Physics-Based Performance Mapping of Ultra-High-Voltage SiC IGBTs and MOSFETs With Machine-Learning Acceleration
Year: 2026
Type: preprint
Abstract: Ultra-high-voltage SiC IGBTs, SiC MOSFETs, series-connected Si IGBTs, and series-connected SiC MOSFETs are compared under hard-and soft-switching using a self-consistent electrothermal framework with a machine-learning surro-gate to accelerate the generation of comparison maps. Three different analyses are performed: maximum-current capability mapping, deterministic minimum-loss device selection, and machine-learning-assisted generation of optimal-device maps with deterministic correction near transition boundaries. For the 20 kV class, the SiC IGBT provides the highest current capability in the low-and medium-frequency range under the zero-voltage-switching approximation, while hard switching and lower carrier lifetimes reduce this advantage, highlighting the superior performance of the series-connected SiC MOSFET configuration. The hybrid machine-learning approach achieves a 2.1× speed-up in map generation , excluding one-time training. This framework provides an efficient first-stage technology-screening method before detailed compact-model extraction and (experimental) converter-level verification.
Source: Research Square
Author: Nazareno Donato
Institution: University of Cambridge
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7223120656