Decoupling Electrical and Thermal Aging in SiC MOSFETs via Interpretable Frequency-Domain Signatures of Observed Thermal Impedance

The coupling of electrical and thermal aging in SiC MOSFETs is a major challenge for accurate online condition monitoring, as decoupling their individual effects remains a primary technical difficulty. This work proposes a method based on the observed thermal impedance (Z_{\text{th,obs}}), a measurable quantity that contains distinct frequency-domain signatures of both aging mechanisms. Based on this physical principle, a Physics-Informed Feature Distillation framework is presented. The framework uses a knowledge distillation strategy, where a high-capacity Teacher model is first trained on a detailed feature atlas to automatically find the most informative frequency-domain features for aging diagnosis. This knowledge is then transferred to a lightweight Student model, which achieves high accuracy using only this distilled feature set. The proposed methodology is thoroughly tested through simulations and experiments on physically aged devices. Results show that the framework accurately decouples the aging mechanisms, is robust against measurement noise, and is computationally efficient, making it highly suitable for real-time applications. This work provides a practical and interpretable method for online condition monitoring of SiC MOSFETs, paving the way for enhanced operational reliability in SiC power electronic systems through the extraction of decoupled aging indicators.