. plained by the thermal inertia of the test setup, but also by the
The required pumping power of 1 mW is low for an average heat temperature sensitivity of the MOSFET. With increasing tempera-
transfer coefficient of 6170 W/m2 · K. Limited by the maximum ture the ohmic resistance of the MOSFET decreases along with its
MOSFET temperature of 163 ◦ C and coolant inlet temperature of heat generation. At the beginning of the coolant shutdown, the
22.5 ◦ C, the maximum measured power distribution was 51 W. heat generation was 11.0 W; at t = 25 s, the power generation
Under these conditions, the MOSFET temperature exceeded the was reduced to 3.0 W.
boiling temperature of the coolant of 100 ◦ C and steam bubbles Due to the heat capacities of the cooling chamber, the inside
were detected at the transparent outlet tubes. The steam bub- remaining coolant, and the convective heat loss to the ambiance
bles collapsed before passing the mounted outlet thermocouples. the temperature rise is moderate. However, the time period of
This ensured reliable measurements of the thermocouples at the several seconds, until a significant temperature rise is detected,
outlet. Also, steady-state coolant outlet temperatures were mea- is sufficient for an adequate response of a control unit. Under
sured. The boiling conditions inside the cooling chamber did shutdown conditions steam bubbles cannot be removed. This
not cause instabilities of the cooling system or semiconductor. may cause pressure rise and thus, damage to the cooling chamber
A possible explanation is the impinging liquid jet, which pushes if the temperature rises above 100 ◦ C. Thus, in order to prevent
the steam bubbles out of the cooling chamber. boiling of the water inside the cooling chamber the MOSFET was
operated up to a maximum temperature of TM = 100 ◦ C.

E. Failure Scenario
The failure scenario is an important aspect in electrical com- F. Efficiency
ponent cooling. To evaluate the thermal response of a power To evaluate the efficiency of the impinging jet cooling, Fig. 17
electronic control unit the temporal temperature rise of the MOS- compares the required pumping power to the heat transfer coef-
FET in the case of a coolant supply shutdown is presented. The ficient. In general, the heat removal rate of the cooling depends
MOSFET was operated under steady-state conditions and the inlet on the temperature difference between the coolant inlet temper-
pipe was blocked. Before shutdown, the cooling device was sup- ature and surface temperature, as well as on the heat transfer co-
plied with a constant coolant flow rate of V̇ = 30 mL/min at an efficient. In common applications, the coolant inlet temperature
inlet temperature of TI = 22.0 ◦ C. The electrically insulated sec- is restricted by the cooling cycle and its conditioning system.
ond setup was chosen because it is the more application-oriented The heat transfer coefficient correlates with the coolant flow
and it also reflects the worst case scenario. The measurement rate, which is determined by the pumping power. Characteriz-
result are presented in Fig. 16. ing the system efficiency by the required pumping power and
4236 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 33, NO. 5, MAY 2018

requirements to the cooling liquid. The dimensionless heat trans-

fer coefficient (Nusselt number) was found to increase with flow
rate (or Reynolds number) in two different regimes. For higher
flow rates (1000 < Re < 4000), the classical power-law depen-
dency for laminar jet impingement was found to be well-suited
for the prediction of the cooling performance. For lower flow
rates (Re < 1000), the Nusselt number increased much stronger
with Reynolds number. Besides the experimental analysis of
the heat transfer characteristics the results of numerical simu-
lations provided insights into fluid dynamics and wall jet de-
velopment inside the cooling chamber, as well as its influence
on the heat transfer. Numerical investigations of pressure drop
and pumping power reflected the relatively low energy require-
ments in terms of pumping power to achieve sufficient cooling
performances.
Fig. 17. Results of the numerical and experimental investigations to evaluate The experimental and numerical investigations of the heat
the efficiency of the impinging jet cooling approaches (of the electrically non-
insulated (Case A) and electrically insulated cooling chamber (Case B)) using transfer characteristics and cooling performances illustrate the
nozzle lengths ranging between 12 and 24 mm. Comparison of the required potential of direct impinging jet cooling. Even for small coolant
pumping power P (numerical results) with respect to the achieved heat transfer flow rates the cooling concept achieved adequate heat trans-
coefficients ᾱ (experimental results).
fer coefficients and heat dissipation rates. The compact design
of the injection chamber and its attachment to the MOSFET top
heat transfer coefficient allows for a temperature-independent
cover minimize the thermal resistance. To fulfill the require-
evaluation of the cooling performance.
ments of future applications further improvements of the heat
Fig. 17 illustrates the influence of the nozzle length on the
transfer coefficient of impinging jets should be achieved. For
efficiency. In the range of ᾱ = 4300 W/m2 ·K to 8600 W/m2 ·K
example, the number of impinging jets could be increased and
the nozzle with a length of LN = 24 mm requires more pumping
the bottom chamber surface could be equipped with microfins.
power to achieve the same heat transfer coefficients compared
Furthermore, lower thermal resistance of the TIM could increase
to the nozzle with a length of 12 mm. Reducing nozzle length
the heat removal rates of the cooling chamber.
increases the efficiency of the impinging jet cooling approach.
The present cooling approach offers the opportunity of indi-
The numerical investigations proved that the nozzle length of
vidual cooling of heat dissipating components of power elec-
12 mm (with a nozzle diameter of 0.6 mm) is sufficient to
tronic modules, such as MOSFETs or IGBTs. Due to separately
develop parabolic velocity profiles at the inlet. Beyond the heat
installed cooling chambers, new flexibility in design of PCBs
transfer coefficient of ᾱ = 8600 W/m2 · K no efficiency benefits
and DCBs can be realized. Their layout is no longer restricted
of the short nozzle can be obtained.
by a planar heat sink. Also, nonhomogeneous heat conductivity
Furthermore, Fig. 17 highlights the efficiency drawback of
between each MOSFET and the heat sink can be avoided. Further-
the additionally installed electrical insulation (TIM). The re-
more, the achieved high heat transfer coefficients allow further
quired pumping power of the investigated cases [electrically
reduction of the coolant flow rate. This leads to higher coolant
noninsulated (Case A) and electrically insulated cooling cham-
outlet temperatures, which offers benefits on the heat removal
ber (Case B)] using the inlet nozzle with a length of LN = 12
of cooling cycle to the ambiance, and thus, offers additional
mm is in the same range. Due to the additional thermal resis-
potential for downsizing.
tance of the TIM, the heat transfer coefficients of Case B are
smaller. This leads to a significant reduction of the cooling effi-
ciency in Case B compared to the efficiency of the electrically REFERENCES
noninsulated cooling approach in Case A. [1] A. J. Robinson, “A thermal-hydraulic comparison of liquid microchannel
and impinging liquid jet array heat sinks for high-power electronics cool-
ing,” IEEE Trans. Compon. Packag. Technol., vol. 32, no. 2, pp. 347–357,
V. CONCLUSION Jun. 2009.
In this paper, a new approach of direct liquid cooling of a [2] S. Kang, “Advanced cooling for power electronics,” in Proc. Int. Conf.
Integr. Power Electron. Syst., Nuremberg, Germany, Mar. 2012, pp. 8–8.
MOSFET semiconductor with a single impinging jet was pre- [3] V. Benda, “Power semiconductors—State of the art and future trends,”
sented. The cooling concept focused on hot spot removal and Global J. Technol. Optim., vol. 2, pp. 29–36, Jan. 2011.
the reduction of thermal resistances. The design of the coolant [4] J. Vobecky, “Future trends in high power devices,” in Proc. Int. Conf.
Microelectron., Niš, Serbia, May 2010, pp. 16–19.
chamber and the positioning of the nozzle center was based [5] J. Biela, M. Schweizer, S. Waffler, B. Wrzecionko, and J. W. Kolar, “SiC
on IR-thermography measurements of the spatial temperature vs. Si—Evaluation of potentials for performance improvement of power
distribution of the MOSFET, illustrating the hot spots of the semi- electronics converter systems by SiC power semiconductors,” IEEE Trans.
Ind. Electron., vol. 58, no. 7, pp. 2872–2882, Jul. 2011.
conductor. Two different concepts, with and without electrical [6] J. Schulz-Harder, “Review on highly integrated solutions for power elec-
insulation between the MOSFET and the coolant chamber were tronic devices,” in Proc. 5th Int. Conf. Integr. Power Syst., Nuremberg,
tested. The TIM material with a thickness of merely 200 μm was Germany, Mar. 2008, pp. 1–7.
[7] D. Saums, “Vehicle electrification thermal management challenges and
found to significantly contribute to the total thermal resistance solutions overview,” in Proc. MEPTEC Therm. Manage. Workshop, San
(30% and more). Omitting an electrical insulation sets special Jose, CA, USA, Mar. 2011.
JÖRG et al.: DIRECT SINGLE IMPINGING JET COOLING OF A MOSFET POWER ELECTRONIC MODULE 4237

[8] S. G. Leslie, “Cooling options and challenges of high power semiconduc- Silvano Taraborrelli (S’16) received the Master’s
tors,” Electron. Cooling, vol. 12, no. 4, pp. 20–27, Nov. 2006. degree (Laurea) in electrical engineering from the
[9] J. Valenzuela, “Liquid cooling for high-power electronics,” Power Elec- Politecnico di Torino, Turin, Italy, in 2010.
tron. Technol., vol. 31, pp. 50–56, Feb. 2005. Since 2012, he has been working as a Research
[10] J. C. Howes, D. B. Levett, S. T. Wilson, J. Marsala, and D. L. Saums, “Cool- Assistant in the Institute for Power Electronics
ing of an IGBT drive system with vaporizable dielectric fluid (VDF),” and Electrical Drives, RWTH Aachen University,
Twenty-fourth Annu. IEEE Semicond. Thermal Meas. Manage. Symp., pp. Aachen, Germany, in the Power Electronic Group.
9–15, Mar. 2008. Since 2013, he has been a participant of the mobilEM
[11] B. W. Webb and C. F. Ma, “Single-phase liquid jet impingement heat Graduate School. His research interests include dc–
transfer,” Adv. Heat Transf., vol. 27, no. 17, pp. 105–217, 1995. dc converter with wide voltage range for automotive
[12] J. H. Lienhard, “Liquid jet impingement,” Annu. Rev. Heat Transf., vol. 6, application.
pp. 199–270, 1995.
[13] S. V. Garimella, “Heat Transfer and flow fields in confined jet impinge-
ment,” Annu. Rev. Heat Transf., vol. 6, pp. 413–494, 2000. Garikoitz Sarriegui received the B.Sc. and M.Sc.
[14] D. J. Womac, S. Ramadhyani, and F. P. Incropera, “Correlation equations degrees in automatization and industrial electron-
for impingement cooling of small heat sources with single circular liquid ics from the University of Mondragon, Mondragon,
jets,” ASME J. Heat Transf., vol. 115, pp. 106–115, 1993. Spain, in 2007 and 2010, respectively. He is cur-
[15] D. J. Womac, F. P. Incropera, and S. Ramadhyani, “Correlation equations rently working toward the Ph.D. degree in SiC and
for impingement cooling of small heat sources with multiple circular liquid GaN semiconductors at RWTH Aachen University,
jets,” ASME J. Heat Transf., vol. 116, pp. 482–486, 1994. Aachen, Germany.
[16] D. C. Wadsworth and I. Mudawar, “Cooling of a multichip electronic Since 2011, he has been a Research Associate
module by means of confined two-dimensional jets of dielectric liquid,” in the Institute for Power Electronics and Electrical
ASME J. Heat Transf., vol. 112, pp. 891–898, 1990. Drives, RWTH Aachen University. His current re-
[17] T. Brunschwiler et al., “Direct liquid jet-impingement cooling with mi- search interests include power electronic converter
cronsized nozzle array and distributed return architecture,” in Proc. Inter- design and implementation, and wide bandgap semiconductor devices.
soc. Conf. Thermomech. Phenom. Electron. Syst., 2006, pp. 693–699.
[18] S. Narumanchi, M. Mihalic, G. Moreno, and K. Bennion, “Design of
light-weight, single-phase liquid-cooled heat exchanger for automotive Rik W. De Doncker (M’87) received the Ph.D.
power electronics,” in Proc. Intersoc. Conf. Therm. Thermomech. Phenom. degree in electrical engineering from the Katholieke
Electron. Syst., 2012, pp. 693–699. Universiteit Leuven, Leuven, Belgium, in 1986.
[19] Y. Han, B. L. Lau, G. Tang, X. Zhang, and D. M. W. Rhee, “Si-based In 1987, he was appointed as a Visiting As-
hybrid microcooler with multiple drainage microtrenches for high heat sociate Professor at the University of Wisconsin,
flux cooling,” IEEE Trans. Compon., Packag., Manuf. Technol., vol. 7, Madison. After a short stay as an Adjunct Researcher
no. 1, pp. 50–57, Jan. 2017. with Interuniversity Microelectronics Centre, Leu-
[20] K. Gould, S. Q. Cai, C. Neft, and A. Bhunia, “Liquid jet impingement ven, he joined, in 1989, the Corporate Research
cooling of a silicon carbide power conversion module for vehicle appli- and Development Center, General Electric Company,
cations,” IEEE Trans. Power Electron., vol. 30, no. 6, pp. 2975–2984, Schenectady, NY. In 1994, he joined Silicon Power
Jun. 2015. Corporation, a former division of General Electric,
[21] S. Narumanchi, M. Mihalic, and K. Kelly, “Thermal interface materi- Inc., as the Vice President of technology. In 1996, he became a Professor at
als for power electronics applications,” in Proc. IEEE Intersoc. Therm. RWTH Aachen University, Aachen, Germany, where he currently leads the In-
Thermomech. Phenom. Electron. Syst., Sep. 2008, pp. 395–404. stitute for Power Electronics and Electrical Drives. Since 2006, he has been the
[22] J. Schulz-Harder, K. Exel, and A. Meyer, “Direct liquid cooling of power Director of the E.ON Energy Research Center, RWTH Aachen University.
electronics devices,” in Proc. 4th Int. Conf. Integrated Power Syst., Naples, Dr. De Doncker was the President of the IEEE Power Electronics Society
Jun. 2006, pp. 1–6. (PELS) in 2005 and 2006. He was the founding Chairman of the German IEEE
[23] G. Engelmann, T. Senoner, H. van Hoek, and R. W. De Doncker, “A sys- Industry Applications Society PELS Joint Chapter. In 2002, he received the
tematic comparison of various thermal interface materials for applications IEEE IAS Outstanding Achievement Award. In 2008, he received the IEEE
with surface-mounted (DirectFETTM ) MOSFETs,” in Proc. IEEE 11th PES Nari Hingorani Custom Power Award. In 2009, he led a VDE/ETG Task
Int. Conf. Power Electron. Drive Syst., 2015, pp. 112–117. Force on Electric Vehicles. In 2010, he received an honorary doctor degree
[24] W. Rohlfs, C. Ehrenpreis, H. D. Haustein, O. Gabrecht, and R. Kneer, of TU Riga, Latvia. In 2013, he received the IEEE William E. Newell Power
“Influence of the local flow acceleration on the heat transfer of submerged Electronics Award.
and free-surface jet impingement,” in Proc. Int. Heat Transf. Conf., Kyoto,
Japan, 2014, pp. 3179–3192.
[25] International Rectifier, “DirectFETTM Power MOSFET Datasheet,” Reinhold Kneer received the Doctoral degree
Feb. 2014. [Online]. Available: www.irf.com (Dr.-Ing.) in mechanical engineering specialized in
[26] Electronics Group of Henkel: Bergquist Company, “Gap PAD 5000S35 atomization processes in combustion engines from
datasheet,” Mar. 2016. [Online]. Available: www.bergquistcompany.com the Karlsruhe Institute of Technology, Karlsruhe,
Germany, in 1994.
Since 2004, he has been the Head of the Institute of
Heat and Mass Transfer, RWTH Aachen University,
Aachen, Germany. The Institute of Heat and Mass
Transfer focuses on fundamental research projects
on the basis of transport processes, boundary layer
processes, and phase changes.
Johannes Jörg received the Master’s degree
(Dipl.-Ing.) in mechanical engineering specialized in
energy engineering from the RWTH Aachen Univer- Wilko Rohlfs received the Diploma in mechanical
sity, Aachen, Germany, in 2013. engineering in 2009 and the Ph.D. degree in 2015. In
Since 2013, he has been a participant of the mo- 2014 he received his PhD in economics from RWTH
bilEM Graduate School and working as a Research Aachen University.
Assistant in the Institute of Heat and Mass Transfer, He is currently a Researcher and Leader of the
Aachen, Germany. He contributes to the development research group “Convective Transport Processes” at
of contactless temperature measurement systems. the Institute of Heat and Mass Transfer, RWTH
His research interest includes convective heat trans- Aachen University, Aachen, Germany. His major re-
fer focused on impinging microjet cooling and jet search topics in fluid dynamics are falling liquid films
interaction. and jet impingement.