Due to its superior electrical and thermal characteristics, silicon carbide power modules will soon replace silicon modules to be mass-produced and implemented in all-electric and hybrid-electric vehicles (HEVs). Redesign of the power modules will be required to take full advantage of these newer devices. A particular area of interest is high-temperature power modules, as under-hood temperatures often exceed maximum silicon device temperatures. This review will examine thermal packaging options for standard Si power modules and various power modules in recent all-electric and HEVs. Then, thermal packaging options for die-attach, thermal interface materials (TIM), and liquid cooling are discussed for their feasibility in next-generation silicon carbide (SiC) power modules.
Worldwide efforts in minimizing CO2 emissions, resulting from growing concerns about the effects of greenhouse gases on climate, have been a major driver in increasing electrification. As societies gravitate toward a more electricity-based future, power electronics, which enable the transmission and conversion of electrical energy, will be used more and more. The majority of electrical energy used has been through some form of conversion, whether it is AC to DC conversion, rectification, or inversion. The U.S. Advanced Research Projects Agency-Energy (ARPA-E) predicts up to 80% of electrical energy in the U.S. is estimated to pass through power electronics by the year 2030—more than a 160% increase from 2005 . Increased popularity and usage of technologies such as alternative energy sources (fuel cells, wind turbines, and photovoltaics), energy storage, public transportation, and particularly hybrid and electric vehicles result largely from this concerted effort and are driving the need for better power electronics technologies.
Power electronics can be packaged discretely or in modules. As power levels increase, they are packaged on organic printed circuit boards, lead frames, and then electrically insulating ceramic substrates. These design changes result from thermal management requirements that necessitate that high power packages be mounted directly on heat sinks, which in turn requires improved electrical isolation. Higher voltage and current levels elevate heat dissipation loads, requiring more aggressive cooling solutions. Effective thermal management is needed to prevent catastrophic failure and to decrease failures caused by high temperatures and temperature cycling.
Power packaging evolution has been gradual and incremental. However, as silicon devices have been nearing their theoretical limit , two high-performance alternatives in the form of gallium nitride (GaN) and silicon carbide (SiC) have emerged and disrupted the power packaging landscape, and conventional packaging solutions are now limiting these devices. These wide bandgap (WBG) semiconductors require more energy to move electrons from valence bands to conduction bands, resulting in the enhancement of various properties as demonstrated in Fig. 1. WBG devices have much higher theoretical breakdown voltages, improved efficiencies, increased switching frequencies, and high-temperature performance . Increased breakdown voltages allow devices to handle higher power levels, or result in smaller devices at fixed blocking voltages. Higher efficiencies will lead to lower energy consumption; mass implementation of technologies enabled by WBG technologies could reduce U.S. electricity consumption by over 20% . Faster switching allows for smaller passives (capacitors, inductors, and transformers)  and smoother output curves. Wide bandgap devices can also allow for both higher operating temperatures and improved thermal conductivities compared to Si. Their temperature-robust nature permits operation in extreme temperature environments (>200 °C), elimination of liquid cooling under certain conditions, and reduction of heat sink size due to reduced heat dissipation. Currently, WBG devices are projected to encroach significantly upon the Si-dominated power device market by 2023, with GaN superseding low voltage applications (<600 V) and SiC higher voltage applications (>600 V) . Both GaN and SiC devices are commercially available, although vertical GaN devices are not yet available commercially. Lateral GaN devices are typically in the 600–650 V range, but SiC devices in the kV range are already commercially available .
Despite GaN's superior performance in all categories except thermal conductivity, it is held back by its lateral architecture. GaN devices are usually grown on substrates such as silicon, silicon carbide, and sapphire, as researchers were initially unable to grow them on GaN substrates. Currently, the only bulk GaN wafers are small and prohibitively expensive . Growing on these foreign substrates causes problems, including high defect densities and stresses caused by lattice and coefficient of thermal expansion (CTE) mismatches , and it limits these devices to lateral architectures. For these lateral devices, device level thermal management becomes a much more important issue, and issues such as near-junction heat spreading, thermal boundary resistances, and device-level cooling are of great interest . Lateral GaN devices also suffer various reliability issues, such as undesirable breakdown mechanisms that can cause catastrophic damage when exceeding certain voltages . Although researchers have recently been able to fabricate GaN-on-GaN devices that operate much closer to the theoretical limits than today's lateral structures , GaN device maturity lags behind its SiC counterparts. SiC devices have been commercially available since 2001, whereas vertical GaN devices are still not. In addition to its maturity, SiC's lower thermal conductivity and the research into characterizing very high-temperature devices (200 °C has already been characterized with plans for higher temperatures) provide attractive qualities for SiC adoption and implementation [6,7]. This led to an industry focus on SiC, and companies including Mitsubishi and Toyota have since tested SiC devices in public transportation and electric vehicles with significant increases in efficiency [11,12].
Wide bandgap devices can be smaller than their silicon counterparts because of their increased breakdown voltages, which then decrease the on-resistance. However, the volumetric heat generation may increase despite the increase in efficiency as a result of the decrease in size and the resulting smaller packages. These factors combine to create power electronics modules that can see higher heat generation per unit volume than Si-based modules. The packaging materials that were sufficient for Si devices are no longer sufficient for WBG technology. In order to take full advantage of the benefits offered by these newer semiconductor devices, packaging materials with improved reliability at higher temperatures, while maintaining cost-effectiveness, are required.
In this review, we assess the thermal packaging of Si and SiC device-based traction inverter power electronics, which can be seen in Fig. 2 highlighted in red. Current power modules and packaging materials are discussed in Sec. 2. Hybrid electric vehicles (HEV) and fully electric vehicle (EV) power module evolution over roughly the past decade are subsequently investigated. Options to address gaps in thermal management of SiC power electronics, particularly high-temperature die-attach, high-conductivity thermal interface materials (TIM), and high-performance liquid cooling, are surveyed. We then discuss the thermal management research needs for future power electronics systems.
Conventional Silicon Devices Power Packaging
Beyond performance, the most important factors in power electronics packaging are cost and reliability. Power electronics package reliability may be improved by decreasing maximum temperatures and temperature swings to mitigate thermally accelerated failures, or utilizing materials better suited for higher temperature operation. Temperature swings cause cyclic stresses between the various layers of the package because of their different CTE, which can lead to failure, depending on the severity and frequency of these cycles. Ciappa presented a comprehensive overview of power module failure mechanisms, such as wire bond-lift off and solder joint or ceramic cracking .
A typical power electronics package can be seen in Fig. 3. The semiconductor devices, which generate the majority of the heat during conduction and switching, are connected at the top with wires and at the bottom with a die-attach technology such as a tin-based lead-free solder. Aluminum is the most common metallization on high power devices, and because of the difficulties associated with bonding aluminum to anything except itself, aluminum wires are commonly selected. In addition to the ease of bonding, its high electrical conductivity makes it an obvious choice for topside interconnections . Most power devices are vertical, so the die-attach must be both an electrical interconnection and thermal path, as opposed to only a thermal pathway as in microelectronic device packaging. They also provide a mechanical bond to the substrate. The substrate provides the electrical insulation necessitated by mounting the devices directly on a heat sink in case of air cooling or cold plate for liquid cooling.
Ceramics are a natural choice for the substrate because of their simultaneously excellent electrical insulation and high thermal conductivities. The ceramic is sandwiched between two metal layers—the top provides an electrical ground for the semiconductor devices, and the bottom provides additional heat spreading, decreases stresses in the ceramic due to thermal expansion mismatch, and connection to the baseplate. Copper or aluminum is attached to a ceramic such as Al2O3 to create a direct bond copper (DBC) or direct bond aluminum substrate, but active metal brazing can also be used to join the metal and ceramic. Alumina is an often-used ceramic, despite its low thermal conductivity because of its “low cost and mature mass production” . Other common ceramics used include beryllia, aluminum nitride, and Si3N4. AlN is a more promising candidate because of beryllia's health and environmental hazards, its own CTE of 4.6 ppm/°C, and its high thermal conductivity of 150 W/m K . Si3N4, which is often used for active metal brazed substrates, is another substrate option that can be used for higher power or extreme temperature swing environments . Relevant properties of these ceramics can be seen in Table 1.
The substrate is connected to a thick copper or aluminum baseplate/heat spreader, often with solder. Copper or aluminum is typically chosen as the baseplate material. CTE mismatch between the ceramic and baseplate can be mitigated by using metal matrix composites (MMCs). MMCs are typically made with a base of copper or aluminum because of their excellent thermal conductivity and aluminum's superior specific thermal conductivity . Reinforcing with additives such as SiC, carbon, diamond, and carbon nanotubes (CNT) have allowed tailoring of CTE, while maintaining high thermal conductivity for improved and more reliable thermal management [18,19]. Examples of MMCs are presented in Table 2. A TIM is applied between the baseplate and the heat sink to decrease the contact resistance . Solders can also be used as a thermal interface material. Yet despite their higher thermal conductivity than TIMs, the latter are typically preferred due to their better mechanical compliance. This approach can be seen in the power electronics module of the 2004 Toyota Prius .
Evolution of Thermal Packaging in Hybrid and Fully Electric Vehicles
The progress in the hybrid electric vehicle industry shows the changes in advanced power electronics packaging and cooling. Tracking these improvements demonstrates incremental advances in achieving solutions that simultaneously have low thermal resistance, high reliability, and low cost while increasing power density. This section reviews several hybrid and fully electric vehicles of the last decade and their thermal packaging strategies.
The 2004 Toyota Prius uses conventional power packaging methods as seen in Fig. 4 . The most notable difference between the packaging in Fig. 3 and the 2004 Prius is the use of aluminum, which has a superior specific thermal conductivity. Design engineers may prefer aluminum's low density and good thermal conductivity over copper, particularly in electric vehicle applications, where weight can decrease fuel efficiency. The thermal design variation comes in the selection of the solder, ceramic, and TIM. These are a lead-free solder, AlN, and a ZnO thermal paste, respectively.
Figure 5 illustrates the power electronics packaging used in the 2008 Lexus LS 600 h . The 2008 LS 600 h implements double sided cooling, which can significantly decrease thermal resistance by cooling both sides; Liang showed a reduction in thermal resistance by almost 40% compared to a traditional module . This also replaces the topside interconnections with copper spacers and plates. The spacers are made to address the varying power device thicknesses. The typical DBC or direct bond aluminum ceramic structure is not seen in this LS 600 h's packaging. The emitter and collector copper plates address the current handling and heat spreading that the topside of the DBC structure provides in a typical package, while the Si3N4 tile addresses electrical isolation. Eliminating the metallic ceramic layer decreases costs and the failures that come with it (e.g., delamination and cracking of the ceramic). However, thermal greases come with large thermal resistances compared to the rest of the package, which will severely lessen the increase in thermal performance brought by double-sided cooling architecture utilization . The 2013 Toyota Camry utilizes a similar design.
The 2010 Toyota Prius modified the 2004 design by decreasing the distance between device and coolant nearly 60% from 9.0 mm to 3.8 mm . The modified packaging can be seen in Fig. 6. The thermal interface material and the base plate were removed to decrease package thickness. Elimination of these will have clear improvements in thermal performance, as interfaces materials (particularly greases or other standard TIMs) have high thermal resistances, but the decrease in lateral heat spreading can also increase the hotspot intensity. 3 mm holes can be seen in the aluminum layer below the ceramic, which were introduced for stress relaxation caused by CTE mismatch between aluminum and AlN .
Nissan's fully electric 2012 Leaf utilized an inexpensive package seen in Fig. 7 . A copper–molybdenum buffer plate is attached to the device using a lead-free solder. This is to grade the coefficients of thermal expansion more gradually and relax the resultant stresses that the device sees. The buffer plate is mounted on a copper plate, which serves as the bottom electrical pathway. The ceramic has been replaced by an insulating pad. Thermal grease is applied to both sides of the pad to decrease contact resistances. The 2012 Leaf decreases costs by using an insulating pad instead of a DBC ceramic, but this increases thermal resistance beyond that of a ceramic-based assembly. This setup can also improve transient thermal performance.
Figure 8 demonstrates the packaging used for the 2014 Honda Accord inverter . The baseplate has been removed from the package, providing a compromise between the 2010 versus 2004 Prius in distance from die to coolant. Similar to the Lexus 2008 LS 600 h, copper is used near the device. A possible explanation for the use of copper near the device is to increase heat spreading near the device, where it most impacts thermal resistance. Further away from the device—particularly the heat sink, where the cold plate is often integrated into the housing—aluminum is usually used. A moderate source of thermal resistance was, as expected, the ceramic, which despite the use of Si3N4, will generally underperform with respect to most metals used for heat transfer applications. However, the convective resistance was found to be the highest thermal bottleneck with the elimination of the TIM.
The second generation Chevrolet Volt was introduced by General Motors in 2016, which utilized a double-sided cooling approach seen in Fig. 9 . Similar to the 2008 Lexus LS 600 h module and the plug-in hybrid Cadillac CT6, the power electronics module uses double-sided cooling . However, unlike either of these modules, the Volt uses ceramics for electrical isolation instead of an isolating pad. Note that although no spacers were illustrated , it is likely that they were utilized to address insulated gate bipolar transistor (IGBT)/diode height discrepancies. Another large change from other HEV/EV models is the heat sink. Most prior heat sinks examined (when available) used straight aluminum channels. However, the 2016 Volt used a copper cold plate with staggered semi-cylindrical protrusions so that the liquid pathway resembles a continuous “s.” A possible reason is that the prior cold plates were often integrated into the surrounding aluminum housing, whereas the Volt's cold plate appears to be separate. Therefore, it is likely that the use of copper can be justified as it will not increase the weight as drastically as when used in the Volt (Fig. 10).
The cooling enhancements are (when available) usually longitudinal fins although the Chevy Volt has meandering fins. These fins were also manufactured using different methods, including machining, brazing, and casting. The coolant was water/ethylene glycol mixture to allow operation in ambient temperature range up to 105 °C. (H)EV modules are gradually evolving toward thinner single-sided cooling modules and/or high-performance double-sided cooling. But the implementation of SiC devices will require adjusting or reimagining thermal packaging approaches from die-attach to the entire cooling system.
Despite the lag between R&D and mass production, examining the power modules of recent (H)EVs can provide insight into the direction the industry is headed. Each power module uses (when specified) tin-based and lead-free solders. Additionally, thermal grease use is pervasive in double-sided modules. Thermal greases are cheaper and not mechanically constraining, which may increase reliability. In these modules, the thermal resistance is already dramatically decreased, so the penalty of using greases is lessened. There is a general trend toward double-sided modules, seen in Toyota's fourth generation power conversion resulting in smaller form factor, decreased thermal resistance, and improved efficiencies . In single-sided cooling modules, there appears to be a migration away from modules using TIMs with the combination of elimination of various layers. When TIMs are absent, the largest bottleneck is the convective resistance.
Advanced Packaging Options for Silicon Carbide
SiC devices display stable operation at much higher temperatures than Si devices. Therefore, as SiC devices see operational temperatures exceeding 200 °C and accompanying increased temperature swings, next generation modules will require materials with better high-temperature performance and reliability. And as package sizes decrease (Toyota has demonstrated an 80% size reduction in their power conversion unit utilizing SiC devices ), previous thermal management solutions will not be sufficient to maintain operable device temperatures. Three areas have seen significant advancements as power electronics packaging evolves to provide proper thermal management to silicon carbide device-based packages—die-attach, TIMs, and heat rejection technologies. Promising solutions for each are discussed and examined for their performance and practicality with respect to high-temperature, high-power packaging, with specific considerations for automotive applications.
Alternatives for industry standard tin-based solders are needed as they have poor high-temperature performance above IGBT temperatures of around 125 °C. These alternatives need to consider melting temperature and wettability for manufacturability, and thermal fatigue resistance and CTE mismatch with the terminal, die, and substrate for reliability. Figure 11 shows the melting temperature and CTE of various die-attach materials, which can serve as benchmarks for feasibility and reliability for new high-temperature die-attach materials. Note that CTE and melting temperature are not the only metrics for the reliability of a die-attach. Mechanical properties such as the modulus of elasticity, ductility, and yield strength also influence reliability. Other features such as porosity can impact properties such as density and elastic modulus , which in turn impact reliability. Additional important characteristics of die-attach materials are the electrical and thermal conductivity . Khazaka et al. discussed various options including high-Pb solder alloys, lead-free solders, transient liquid-phase (TLP) bonding, and nanosilver paste sintering .
High-Pb solder alloys have excellent wettability, high ductility, melting points up to 300 °C, and excellent reliability—all desirable characteristics for a solder. Overall reliability results partly from their high melting temperatures (Tm), but also from their resistance to thermal fatigue, electromigration, and the formation of intermetallic compounds . Note that although Tm is not the definitive measure of reliability, using solders under 0.3–0.5 Tm will generally prevent rapid creep [34,37]. In addition to these characteristics, when legislation began phasing out high-Pb solders, other high-temperature alternatives did not have sufficient long-term field data to confirm long-term reliability. However, health hazards of lead  resulted in phasing out its use in consumer electronic and electrical equipment. High-Pb solders are still used in certain areas such as military and aerospace applications, but health concerns prevent these die-attach options from being widely used. Properties of various die attach options including high-Pb solders can be seen in Fig. 12.
Lead-free solders present an alternative for high-temperature die-attach. Gold-based alloys, including AuSn, AuGe, and AuSi, have gained momentum due to their eutectic melting points of 280 °C, 356 °C, and 363 °C, respectively. Although they are also suitable from an engineering perspective, gold-based solders are cost-prohibitive and therefore not an acceptable choice for use in mass production. They could be employed in niche applications where cost is not a limiting factor . These solders are also very stiff, which can transfer stresses to the device and cause die cracking. Other alloys, including zinc and bismuth, have been studied for their potential as well [39,40], but they come with issues including high processing temperatures during application, and poor thermal and electrical conductivity. Furthermore, lead-free solder options can see excessive intermetallic formation, which can decrease joint strength. An and Qin looked into the effect of intermetallics on solder joint microcracking both experimentally and with micromechanical finite element models .
Transient liquid-phase bonding uses a low melting point metal, either placed or allowed to flow between two higher melting point metals. A typically large, constant pressure (ranging from ∼1 kPa to ∼100 MPa) is applied to ensure contact, accommodate non-co-planarity, and promote bonding. As the temperature exceeds the intermediate layer's melting temperature, the thin layer of intermediate metal liquefies. The liquid phase promotes high diffusion rates, and the die-attach subsequently solidifies then homogenizes. The end result is an alloy with a melting temperature much higher than the interlayer metal's melting temperature, which can even exceed 1000 °C in some cases. Significant advantages are the low bonding temperatures and mechanical strength similar or higher than the substrate materials. Cook and Sorenson listed disadvantages including its time-consuming nature and the specialized equipment this method requires , although there has been research into high-throughput solutions. TLP bonding also causes intermetallic formations which lower bond strength and ductility, and continued phase evolution can be accompanied by voiding.
The final die-attach technology examined is silver sintering. Silver has excellent thermal and electrical properties, but its melting point is too high (961 °C) to use without risking damage to electronic components. Silver based solders take advantage of these excellent properties. Sintering further improves silver's performance as a die-attach because it takes advantage of silver's thermal and electrical conductivity, as well as its high melting temperature, while not exposing electronics to these high temperatures normally needed to melt the die-attach. The process uses silver pastes which are pressurized up to 40 MPa and heated to sinter the silver particles together . Using microscale or smaller particles decreases the temperature to which the die-attach needs to be heated to 0.2–0.4 times silver's melting point, after which the melting point of the die-attach rises to the melting point of bulk silver . Nanocopper sintering has also recently been explored for power electronics packaging [44,45]. Its use can overcome the barriers of high cost and electrochemical migration, both of which are seen in silver sintered joints. However, the technology is not as mature as silver and requires more investigation. One of the factors preventing mass implementation of silver sintering is the need for specialized pressure sintering equipment. Two developments that have improved the high pressure requirements needed—nanosilver particle sintering and “pressureless” sintering. Nanosilver sintering (where particles are typically smaller than 100 nm) decreases the pressure needed to 1 MPa or less. “Pressureless” sintering does not require any pressure, but comes at the cost of decreased reliability when compared to pressurized processes. Silver sintering has been shown to have better long-term reliability compared to leaded or lead-free solders with respect to die shear strength, thermal/electrical conductivity, and cycles until failure. The prominent thermal fatigue failure mechanism results from the merging of microcavities at grain boundaries. Its porous nature leads to creep at elevated temperatures as the microstructure continues to evolve and coarsen. Furthermore, similar to gold solders, silver's stiffness may limit die size because of the risk of cracking. Despite these problems, companies such as Infineon Technologies and Semikron Eletronik have begun using sintered silver joints as part of their products .
Thermal Interface Materials.
The largest thermal management barriers in traditional power electronics modules were the TIM and ambient heat rejection via air cooling. With the incorporation of liquid cold plates and high heat flux removal techniques, TIMs have become the largest bottleneck to thermal management . Narumanchi et al. simulated a Toyota Prius inverter to show the effect of thermal interface material conductivity . Their baseline case showed 23 °C temperature rise across the TIM for a heat flux of 95–120 W/cm2 and show that the rise across the interface can become an order of magnitude smaller with significant TIM improvements. Mahajan et al. reviewed the design characteristics of TIMs and some commonly used materials . In order to increase both performance and reliability, a TIM's effective thermal conductivity should be maximized via reduction of contact and bulk resistances, and its reliability should be ensured via testing and understanding of failure mechanisms. It is also important to consider a variety of factors including surface finish, thermal performance, long-term reliability, and cost.
Commonly used materials include greases, phase change materials (PCM), and gels . Thermal greases are made of a polymer base with a ceramic or metallic filler. They do not cure, have high bulk thermal conductivity, and cost less than other options. Disadvantages include pump-out and spillage during applications. Pump-out occurs when the thermal expansion pushes the grease out of the interface. Application can be messy, with the possibility of spillage on other potentially sensitive components. Phase change materials mitigate the problems with thermal greases discussed earlier. PCMs melt at some temperature below the device temperature, which mitigate the temperature rise due to transient heat fluxes , while increasing thermal conductivity as the molten PCM fills microscopic voids. They are easier to apply than greases because they are solid as opposed to a viscous paste, and require no cure. Gels are typically silicone with embedded ceramic or metallic particles. Their thermal conductivity is typically lower than that of thermal greases. They require curing, after which they do not suffer pump-out or similar effects. Alternatives for popular, typically tin-based solders are needed as they have poor high-temperature temperature, high-power packaging, with specific considerations for automotive applications. Several examples of commercial TIMs are given in Table 3.
TIM research has focused on three primary areas to enhance performance: using novel high-performance fillers, studying TIMs on the micro and nanoscale, and developing TIMs based on carbon allotropes [48–55]. Fillers such as aluminum nitride, silver nanowires, carbon nanotubes (CNTs), and alumina micro particles can be seen. Some TIMs utilize a combination of micro and nano fillers to form percolating networks such as Sanada et al. and Zhang et al. [51,52]. The latter obtained thermal conductivities up to 140 W/m K—comparable to high-end solders. The use of various carbon allotropes such as graphene, graphite, and carbon nanotubes are discussed in Refs.  and [53–55]. Carbon nanotubes are the most promising of all the carbon allotropes and have emerged at the forefront of TIM research. Their axial thermal conductivity reaches theoretical values of 6600 W/m K, making vertically aligned (VA) CNTs a promising option . They are mechanically compliant (which decreases stresses caused by CTE mismatch) and are chemically stable over a wide range of temperatures. There has also been some work with randomly distributed nanotubes . Cola  discusses the three most promising CNT arrays: one-sided interfaces, two-sided interfaces, and CNT-coated foil interfaces. The first type is grown on a substrate, and the ends of the CNTs are in contact with but not bonded to the opposite surface. The second type has two arrays grown on both sides of the substrate, after which the two substrates are pressed together and the nanotubes entangle in each other. The last type is grown on both sides of a foil, which allows high quality CNT growth at around 700 °C without other components seeing these extreme temperatures. Experimental values of 7 mm2 K/W, 4 mm2 K/W, and 8 mm2 K/W for each of these, respectively have been reported. It is concluded that further improvements, such as maximizing CNT-surface contact and decreasing interfacial resistances between the CNT arrays and the substrate, could potentially help achieve the theoretical resistance limit of approximately 0.1 mm2K/W (approximately two orders of magnitude better than high end conventional thermal greases ). Several experimental TIM research results demonstrating the enhancements mentioned are summarized in Table 4.
High Heat Flux Cooling.
High-power power electronics deal with high voltages and currents, and the large amounts of power result in elevated heat generation and the necessity for efficient cooling. For example, the 2010 Prius can supply 60 kW of power to the motor, and the 2800 LS 600 h's motor is rated for 110 kW. Even at higher efficiencies of 96%, these modules will generate 2.4 and 4.4 kW of waste heat, respectively . Device-level heat fluxes are currently 100–150 W/cm2 and are projected to rise to 500 W/cm2 with next-gen devices . Air cooling can be used as seen in BMW's Mini E , but higher power densities typically require the enhanced performance of liquid cooling. All of the EV power electronics modules explicitly reviewed above utilized single-phase liquid cooling, so as package volumes decrease sharply as a result of SiC device implementation , compact, and high-performance cooling options beyond single-phase cooling must be investigated. Efficient cooling can increase device and package lifespan.
High-performance cooling can also improve output power ratings, as shown by Bhunia et al., who demonstrated the performance increase of a 1200 V/150 A IGBT inverter when cooling with jet impingement compared to conventional air cooling . Simultaneously, it is important to consider weight, volume, reliability, and cost tradeoffs that accompany high heat flux cooling technologies. The higher heat removal capabilities of liquids, which can be augmented by porous media , enhanced coolants , or micro fins , come at the cost of an added coolant loop that brings complexities such as sealing and added weight/volume to the design. High performance cooling methods can be combined with proper fluid distribution design to appropriately target hotspots [63,64].
Commonly researched high heat flux cooling methods for power electronics are jet impingement cooling, spray cooling, and mini or microchannels. Each of these can also be used with two-phase flow. Sharar reviewed the benefits of two-phase flow, first and foremost among them the increased heat transport ability because of latent heat of vaporization. Another advantage mentioned is that flow boiling requires lower pumping power because of its improved thermal performance and resulting lower required flowrate . Wang et al. simulated single-phase and two-phase cooling of a Toyota Prius motor inverter and showed that the two-phase cooling resulted in decreasing the maximum temperature difference from 32 °C to 3.9 °C degrees, exhibiting the ability of two-phase flow's ability to reduce temperature gradients . Despite the obvious advantages, problems accompanying two-phase flows, such as critical heat flux (CHF) and flow instabilities have prevented widespread adoption. The three aforementioned methods have each demonstrated heat flux removal capabilities of approximately 1000 W/cm2 or more . There has also been work on combining multiple types of cooling schemes. For example, Ditri et al. used impinging jets to cool 4 GaN HEMTs and then directed the fluid through microchannels toward the exit. Sung and Mudawar directed slot jets into microchannels. Coursey et al. used spray cooling on open microchannels [67–69]. All of the aforementioned strategies are attractive options for high heat flux cooling, and each of these methods is accompanied by different advantages and disadvantages.
Jet impingement directs jets toward the surface that needs to be cooled as seen in Fig. 13(a). Multiple jets can be used to improve temperature uniformity. The high heat transfer capability is due to the thin boundary layer formed near the hot surface. The cons of this method include difficulties with fluid removal, uneven cooling due to boundary layer thickness increase, and its aggressive nature, which may result in surface erosion over extended periods . There are further subtypes of jet impingement—submerged, where the jet is surrounded by a liquid, and free jet, where the jet is surrounded by air. Increasing system turbulence by surface enhancements or by increasing the jet's turbulence has been used to increase heat transfer performance . Two-phase jet impingement has been investigated, and several conclusions run contrary to expectations, including improved temperature uniformity with a single-jet configuration instead of multi-jet array and an almost negligible effect on pressure-drop .
Spray cooling (Fig. 13(b)) directs a spray of droplets toward the heated surface. The droplets impinge on the surface, where they can agitate the liquid film formed, or evaporate. Structured surfaces are the primary enhancement method for sprays . Sprays can provide more uniform cooling than jet impingement, but they come with downsides including difficulties with fluid removal and management, inherently high driving pressure requirements to generate a spray, nozzle clogging, and extensive parameters to control for in design . Mudawar  mentioned the unpredictability of spray nozzles, noting that nozzles would often have different spray patterns despite being produced in the same manner. Two-phase spray cooling lacks complete models that are both accurate and widely applicable. Also, these models do not take into account coupled effects such as substrate heat conduction. And the impact of variables on CHF such as dissolved gases and spray angles needs further investigation .
Mini and microchannels (Fig. 13(c)) can be arbitrarily defined as having hydraulic diameters of 200 μm–3 mm and 10–200 μm, respectively . Since the convective heat transfer coefficient is inversely related to channel width, microchannels can obtain extremely high heat fluxes . This type of cooling is generally more compact than jet impingement or spray cooling—Mandel et al. were able to obtain heat densities approaching 490 W/cm3 with bonded SiC microchannel test chips . However, they can have significant temperature gradients, stresses from the high driving pressures necessary to pump the coolant through the small channels, and fouling (which causes increased thermal resistance and may lead to clogging and system failure) . Major barriers to implementing two-phase microchannels are twofold: instabilities and critical heat flux. Instabilities are caused by rapid bubble growth, and the resulting temperature and pressure variations can be both high-frequency, large-magnitude changes as seen in Ref. . Instabilities can be minimized with modifications like inlet restrictions or expanding microchannels. Dryout can be delayed with artificial nucleation sites . The research into the transient behavior of boiling flows in microchannels is scarce, although researchers such as Huang et al. have conducted experiments characterizing the transient thermal behavior of microchannel evaporators . The lack of information on transient behavior is another barrier as practical loads are typically not steady-state.
While typical power electronic packages are cooled as seen in Fig. 3, there has been focus on decreasing package device-to-coolant resistance. This decrease in overall resistance can be achieved in two manners—by selecting materials with increased thermal conductivity or by eliminating some of the package layers. Removal or combination of layers can decrease resistance by eliminating contact resistances and decreasing total thickness. Bennion et al. eliminated the TIM by combining the baseplate and the heat sink and removing the baseplate . Kelly et al. and the 2010 Toyota Prius cooled the ceramic substrate directly [23,81]. Yin et al. examined using minichannels to cool the electrically insulating ceramic . Lee et al. modeled single- and two-phase liquid cooling using microchannels of the silicon carbide substrate of a GaN-on-SiC HEMT . Recently, embedded cooling has also been researched for vertical power devices, potentially by using microchannels etched into the semiconductor [83,84] or incorporating microchannels into the electrode, or by integrating a silicon cooling structure at the device level . Electrical simulations have shown no negative side effects with respect to electrical performance , although manufacturability, modularity, and cost are potential barriers to adoption despite its excellent heat transfer performance. It is important to remember that one of the functions of power electronics packaging is to spread heat. Removal of layers decreases overall thermal resistance, but may decrease heat spreading, requiring even higher heat flux removal methods. The aforementioned papers utilized increasingly aggressive cooling as the spreading layers decreased, going from large channels to jet impingement, minichannels, and then microchannels. Additional thermal resistances between the coolant and the chip can be eliminated by directly cooling the chip. Direct or immersion cooling is where the coolant liquid is in direct contact with the semiconductor device. An overview of direct-immersion cooling techniques can be seen in Ref. . While immersion cooling decreases the overall thermal resistance, problems such as coolant leakage and compatibility requirements make for noteworthy shortcomings, resulting in a preference toward indirect cooling.
Needs and Future Prospects for Thermal Packaging of Power Electronics
As hybrids and fully electric vehicles gain popularity, SiC will move from testing and development to mass production and implementation. Its high operating temperature and high frequency properties will require redesign of the package, including the use of high-temperature materials, smaller passives for high-frequency operation, and minimized system parasitics . SiC devices, which can withstand temperatures in excess of 200 °C, will enable an often-stated goal of combining vehicle cooling loops into one coolant loop with an inlet temperature of 105 °C, which will yield significant space and cost savings . This can be taken another step by combining all EV coolant loops . But the combination of fluid loops means that the packaging will see minimum temperatures of 105 °C and up to or exceeding 200 °C, whereas the maximum junction temperature of silicon-based units is 125 °C. Power cycling will then cause even larger temperature swings, requiring more reliable packaging materials that can function at ever-increasing maximum temperatures. Three critical components were identified in technological gaps for next-gen power modules: die attach materials, thermal interfaces, and high heat flux cooling techniques. Currently, there is no obvious solution for each of these discussed packaging alternatives.
As temperatures exceed reliable operating temperatures of tin-based solders, there is still no drop-in solution for reliable high-temperature die-attach except high-lead solders. Silver sintering is the current frontrunner, especially as ambient temperatures continue to increase. Other options may have more desirable mechanical qualities than silver sintering, but their melting temperatures will be approached or exceeded as automotive applications continue to see higher temperatures . Even at 200 °C, the majority of high-temperature solders will not reliably function. But lack of equipment for applying simultaneous high temperatures and pressures and failure caused by microstructure evolution must first be addressed before silver sintering can become the new standard die-attach. Further research into pressureless sintering and delaying or preventing microstructure evolution might push silver sintering into this position.
Thermal interface materials are still the bottleneck in further reduction of thermal resistance in most power electronics packages. Reducing the use of TIMs is attractive from a thermal perspective, but complete elimination will not be possible. Minimizing usage when possible is an attractive strategy, as demonstrated by the direct substrate cooled packages. High thermal conductivity fillers and carbon-based materials should be further investigated as both an enhancement and a basis for TIMs.
Double-sided cooling architectures appear adequate to provide the cooling capability for SiC modules for the current generation, although high heat flux removal technologies should be investigated for next-generation modules. Two-phase cooling methods should be further investigated, especially from the standpoint of having a single coolant loop. However, industry adoption of these technologies will require demonstration and understanding of their long-term reliability. New liquid coolants suitable for reliable operation temperatures in the range 150–250 °C can significantly advance thermal packaging for future automotive applications.
This research is supported by the members of the High-Power, High-Temperature Electronics for Electric Cars Consortium of the 3D Systems Packaging Research Center at the Georgia Institute of Technology.