Defect Classification Using Single-Class and Multi-Class Classifier Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated defect classification systems in the semiconductor industry face challenges in achieving high purity of classification due to classification uncertainty, particularly in regions of overlap between defect classes and at the outer borders of class ranges, leading to rejected defects that require human intervention.
Innovation Solution
The implementation of a method that uses both single-class and multi-class classifiers, along with confidence threshold adjustments, to identify defects and assign them to classes, with the multi-class classifier handling overlap regions as non-decidable defects and the single-class classifiers verifying known defects, while optimizing purity and rejection rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a multi-class classifier is used to classify defects into different types, then the classification coverage is improved, but the classification purity deteriorates due to overlap regions between defect classes
Solution Approach 1:
The patent segments the classification task into two distinct parts: a multi-class classifier that handles the broad classification coverage, and a separate single-class classifier that specifically targets the overlap regions to ensure high purity. This segmentation allows each classifier to specialize in its strength, resolving the contradiction between coverage and purity.
Solution Approach 2:
The single-class classifier acts as an intermediary between the multi-class classifier and the final classification decision. It mediates the classification process by verifying predictions in uncertain overlap regions, thereby improving purity without compromising the overall coverage provided by the multi-class classifier.
2Measurement precision
If the confidence threshold is increased to improve classification purity, then the rejection rate increases, requiring more human intervention
Solution Approach 1:
The patent applies different confidence threshold strategies to different defect regions. In overlap regions, the single-class classifier uses locally optimized thresholds that account for the specific characteristics of each defect type, rather than applying a global threshold. This local quality approach maintains high purity while minimizing unnecessary rejections.
Solution Approach 2:
The system dynamically adjusts the confidence threshold parameter based on the defect characteristics and the classifier's confidence score. By changing this parameter adaptively rather than using a fixed threshold, the system optimizes the balance between purity and rejection rate for different defect scenarios.
3Measurement precision
If multiple classifiers are used to verify defect classifications, then the classification purity is improved, but the system complexity increases
Solution Approach 1:
The patent implements a dynamic classification system where the single-class classifier is selectively applied based on the multi-class classifier's confidence score. Rather than always using both classifiers, the system dynamically determines when verification is needed, reducing unnecessary computational complexity while maintaining high purity where it matters most.
Solution Approach 2:
The multi-class classifier performs a preliminary classification that handles the majority of clear-cut cases, preparing the data for the single-class classifier. This preliminary action allows the more complex single-class classifier to focus only on the uncertain cases, effectively reducing the overall system complexity while maintaining high purity.
Data Source
AI summary
A method for defect analysis includes identifying single-class classifiers for a plurality of defect classes, the plurality of defect classes characterized by respective ranges of inspection parameter values. Each single-class classifier is configured for a respective class to identify defects belonging to the respective class based on the inspection parameter values, while identifying the defects not in the respective class as unknown defects. A multi-class classifier is identified that is configured to assign each defect to one of the plurality of the defect classes based on the inspection parameter values. Inspection data is received, and both the single-class and multi-class classifiers are applied to the inspection data to assign the defect to one of the defect classes.