Classification and Recognition Methods for Wafer Defect Detection

Based on the defect image information obtained from wafer scanning, this method compares defect features by combining the characteristics of defect images within and between different layers. This method mainly uses machine learning algorithms to extract and classify the characteristic information of defect images. Through learning and training on the features of scanned defects, and in combination with the features of training defect images, it conducts classification training and learning for defects. It compares and identifies newly added types of defect images. For small defect type images, it compares and identifies the differences between the defect images and the reference images. By applying the nine - grid defect division method, it detects and classifies tiny and subtle defects, and detects the defect difference information in the defect images. By extracting the SAM features of the online scanned defect images, it obtains the defect features of the wafer images. Based on the defect scanning classification results, it conducts a correlation analysis between the scanned defects and the product yield to locate the problematic process. This method can automatically locate the scanned defects according to the characteristic information of the scanned defects and the product yield, thereby improving the product yield.