The traction inverter of a battery electric vehicle converts the DC voltage of the battery into three-phase currents for the motor, and its losses subtract directly from driving range. For two decades the switch of choice was the silicon insulated-gate bipolar transistor with an antiparallel diode. Silicon carbide (SiC) MOSFETs now replace it in a growing number of production vehicles, a change anticipated in early automotive work on the technology (Hamada et al., 2015) and documented in surveys of traction inverter design (Reimers et al., 2019). Gallium nitride (GaN) transistors serve mainly in chargers and DC-DC converters at present and are under study for the inverter itself.

The benefit is greatest at partial load, where a vehicle spends most of its time. A unipolar MOSFET has no fixed on-state voltage drop and no tail current at turn-off, so both conduction and switching losses fall under light load in a way that a bipolar device cannot match (Biela et al., 2011). Realising the benefit over a vehicle life of many thousands of operating hours and a very large number of temperature swings is a separate matter. Field experience with power electronics has long identified the semiconductor module as a leading source of converter failures (Yang et al., 2011), and the new devices bring failure mechanisms of their own. This review examines those mechanisms and the packaging developments intended to contain them.

What follows is a narrative review. It relies on textbooks of device physics and reliability, on widely cited review and tutorial papers, on original papers that first described particular degradation mechanisms, and on the automotive qualification documents in common use. It does not attempt a systematic search or a quantitative synthesis, and it reports no test data of its own. Device topics are treated first, packaging second, and qualification and system effects third, because the packaging choices are best understood as answers to the device and thermo-mechanical problems that precede them.

The 4H polytype of SiC has a bandgap of about 3.26 eV against 1.12 eV for silicon, a critical electric field roughly ten times higher, and a thermal conductivity about three times higher (Kimoto and Cooper, 2014). A higher critical field permits a thinner and more heavily doped drift region for a given blocking voltage, and the specific on-resistance of a unipolar device falls accordingly (Baliga, 2008). The practical result is a MOSFET rated at 1200 V with conduction losses that would require a bipolar structure in silicon. GaN has a bandgap of about 3.4 eV and owes its performance to the two-dimensional electron gas that forms at an AlGaN/GaN heterojunction, which gives the lateral high-electron-mobility transistor its low resistance and very small capacitances (Mishra et al., 2002).

Early assessments of the two materials (Hudgins et al., 2003) and later surveys (Millán et al., 2014; She et al., 2017) agree on the division of territory. SiC, with vertical devices on native substrates, suits the voltage and current levels of the traction inverter, particularly as battery voltages rise towards 800 V. Commercial GaN power transistors are lateral devices grown on silicon and are concentrated at blocking voltages up to about 650 V (Jones et al., 2016). Kaminski and Hilt (2014) caution that the two should not be treated as interchangeable: their structures, failure modes and maturity differ in ways that matter more to the designer than the similarity of their bandgaps.

The reliability questions raised by wide-bandgap devices are of two kinds: those rooted in the chip itself and those that arise because a smaller, hotter and stiffer chip loads the package differently. The first kind is taken here and the second in the next section. Table 1 collects the principal mechanisms with the responses discussed in the literature.

The conventional power module solders the chip to a ceramic substrate clad with copper, connects its top side with aluminium wire bonds and solders the substrate to a baseplate. Its wear-out mechanisms under thermal cycling, chiefly bond-wire lift-off and solder fatigue, are well characterised (Ciappa, 2002; Lutz et al., 2011). Figure 1 relates each element of the module to its characteristic failure mechanism. SiC aggravates most of them. The chip is smaller, so heat flux is higher; the material is considerably stiffer than silicon, so more of the thermal mismatch strain is taken up by the joint beneath it; and the designer is tempted to run hotter because the semiconductor permits it.

The reviews of Chen et al. (2017) and Lee et al. (2020) describe a consistent set of responses. In die attach, solder is giving way to sintered silver, whose melting point of 961 °C lies far above any operating temperature and whose thermal and electrical conductivities exceed those of solder alloys; Siow (2014) reviews its processing and remaining questions, such as porosity control and behaviour on different metallisations. On the top side, aluminium wires are being replaced by copper wires and ribbons, copper clips, or planar and flexible-foil interconnects that also remove heat from the upper face of the chip. For the substrate, silicon nitride joined by active metal brazing offers higher fracture toughness than alumina or aluminium nitride and tolerates the wider temperature swings.

Two further trends are electrical and thermal in motive. Wide-bandgap devices switch in tens of nanoseconds, and any inductance in the commutation loop converts the current slope into overshoot and ringing. Module layouts have accordingly moved towards laminated or stacked current paths, integrated decoupling capacitors and separate Kelvin source connections for the gate circuit (Chen et al., 2017). On the thermal side, direct liquid cooling of a pin-fin baseplate removes the thermal interface material, and double-sided cooling, made possible by planar interconnects, extracts heat from both faces of the chip. Transfer-moulded packages, which encapsulate the assembly in epoxy in place of silicone gel inside a frame, have become common in automotive modules (Reimers et al., 2019).

Automotive components are qualified against agreed stress-test catalogues. Discrete devices follow AEC-Q101 (Automotive Electronics Council, 2021), and power modules for vehicle converters are commonly qualified to the ECPE guideline AQG 324, which has been extended with provisions specific to SiC (ECPE, 2021). Power cycling is central to both. The empirical lifetime relations in use descend from the work of Held et al. (1997) on IGBT modules for traction, which expressed the number of cycles to failure as a function of junction temperature swing and mean temperature.

Two difficulties attend the transfer of these methods. Lifetime models fitted to soldered, wire-bonded silicon modules cannot be assumed to hold for sintered, clip-bonded SiC modules, whose dominant failure site may have moved. And power cycling of a SiC MOSFET is complicated by the threshold drift described earlier, which disturbs the temperature-sensitive electrical parameter normally used to infer junction temperature during the test. Wang et al. (2013) argue more generally for a physics-of-failure approach in which lifetime is designed from mission profile and mechanism, not demonstrated afterwards by a fixed test list.

Reliability does not end at the module terminals. The steep voltage edges that reduce switching loss also travel down the motor cable. Morya et al. (2019) review the effects on the drive: reflected-wave overvoltage at the motor terminals, uneven voltage distribution across the first turns of the winding with attendant risk of partial discharge, bearing currents driven by common-mode voltage, and increased conducted and radiated electromagnetic interference. The close integration of inverter and motor in a vehicle shortens the cable and eases the first of these, but the others remain. In practice the switching speed of a traction inverter is often set below what the devices could achieve, which means that part of the theoretical loss advantage is traded for insulation life and electromagnetic compatibility.

Several matters are unsettled in the published record. Validated lifetime models for the new material combinations are scarce, and those that exist are mostly held by manufacturers. Accelerated tests apply regular cycles, whereas a vehicle imposes an irregular mixture of short and long, shallow and deep temperature excursions, and the rules for accumulating damage across such a mixture are not well established for sintered joints. The interaction of mechanisms, for example whether threshold drift alters current sharing between paralleled chips and thereby their thermal loading, has received little study. For GaN, the path to the traction inverter depends on higher-voltage devices or multilevel topologies and on packages able to exploit a lateral chip. Finally, condition monitoring in service, which would allow maintenance to be based on measured degradation, remains largely a research topic (Wang et al., 2013).

The literature presents wide-bandgap traction converters as a technology whose electrical case is made and whose reliability case is still being assembled. For SiC, the device-level concerns of oxide stability, bipolar degradation and short-circuit withstand are understood well enough to be managed by processing, screening and protection. The larger shift is in the package, where sintered joints, copper and planar interconnects, tougher substrates and double-sided cooling are displacing a construction optimised for silicon. Qualification practice is adapting, but lifetime prediction for the new material sets under real driving profiles rests on a thinner evidential base than designers had for the IGBT module, and that is where further published work would be most valuable.

To be declared by the author(s) before publication.

To be declared by the author(s) before publication.