Skip to Main Content
Spatial defect patterns generated during integrated circuit (IC) manufacturing contain valuable information on the fabrication process and can help engineers identify the root causes of any defect. Classification of these defect patterns is crucial to improving reliability and yield during IC manufacturing. Accurate classification requires good feature selection in order to assist in identifying the defect cluster types. In this paper, we demonstrate that the linear Hough transformation, the circular Hough transformation incorporating the cover ratio approach, and the zone ratio approach, when used as feature-extraction techniques, are able to distinguish lines, various solid circle-like cluster patterns such as blobs and bull's-eyes, and various hollow circle-like cluster patterns such as rings and edges. On the basis of these features, in this paper we provide a comprehensive evaluation of several data-mining classification approaches in terms of performance and accuracy. The results obtained using both artificial and real manufacturing data demonstrate the potential of this approach for analyzing general defect patterns that are generated during the IC fabrication process.