Hybrid deep learning and physics-based framework for grayscale photolithography mask generation

Nanotechnologies make fast progress both in miniaturization and in functionality. To support these advancements, it is necessary to develop 3D patterning methods usable at an industrial scale. The grayscale photolithography appears as a promising technique for this goal. However, some challenges should be overcome, especially the grayscale photolithography mask design. Today, only two main methods exist to generate and optimize these masks: physics-based optimization (PBO) and deep-learning optimization (DLO). These two have different advantages: accuracy and physics consistency for PBO and low computation costs and innovative solutions for DLO. To combine the benefits of each method, hybrid optimization (HO) is introduced in this paper. This method uses different parts of the PBO and the DLO to create a new optimization method that strikes a balance between PBO and DLO. This paper shows that HO provides a model 5 times faster than PBO and up to 18% more accurate than DLO. These results demonstrate the potential of the HO method, which offers a good balance between speed and accuracy. It represents a new approach to generating grayscale photolithography masks.