Wafer Image Segmentation Using Geometry-Based Edge Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current wafer inspection methods, such as intensity-based and projection-based segmentation, are prone to instability and misclassification of segments, leading to inadequate defect detection due to intensity variations and irregular segment boundaries, which affects the adaptability to local noise and accuracy of defect inspection.
Innovation Solution
A computer-implemented method that determines pixel statistics, assigns pixels to segments based on these statistics, detects edges, generates an edge map, and applies these segments and edge maps to accurately segment the image, enhancing defect detection by stabilizing the segmentation process and improving correlation with the underlying physical structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If intensity-based segmentation is used to group similar intensity pixels together, then the segmentation process is simple to implement, but the segmentation becomes unstable and prone to misclassification due to intensity variation from job-to-job during runtime
Solution Approach 1:
The patent transitions from intensity-based parameters to geometry-based parameters for segmentation. Specifically, it uses the sum of absolute differences (SAD) metric and correlation coefficient between projection patterns to segment pixels, replacing the unstable intensity-based approach. This parameter change makes segmentation more reliable by focusing on geometric properties that remain consistent across different jobs and intensity variations.
2Ease of manufacture
If projection-based segmentation is used to separate segments based on relative projected intensity, then the segmentation process is simple, but the segmentation result fluctuates along the side of the underlying physical structure pattern causing mis-segmentation
Solution Approach 1:
The patent applies local quality by using geometry-based segmentation that adapts to local patterns in the wafer image. Instead of applying a global projection-based segmentation that fluctuates along physical structure boundaries, the method calculates geometry metrics (SAD and correlation) locally for each pixel based on its neighborhood, allowing accurate segmentation that respects local pattern variations while maintaining overall simplicity.
3Reliability
If median intensity-based segmentation is used to separate array region from page break region, then the segmentation is more stable, but the segment boundaries become irregular and do not correlate well with the underlying physical structure pattern
Solution Approach 1:
The patent moves from one-dimensional intensity-based segmentation to two-dimensional geometry-based segmentation. By calculating SAD and correlation coefficients across both x and y directions and combining these geometric dimensions, the method produces regular, grid-like segment boundaries that align with the underlying physical structure, while maintaining the stability benefits of median-based approaches.
Data Source
AI summary
Methods and systems for segmenting pixels for wafer inspection are provided. One method includes determining a statistic for individual pixels based on a characteristic of the individual pixels in an image acquired for a wafer by an inspection system. The method also includes assigning the individual pixels to first segments based on the statistic. In addition, the method includes detecting one or more edges between the first segments in an image of the first segments and generating an edge map by projecting the one or more edges across an area corresponding to the image for the wafer. The method further includes assigning the individual pixels to second segments by applying the first segments and the edge map to the image for the wafer thereby segmenting the image. Defect detection is performed based on the second segments to which the individual pixels are assigned.