Prediction of Key Metrics of Stacked Nanosheet nFETs using Genetic Algorithm-based Neural Networks

In this paper, we demonstrate the prediction of important figures of merit (FoMs) including threshold voltage (Vth), subthreshold swing (SS), on-state (Ion) and off-state (Ioft) current, of vertically stacked lateral nanosheet field-effect-transistors (NSFET) using 1) an artificial neural network generated by genetic algorithm (GA) and 2) a conventional multi-layer neural network (NN). Our work shows that the trained GA-based NN has a great capability of predicting FoMs with an average of coefficients of determination at 0.992, which is better than that of the trained multi-layer neural network at 0.987. Additionally, GA-based NN has a significant reduction of calculation time by 80% compared with that of multi-layer NN under the same computing power, which indicates the possibility to reduce the computational cost by using the auto-machine learning approach for TCAD simulation.