Semiconductor Image Segmentation with Anchor Features for Low-Contrast
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor inspection methods face challenges in accurately measuring and segmenting 3D semiconductor structures due to high computational effort, imaging noise, and low image contrasts, which complicates the generation of training data for machine learning algorithms, especially for features like HAR structures.
Innovation Solution
A method involving the selection of an anchor feature with high image contrast, using a transfer property to determine contours in cross-section images, reduces user interaction and enhances segmentation and annotation robustness, allowing for high-throughput and accurate inspection of semiconductor structures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional segmentation methods are used for 3D semiconductor structures, then measurement precision can be maintained, but computational effort increases significantly
Solution Approach 1:
The patent applies segmentation by dividing the complex 3D semiconductor structure analysis into multiple 2D cross-section images. Instead of processing the entire 3D volume at once, the method slices the structure into multiple 2D sections, segments features in each section independently, and reconstructs the 3D information. This reduces computational complexity while maintaining measurement precision through the use of anchor features and transfer properties across sections.
2Measurement precision
If traditional segmentation methods are used for 3D semiconductor structures, then measurement precision can be maintained, but productivity decreases due to high computational effort
Solution Approach 1:
The patent implements preliminary action by pre-selecting anchor features in each 2D cross-section that serve as reference points for subsequent segmentation. These anchor features are identified beforehand and used to establish transfer properties to adjacent sections, enabling faster and more accurate segmentation without requiring intensive computational processing of the entire 3D volume, thus improving productivity while maintaining precision.
3Measurement precision
If machine learning training data is generated from 3D semiconductor structures, then inspection accuracy can be improved, but the generation process requires excessive computational resources and time
Solution Approach 1:
The patent applies copying by generating 2D cross-section images that represent slices of the 3D semiconductor structure. These 2D copies are then segmented and annotated to create training data for machine learning algorithms. Instead of working with the full 3D data, the method creates and processes multiple 2D copies, significantly reducing the computational resources and time required for training data generation while maintaining the essential structural information needed for accurate inspection.
4Productivity
If cross-section images are used for segmentation, then productivity can be improved through faster processing, but image noise and low contrast reduce measurement precision
Solution Approach 1:
The patent introduces an intermediary approach by using anchor features as mediators between noisy 2D cross-section images and the final segmentation results. These anchor features serve as reliable reference points that bridge the gap between the noisy image data and the desired accurate segmentation, enabling fast processing while maintaining measurement precision through the transfer property mechanism.
Data Source
AI summary
A method for segmentation of images using anchor features utilizes prior knowledge and can also be applied to semiconductor features with poor image contrast. The method can be more flexible and robust and involve less user interaction than conventional segmentation methods. With a system incorporating a method of the disclosure, an inspection task of semiconductor objects of interest can be improved and training data for training a machine learning method can be provided.