The downscaling of device dimensions in semiconductor manufacturing has meant that critical defect sizes have become smaller and smaller. This makes it more likely that the highly sensitive optical wafer inspection tool used for detecting small defects will erroneously detect process variations as defects, and generate a large amount of ‘nuisance’ information. Therefore, the scanning electron microscope (SEM)-based review tool used needs to automatically discriminate between defects and nuisance information. To identify nuisance information, the absence of defects in the SEM image needs to be accurately detected through an inspection process using the review tool. We propose a defect detection method using (a) the integration of multiple comparison detection (IMCD) results to minimize the number of defect candidates and (b) discrimination based on a normal patch image model (DNPM) to judge whether a candidate is a defect or normal. An evaluation using SEM images of a processed wafer revealed that combining the IMCD and DNPM methods achieves a nuisance information discrimination rate of 84.4% and a defect detection rate of 93.3%, which are higher rates than those achieved by the one-class support vector machine. The proposed methods automatically collect defect images efficiently even when a great deal of nuisance information is produced by the optical wafer inspection tool, and enable manual visual checks to be reduced.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.