An artificial neural network based electro-thermal behavioral model for SiC MOSFETs operating under wide temperature range and short-circuit condition

To predict the switching behaviors of SiC MOSFETs in complex circuit systems, a hybrid modeling method combining an artificial neural network (ANN)-based electrical model with a thermal network model is proposed. The proposed ANN-based electrical model is trained using drain current (I_DS) measurements obtained from single-pulse short-circuit (SC) tests under high drain-source voltage (V_DS) conditions to improve transient simulation accuracy. Through comprehensive comparison between simulation and experimental results, the proposed ANN-based electrical model achieves dynamic parameter errors ranging from 0.63 % to 18.9 % across a wide temperature range (25–100 °C), achieved by optimized network topology, weights, and bias selection. Furthermore, the proposed hybrid model enables transient behavior prediction under extreme SC conditions, achieving errors between simulated and measured peak SC currents ranging from 1.22 % to 9.9 %. Furthermore, validation tests conducted at 100 °C and 125 °C confirm the model's reliable temperature prediction performance, with all dynamic parameter errors remaining below 13.2 % at 100 °C and 41.65 % at 125 °C, respectively.