Lithographic Simulator Based on Deep Learning with Graph Input

Abstract

This paper discuss a simple deep neural network which aimed to finish the simulation of the lithographic process. It can be finalized by a more comprehensive model formed by combined networks each for different parts of lithographic process. The advantage of the DNN is that it uses a graph input as the representation of the layout. As a result it can be easily combined with the current industrial software. Furthermore, this DNN can be applied reversely to generate a regularized pattern from data of current commercial ILT package. It will at least improve the manufacturability of ILT results generated by the current commercial package.