Parallel Pattern Matcher CS267 Final Project

As the critical dimension in optical lithography continues to shrink and integrated circuit masks become more complex, it is becoming more important to determine where the layout is affected the most by non-ideal process conditions. If the most problematic shapes can be identified and represented in a simple way, then these test patterns can be used to locate areas in any layout that are the most sensitive to these process effects. After the ‘hot spots’ have been found, the designer can use these results to go back and alter the geometry to reduce the sensitivity to these effects. This procedure will reduce the effects of imperfect optics on the printing of the image from the mask to the wafer, improving the yield of the design. Locations of interest can be filtered to include only points along edges, line ends, inside corners, and/or outside corners. Additional information on the pattern matching theory and validation, applications of pattern matching, and pattern matching algorithms is available from SPIE and ISQED. Some of the pattern matching concepts and algorithms are similar to those used in Optical Proximity Correction (OPC). The general pattern matching method presented here can be used to search for matches between any user-defined bitmap image and any set of input polygons on multiple layers involving complex floating-point number weights. A typical 128 by 128 pixel pattern yields reasonable speed vs. accuracy and is scaled to represent a radius of several wavelengths. Examples of pattern matching runs involving lithographic lens aberration patterns matched to phase-shifting masks are shown in Figure 1 and Figure 2. The pattern matching algorithms have been optimized for efficient processing of full chip layouts, such as the one shown in Figure 3. Frank Gennari has been working on the pattern matching system as Ph.D. research for the past three years. The goal of this paper is not to describe the pattern matcher work in great detail but to explain how we converted a large sequential program into a scalable parallel program.