부산대학교 도서관

Lithography hotspot detection plays an important role in the manufacturability design of integrated circuits, which affects the yield of the final product. In this paper, based on deep learning technology named Yolov5, a lithography hotspot detection method is proposed, which can localize the lithography hotspots rapidly. Using ICCAD 2012 contest benchmark 1 and the OPC post-processed simulation graphics as the datasets to conduct the experiments, the results show that the proposed algorithm can effectively improve the detection efficiency, with the recall of 96.3%, the precision of 95.3%, the average detection time of 0.7 h/mm 2 .