Defect Classification GUI for Semiconductor Image Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual classification of defect images in semiconductor manufacturing is time-consuming and labor-intensive, and existing automatic classification methods often produce incorrect results due to insufficient resolution or classification criteria.
Innovation Solution
A GUI and classification apparatus that hierarchically organizes images based on feature calculations, allowing users to input class information and automatically add it to images, reducing user load and improving classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual classification is performed by users visually recognizing and adding classes to each image, then classification accuracy can be maintained, but user workload increases and classification time is excessive
Solution Approach 1:
The classification process is segmented into two distinct phases: (1) automatic classification by the ADC using image processing and neural networks to generate preliminary classification results, and (2) manual verification and correction by users only for misclassified images. This segmentation allows the system to leverage automated processing for speed while maintaining accuracy through targeted manual intervention.
Solution Approach 2:
The system performs preliminary automatic classification using the ADC before manual review. The neural network pre-processes and classifies images based on learned patterns from training data, providing a first-pass classification that handles the majority of images correctly, thereby reducing the volume of images requiring manual verification.
2Productivity
If automatic classification is performed using neural networks and image processing, then classification speed increases, but classification accuracy deteriorates due to insufficient resolution or incorrect classification criteria
Solution Approach 1:
The system implements a feedback mechanism where users can review and correct automatic classification results. Misclassified images are identified and presented to users for verification, and the corrected classifications are fed back into the system to improve the neural network's classification criteria and accuracy over time.
Solution Approach 2:
The system introduces an intermediary review step between automatic classification and final results. Users act as intermediaries to verify and correct classifications, bridging the gap between automated speed and manual accuracy, ensuring that final classifications meet required accuracy standards.
3Measurement precision
If users must select and add classes to each individual image from a large number of images, then precise classification can be achieved, but the operational complexity and user burden increase significantly
Solution Approach 1:
The system merges multiple classification operations into a single automated process. Instead of requiring users to individually process each image, the ADC automatically classifies batches of images using neural networks, and users only need to review and correct results in a consolidated interface, significantly reducing operational complexity.
Solution Approach 2:
The system enables self-service automatic classification where the ADC independently performs image processing, feature extraction, and classification without requiring manual intervention for each image. The system serves itself by automatically handling the classification workflow, with users only involved when correction is needed.
Data Source
AI summary
Provided is a GUI including: an unadded pane region that hierarchically displays folders which are sets of images having no class information added thereto; an image pane region that displays the images displayed in the unadded pane region, the displayed images having no classification added thereto; and a class pane region that displays images having classification added thereto, wherein by externally inputting class information for one image having the class information added thereto, the input class information is displayed.