Doctoral Dissertation
Wide-bandgap semiconductors for next-generation power electronics systems
Wide-bandgap power semiconductors promise to reshape the power electronics landscape, opening completely new use cases and increasing efficiency and power density in existing ones. Most notably, gallium nitride (GaN) and silicon carbide (SiC) were successfully commercialized in the past decades, with theoretical benefits over silicon of multiple orders-of-magnitude. When combined with soft-switching techniques and topologies, these wide-bandgap materials have the potential to move power conversion to MHz operating frequencies, radically shrinking power converters and enabling new fabrication methods with the frequency-driven reduction of passive component requirements. Unfortunately, soft-switched converters built at MHz frequencies have consistently underperformed their modeled efficiency, as this work shows for three DC-RF inverters at high- and very-high-frequency. These inverters have measured semiconductor losses nearly an order-of-magnitude greater than expected from manufacturer-provided simulation models, a discrepancy that demands investigation. These losses are attributed to the process of resonantly charging and discharging the output capacitance (Coss) of the power semiconductors, a loss mechanism termed “soft-switching losses” or “Coss losses.” Our measurements constitute the first recognition of this problem in GaN HEMTs, and these initial measurements are then extended to SiC and Si MOSFETs, finding dependencies and scaling laws for each device class. To complete the understanding of losses at high-frequencies, the well-understood phenomenon in GaN HEMTs of dynamic on-resistance is then revisited. Our work conclusively shows that dynamic on-resistance cannot be accurately characterized using the standardized double-pulse-test, and uses the underlying physics to determine the parameters that must be controlled for accurate reporting. Using this measurement framework, this work extends the dynamic on-resistance measurements to MHz frequencies for the first time, finding that the majority of the dynamic effects in soft-switched converters occur below 1 MHz for the tested device. With both off-state and on-state losses precisely understood at MHz frequencies, the promise of high-frequency power conversion can finally be realized. While adopted widely in cell phones, inductive wireless power transfer for higher-value applications (e.g. electric vehicles) is beset by both low performance and high cost due to the limitations of litz wire. At 6.78 MHz, the first international industrial, scientific, and medical (ISM) band above 200 kHz, litz wire can be completely eliminated, paving the way to low cost, small, light, and high-performance systems. A 1 kW DC-DC converter that transfers power across a 2 cm gap with 6.6 cm diameter coils at over 95% efficiency is demonstrated, a new benchmark in power density and efficiency for MHz-frequency wireless power transfer. This performance would, plainly, not have been possible without the identification and quantification of Coss losses. Our future power, transportation, and computing infrastructures are dependent on the implementation of wide-bandgap power semiconductors to reduce size, weight, and cost while increasing efficiency to address the climate challenge. This thesis is our small contribution to meaningfully improving these semiconductors and showing what’s possible for the next generation of power conversion.