Single event effects prediction of MOSFET device using deep learning

Abstract

Single event effect (SEE) is an important problem in the reliability research of integrated circuits. The study of SEE of traditional MOSFET devices is mainly based on simulation software, which is characterized by slow simulation speed, large computation and time-consuming. In this paper, a SEE research method based on deep learning is proposed. The method relies on 28 nm MOSFET. The complete drain transient current pulse, transient current peak value and total collected charge can be obtained in a short time by inputting relevant parameters that affect the SEE. The accuracy of the network for predicting transient current peak and total collected charge is 96.95% and 97.53% respectively, and the mean goodness of fit of the network for predicting the drain transient current pulse curve is 0.985. Compared with TCAD Sentaurus software, the simulation speed is increased by 5.89 × 103 and 1.50 × 103 times respectively. This method has good prediction effect and provides a new possibility for the study of SEE.