A Data-Driven Lifetime Prediction Method for Thermally Aged SiC MOSFET Applications

Power semiconductor switches, such as Metal-Oxide Semiconductor Field-Effect Transistor (MOSFETs), are widely utilized in solid-state power controllers (SSPC) of electric vehicles, aircrafts and trains. Predictive Health Monitoring (PHM), coupled with the reliability analysis of MOSFETs, are of ut-most significance in power electronic systems. Among various PHM indicators, the ON-state resistance of MOSFETs stands out as a vital and indicative harbinger of failure. This paper introduces a data-driven methodology employing a Long-Short Term Memory (LSTM) algorithm to predict the variations of the ON-state resistance. The experimental dataset was derived from subjecting the MOSFET to power cycling under thermal stress conditions. Furthermore, the model's efficacy was scrutinized utilizing a minor fraction of the dataset for training the LSTM algorithm, showcasing robust performance. Additionally, the proposed model was validated across diverse MOSFET degradation datasets, affirming its universal applicability.