Improved Non-parametric Subtraction for Detection of Wafer Defect

Automated defect inspection for wafer has been developed since 1990 ´s to replace defect detection by human eye for low-cost and high-quality. Defects are detected by comparing an inspected die with a reference die in application of wafer defect inspection. Referential methods compare with reference image by computing the intensity difference pixel by pixel between a reference image and an inspected image or measuring the similarity between two images using normalized cross correlation or eigen value. These methods are problematic for defect detection due to illumination change, noise and alignment error. To reduce the sensitivity of illumination change and noise, the new image subtraction called non-parametric subtraction was proposed. Non-parametric subtraction can solve problem about illumination change and noise, but sensitivity of alignment remains unsolved. This paper introduces new approach less sensitive to alignment using non-parametric subtraction for wafer defect inspection.