Iterative Defect Classification for Semiconductor Fabrication
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current defect classification methods in semiconductor fabrication are slow and prone to human error due to manual review of SEM images, which limits the classification rate and accuracy, especially when dealing with large numbers of defects.
Innovation Solution
An iterative defect classification method and system that uses image acquisition, automated grouping, initial manual classification, automated classification with a generated classifier, identification of low-confidence defects for additional manual review, and repeated classification to increase confidence levels, ultimately achieving high-confidence determination of defect types.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual classification of defects is performed by users reviewing SEM images, then classification accuracy can be maintained through human judgment, but the classification speed is slow (less than 1000 defects per hour) and the process is highly time-consuming
Solution Approach 1:
The patent replaces the manual mechanical review process with an automated machine learning classification system. The system uses trained classifiers to automatically categorize defects based on image analysis, eliminating the need for human reviewers to manually examine each SEM image. This substitution dramatically increases classification speed while maintaining accuracy through iterative training with manually classified samples.
Solution Approach 2:
The system enables self-service classification where the automated classifier independently processes and categorizes defects without requiring continuous human intervention. Once trained, the system can autonomously classify new defects by comparing them against the trained model, freeing users from the time-consuming manual review process while preserving classification quality.
2Quantity of substance
If a large number of defects (greater than 4000) are classified manually, then comprehensive defect coverage is achieved, but the time required for complete classification becomes excessively long and the process becomes impractical
Solution Approach 1:
The system performs preliminary manual classification of a representative subset of defects to create training data before automated classification begins. This preliminary action establishes the classification framework and criteria that the automated system will subsequently apply to the entire defect population, enabling rapid processing of large volumes while maintaining consistent classification standards.
Solution Approach 2:
The classification process is segmented into distinct phases: initial manual classification of training samples, automated classifier training, and bulk automated classification. This segmentation allows the system to handle large defect volumes efficiently by separating the time-intensive manual work from the rapid automated processing, achieving both comprehensive coverage and reasonable time investment.
3Reliability
If manual defect classification is performed to ensure accurate defect type identification, then classification reliability can be maintained, but the process becomes susceptible to human error and is highly operator-intensive
Solution Approach 1:
The system incorporates feedback mechanisms where manually classified defects are used to train and refine the automated classifier. The classifier learns from correct manual classifications and continuously improves its accuracy. This feedback loop maintains high reliability in defect type identification while progressively reducing operator intensity as the system becomes more proficient.
Solution Approach 2:
The system creates a digital model (classifier) that copies and encodes the classification expertise from manual reviewers. Instead of relying on human operators to consistently apply classification criteria, the system replicates this expertise in software form, eliminating human error and reducing operator intensity while maintaining the reliability of accurate defect type identification.
4Productivity
If automated classification systems are implemented to increase classification speed, then productivity improves, but the system may lack the judgment and adaptability of human reviewers in complex classification scenarios
Solution Approach 1:
The classification system is designed to be dynamic and adaptive rather than static. The classifier can be retrained with new manually classified samples to adapt to emerging defect types or changing classification criteria. This dynamic capability allows the automated system to maintain high productivity while developing the judgment and versatility needed for complex classification scenarios through continuous learning.
Data Source
AI summary
Defect classification includes acquiring one or more images of a specimen including multiple defects, grouping the defects into groups of defect types based on the attributes of the defects, receiving a signal from a user interface device indicative of a first manual classification of a selected number of defects from the groups, generating a classifier based on the first manual classification and the attributes of the defects, classifying, with the classifier, one or more defects not manually classified by the manual classification, identifying the defects classified by the classifier having the lowest confidence level, receiving a signal from the user interface device indicative of an additional manual classification of the defects having the lowest confidence level, determining whether the additional manual classification identifies one or more additional defect types not identified in the first manual classification, and iterating the procedure until no new defect types are found.