Least-Squares Image Alignment for Semiconductor Mask Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional pattern inspection apparatuses face challenges in accurately correcting for systematic position errors such as stage misplacement, expansion, contraction, and rotation, which are essential for detecting minute defects in semiconductor mask patterns, due to limitations in existing image registration methods that fail to adequately compensate for these distortions.
Innovation Solution
A pattern inspection apparatus and method utilizing a least-squares method with a mathematical model that aligns optical and reference images pixel by pixel, calculating parameters for parallel shift, expansion and contraction, rotation, gray-level offset, and image transmission loss, and iteratively recalculating parameters to generate a corrected image, thereby improving alignment accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image registration methods are used, then the inspection process can be completed, but the alignment accuracy is insufficient to detect minute defects
Solution Approach 1:
The patent transforms the image registration problem from pixel-level parameter estimation to feature-point-level parameter estimation. By extracting characteristic points and calculating their displacement vectors, the system determines transformation parameters (translation, rotation, scaling) that accurately represent systematic errors. This parameter transformation enables precise compensation of distortion while maintaining computational efficiency, resolving the contradiction between registration accuracy and defect detection reliability.
2Measurement precision
If complex image registration algorithms are applied to correct all types of errors, then alignment precision improves, but calculation time increases and random errors are amplified
Solution Approach 1:
The patent extracts only the essential information needed for registration by identifying and utilizing characteristic points (features) in the image. Instead of processing all pixels, the system selects a subset of distinctive points that adequately represent the geometric transformation. This extraction approach reduces computational complexity while maintaining alignment precision, as the characteristic points provide sufficient constraints to determine transformation parameters without requiring exhaustive pixel-wise processing.
Solution Approach 2:
The patent segments the image registration process into distinct stages: (1) extracting characteristic points from the image, (2) calculating displacement vectors between corresponding points, (3) computing transformation parameters from these vectors, and (4) applying the transformation. This segmentation allows each stage to be optimized independently and enables parallel processing, significantly reducing operation time while preserving alignment precision.
3Productivity
If simple shift models are used for image registration, then calculation is fast, but systematic errors from distortion and rotation cannot be compensated
Solution Approach 1:
The patent introduces dynamic transformation parameters that adapt to the actual error characteristics in the image. Instead of using a fixed simple shift model, the system calculates transformation parameters (including translation, rotation, and scaling components) based on the measured displacement vectors of characteristic points. This dynamic approach allows the registration algorithm to automatically adjust to various distortion patterns while maintaining computational efficiency, resolving the contradiction between processing speed and error compensation capability.
Data Source
AI summary
A pattern inspection apparatus includes a first unit configured to acquire an optical image of a target workpiece to be inspected, a second unit configured to generate a reference image to be compared, a third unit configured, by using a mathematical model in which a parallel shift amount, an expansion and contraction error coefficient, a rotation error coefficient, a gray-level offset and an image transmission loss ratio are parameters, to calculate each of the parameters by a least-squares method, a forth unit configured to generate a corrected image by shifting a position of the reference image by a displacement amount, based on the each of the parameters, and a fifth unit configured to compare the corrected image with the optical image.