Integration of Machine Learning with Statistical Variation Analysis for Ferroelectric Transistor (FE-MOSFETs)

Abstract

This paper investigates a comparative analysis of technology computer-aided design (TCAD) versus machine learning (ML) technique for ferroelectric-based substrate metal oxide semiconductor field effect transistor (FE-MOSFET), which shows the low power energy storage device and ML algorithms reduce the time or overall process. The simulations carried out through TCAD require approximately 44–46 days, encompassing variations in input parameters like gate work function ([Formula: see text]), doping concentration ([Formula: see text]), channel doping ([Formula: see text]), gate-to-source voltage ([Formula: see text]), and drain-to-source voltage ([Formula: see text]). In order to lower the computing cost of numerical TCAD device simulations, a new ML-assisted technique is provided for studying the FE-MOSFET. To reduce the runtime of physics-based TCAD by about 10–12[Formula: see text]h for each iteration, a ML-based prediction alternative is created. The proposed combination of TCAD device simulation and ML algorithms is the future of the next generation of electronics.

Keywords

• Ferroelectric base metal oxide semiconductor field effect transistor (FE-MOSFET)
• machine learning (ML)
• artificial neural network (ANN)
• weighted mean absolute percentage error (WMAPE)