Post Exposure Bake Physics-informed Fourier Laplace Neural Network for Post Bake Simulation

Post exposure bake (PEB) is a widely used technique in advanced lithography, with its simulation being a critical step in the lithography simulation. PEB simulation involves solving several complex partial differential equations, leading to substantial computational time consumption that hinders its application to full-chip simulation. Deep learning (DL) methods have demonstrated great potential in accelerating PEB simulation. However, the lack of domain knowledge of PEB limits the ability of DL-based approaches to enhance simulation accuracy. In this paper, we propose a post exposure bake physics-informed Fourier Laplace neural network (PBFLN) for PEB simulation. This approach utilizes GPU-enabled PEB simulation as a layer within the Fourier Laplace neural network framework and iteratively conducts the prediction and correction of PEB reactant production distribution. With this approach, the critical dimension (CD) error of the resist profile between the commercial rigorous simulation tool and DL-based method is decreased by 45% compared to the previous state-of-the-art method, and computation time is reduced by a few orders of magnitude compared to commercial rigorous simulation tools. By gearing up DL with the physical knowledge of PEB, PBFLN provides a novel and efficient PEB simulation solution.