EUV Mask Inspection Pre-classification via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems for inspecting masks for EUV lithography are inefficient in identifying and classifying potential defects, leading to unnecessary checks and prolonged processing times.
Innovation Solution
A multi-stage system comprising a first partial system for defect identification, a second partial system for pre-classification using automated image analysis or machine learning, and a third partial system for confirmatory checks, with a database for pre-classified defects to reduce the number of defects requiring further inspection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all potential mask defects are subjected to confirmatory checks, then measurement precision and reliability are improved, but productivity is reduced due to prolonged processing times
Solution Approach 1:
The inspection system is divided into multiple partial systems with different functions: a first partial system for initial defect detection, a second partial system for pre-classification using automated image analysis, and a third partial system for confirmatory checks. This segmentation allows each system to specialize in specific tasks, improving overall efficiency while maintaining accuracy for critical defects.
Solution Approach 2:
The second partial system performs preliminary classification of potential defects before they reach the third partial system for confirmatory checks. By pre-classifying defects based on automated image analysis, the system identifies which defects require further inspection and which can be discarded, thereby reducing the number of defects subjected to time-consuming confirmatory checks while maintaining measurement precision for critical cases.
2Productivity
If automated image analysis and machine learning are used for pre-classification, then productivity is improved by reducing the number of defects requiring further inspection, but device complexity increases
Solution Approach 1:
The second partial system acts as an intermediary between the first partial system (initial detection) and the third partial system (confirmatory checks). It uses automated image analysis and machine learning algorithms to pre-classify potential defects, filtering out false positives and low-priority defects before they reach the complex confirmatory check system. This intermediary layer reduces the workload on subsequent systems while maintaining overall system reliability.
3Measurement precision
If a multi-stage inspection system is implemented, then measurement precision is improved through pre-classification, but loss of time increases due to multiple inspection stages
Solution Approach 1:
The second partial system performs preliminary classification of potential defects using automated image analysis and machine learning before they undergo confirmatory checks. This pre-classification quickly identifies and filters out false positives and low-priority defects, so that only a small subset of potentially critical defects proceeds to the time-consuming third partial system. The time invested in preliminary action is offset by the significant reduction in subsequent confirmatory checks.
Solution Approach 2:
Different inspection stages apply different levels of scrutiny and analysis methods appropriate to their specific function. The first partial system performs rapid initial detection, the second partial system applies automated image analysis and machine learning for pre-classification, and the third partial system conducts detailed confirmatory checks only on selected defects. This local quality approach ensures that time and resources are concentrated where they are most needed while maintaining overall measurement precision.
Data Source
AI summary
A pre-classification of potential mask defects on the basis of machine learning is provided during the inspection of a mask for EUV lithography.