Semiconductor Defect Detection Using AI and Human Expert Review
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing semiconductor detection methods are inadequate in accurately and efficiently identifying defect points, often missing real defects or misclassifying them due to limitations in analyzing wafer images and layout designs.
Innovation Solution
A detection method utilizing a machine learning algorithm, specifically a convolutional neural network, that superimposes and compares layout and scan graphics to extract non-overlapping patterns, encodes these patterns into sample coded data, and uses this data to establish and train a detection model library for accurate defect point detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional automated defect detection methods are used to compare wafer images with layout designs, then the detection process is simple and fast, but the detection accuracy is low and many real defect points are omitted
Solution Approach 1:
The patent introduces an intermediary human expert in the loop to review and verify defect detection results. The system combines automated AI defect detection with human expert judgment, where the human expert acts as a mediator to confirm or reject detected defects, thereby improving detection accuracy while maintaining reasonable system complexity through hierarchical processing
Solution Approach 2:
The detection process is segmented into multiple stages: initial automated defect detection using AI algorithms, followed by selective human expert review of suspicious cases. This segmentation allows the system to use simple automated methods for most cases while applying more complex human judgment only where needed, improving overall accuracy without proportionally increasing system complexity
2Measurement precision
If more comprehensive analysis methods are used to fully analyze wafer images and layout designs, then defect detection accuracy improves, but the detection time and computational resources increase
Solution Approach 1:
The system performs partial comprehensive analysis by using AI algorithms to quickly identify suspicious defect regions, then applies more resource-intensive human expert review only to those specific regions rather than analyzing the entire wafer image comprehensively. This partial application of excessive analysis resources improves accuracy while controlling detection time
Solution Approach 2:
The AI defect detection system automatically performs initial screening and prioritization of defect candidates, serving itself to identify which regions require further human review. This self-service capability reduces the time humans need to spend on routine analysis, improving overall detection efficiency while maintaining high accuracy through targeted human review
Data Source
AI summary
Embodiments of the present disclosure provide a detection method and apparatus, an electronic device, and a storage medium. In one form, the detection method includes: providing a layout graphic and a scan graphic; superimposing and comparing the layout graphic and the scan graphic, and extracting a sample non-overlapping pattern; encoding the sample non-overlapping pattern, to form sample coded data; using the sample coded data as input data of machine learning, to obtain a detection model library; and detecting a defect point of a to-be-detected device by using the detection model library. The present disclosure can improve the accuracy of defect point analysis, thereby accelerating the development of technology and improving the production efficiency.