Prediction of single event effects in SiC MOSFET devices based on deep learning

Single event effects (SEE) are a critical reliability concern for Silicon carbide (SiC) MOSFETs, particularly in aerospace applications. While traditional Technology computer-aided design (TCAD) simulations offer accurate SEE prediction, they are computationally intensive and require specialized knowledge. This paper proposes a novel data-driven prediction method with SRIM-TCAD integrated modeling. First, a dataset of 52,920 SEE events in SiC MOSFETs is constructed, considering diverse environment temperatures, heavy ion energies, drain bias voltages, incidence angles, incidence positions, and incidence locations. Then, two different deep learning models are adopted: one to predict the drain transient current pea (<i>I</i>₀) and total collected charge (<i>Q</i>₀), and another to predict the drain transient current pulse. Residual deep neural network (RDNN) is used for predicting the<i>I</i>₀and<i>Q</i>₀. Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU) is applied for the drain transient current pulse. A symmetric log-reciprocal data scaling technique is proposed and applied during preprocessing for both models. Experimental results show that the RDNN achieved an<i>R</i>²of 0.99864 for<i>I</i>₀and<i>Q</i>₀prediction, while the CNN-GRU model predicted the drain transient current pulse with an<i>R</i>²of 0.99783. These models provide a prediction speed-up of approximately five to six orders of magnitude compared to TCAD simulations. The proposed method demonstrates high accuracy and significant computational cost reduction, offering an alternative for SEE prediction in SiC MOSFETs and potentially other semiconductor devices.