Modeling and simulation of semiconductor devices using machine learning

Abstract

This paper delves into the modeling and simulation of semiconductor devices through the utilization of machine learning. We propose an approach leveraging machine learning to model device performance and physical characteristics based on semiconductor device parameters, subsequently employing these models for optimization and simulation. To elaborate further, we employ Bayesian optimization methods to maximize the performance of a state-of-the-art 3-nanometer node nanosheet field-effect transistor within a five-dimensional design space. Furthermore, through the application of physics-informed machine learning, we efficiently model and simulate the spatial physical attributes of nanowires, which represent the next-generation gate-all-around device. This study introduces and proposes a novel methodology for effectively addressing the challenges of modeling increasingly complex semiconductor devices by harnessing the latest advancements in machine learning techniques.