Context-Based Defect Inspection Using Probabilistic Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional defect inspection systems in semiconductor fabrication require significant expertise and are inefficient in handling the increasing complexity of semiconductor structures, leading to variability in defect detection accuracy and the risk of missing novel defect types.
Innovation Solution
A context-based inspection system that utilizes probabilistic context maps generated by unsupervised classifiers during runtime, combined with supervised classifiers, to identify defects of interest (DOIs) in semiconductor samples, improving sensitivity and reducing the reliance on operator expertise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional detection techniques are used, then operator expertise is required to set up inspection recipes, but this leads to variability in defect detection accuracy and inability to handle novel defect types
Solution Approach 1:
The inspection system performs self-learning by automatically analyzing inspection images to generate context maps and train classification models without requiring manual setup by operators. The system autonomously adapts to different defect types and structures, eliminating the need for expert operators to configure inspection recipes while maintaining high detection accuracy.
Solution Approach 2:
The patent replaces the manual mechanical process of expert operators configuring inspection parameters with an automated machine learning system. The unsupervised and supervised classifiers automatically learn defect patterns from images, substituting human expertise with algorithmic analysis that can handle novel defect types without retraining.
2Adaptability or versatility
If conventional detection techniques are used, then simple defect types can be detected, but the system cannot effectively handle the increasing complexity of semiconductor structures with multiple parameters
Solution Approach 1:
The system transforms the inspection approach by changing from manual parameter configuration to automated parameter learning. The machine learning models automatically determine optimal inspection parameters and context features based on the specific semiconductor structure being inspected, enabling the system to adapt to increasing structural complexity without requiring operators to manage numerous parameters manually.
Solution Approach 2:
The patent segments the inspection process into distinct computational stages: generating context maps from inspection images, training unsupervised classifiers to identify normal variations, and applying supervised classifiers to detect defects. This segmentation allows the system to handle complex parameters systematically by processing them through specialized algorithmic modules rather than requiring holistic human expertise.
3Reliability
If traditional inspection methods are used, then setup time can be reduced, but detection sensitivity and ability to identify novel defect types deteriorates
Solution Approach 1:
The system performs preliminary learning actions by automatically generating context maps and training classification models during the inspection setup phase without requiring manual intervention. This preliminary automated action prepares the system for high-sensitivity detection of novel defect types while eliminating the time-consuming manual setup process, as the system self-configures based on the inspection images provided.
Data Source
AI summary
A context-based inspection system is disclosed. The system may include an optical imaging sub-system. The system may further include one or more controllers communicatively coupled to the optical imaging system. The one or more controllers may be configured to: receive one or more reference images; receive one or more test images of a sample; generate one or more probabilistic context maps during inspection runtime using an unsupervised classifier; provide the generated one or more probabilistic context maps to a supervised classifier during the inspection runtime; and apply the supervised classifier to the received one or more test images to identify one or more DOIs on the sample.