Abstract
In recent years, the semiconductor industry has been a high-growth business. Yield improvement is more and more important for the process of semiconductor manufacturing since it is an important element to ensure the profitability. Defect spatial pattern impedes yield and it can reveal the root cause of the occurred defects. For this consideration, this research intends to develop a defect spatial pattern methodology for yield enhancement. Defect spatial patterns are classified into seven categories including Curve Type, Line Type, Local Type, Ring Type, Radial Type, Repeat Type and Die Edge Type. To achieve the purpose of recognition, the coordinate data are transformed into image format and image processing techniques are applied to eliminate the random defects. Clustering methods are then applied to locate systematic defects. Minimum rectangle area method is collected cluster defects for further analysis. The feature extraction procedure in this research is adopted by wavelet transform algorithm. By setting feature vectors into feature space and applying hierarchical agglomerative algorithm to clustering, the result is expected to reach the goal of description and classification. The proposed methodology is verified with industrial data from a famous semiconductor company in Taiwan. The average accuracy of recognition is 95%.The experimental results show that this method can successfully recognize the defect spatial patterns and enhance the yield in semiconductor manufacturing.