Predicting CNT TFET Output Characteristics Using Deep Learning Techniques

Abstract

A deep neural network (DNN) model is proposed to predict the drain-source current <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(I_{DS})$</tex> and the high frequency response in terms of cut-off frequency <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(f_{T})$</tex> of carbon nanotube (CNN) tunnel field-effect transistor (TFET). The dataset required to train the model is prepared by a prior simulation using a previously developed MATLAB physics-based simulator where the device physical parameters are varied. Different activation functions are implemented in the DNN model and examined to achieve the best performance for the model. The trained model is found to predict output characteristics for devices with new input parameters that were not used in training with minimal mean squared error (MSE). The NN model is also compared to other machine learning models, and results show its superiority in terms of achieving the lowest MSE.