Artificial Intelligent Matching For Scratches Of Semiconductor Wafers Based On A K-NN Algorithm

Wafers are backplanes for integrated circuits(IC).The multi-layered and high-densitycomponents in the IC are based on wafers.The production of IC is critical for wafers flatness.Chemical mechanical polishing(CMP)is a key technology for wafer planarization in the semiconductor industry.However,there are many factors affect wafers flatness and cause scratches on wafers.Scratches are one of the major factors in the loss of chips output and quality.In this paper,we use skeletonization method and K Nearest Neighbor(K-NN)algorithm to intelligently identify and match the scratches,and apply coordinate transformation as a mathematical calculation method to improve based on the mathematical characteristics of scratches.In general,matching scratches is performed manually with simple equipment,which is poorly consistent and time-consuming.Combined with existing work experience,this paper introduces a new pattern recognition matching method.First,binarization and skeletonization methods are used to obtain the single-pixel binary curve segments(main lines);then these main lines are initially grouped using a clustering method to extract some basic geometric features of the mainline endpoints,such as distance,slope,curvature,etc.Finally,the K-NN algorithm is used to intelligently match the same type of mainline.Specifically,it is notable to mention that the application of binarized scratches images makes the data simple.On the basis of binarazation,skeletonization makes it easier to extract the geometric features of scratches.Besides,a dynamic coordinate system is introduced before extracting the features of the main lines.There is an error matching caused by the small difference in slope when the tangent slope is large.The dynamic coordinate system introduced in the feature extraction greatly simplifies the setting of the matching threshold.In summary,we propose a new pattern recognition matching method,which combines the skeletonization and K-NN algorithm to make the data more operational,features more easily extracted,and the geometric meaning is more obvious.Besides,this matching method is applied to the wafer scratch matching problem in semiconductor manufacturing.Experimental results show that the wafer scratch matching method proposed in this paper meets the manual matching requirements and the matching consistency is good.