Defect clustering and classification for semiconductor devices

Abstract

In semiconductor manufacturing, cracks, scratches, contaminants, process variations, and errors by operators or equipment are all production problems that cause defects on wafers. It is highly desirable to find and classify defect signatures recorded in the data sets for final electrical tests and to correlate these signatures with causal data residing in the corresponding process and engineering data sets. In the first part of this paper, we explore the first of these issues-automating the identification of defect clusters on individual wafers within and across lots. Selection of the right number of clusters is achieved by incorporating the Calinsky and Harabasz (CH) index. We observe that the single-link clustering algorithm works the best in detecting the right number and shape of defect clusters. In the second part of this paper, we attempt to classify clustered defect patterns using an analytical tool, called the Hough Transformation (HT), proposed by Cunningham and MacKinnon, to identify scratch defects on semiconductor wafers. We observe that HT works well in detecting a variety of scratches, including diagonal scratches and line scratches.