AI Computational Lithography

Abstract

Machine learning based computational lithography is intended to accelerate the speed of the solutions significantly. There are three critical aspects of AI computational lithography: (1). The feature vector design, (2). The approximate mapping function construction, (3). The model training scheme. Approximate mapping function construction can be realized using forward neural network architecture in theory, model training is an art with the help of mathematical understanding, while feature vector design must achieve optimal resolution, sufficiency and efficiency simultaneously. To pave the way of successful AI computational lithography implementation, we have designed physics based optimal feature vector for AI computational lithography. By combining this feature vector design method with deep neural network architecture, a universal machine learning based computational lithography framework can be established.