A general toolkit for advanced semiconductor transistors: from simulation to machine learning

Abstract

This work presents an overview of a set of inhouse-built software intended for state-of-the-art semiconductor device modelling, ranging from simulators to post-processing tools and prediction codes based on statistics and machine learning techniques. First, VENDES is a 3D finite-element based quantum-corrected semi-classical/classical toolbox able to characterise the performance, scalability, and variability of transistors. MLFoMPy is a Python-based tool that post-processes IV characteristics, extracting the most relevant figures of merit and preparing the data for subsequent statistical or machine learning studies. FSM is a variability prediction tool that also pinpoints the most sensitive regions of a device to the source of fluctuation. Finally, we also describe machine learning-based prediction tools that were used to obtain full IV curves and specific figures of merit of devices suffering the influence of several sources of variability.