Design 2D Field-Effect Transistors by Machine-Learning-Assisted Quantum Transport Approach

A machine-learning (ML) -assisted quantum transport device simulation framework is developed for multiscale modeling from interface to device and circuit made of 2D semiconductors. The results provide insight into the mechanisms enabling low contact resistance at semimetal-TMDC semiconductor interfaces and the scalability and potential performance advantages of 2D FETs from a quantum transport perspective. Machine learning methods significantly improve the simulation and design efficiency by facilitating multiscale simulations from ab initio densityfunctional simulations to device quantum transport simulations, automating design optimization of 2D FETs with multiple design targets, including power, speed, and variability. The methods also enable exploring the potential of 2D FETs in compute-in-memory applications through developing a hybrid physics and ML-based device model of 2D FETs.