Defect Size Estimation Using ML Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques are unable to accurately determine the size and type of defects in semiconductor fabrication, particularly for defects smaller than 80 nm or larger than 200 nm, leading to inaccuracies in defect size estimation within 15-20% of the actual size.
Innovation Solution
A system and method utilizing machine learning techniques, including image processing and classification, to receive training images, generate a machine learning classifier, filter and binarize product images, perform morphological operations, and refine defect size estimates based on both algorithm-estimated sizes and defect type classifications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional techniques are used to estimate defect sizes by computing pixel area, then the process is simple and fast, but the measurement precision is insufficient for defects outside the 80-200 nm range
Solution Approach 1:
The inspection system segments the defect analysis process into multiple stages: initial pixel-based estimation, machine learning classification of defect type, and refined size determination based on defect category. This segmentation allows the system to apply different methods for different defect types, improving overall precision while managing complexity.
Solution Approach 2:
The system performs preliminary defect type classification using machine learning before final size measurement. By preliminarily identifying whether a defect is resist-related, substrate-related, or contaminant-related, the system can then apply appropriate measurement algorithms tailored to each defect type, thereby improving measurement precision across various defect sizes.
2Loss of information
If conventional pixel area computation is used, then the method is easy to implement, but it cannot determine the type of defect being inspected
Solution Approach 1:
The system introduces machine learning classifiers as an intermediary between image acquisition and defect characterization. These classifiers analyze image features to determine defect types (resist, substrate, contaminant), preserving defect type information that would otherwise be lost in simple pixel area computation.
Solution Approach 2:
The patent replaces the simple mechanical pixel-counting method with a more sophisticated system combining machine learning classification and advanced image processing. This substitution enables defect type identification while maintaining automated operation, though it increases system complexity.
3Manufacturing precision
If simple pixel area multiplication is used for size estimation, then the calculation is fast and simple, but the manufacturing precision requirement of 15-20% accuracy cannot be met for all defect sizes
Solution Approach 1:
The system applies local quality by using different measurement algorithms and processing steps tailored to specific defect types. For example, resist defects may require different size estimation methods compared to substrate defects or contaminants, allowing each defect category to be measured with optimal precision appropriate to its characteristics.
Solution Approach 2:
The system changes parameters dynamically based on defect type classification. Once a defect is classified, the system adjusts measurement parameters, image processing filters, and calculation methods to optimize size estimation accuracy for that specific defect category, thereby achieving the required 15-20% precision across diverse defect types.
Data Source
AI summary
A system for characterizing a specimen is disclosed. In one embodiment, the system includes a controller configured to: receive training images of one or more defects of the specimen; generate a machine learning classifier based on the training images; receive product images of one or more defects of a specimen; determine one or more defect type classifications of one or more defects with the machine learning classifier; filter the product images with one or more smoothing filters; perform binarization processes to generate binarized product images; perform morphological image processing operations on the binarized product images; determine one or more algorithm-estimated defect sizes of the one or more defects based on the binarized product images; and determine one or more refined estimates of one or more defect sizes of the one or more defects based on the one or more algorithm-estimated defect sizes and the one or more defect type classifications.