A multiphysics SET modeling method based on machine learning

Abstract

With continuous downscaling of transistor sizes, the sensitivity to single event effect (SET) has become one of the most important reliability issues for aerospace integrated circuits. Besides the SET, integrated circuits will be affected by multiphysics such as temperature and voltage when working in space. Currently, the commonly used modeling methods are based on physical mechanisms and the double exponential pulse current. However, both methods are hard to build an accurate SET current model when various variables are considered. In this article, a novel machine learning modeling method of multiphysics SET is proposed, using intelligent algorithms to optimize the network also. With this method, we can obtain a reasonable and accurate multiphysics SET model based on neural network with single hidden layer, and there is no need to consider complex physical mechanisms. The model data is collected from TCAD simulation. Ant colony algorithm was used to optimize the initial values of the network. The RMS (root mean square error) of modeling result is less than 2%.