Adaptive ROI Classification for SEM Critical Dimension Metrology
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current techniques for aligning patterns of interest (POIs) and regions of interest (ROIs) in scanning electron microscopy (SEM) images are unreliable due to variations in SEM-to-SEM alignment and structural variations in semiconductor specimens, leading to inaccurate ROI placement and failure to include the entire region required for measurements.
Innovation Solution
A system and method using machine learning techniques to generate adaptive regions of interest (ROIs) that can correct for process and structural variations, employing a machine learning classifier to accurately place ROIs based on training images and product images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional POI alignment techniques are used to place ROIs in SEM images, then the process is simple and straightforward, but the alignment accuracy deteriorates due to SEM-to-SEM alignment variations and structural variations in specimens
Solution Approach 1:
The patent replaces the mechanical alignment system (POI-based coordinate transformation) with a machine learning-based semantic segmentation system. Instead of relying on physical coordinate transformations between SEM images, the system uses deep learning models to automatically identify and segment relevant structures in the target image, thereby achieving accurate ROI placement without being affected by alignment variations
Solution Approach 2:
The patent changes the fundamental parameters used for ROI placement from fixed geometric coordinates (based on POI alignment) to adaptive semantic features (identified by machine learning). The system transforms the input images through multiple processing stages including normalization, augmentation, and feature extraction, ultimately generating ROIs based on learned semantic patterns rather than predetermined geometric parameters
2Adaptability or versatility
If fixed POI structures are used for alignment, then the method is consistent and repeatable, but it fails to account for process variations and structural variations in the specimen
Solution Approach 1:
The patent implements a dynamic ROI generation system where the region of interest is not fixed but adapts to each target image through machine learning. The system dynamically adjusts ROI boundaries, size, and position based on the specific structural variations present in each specimen, allowing the measurement system to respond flexibly to process variations while maintaining measurement accuracy
Solution Approach 2:
The patent introduces machine learning models as an intermediary between the input SEM images and the final ROI placement. This intermediary layer processes the images through multiple transformations including normalization, augmentation, and semantic segmentation, enabling the system to bridge the gap between fixed measurement protocols and variable specimen structures
3Measurement precision
If manual ROI placement based on POI location is used, then the system is easy to operate, but misalignment occurs leading to incomplete measurement regions
Solution Approach 1:
The patent implements a self-service ROI generation system where the machine learning model automatically identifies and places ROIs without requiring manual intervention. The system performs self-correction by comparing predicted structures with actual image features and adjusting ROI boundaries accordingly, thereby maintaining measurement precision while eliminating the need for manual alignment operations
Data Source
Figure 1
Figure 2
Figure 3A~3B
AI summary
A metrology system is disclosed. In one embodiment, the system includes a characterization sub-system configured to acquire one or more images of a specimen. In another embodiment, the system includes a controller configured to: receive one or more training images of a specimen from the characterization sub-system; receive one or more training region-of-interest (ROI) selections within the one or more training images; generate a machine learning classifier based on the one or more training images and the one or more training ROI selections; receive one or more product images of a specimen from the characterization sub-system; generate one or more classified regions of interest with the machine learning classifier; and determine one or more measurements of the specimen within the one or more classified regions of interest.