Hotspot Detection with Machine Learning Based on Pixel-Based Feature Extraction

The complexity of physical verification increases rapidly with fast shrinking technology nodes. Considering only design rule checking (DRC) constraints or lithography models cannot capture the side effects in fabrication process well. Thus, it is desirable to consider a more general problem various types hotspots. In this paper, we apply machine learning which based on pixel-based feature extraction deal generalized hotspot detection problem. First, two-dimensional discrete Fourier transformation-based pixel method proposed alleviate shifting effect and produce stable features. Then, pattern-based layout scanning approach developed enhance program efficiency while preserving good accuracy. Finally, two false alarm reduction strategies effectively reduce number detected nonhotspots further improve accuracy position. Experimental results industrial benchmarks show that our algorithm outperforms three competitive works terms accuracy, rate, efficiency, time.