PSCaps: High-Performance Pose-Sensitive Layout Hotspot Detector based on CapsNet

Advanced technology nodes face challenges with Design Rule Violations (DRVs), primarily due to the possibility of nm-level small variations that can lead to the occurrence of DRVs. Various Machine Learning (ML) techniques have been introduced to detect whether the layout design conforms to manufacturing rules, thus alleviating the time-consuming challenge associated with traditional lithography simulations. However, existing ML models still face challenges in detecting layout violations where there are pose variations among geometric shapes in the layout. In this study, we propose a hotspot detector called PSCaps based on the CapsNet, which considers the pose information of geometric shapes in the layout. The method effectively captures spatial information and hierarchical structures between geometric shapes in the layout. Through the dynamic routing mechanism, the model adaptively learns the relationships and weight allocations between different capsules. Additionally, we employ multiple data augmentation methods to alleviate the problem of imbalanced hotspot and non-hotspot data in the open-source dataset. The benchmarks of ICCAD-2012 and ICCAD-2019 are used to validate our method. The experimental results demonstrate that our proposed hotspot detector outperforms other state-of-the-art works.