Neural Network Feature Transformation for Wafer Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The miniaturization of semiconductor integrated circuits has led to a need for high-speed and high-accuracy detection of minute defects on wafers, which is challenging due to variations in defect size, shape, and image brightness and contrast caused by exposure conditions.
Innovation Solution
An information processing apparatus comprising a first neural network for extracting features from query and search object images, a processing circuitry for calculating similarity scores and feature transformation information, and a user interface for determining feature transformation based on user feedback, which is then used to update the neural networks for improved similarity detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional defect detection methods are used on semiconductor wafers, then the detection process can be performed, but the accuracy is insufficient due to variations in defect size, shape, and image brightness and contrast
Solution Approach 1:
The patent transforms wafer images into frequency domain representations using Fourier transform, changing the parameter space from spatial domain to frequency domain. This parameter transformation enables better separation of defect signals from background variations, improving detection accuracy despite variations in defect size, shape, and image conditions
Solution Approach 2:
The patent introduces an intermediary process (frequency transformation and feature extraction) between the raw wafer image and defect detection. By using frequency domain analysis as an intermediary, the system can effectively handle variations in image brightness and contrast that directly affect conventional detection methods
2Manufacturing precision
If the semiconductor processing technology advances and miniaturization increases, then more advanced processing is achieved, but the detection of minute defects becomes more difficult
Solution Approach 1:
The patent moves the detection problem from two-dimensional spatial analysis to two-dimensional frequency analysis, effectively changing the dimension of analysis. This dimensional transformation in the frequency domain allows minute defects to be distinguished from background noise more effectively, enabling detection capability to keep pace with miniaturization
3Measurement precision
If user feedback is incorporated to refine feature transformation, then detection accuracy improves, but the processing time and complexity increase
Solution Approach 1:
The patent implements a feedback mechanism where user corrections to similarity检测结果 are used to refine the feature transformation process. This feedback loop allows the system to learn from user expertise and continuously improve detection accuracy, with the transformed features being reused in subsequent detections to minimize time loss
Solution Approach 2:
The patent performs feature transformation in advance, converting wafer images to frequency domain representations before the actual defect detection process. This preliminary action prepares the data in an optimized format that facilitates faster and more accurate similarity comparison, reducing processing time during the detection phase
Data Source
AI summary
An information processing apparatus includes a first neural network that extracts a feature of a query image and features of search object images; a processing circuitry detects a degree of similarity between the search object images and a query image based on the feature of the query image and the features of the search object images, and calculates a score of each of the search object images based on the degree of similarity between each of the search object images and the query image and feature transformation information relating to the degree of similarity between each of the search object images and the query image; a second neural network that outputs the feature transformation information; and a user interface that determines by a user the feature transformation information of each of the search object images based on the degree of similarity or the scores of the search object images.