Wafer Defect Review Using Single-Image ML Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing complexity and feature density of semiconductor wafers require more processing steps and traditional defect inspection methods, which are inefficient and often necessitate multiple high-resolution images for defect detection and classification, slowing down the manufacturing process.
Innovation Solution
A machine learning-based approach that uses a single coarse SEM image to identify and classify defects by comparing it with a reference image, generating a fine-resolution image through digital image processing and patch substitution, eliminating the need for additional high-resolution imaging.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple high-resolution images are taken for defect detection and classification, then measurement precision is improved, but productivity deteriorates
Solution Approach 1:
The patent segments the defect review process into two distinct stages: a fast coarse review stage using low-resolution images for initial defect detection, and a targeted fine review stage using high-resolution images only for suspected defects. This segmentation allows the system to maintain high throughput during the coarse review phase while ensuring measurement precision is achieved during the fine review phase for classification.
Solution Approach 2:
The patent applies local quality by providing different image resolutions to different parts of the review process. Low-resolution images are used for the majority of the wafer surface during coarse review, while high-resolution images are applied locally only to regions containing suspected defects during fine review. This ensures that high measurement precision is achieved where needed without sacrificing overall productivity.
2Manufacturing precision
If multiple high-resolution images are taken for defect classification, then manufacturing precision is improved, but loss of time increases
Solution Approach 1:
The patent performs preliminary action by conducting a coarse review of the entire wafer using low-resolution images to identify and flag suspected defects before proceeding to fine review. This preliminary screening step filters out the majority of the wafer surface that does not contain defects, so that high-resolution imaging and detailed classification are performed only on a small subset of suspected defect regions, significantly reducing the total review time while maintaining classification accuracy.
Solution Approach 2:
The patent applies partial action by applying high-resolution imaging and detailed classification procedures only to the portion of the wafer that contains suspected defects, rather than performing these resource-intensive operations on the entire wafer surface. This partial application of high-resolution analysis maintains manufacturing precision for defect classification while minimizing the loss of time.
3Productivity
If a single coarse image is used for defect review, then productivity is improved, but measurement precision deteriorates
Solution Approach 1:
The patent applies dynamics by making the image resolution dynamic and adaptive based on the review stage and defect suspicion level. The system starts with low-resolution images for the coarse review stage to maintain high productivity, then dynamically transitions to high-resolution images for the fine review stage when defects are suspected. This dynamic adjustment of image quality ensures that measurement precision is improved at the appropriate moment in the process without sacrificing overall productivity.
4Reliability
If traditional defect inspection methods are used, then reliability is maintained, but device complexity increases
Solution Approach 1:
The patent merges the coarse review and fine review processes into a single integrated defect inspection system. The system combines low-resolution and high-resolution imaging capabilities, along with automated defect detection and classification algorithms, into one unified device. This merging maintains reliability by ensuring both review stages are performed systematically while actually reducing device complexity compared to having separate independent systems for coarse and fine review.
Data Source
AI summary
This disclosure is directed to solutions of detecting and classifying wafer defects using machine learning techniques. The solutions take only one coarse resolution digital microscope image of a target wafer, and use machine learning techniques to process the coarse SEM image to review and classify a defect on the target wafer. Because only one coarse SEM image of the wafer is needed, the defect review and classification throughput and efficiency are improved. Further, the techniques are not distractive and may be integrated with other defect detecting and classification techniques.