Neural Network Device for SEM Image Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image processing methods for scanning electron microscope (SEM) images are time-consuming and lack consistency in detecting target objects, hindering efficient analysis in fields like semiconductor wafer inspection and circuit design.
Innovation Solution
A neural network device is employed, comprising a pre-processor to select and crop target images from SEM images, a neural network processor to infer target objects using a segmentation model, and a post-processor to merge crop detection images into a predicted image of the same size as the original SEM image, enhancing analysis efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image processing methods are used on SEM images, then target objects can be detected, but the processing time is excessively long
Solution Approach 1:
The patent divides the SEM image into multiple frequency bands using wavelet transform, separating the image into approximation components and detail components. This segmentation allows parallel processing of different frequency components, significantly reducing processing time while maintaining detection accuracy. The neural network processes each frequency band independently and combines results, achieving both speed and precision.
2Reliability
If conventional image processing methods are used, then target objects can be detected, but analysis consistency is insufficient
Solution Approach 1:
The patent implements a feedback mechanism where the neural network processes multiple frequency bands and iteratively refines detection results. The system compares detection outcomes across different frequency components and adjusts processing parameters to improve consistency. This feedback loop ensures reliable and consistent analysis across varying SEM image conditions.
3Quantity of substance
If the entire SEM image is processed at once, then complete image analysis is achieved, but processing time increases significantly
Solution Approach 1:
The patent segments the SEM image into multiple frequency bands and processes them in parallel using wavelet transform. This allows the system to analyze the complete image content while maintaining high processing speed through parallel computation of different frequency components, rather than sequential processing of the entire image.
Solution Approach 2:
The patent applies processing to specific frequency bands that contain the most relevant information for target object detection. By focusing computational resources on the most informative frequency components rather than uniformly processing all frequencies, the system achieves efficient processing without sacrificing detection accuracy.
Data Source
AI summary
A neural network device includes: (1) a pre-processor configured to select target images from scanning electron microscope (SEM) images, based on frequencies respectively corresponding to the SEM images, and crop each of the target images into a plurality of cropped images; (2) a neural network processor configured to generate a crop detection image by inferring a target object from each of the plurality of cropped images by using a segmentation model trained to detect the target object in each of the plurality of cropped images; and (3) a post-processor configured to merge crop detection images with each other in a same size as the SEM images, based on position information of the plurality of cropped images.