Methodology for Lithography Hotspot Detection using ResNet50V2 and Model soups.

In the semiconductor industry, a continuous trend of reducing chip size to maintain competitiveness has led to a significant increase in the complexity of the chip design. This paper presents an efficient approach, focusing specifically on the lithography process, known for its potential pattern distortions called lithography hotspots. The methodology proves to be highly effective in detecting lithography hotspots using ResNet50V2 and model soups. To fine-tune the model and address the challenge of class imbalance in the dataset, we introduce data augmentation techniques that rebalance the data while preserving geometric information. The experimental results obtained on the ICCAD-2012 benchmark dataset demonstrate the effectiveness of our proposed methodology. We achieved impressive results, with an average precision of 0.908 and f1 score of 0.926. These outcomes highlight the practical potential of our methodology for applications in the semiconductor industry.