Deep neural network for modeling displacement Talbot lithography

Displacement Talbot Lithography (DTL) has emerged as a critical patterning solution for various photonic applications. While enabling an effectively unlimited depth of focus and a large exposure field, the unique image formation of DTL also imposes critical modeling demand. Traditional electromagnetic solvers are accurate but computationally heavy. In this work, we present a deep learning approach using Convolutional Neural Networks (CNNs) to predict DTL aerial images without direct Maxwell equation solving. This method accelerates the computation by more than a factor of 600 while maintaining high fidelity. This work significantly lowers the barrier for DTL modeling, facilitating its broader adoption in various nanomanufacturing scenarios.