Adaptive Digital Filtering for Small Defect Optical Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current digital filter techniques in optical characterization systems are limited by a predefined set of templates, which are insufficient for detecting smaller defects, leading to increased noise and nuisance rates, and fail to enhance detection sensitivity.
Innovation Solution
A system and method that utilizes a controller with processors to apply a convolution filter and adjust it using a machine learning classifier to generate an optimized digital filter, enhancing defect detection in optical inspection systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If predefined digital filter templates are used, then the system is simple to operate, but defect detection sensitivity is insufficient for smaller defects
Solution Approach 1:
The patent implements dynamic filter adjustment by training machine learning classifiers (e.g., support vector machines, neural networks) to automatically optimize digital filter coefficients based on defect characteristics. The system transitions from static predefined filters to dynamic adaptive filters that adjust parameters in real-time according to the specific defect being detected, thereby improving detection sensitivity without requiring manual intervention.
Solution Approach 2:
The system changes filter parameters (coefficients, thresholds, kernel sizes) based on defect size, type, and location. By using machine learning to predict optimal parameter sets for different defect scenarios, the system adapts its filtering characteristics dynamically, enabling high sensitivity for both small and large defects while maintaining system simplicity through automated parameter selection.
2Reliability
If predefined digital filter templates are used, then the device complexity is low, but nuisance rates increase due to insufficient detection sensitivity
Solution Approach 1:
The system incorporates feedback loops where detection results (including nuisance detections) are fed back to the machine learning classifiers, which then adjust filter parameters to improve future detections. This closed-loop approach continuously optimizes detection reliability by learning from both true defects and nuisance signals, reducing false positives while maintaining system automation.
Solution Approach 2:
The machine learning-based filter adjustment system operates autonomously without requiring manual tuning or intervention. The system self-optimizes by automatically training classifiers on defect data and generating appropriate filter parameters, thereby improving reliability while keeping the operational complexity low for end users.
3Measurement precision
If standard optical inspection is used for small defects, then the system is simple, but noise increases and detection becomes difficult
Solution Approach 1:
The system applies preliminary filtering and enhancement operations before final defect detection. Machine learning classifiers pre-process images by predicting optimal filter parameters and applying targeted filtering to enhance defect signals while suppressing noise specific to small defect characteristics, thereby improving detection precision before the actual detection step.
Solution Approach 2:
The system applies different filtering strategies and parameters to different regions of the image based on local characteristics. Machine learning models identify regions containing small defects and apply specialized filtering only to those areas, preserving signal quality while reducing overall noise impact and maintaining system simplicity in non-problematic regions.
Data Source
AI summary
A system for enhancing defect detection in optical characterization systems using a digital filter is disclosed. The system may include a controller including one or more processors configured to execute a set of program instructions. The set of program instructions may be configured to cause the one or more processors to: acquire one or more sample images, the one or more sample images including one or more difference images, the one or more sample images including one or more photomask images, the one or more difference images including defect data; generate one or more filtered images by applying a digital filter to each of the one or more sample images, the digital filter including a convolution filter including one or more convolution filter coefficients; and adjust the applied digital filter using a machine learning classifier, the adjusted digital filter configured to enhance defect detection of an inspection sub-system.