LAUSR

Scratches, those usually generated during polishing the silicon wafer surface, are one of the major yield loss factors in semiconductor manufacturing industry. In order to determine the source of the scratches in real time and reduce the yield loss, it is critical for manufacturers to match and identify the same type of scratches automatically. In this paper, an improved K nearest neighbors (KNN) algorithm to address this issue is presented. Firstly, a skeleton extraction method is used to depict the main lines of scratches. Then the clustering protocol is applied as a preliminary step to group these main lines so that some essential endpoints features of main lines, such as distance, slope and curvature, can be extracted. During feature extraction, a dynamic coordinate system is introduced and this greatly reduces the distortions arise due to the magnitude of tangent difference. An intelligent matching of similar scratches MSML-KNN algorithm is formulated. The experimental results show that the proposed matching method for wafer scratches has a good adaptability and robustness.