Graphene Nanoribbon FET Compact Model on the Basis of ANN Configuration Applicable in Different Spice Levels

Abstract

The artificial neural networks (ANNs) are widely utilized as a powerful approximator for a vast number of complicated nonlinear functions in many fields in different domains. Multi-layer perceptron (MLP) as a unique strategy in this work has been highlighted making an analytical drain current prediction of graphene nanoribbon field-effect transistor (GNRFET) simple and efficient having a linear dependence on the fundamental variables. The MLP structure in this work is configured with three hidden layers and 5-dimensional inputs giving the best result after performing different experiments. Target output as the key parameter is actually the numerically calculated drain current by the Non-Equilibrium Green Function (NEGF) method which is commonly used for the simulation of nanoscale devices. The critical parameters in the cases of gate oxide thickness, gate length, number of carbon atoms across the channel, gate voltage, and drain voltage are selected as the dimensions of the input vector impacting the obtained drain current by the ANN. The comprehensive comparison between the NEGF approach and the proposed ANN-based model revealed an excellent match and correlation between them. As a result, this model can be taken into consideration as a suitable tool in the different spice levels owing to the saving the running time and also increase in the efficiency of nanoscale circuits based on the GNRFET structure.