A novel clustering and declustering algorithm for fuzzy classification of wafer defects

A method has been developed for enhancing the efficiency and accuracy of wafer defect analysis for yield improvement. This multi-step fuzzy algorithm has been developed for automatic clustering and classification of wafer defects. The algorithm utilizes a combination of new and existing feature measurements to identify and match defects with those referenced in a defect classes library. The process is more efficient than other approaches like pair-wise K-Nearest Neighbor (K-NN) classifiers and other fuzzy methods, which can be computationally very expensive. The algorithm also offers improved accuracy and the ability to decluster defects in cases where more than one overlap.