AI Wafer Inspection Using Multi-Light Die Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing wafer inspection methods face challenges in high-throughput inspection due to high equipment costs and the need for continuous updates of reference images, and they struggle to adapt to diverse defect types and changing production processes.
Innovation Solution
A wafer inspection system utilizing an illumination device with multiple light sources and a sensing device to obtain image data, combined with a deep learning model that classifies dies into categories without requiring reference images, by training on image data to identify similarities and differences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If optical inspection technology is used to compare images with reference images, then defect detection can be performed, but reference image databases must be continually updated and supplemented for different production processes and dies
Solution Approach 1:
The system performs self-learning by automatically training deep learning models on inspection images to establish defect patterns without requiring manual reference image databases. The model continuously improves its defect recognition capability through self-service learning from actual production data
Solution Approach 2:
The patent replaces the traditional mechanical reference image comparison system with an AI-based deep learning model that automatically learns defect patterns. This substitution eliminates the need for manual reference image database creation and maintenance while improving defect detection accuracy
2Measurement precision
If electronic microscope is used for defect inspection, then specific types of defects can be detected, but equipment costs are high and professional operating techniques are required
Solution Approach 1:
The system uses standard optical inspection equipment instead of expensive electron microscopes, replacing high-cost specialized equipment with more accessible and easier-to-operate optical systems that can still achieve effective defect detection through AI enhancement
Solution Approach 2:
The patent replaces the complex electron microscope system with an optical inspection system enhanced by deep learning algorithms. This substitution maintains defect detection precision while significantly reducing equipment costs and operational complexity
3Adaptability or versatility
If different inspection methods are used for different defect types, then various defects can be identified, but inspection methods need to persistently innovate and improve to adapt to changing production processes
Solution Approach 1:
The deep learning model serves as a universal inspection system that can identify multiple types of defects (particles, scratches, pattern defects) using a single unified approach. The model's multi-functionality allows it to adapt to different defect types without requiring separate specialized inspection methods for each defect category
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables efficient and accurate classification of good and defective dies, adapting to various defects and production processes, without the need for constant reference image updates, improving throughput and reducing costs.
Implementation Method 1
The illumination device includes a plurality of light sources, and is configured to illuminate the wafer to be inspected
Implementation Method 2
The sensing device is configured to sense lights reflected from the wafer to be inspected
Data Source
AI summary
A wafer inspection system includes: an inspection platform, accommodating wafer to be inspected including multiple dies, an illumination device, a sensing device, a control module, for each of the dies, controlling the illumination device to illuminate the die with different combinations of light sources in multiple time periods and controlling the sensing device to obtain multiple sets of image data of the die in the time periods, and a computing module, training a deep learning model according to the image data of the dies to determine multiple classification features, classifying the dies according to the classification features to categorize those determined to be same into a same category, defining dies that belong to a first category having a greatest number of dies as first-category dies, and defining dies that do not belong to the first category as second-category dies.