Iterative Image Clustering for Semiconductor Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image clustering methods for semiconductor circuit patterns face challenges in accurately classifying complex patterns due to scale heterogeneity, leading to human errors, high costs, and inefficient classification speeds, particularly in detecting defects on wafers.
Innovation Solution
An iterative clustering method using a computing device that converts images into feature vectors, performs initial clustering, and repeats clustering with adjusted parameters until all clusters meet a reference score, incorporating pre-processing steps like removing transparency, adjusting resolution, and using reinforcement learning to optimize clustering parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If manual classification by operator is used to reduce inspection points, then work time is reduced, but human errors exist and classification accuracy deteriorates
Solution Approach 1:
The patent replaces manual operator classification with an automated image clustering system that uses machine learning algorithms to classify circuit pattern images. The system extracts features from images and automatically groups them into clusters without human intervention, eliminating human errors while maintaining efficient processing speed.
Solution Approach 2:
The clustering system performs self-service by automatically evaluating cluster quality using silhouette scores and iteratively optimizing clustering parameters without requiring operator intervention. The system autonomously adjusts clustering parameters to improve classification accuracy while maintaining efficient processing.
2Extent of automation
If conventional clustering functions are used to classify circuit pattern images, then automation is achieved, but clustering performance is low causing misclassification errors
Solution Approach 1:
The patent implements dynamic clustering parameter adjustment where the system automatically modifies clustering parameters based on silhouette score evaluation. The system iteratively adjusts parameters such as cluster count and feature weighting to optimize clustering accuracy, transitioning from static conventional clustering to dynamic adaptive clustering.
Solution Approach 2:
The system incorporates feedback mechanisms by evaluating cluster quality using silhouette scores and using this feedback to iteratively optimize clustering parameters. The silhouette score measures cluster separation quality, and this feedback drives continuous improvement of clustering accuracy until convergence is achieved.
3Measurement precision
If comprehensive inspection of all circuit pattern points is performed, then inspection accuracy is improved, but the number of comparison points becomes excessive increasing work time
Solution Approach 1:
The patent segments the comprehensive inspection task into clustered groups of similar circuit pattern images. By grouping images with similar characteristics into clusters, the system enables representative sampling within each cluster, reducing the total number of images requiring detailed inspection while maintaining high inspection accuracy through cluster-based analysis.
4Productivity
If feature vectors from multiple images are clustered in one step, then processing speed is improved, but scale heterogeneity causes poor clustering performance
Solution Approach 1:
The system dynamically adjusts clustering parameters based on the heterogeneity of feature vectors. By monitoring cluster quality metrics and automatically modifying parameters such as distance thresholds and feature weights, the system adapts to scale variations in the data, maintaining high clustering performance while processing multiple images efficiently.
Solution Approach 2:
The patent changes clustering parameters iteratively to account for scale heterogeneity in feature vectors. The system adjusts parameters such as silhouette score thresholds and clustering algorithm parameters based on the characteristics of the input data, enabling effective clustering despite variations in feature scales across different images.
Data Source
AI summary
A method performed by a computing device for clustering an image according to an embodiment of the present disclosure includes performing a first clustering on feature vectors of the plurality of images, and performing a second clustering for feature vectors belonging to some clusters that do not satisfy a reference score among clusters formed as a result of the first clustering, wherein a clustering parameter of the second clustering and a clustering parameter of the first clustering are different from each other.