Digital Nonlinear Image Filtering for Wafer Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current optical wafer inspection systems face challenges in accurately distinguishing between defects of interest (DOIs) and nuisances due to limited digital linear filter templates, leading to high nuisance rates and low DOI capture rates, especially for small defects near the sensitivity limit.
Innovation Solution
Implementing a digital nonlinear neural network, such as AiryNet™, for image filtering that applies supervised nonlinear techniques using a region-based neural network with a limited number of layers and training parameters, allowing for enhanced defect detection and nuisance reduction by classifying defect and noise subregions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If digital linear filter templates are used for image filtering, then the system is simple to implement, but the nuisance rate is high and DOI capture rate is low
Solution Approach 1:
The patent transforms the filtering approach by changing from linear filter templates to a nonlinear neural network model. This parameter change enables the system to adaptively learn complex defect patterns and noise characteristics, significantly improving DOI capture rate while maintaining low nuisance rate, despite increased computational complexity.
Solution Approach 2:
The patent replaces the traditional mechanical/filter-based linear filtering system with a neural network-based nonlinear filtering system. This substitution allows the system to automatically learn and adapt to various defect types and noise patterns, achieving superior detection performance compared to fixed linear filters.
2Measurement precision
If more digital linear filter templates are used to improve DOI detection, then the detection sensitivity increases, but the system complexity increases
Solution Approach 1:
The patent changes the fundamental parameter of the filtering system from a collection of fixed linear templates to a single trainable nonlinear neural network. This allows the system to achieve high detection sensitivity through adaptive learning rather than through increasing the number and complexity of filter templates.
Solution Approach 2:
The neural network is trained to automatically learn and adapt to the specific characteristics of defects and noise in the inspection data. This self-learning capability eliminates the need for manual design and tuning of multiple complex filter templates, achieving high sensitivity with a unified model.
3Speed
If traditional filtering methods are used, then the processing speed is fast, but the ability to distinguish DOIs from nuisances is poor
Solution Approach 1:
The patent replaces traditional filtering methods with a neural network-based approach that can process images at high speed while simultaneously achieving superior defect classification accuracy. The neural network learns to distinguish DOIs from nuisances through training, maintaining fast processing while dramatically improving discrimination capability.
Solution Approach 2:
The patent changes the processing methodology from fixed linear filtering to adaptive nonlinear filtering using neural networks. This allows the system to maintain processing speed while significantly improving the ability to distinguish between defects of interest and nuisance artifacts through learned feature extraction.
Data Source
AI summary
Methods and systems for determining information for a specimen location are provided. One system includes a computer subsystem configured for inputting one or more images for a specimen location into a region-based neural network configured for performing digital non-linear filtering of the one or more images to thereby generate a filtered image for the specimen location. The computer subsystem is also configured for determining information for the specimen location from the filtered location.