Doping Variability Modeling in 2D Material Channel FETs using Artificial Neural Network

Abstract

Two-dimensional (2D) material-based Field-effect transistors (FETs) are the future for digital technologies due to their excellent scalability and superior material properties compared to conventional silicon material. However, accurately and efficiently modeling these devices remains challenging due to quantum mechanical effects, which complicate transport through 2D materials. In this work, a compact modeling framework for the n-type and p-type 2D material FETs is proposed and implemented using an artificial neural network (ANN). A shallow ANN has been implemented to model the FETs for efficient SPICE implementation and circuit simulation. The ANN has been optimized for the number of hidden layers, the number of neurons, the activation function, the learning rate, and the number of epochs. The neural network model is capable of capturing device performance variations due to doping variations in the channel region, as well as in the source and drain regions.