Specimen Defect Detection Using Color-Grouped Reference Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current defect detection methods in semiconductor manufacturing face challenges due to large color variations among test frames, leading to low sensitivity and high nuisance rates, as well as artifacts from defects on base frames, which affect the accuracy of defect detection in logic areas.
Innovation Solution
Implement a system with a computer subsystem that groups images by color similarity within a die row, generates multiple reference images, and applies a defect detection method to each group separately, using a median-based computed reference and double detection to enhance sensitivity and reduce nuisances.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single reference image is used for defect detection in logic areas, then the inspection process is simple, but the sensitivity is low and nuisance rate is high due to color variations among test frames
Solution Approach 1:
The patent segments the test frames into multiple groups based on color similarity, with each group processed separately to generate its own reference image. This segmentation approach allows the system to adapt to color variations in different regions while maintaining manageable processing complexity for each group.
Solution Approach 2:
The patent changes the parameter of reference image generation by creating multiple reference images from different groups of test frames rather than using a single reference image. This parameter change enables the system to handle color variations effectively while improving defect detection sensitivity.
2Measurement precision
If multiple reference images are generated to handle color variations, then defect detection sensitivity improves, but the processing complexity and computation time increase
Solution Approach 1:
By segmenting test frames into color-similar groups, the patent reduces the computation time required for generating each reference image compared to processing all frames together. Each group can be processed independently and in parallel, improving overall processing efficiency while maintaining high sensitivity.
Solution Approach 2:
The patent performs preliminary clustering of test frames by color similarity before generating reference images. This preliminary action organizes the data in a way that accelerates subsequent processing steps and enables efficient parallel computation of multiple reference images.
3Reliability
If computed reference images are used to reduce nuisance rates, then artifact detection improves, but the system becomes more complex with multiple processing steps
Solution Approach 1:
The patent segments the computation of reference images into multiple independent groupings, where each group's reference image can be computed separately. This segmentation simplifies the overall system architecture by breaking down a complex single-reference computation into manageable independent tasks.
Solution Approach 2:
The patent changes the approach from generating a single reference image to generating multiple reference images with different characteristics. This parameter change improves the system's ability to distinguish real defects from artifacts while keeping each individual reference generation process relatively simple.
Data Source
AI summary
Methods and systems for detecting defects on a specimen are provided. One system performs double detection in which at least one of the reference images compared to a test image is a computed reference image generated from multiple images corresponding to the test image. The other reference image may or may not be computed from more than one of the multiple images. Such a computed reference image may also be a median-based computed reference generated from multiple-median images generated from different subsets of images in a job of images generated by an inspection subsystem for a specimen. Such a system may also group images for a die row on a specimen into different jobs based on color so that different jobs have different color value ranges. Such grouping may also be performed so that each of the jobs includes a number of images greater than a predetermined, minimum job size.