ML Pattern Grouping for Semiconductor Defect Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In semiconductor manufacturing, the shrinking size of integrated circuit components leads to challenges in defect detection accuracy and throughput due to high rates of nuisance defects identified by inspection tools, requiring efficient methods for defect review and classification.
Innovation Solution
A pattern grouping method using machine learning that generates fixed-dimensional feature vectors from defect images, allowing for improved defect identification and classification by reducing operator intervention and enhancing processing efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If inspection tools are used to detect defects in shrinking IC components, then defect detection capability is maintained, but nuisance defect rates increase and throughput decreases
Solution Approach 1:
The patent segments the defect review process by automatically grouping similar defect patterns into clusters, allowing operators to review representative patterns rather than every individual defect. This segmentation reduces the total number of defects requiring manual review while maintaining detection capability.
Solution Approach 2:
The patent introduces an intermediary automated pattern recognition system that processes defect images and groups them by similarity. This intermediary system acts as a filter between the inspection tool and the operator, preprocessing defects to identify and group nuisance patterns before human review.
2Measurement precision
If operators manually review all identified defects, then defect identification accuracy is improved, but review time increases substantially
Solution Approach 1:
The patent performs preliminary actions by automatically grouping defect patterns before operator review. Defects are pre-processed, pre-analyzed, and pre-grouped by similarity metrics, so that operators receive organized clusters rather than raw defect lists, reducing their review time while maintaining accuracy.
Solution Approach 2:
The patent creates representative copies or prototypes of defect patterns within each cluster. Operators review these representative patterns to understand the characteristics of entire groups, rather than examining every individual defect instance, significantly reducing review time while preserving identification accuracy.
3Adaptability or versatility
If feature extraction methods are used to process defect patterns, then pattern recognition capability is improved, but processing complexity increases
Solution Approach 1:
The patent transforms defect patterns into standardized parameter representations (feature vectors) that capture essential characteristics. By changing the representation parameters from raw images to structured feature vectors, the system improves pattern recognition capability while managing processing complexity through dimensionality reduction and standardization.
Data Source
AI summary
A pattern grouping method may include receiving an image of a first pattern, generating a first fixed-dimensional feature vector using trained model parameters applying to the received image, and assigning the first fixed-dimensional feature vector a first bucket ID. The method may further include creating a new bucket ID for the first fixed-dimensional feature vector in response to determining that the first pattern does not belong to one of a plurality of buckets corresponding to defect patterns, or mapping the first fixed-dimensional feature vector to the first bucket ID in response to determining that the first pattern belongs to one of a plurality of buckets corresponding to defect patterns.