Speed Up Quantum Transport Device Simulation on Ferroelectric Tunnel Junction With Machine Learning Methods

As the device size scales down to the nanometer regime, quantum effects play an important role in device characteristics and performance. Quantum transport device simulation based on the nonequilibrium Green's function (NEGF) has been extensively applied to simulate the nanoscale devices. The NEGF simulations, however, can be computationally expensive, especially in the presence of scattering. In this study, a machine learning (ML)-based framework is developed, targeting on replacing the computationally intensive NEGF simulations. This framework first learns a sparse representation of a quantum transport property of interest and then trains a model to describe the quantitative mapping relation between the device parameters and properties. Also, the accuracy is further improved with the application of feature engineering. As an example, a graphene-ferroelectric-metal (GFM) ferroelectric tunnel junction (FTJ) is simulated. The results show that the ML-based framework allows circumventing the NEGF calculation and simultaneously maintaining high accuracy in quantum transmissions and tunneling I-V characteristics. This ML-based framework can be applied to speed up the quantum transport device simulations and enable efficient tunneling device design.