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

Area of Science:

• * Materials Science and Engineering
• * Semiconductor Device Physics
• * Reliability Engineering

Background:

• * Single Event Effects (SEE) pose a critical reliability challenge for Silicon carbide (SiC) Metal-Oxide-Semiconductor Field-Effect Transistors (MOSFETs), especially in demanding aerospace environments.
• * Traditional Technology Computer-Aided Design (TCAD) simulations, while accurate, are computationally prohibitive and require specialized expertise for SEE prediction.
• * The need for efficient and accessible SEE prediction methods is paramount for advancing the use of SiC MOSFETs in high-reliability applications.

Purpose of the Study:

• * To develop a novel, data-driven approach for predicting SEE in SiC MOSFETs.
• * To significantly reduce the computational cost associated with SEE analysis.
• * To offer a highly accurate and faster alternative to conventional TCAD simulations for reliability assessment.

Main Methods:

• * Construction of a comprehensive dataset (52,920 SEE events) encompassing varied environmental temperatures, ion energies, bias voltages, and incidence parameters.
• * Implementation of two deep learning models: Residual Deep Neural Network (RDNN) for peak current (I0) and collected charge (Q0) prediction, and Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU) for transient current pulse prediction.
• * Application of a symmetric log-reciprocal data scaling technique for optimized model preprocessing.

Main Results:

• * RDNN model achieved high prediction accuracy with an R² of 0.99864 for I0 and Q0.
• * CNN-GRU model demonstrated excellent performance in predicting the drain transient current pulse, achieving an R² of 0.99783.
• * The proposed data-driven models offer a speed-up of five to six orders of magnitude compared to traditional TCAD simulations.

Conclusions:

• * The developed data-driven method provides a highly accurate and computationally efficient solution for SEE prediction in SiC MOSFETs.
• * This approach significantly reduces the time and resources required for reliability analysis, making it more accessible.
• * The findings suggest a promising alternative for SEE prediction in SiC MOSFETs and potentially other semiconductor devices, accelerating development and deployment in critical applications.