Defect Classification System with Human-in-the-Loop Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor manufacturing processes face challenges in accurately classifying defects in semiconductor wafers due to false positives and the complexity of defect types, which can lead to operational faults and inefficiencies in high-density integration.
Innovation Solution
A system utilizing hardware-based GUI components and processing and memory circuitry to classify defects by obtaining attribute values, clustering them, and allowing user confirmation for accurate classification, enabling the creation of classes and training of classifiers for improved defect identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated defect classification is used, then productivity is improved, but measurement precision deteriorates due to false positives
Solution Approach 1:
The patent introduces an intermediary review process where defects classified by the automated system are presented to a user for confirmation. The system displays defect images with predicted class labels and confidence scores, allowing human operators to verify and correct classifications. This intermediary step resolves the contradiction by maintaining high automated processing speed while improving accuracy through human oversight for uncertain cases.
Solution Approach 2:
The system dynamically adjusts the confidence threshold parameter to control the balance between automated classification and user review. By changing this parameter, the system can optimize the division of labor between automated processing and human verification, thereby managing both productivity and measurement precision according to specific operational requirements.
2Measurement precision
If comprehensive defect examination is performed, then measurement precision is improved, but loss of time increases due to multiple inspection steps
Solution Approach 1:
The system performs preliminary automated classification of all defects before user review, pre-sorting them by predicted class and confidence level. This preliminary action enables the subsequent user review step to focus only on uncertain or critical cases, thereby maintaining high measurement precision while significantly reducing the total inspection time compared to comprehensive manual examination of all defects.
Solution Approach 2:
The inspection process is segmented into two distinct phases: automated classification for initial defect categorization and user review for verification and correction. This segmentation allows the system to leverage the speed of automated processing for routine tasks while reserving human expertise for complex or uncertain cases, thus optimizing both time and accuracy.
3Measurement precision
If manual defect review is implemented, then measurement precision is improved, but device complexity increases due to additional system components
Solution Approach 1:
The system integrates multiple functions into a unified platform that combines automated defect detection, classification, and user review capabilities. The examination tool serves both as an automated inspection device and as a presentation interface for user verification, reducing the need for separate dedicated systems and thereby limiting the increase in device complexity while maintaining improved measurement precision.
Data Source
AI summary
A system, method and computer software product, the system capable of classifying defects and comprising: an hardware-based GUI component; and a processing and memory circuitry configured to: a. upon obtaining data informative of a plurality of defects and attribute values thereof, using the attribute values to create initial classification of the plurality of defects into a plurality of classes; b. for a given class, presenting to a user, by the hardware-based GUI component, an image of a defect initially classified to the given class with a low likelihood, wherein the image is presented along with images of one or more defects initially classified to the given class with the highest likelihood; and c. subject to confirming by the user, using the hardware-based GUI component, that the at least one defect is to be classified to the given class, indicating the at least one defect as belonging to the given class.