In this paper, we propose a novel methodology for machine learning-based hotspot detection that uses lithography information to build support vector machine (SVM) during its learning process. Unlike previous studies that use only geometric information or require a post-optical proximity correction (OPC) mask, this proposed method utilizes detailed optical information but bypasses post-OPC mask by sampling latent image intensity and uses those points to train an SVM model. The results suggest high accuracy and low false alarm, and faster runtime compared with methods that require a post-OPC mask.

In this paper, we propose novel algorithms for pattern matching which dissects patterns into rectangles based on polygon edges. Unlike other design rule check (DRC)-based pattern matching algorithms, our solution utilizes simple DRC edge length rules to create rectangles for hotspot pattern descriptions. This approach has at least three advantages over other solutions. First, it is faster than other state-of-the-art pattern matching tools. Second, it is intuitive and simple for pattern matching engineers to understand and describe patterns. Third, it scales well for parallel computation. We also show how to improve pattern matching run time using vector space created by an origin rectangle and other reference rectangles inside a pattern bounding box. By adopting the vector concept, we iterate only once or twice when detecting different pattern orientations. Other pattern matching techniques usually iterate eight times (4 rotations × 2 mirrored images) to detect all of the eight different orientations. Our method eliminates these unnecessary iterations.