Research On-line Defect Detection Method For Semiconductor Wafer Spin Coating System

In recent years, the lithography process complexity continues to improve with the rapid development of China's semiconductor industry, the sources of the wafer's defect become more and more diversified, artificial visual inspection is the mainly traditional way of wafer inspection, it has been unable to meet industry needs because the low efficiency of the detection method and susceptible to subjective factors. In order to improve the speed and reliability of wafer defect detection and achieve online visual wafer defect detection, reducing the impact of defects in the semiconductor manufacturing industry, improve product yield, analyzing existing defect detection method based on the combination of computer technology and image processing technology, the research started with the background above.First, in order to obtain high-resolution images of the wafer's surface, choose the appropriate image capture device to design an image acquisition system. In the light source selection, the system use red coaxial light source, it not only overcome the shortcomings of high reflectivity of the wafer surface, but also the red light's longer wavelength supports protection of the photoresist on the wafer surface with maximum. The micron precision limits field of view of the camera, in order to get the full wafer image, and to avoid interference caused by camera shake, to build a three-axis motion platform which responsible for adsorption wafer, moving in accordance with the provisions of the path, to capture the entire complete image of the wafer.Second, the image maybe be impacted by channel bandwidth, noise, etc. when the capture and transmission, according to the characteristics of image noise and experimental comparison algorithm to determine use the improved median filtering method-Multiple Value filter to filter the image noise; to improve the contrast of the image, highlighting the area of interest, after filtering is complete, using histogram specification method for image enhancement, enhance image interpretation and recognition effect.Finally, the template matching algorithm is study in-depth in order to match the all Die unit in the wafer image, consider the characteristics of Die unit distribution and the experimental results comparison algorithm to determine the use of normalized cross-correlation template matching algorithm, the method have the "immunity" to the change of image gray value changes linearly, and has the better robustness to light change. To obtain the differential image by the image subtraction method, and then to compare with a present threshold value, the point is marked as defective if it is greater than the threshold.Verify the proposed methods could detect wafer surface defect effective, precision reaches fifteen micron approximately by a large number of experience about wafer die. The defect detection algorithm can replace human work to quickly and accurately wafer defect detection can be realized in the actual application.