Golden Template Alignment Using ROI Pixel Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The alignment of a golden template with a specimen image for defect detection in patterned device inspection is time-consuming and complex, particularly with high-resolution images, due to the need for pixel-by-pixel comparison.
Innovation Solution
Utilizing machine learning techniques to identify distinct regions across multiple images of patterned specimens, employing an intersection over union algorithm to facilitate efficient alignment, reducing computational load by comparing pixels within these regions instead of the entire image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pixel-by-pixel comparison is performed across the entire image, then alignment precision is improved, but processing time increases significantly
Solution Approach 1:
The patent divides the image into multiple regions of interest (ROIs) that contain distinctive features. Instead of performing pixel-by-pixel comparison across the entire image, the system segments the image into these meaningful regions and performs alignment only on the ROIs. This segmentation approach maintains alignment precision while significantly reducing processing time.
Solution Approach 2:
The patent extracts only the necessary regions containing distinctive features from the entire image. By taking out and focusing only on these critical regions for alignment comparison, the system achieves precise alignment without the computational burden of processing every pixel in the complete image.
2Speed
If machine learning techniques are used to identify regions, then alignment speed is improved, but computational complexity increases
Solution Approach 1:
The patent performs preliminary action by training a machine learning model in advance to recognize and identify regions of interest. Once trained, the model can quickly annotate ROIs in new images without requiring complex real-time computation. This preliminary training phase enables fast alignment by pre-establishing recognition patterns.
Solution Approach 2:
The patent introduces a machine learning model as an intermediary between the image input and alignment processing. This intermediary component automatically identifies and selects regions of interest, acting as a mediator that simplifies the subsequent alignment computation by filtering out irrelevant areas before the main processing step.
Data Source
AI summary
Systems and methods of analyzing aligning a golden template with a scanned image of specimen under manufacture are disclosed herein. A computing system generates a golden template for a patterned specimen design by forming a single coherent image from a first plurality of scanned images of a plurality of specimens manufactured according to the patterned specimen design. The computing system identifies a plurality of regions of interest in the patterned specimen design that is present across a second plurality of scanned images of the plurality of specimens manufactured according to the patterned specimen design. The computing system receives a new scanned image of a new specimen manufactured in accordance with the patterned specimen design. The computing system aligns the golden template with the new scanned image of the new specimen by performing a pixel-by-pixel analysis in the plurality of regions of interest only.