Iterative Defect Classifier Training for Semiconductor Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for setting up defect classifiers in semiconductor inspection are labor-intensive, require extensive expertise, and produce inconsistent results due to the manual construction of decision trees, making them costly and prone to errors.
Innovation Solution
A system and method for optimizing the training set of inspection-related algorithms through iterative learning, where the algorithm learns to improve its performance by selecting a minimal set of instructive defects and refining its classification boundaries, integrating training set acquisition with algorithm tuning to ensure consistent behavior.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual construction of decision trees is used for defect classification, then domain expertise can be incorporated into the classification process, but the process becomes labor-intensive and prone to errors
Solution Approach 1:
The system performs preliminary actions by automatically generating decision trees and classification rules before manual review, preparing multiple candidate classifications in advance that can be evaluated and refined by domain experts, thus reducing the time required for manual construction while maintaining accuracy
Solution Approach 2:
The system implements feedback mechanisms where classification results are evaluated against ground truth data, and the decision tree construction process is iteratively refined based on performance metrics, allowing continuous improvement of classification accuracy while automating the tuning process
2Adaptability or versatility
If manual construction of decision trees is used for defect classification, then customization to specific defect types is possible, but the process becomes costly and inconsistent
Solution Approach 1:
The system provides a universal automated decision tree construction framework that can handle multiple defect types and classification scenarios through parameter configuration rather than manual reconstruction, enabling customization across different applications while maintaining consistent processes and reducing costs
Solution Approach 2:
The system enables customization by allowing users to modify parameters such as defect attributes, classification thresholds, and data weighting factors within the automated framework, achieving adaptability to specific defect types without requiring complete manual reconstruction of the decision tree
3Measurement precision
If extensive training sets are used to improve algorithm performance, then classification accuracy increases, but the complexity of the training process increases
Solution Approach 1:
The system extracts and focuses on the most informative features and defects for training by automatically identifying key discriminative attributes and selecting representative defect samples, reducing the amount of training data needed while maintaining or improving classification accuracy
Solution Approach 2:
The system employs techniques such as ensemble methods and iterative refinement where multiple simpler models are combined or where training is performed in stages with increasing complexity, achieving high accuracy without requiring excessively large or complex training sets from the beginning
Data Source
AI summary
Methods and systems for training an inspection-related algorithm are provided. One system includes one or more computer subsystems configured for performing an initial training of an inspection-related algorithm with a labeled set of defects thereby generating an initial version of the inspection-related algorithm and applying the initial version of the inspection-related algorithm to an unlabeled set of defects. The computer subsystem(s) are also configured for altering the labeled set of defects based on results of the applying. The computer subsystem(s) may then iteratively re-train the inspection-related algorithm and alter the labeled set of defects until one or more differences between results produced by a most recent version and a previous version of the algorithm meet one or more criteria. When the one or more differences meet the one or more criteria, the most recent version of the inspection-related algorithm is outputted as the trained algorithm.