Using machine learning to replace 3D particle-in-cell (PIC) simulation of nanocale vacuum channel transistors (NVCTs).

This work aims to use a neural network instead of PIC simulation to predict device characteristics of an NVCT. An NVCT is a nanoscale vacuum triode in which applying a voltage to the gate causes field-emitted electrons to travel from the emitter through the gate to the anode; such devices could replace solid-state transistors under extreme conditions (e.g., low/high temperatures, radiation, high voltage). 3D PIC simulations are expensive, so we investigate the use of neural networks to predict the results (gate current and anode current) of simulations as a small number of parameters (inputs to the neural network) are varied, including: gate-anode distance, gate-anode voltage, gate-emitter voltage, etc. We use PIC simulations to train the neural network, and to evaluate the neural network performance beyond the training data.