Deep Neural Network for Device Modeling

Abstract

High quality and high accuracy device model is an essential link from device simulation to circuit simulation. However, device model using artificial neural network (ANN) often causes unphysical behaviors. A novel deep neural network (DNN) approach for device modeling is proposed in this article. A novel regression method is used to eliminate unphysical behaviors. The developed DNN model has been verified by a complete set of physical measured data with smooth and accurate predicted results.