Feasibility of gate oxide degradation identification in SiC MOSFETs using convolutional neural network
Year: 2026
Type: article
Abstract: This study investigates the feasibility of identifying gate oxide degradation using a convolutional neural network (CNN) trained with gate current waveforms. A simplified metal–oxide–semiconductor field-effect transistor (MOSFET)-based discussion is first provided to relate changes in oxide/interface states to the input-capacitance-gate-voltage characteristics, and hence, to transient gate current waveforms. A healthy 1.2-kV silicon carbide (SiC) MOSFET and an aged device subjected to high-temperature gate-bias stress were prepared. Gate current waveforms were obtained via double-pulse tests under different input voltages, load currents, and case temperatures. Each validation case was repeated ten times using different random seeds. The turn-off gate-current waveforms achieved mean classification accuracies of 99.8% and 99.5% in the leave-one-voltage-out and leave-one-current-out validations, respectively. They also generally exhibited less variability between runs than the turn-on waveforms. Gradient-weighted class activation mapping (Grad-CAM) visualization indicated that the CNN primarily used waveform intervals corresponding to the negative gate–source-voltage region, where the measured difference in the C iss - V GS characteristics between healthy and aged devices was large. These results demonstrate the potential of CNN-based gate-current condition monitoring for identifying gate oxide degradation states in SiC MOSFETs.
Institution: Chiba Institute of Technology
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,500
OpenAlex ID: W7167570475
