GMUNet-ILT: A lightweight MLP-based network for Inverse Lithography Technology

With the continuous scaling down of critical dimension (CD) in advanced integrated circuits, Resolution Enhancement Techniques (RETs) are employed to improve the printing performance in lithography processes. Inverse lithography, a widely studied RET, precisely controls the printed images on wafers. As a type of computational lithography, inverse lithography often incurs significant computational costs. Therefore, this paper proposes a lightweight and fast deep learning convolutional network architecture for computational lithography, which is based on an improved U-Net incorporating shifted-window multi-layer perceptron and Ghost modules. Our approach achieves relatively low mask error and a runtime of less than 5 seconds.