A Defect Detection Method of Mixed Wafer Map Using Neighborhood Path Filtering Clustering Algorithm

As the wafer process becomes more complex, the probability of mixed-type defective wafer maps is constantly increasing. Therefore, it is necessary to perform effective filtering and denoising processing on the mixed-type defective wafer map to facilitate the identification. We propose a Neighborhood Path Filtering Clustering (NPFC) algorithm in this paper. For determining the number of clusters in the wafer map, we combine the clustering index of the Silhouette coefficient and the similarity index of Mahalanobis distance to propose a clustering similarity index SD. Then, calculating the compactness index between the clusters is judged whether the clusters are merged or not, to obtain the final clustering effect. The greatest advantage of this method is that it avoids the influence of algorithm parameter settings on the filtering effectiveness. Meanwhile, it can still effectively identify clusters of arbitrary shapes. The experimental results show that compared with median filtering, K-mean clustering and adjacency clustering, the NPFC algorithm achieves good results in the filtering and clustering effect.