Image Processing System for Semiconductor Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing machine learning methods for image identification, particularly in semiconductor inspection using scanning electron microscopes, require extensive and time-consuming data preparation and learning processes, which can interfere with production line operations and are prone to errors due to disturbances like contrast reduction, luminance unevenness, and noise.
Innovation Solution
An image processing system and computer program that generate learning data by selecting and combining images that have failed and successfully identified, using a machine learning engine to add specific disturbance information to create corrected collation image data, thereby reducing the amount of learning data and shortening the learning time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large amount of learning data is used to improve identification performance, then identification accuracy is improved, but data preparation time and learning time increase significantly
Solution Approach 1:
The patent extracts only the essential disturbance information from failed identification images by having operators selectively mark specific disturbance regions (such as charging artifacts, noise areas, or distortion zones) rather than using entire images. This extraction approach creates compact learning data that captures the critical failure modes without requiring large volumes of complete images, thereby reducing both data preparation time and learning time while maintaining identification accuracy.
2Measurement precision
If machine learning is performed using a large amount of learning data, then identification performance is improved, but production line operation may be interfered with
Solution Approach 1:
The patent enables learning data to be prepared and processed outside the main production line flow by extracting only the necessary disturbance information from failed images. This allows machine learning to be performed using minimal data extraction rather than processing large volumes of production images, thereby improving identification performance without significantly interfering with production line operations and maintaining productivity.
3Ease of operation
If traditional machine learning methods are used for SEM image identification, then identification is performed, but the system is sensitive to disturbances such as contrast reduction, luminance unevenness, and noise
Solution Approach 1:
The patent applies preliminary action by pre-marking and extracting disturbance information from failed identification images before the actual identification process. Operators selectively mark regions affected by charging, noise, or other disturbances in advance, creating a curated dataset of disturbance patterns. This preliminary extraction enables the identification system to learn and account for these disturbances beforehand, improving robustness against contrast reduction, luminance unevenness, and noise while maintaining ease of operation.
Data Source
AI summary
An object of the present invention is to achieve both suppression of data amount of an image processing system that learns a collation image to be used for image identification using a discriminator and improvement of identification performance of the discriminator. In order to achieve the above object, there is proposed an image processing system including a discriminator that identifies an image using a collation image, the image processing system further including a machine learning engine that performs machine learning of collation image data required for image identification. The machine learning engine searches for a successfully identified image using an image for which identification has been failed, and adds information, obtained based on a partial image of the image for which identification has been failed and which has been selected by an input device to the successfully identified image obtained by the search to generate corrected collation image data.