Characteristics prediction of sub-5 nm nanosheet field effect transistor (FET) using a machine learning approach

Abstract

The field-effect transistor (FET) is a vital component in various electronic devices, including integrated circuits (IC_S), switching modules, and microprocessors. The current technological breakthroughs have enabled the development of N5 (5 nm node) technology for fabricating transistors. Before the production of transistors, it was crucial to engage in modelling and simulation to reduce costs and save time. Hence, developing a methodology for predicting transistor characteristics is essential for minimizing expenses and time in advancing transistor technology. Machine learning (ML) enables data-driven modeling of complex nonlinear systems to gain knowledge and enhance their performance without explicit programming. ML trains machines to optimize the processing and understanding of data. Researchers have conducted several studies to enable ML to acquire knowledge without explicit autonomous programming. However, the previous ML model achieved a coefficient of determination (R²) of only 0.98, or 98%. Here, we report on the use of Technology Computer-Aided Design (TCAD) to generate a dataset that achieves a high predictive performance. The Nanosheet Field-Effect Transistor (NSFET) can be modified by adjusting five essential factors: Gate Length (L_g), Sheet Width (F_w), Sheet Height (F_h), Spacer Length (L_sp), and equivalent oxide thickness (e_ot). An Artificial Neural Network (ANN) is used to forecast various features of NSFET, including Threshold Voltage (V_T), Off-State Current (i_off), Saturation Current (i_sat), and Subthreshold Swing (ss_lop). The results indicate that the ANN model accurately predicts NSFET properties, yielding an R² value of 0.9915 indicating strong correlation within the simulated dataset.