Lithographic Hotspot Detection Using Adaptive Squish Pattern Sampling Combined with Faster R-CNN

In the early stages of integrated circuit layout design, lithographic hotspot detection is employed to identify layout areas that may generate manufacturing defects, thereby improving chip yield. Among various hotspot detection methods, those based on machine learning have received widespread research and application due to their outstanding performance compared to other techniques. The core challenges of lithographic hotspot detection lie in the speed of hotspot data acquisition, the effectiveness of hotspot feature extraction, and the accuracy of hotspot detection. To accelerate hotspot data acquisition, this paper proposes a layout indexing method for hotspot data extraction based on grid indexing, which allows us to quickly obtain layout data. To obtain effective and streamlined hotspot features, we employ the concept of adaptive squish pattern sampling and expand its application scope to achieve a larger lithographic hotspot detection range. To construct a more precise detection model, we use the Faster R-CNN model and input the extracted hotspot features for hotspot detection. We use the benchmark data from the ICCAD 2012 competition for testing and achieve satisfactory results.