TCAD-Informed Surrogate Models of Semiconductor Devices

Abstract

Extensive research has been conducted over the last half-century to develop models of semiconductor devices for use in circuit analysis and simulation. Such models typically fall into one of two categories: “Cheap” analytical models that can be solved quickly but introduce significant error, and “expensive” physics-based models that achieve high accuracy at the price of prohibitive computation time. As electronic circuits grow to contain billions of active devices, there is a pressing need for new models that are both accurate and fast to compute.

In this thesis, we introduce Semiconductors.jl, a new semiconductor simulation tool written in the Julia programming language. We use Semiconductors.jl to implement performant surrogate models that approximate the behavior of fine-grained technology computer-aided design (TCAD) device models using a coarsified grid. The resulting surrogate models are shown to approximate the current-voltage characteristics of the fine-grained models to within a maximum error of 0.1% while using less than one tenth as many discretization nodes as the fine-grained baseline model.