Clip Clustering for Early Lithographic Hotspot Classification

Early lithography hotspot detection is critical to improving manufacturing yield. EDA tools have been proposed to detect potentially problematic patterns during physical design and physical verification stages. Considering that detailed lithography simulation has a high computational cost for full-chip scale, the pattern matching method proved to be a fast solution with good accuracy due to a set of pre-characterized patterns as input. It is proposed a clip clustering method for pattern classification in early detection of lithography hotspots. We focus at both clip representation and clip clustering stage that is the major challenge of this method. It was performed experiments on 2016 ICCAD contest benchmark suite, and results show the efficiency of our clustering approach. The algorithm supports both area and edge constrained clustering. Our solution generates on average 9.4% fewer clusters than the contest winner while staying within 16% range on average from the state of the art algorithms. Moreover, Our clustering pattern-driven layout strategy outperforms the 2016 ICCAD winner on runtime by up to 60%.