Learning-Based Die-to-Die Mask Inspection for Sub-Micron Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current die-to-die mask inspection methods struggle to accurately detect ultra-precision defects of 1 μm or less due to vibrations and difficulties in distinguishing between inspection errors, mask information, and actual defects.
Innovation Solution
A learning-based die-to-die mask inspection apparatus and method that uses an image sensor to acquire images of dies, a model generation unit to generate a clean mask using a pre-trained model, and a processor to detect defects by comparing crop data with the clean mask, thereby isolating and identifying defects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If die-to-die mask inspection is performed using image difference with known position information, then defects can be found, but ultra-precision defects of 1 μm or less cannot be inspected due to stage vibrations and difficulty in distinguishing inspection errors from actual defects
Solution Approach 1:
The patent performs preliminary actions by acquiring multiple images of the same die region under different conditions (different illumination angles, different wavelengths, or different focus positions) before conducting the actual defect detection. This preliminary multi-image acquisition allows subsequent analysis to distinguish between real defects and artifacts caused by stage vibrations or measurement errors, thereby improving both measurement precision and result reliability for ultra-precision defects of 1 μm or less
Solution Approach 2:
The patent creates multiple copies of the die region images under different imaging conditions. By comparing these copied images taken at different times or under different parameters, the system can identify consistent features that represent actual defects while filtering out transient variations caused by vibrations or measurement errors, thus enhancing the reliability of defect detection for sub-micron features
2Productivity
If traditional die-to-die inspection method is used, then inspection can be performed, but it is difficult to determine whether the inspection result is an error caused by vibrations, mask information, or a defect
Solution Approach 1:
The patent implements feedback mechanisms by analyzing multiple images and comparing results across different imaging conditions. The system uses this feedback loop to iteratively refine defect identification, cross-validating findings across multiple image sets to distinguish true defects from artifacts. This multi-conditional analysis provides feedback that confirms or refutes potential defect locations, ensuring accurate defect identification without sacrificing inspection speed
Solution Approach 2:
The patent applies partial action by focusing computational resources on regions of interest identified in preliminary image comparisons. Rather than analyzing entire die regions uniformly, the system performs excessive analysis only on suspicious local areas where defects may be present, comparing these specific regions across multiple images. This approach maintains high productivity while improving defect identification accuracy by concentrating computational effort where it is most needed
Data Source
AI summary
The present invention relates to a learning-based die-to-die mask inspection apparatus and method. The learning-based die-to-die mask inspection apparatus includes an image sensor that acquires images of dies of a mask, a model generation unit that generates a clean mask using a pre-trained model, and a processor that generates crop data of corresponding pairs for the same region of the dies from the images acquired by the image sensor, inputs the crop data to the model generation unit, receives the clean mask from the model generation unit, and then detects a defect in each of the dies through the crop data and the clean mask.