基于人工神经网络的封装SiC MOSFET短路耐受时间筛选方法

Developing an effective methodology to enhance the uniformity of short-circuit withstand time (SCWT) in silicon carbide (SiC) MOSFETs is crucial for ensuring device reliability and consistency in power electronic systems. This letter presents a detailed analysis of SCWT variations caused by inevitable process-induced deviations and introduces a new screening approach based on artificial neural network (ANN) technology. A two-hidden-layer ANN model is constructed using characteristic parameters extracted from SiC MOSFETs. The trained model accurately predicts the SCWT of TCAD-simulated SiC MOSFETs, achieving a maximum error of less than 15% and an average error of only 2%. This proposed method effectively identifies and removes devices with shorter SCWT without compromising the performance of reliable devices, thereby enhancing post-fabrication consistency for packaged devices.