Workpiece Inspection Using Chi-Square Probability for Noisy Edge Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current pattern inspection techniques face challenges in accurately detecting ultrafine defects in semiconductor and LCD photomasks due to difficulties in extracting contour lines from noisy optical images, particularly at gently sloped edges and complex profile shapes, and the limitations of using filters designed for specific shapes.
Innovation Solution
A workpiece inspection method and apparatus that compares optical images to reference images by searching for pixel groups with similar grayscale values, calculating probability values using a chi-square distribution, and determining defect presence based on these values, without relying on shape-matched filters, allowing for enhanced detection of defects in various angled and complex pattern edges.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If shape-matched filters are used for defect detection, then detection accuracy for specific pattern shapes is improved, but adaptability to various pattern shapes and edges deteriorates
Solution Approach 1:
The patent replaces shape-specific filters with a universal probability calculation mechanism that works for all pattern shapes. The chi-square distribution-based probability value calculation is applicable to any pixel regardless of the underlying pattern geometry, enabling the same defect detection algorithm to handle various pattern shapes without requiring shape-matched filters.
Solution Approach 2:
The patent changes the detection parameter from shape-based filter responses to grayscale value-based probability calculations. By using pixel grayscale values and their statistical distribution (chi-square distribution) instead of shape-matched filter outputs, the system achieves shape-independent defect detection while maintaining high accuracy.
2Measurement precision
If contour line extraction is performed on noisy optical images, then defect detection capability is improved, but reliability deteriorates due to extraction errors at gently sloped edges and complex profiles
Solution Approach 1:
The patent extracts only the essential information (pixel grayscale values) needed for defect detection while discarding the problematic intermediate step of contour line extraction. By directly comparing grayscale values between optical images and reference images, the system avoids extracting contour lines from noisy images, thereby eliminating extraction errors at gently sloped edges and complex profiles.
Solution Approach 2:
The patent introduces probability values based on chi-square distribution as an intermediary between raw pixel grayscale values and defect detection decisions. This statistical intermediary provides a reliable framework for determining whether grayscale differences represent actual defects or merely noise, significantly improving detection reliability.
3Measurement precision
If noise models and detection thresholds are optimized, then detection accuracy is improved, but ease of operation deteriorates due to complex optimization requirements
Solution Approach 1:
The patent enables the system to automatically determine detection thresholds through statistical analysis of the reference image data itself. By calculating the chi-square distribution from the reference image pixel grayscale values, the system self-determines appropriate threshold values without requiring manual optimization or external calibration, significantly improving ease of operation.
Data Source
AI summary
A workpiece inspection apparatus includes a search unit for finding from an input reference image a first pixel group which contains a certain pixel and a second pixel group having grayscale values within a threshold, and a probability acquisition unit responsive to receipt of an optical image of a workpiece being tested and the reference image, for obtaining a probability value based on a grayscale value of the certain pixel, a grayscale value of its corresponding pixel in the second pixel group, a grayscale value of an optical image pixel corresponding to the certain pixel, and a grayscale value of an optical image pixel corresponding to a pixel in the second pixel group corresponding to the certain pixel. The probability value is used to determine whether a defect exists at the optical image pixel corresponding to the certain pixel.