Litho-AsymVnet: super-resolution lithography modeling with an asymmetric V-net architecture

In this study, we have proposed a Litho-AsymVnet framework to perform end-to-end superresolution lithography modeling. Our Litho-AsymVnet framework with an asymmetric autoencoder architecture takes in a lower resolution mask pattern image as input and produces a 6× higher resolution resist pattern image as output. To address the boundary pixel errors, we have proposed a “trimming” method and concentric binary cross-entropy loss function to achieve a good trade-off between prediction accuracy and runtime. The experimental results show that our proposed framework produces a high quality prediction of resist pattern compared with the prior work.