SEM Image Distortion Correction via ML Patch Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current distortion correction techniques for SEM images are inadequate due to reliance on local alignments of smaller patches, which can be flawed, especially in cases of sparse or repeating patterns, leading to imperfect defect detection in IC manufacturing.
Innovation Solution
A method and apparatus for correcting distortion in inspection images using a machine learning-based alignment evaluation algorithm that determines local alignment results and selects an alignment model to correct distortions, minimizing the influence of flawed data by excluding outliers and fitting an alignment model that minimizes the L0 norm of distances between local alignment results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If local alignment methods are used for distortion correction, then the process can be simplified, but the accuracy deteriorates due to reliance on potentially flawed local alignment results
Solution Approach 1:
The patent introduces a feedback mechanism where alignment models are evaluated using a held-out set of local alignment results. The evaluation metric (e.g., correlation coefficient, RMSE) provides feedback on model quality, allowing selection of the best model that generalizes well to unseen data, thus preventing over-reliance on potentially flawed local alignments.
Solution Approach 2:
The patent performs preliminary actions by dividing the image into multiple patches and pre-computing local alignment results before selecting an alignment model. This preliminary processing enables systematic evaluation and selection of robust alignment models that minimize the impact of any single flawed local alignment.
2Device complexity
If traditional distortion correction is used, then the method is simpler, but defect detection accuracy deteriorates due to misalignment
Solution Approach 1:
The patent segments the inspection image into multiple overlapping patches, allowing independent alignment analysis for each region. This segmentation enables the system to identify and correct distortions locally while maintaining global consistency, thereby improving defect detection accuracy without requiring a completely new correction method.
Solution Approach 2:
The patent changes the parameters of the alignment model selection process by using multiple candidate models with different complexity levels and evaluation metrics. This parameter optimization ensures that the selected model achieves the best balance between computational cost and defect detection accuracy.
3Measurement precision
If alignment models are selected based on fit to local alignment results, then measurement precision improves, but device complexity increases
Solution Approach 1:
The patent uses evaluation metrics (correlation coefficient, RMSE) as feedback mechanisms to automatically assess alignment model quality. This feedback enables automated model selection based on quantitative performance data, improving alignment precision while keeping the complexity manageable through objective, measurable criteria.
Solution Approach 2:
The patent creates multiple candidate alignment models that can be evaluated and selected. By preparing and comparing multiple models rather than relying on a single complex model, the system achieves high precision through model selection while distributing complexity across multiple simpler, evaluable alternatives.
Data Source
AI summary
An improved systems and methods for correcting distortion of an inspection image are disclosed. An improved method for correcting distortion of an inspection image comprises acquiring an inspection image, aligning a plurality of patches of the inspection image based on a reference image corresponding to the inspection image, evaluating, by a machine learning model, alignments between each patch of the plurality of patches and a corresponding patch of the reference image, determining local alignment results for the plurality of patches of the inspection image based on a reference image corresponding to the inspection image, determining an alignment model based on the local alignment results, and correcting a distortion of the inspection image based on the alignment model.