Research on accurate recognition of arbitrarily selected mark patterns in alignment of lithography

The alignment of mask and wafer is a very important step in the process of lithography. When a specific pattern in the exposure image is selected as the alignment mark, the traditional automatic alignment methods which are based pre-set markers and image processing are not suitable. To address this issue, we propose a novel accurate image recognition method for arbitrarily selected mark patterns, which combines Support Vector Machine (SVM) with feature extraction to achieve adaptive switching of alignment templates. Firstly, based on the distinct linear contour features of silicon wafer exposure patterns, which lack color and texture characteristics, we extract Histogram of Oriented Gradients (HOG) features from the images to construct feature vectors ; then, we select the optimal SVM kernel function through experimental comparisons, and select regions of interest on silicon wafer exposure images for testing ; finally, we utilize HU-based shape features for secondary matching and recognition decisions. The experimental results demonstrate that the proposed method achieves a recognition accuracy of 100%, enabling the implementation of adaptive alignment template selection and switching.