SEM Image Cascade Denoising for Defect Inspection Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing denoising algorithms, including machine learning and deep learning techniques, fail to effectively remove noise from scanning electron microscope (SEM) images, and training these models is challenging due to the lack of a ground truth image without noise.
Innovation Solution
A cascade denoising technique is employed, involving a non-machine learning based denoising algorithm followed by a machine learning based denoising algorithm, with a denoised reference image serving as a ground truth for training the machine learning model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single denoising algorithm is used on SEM images, then the processing is simple and fast, but the denoising performance is insufficient and noise remains in the images
Solution Approach 1:
The denoising process is segmented into multiple sequential stages: a first denoising algorithm processes the original SEM image to produce an intermediate denoised image, which then serves as input to a second denoising algorithm to produce the final denoised image. This multi-stage segmentation allows each algorithm to specialize in different aspects of noise removal, achieving superior overall denoising performance compared to single-algorithm approaches.
2Measurement precision
If machine learning based denoising algorithms are used, then denoising performance can be improved, but training these models is challenging due to lack of ground truth images without noise
Solution Approach 1:
A non-machine learning based denoising algorithm is applied as a preliminary step to generate an intermediate denoised image from the noisy SEM image. This intermediate result serves as a surrogate ground truth for training the machine learning based denoising algorithm, enabling the model to learn effective denoising patterns without requiring access to actual noise-free reference images that do not exist in practice.
3Measurement precision
If multiple denoising algorithms are applied in sequence, then the images become cleaner and defect detection accuracy improves, but the processing time and computational resources increase
Solution Approach 1:
Different denoising algorithms are selected for the first and second processing stages, each with different computational characteristics. The first algorithm is chosen for its ability to quickly remove prominent noise, while the second algorithm is selected for its effectiveness in removing residual noise. This parameter variation in algorithm selection optimizes the balance between processing time and final image quality, achieving high defect detection accuracy without excessive computational overhead.
Data Source
AI summary
An improved systems and methods for generating a denoised inspection image are disclosed. An improved method for generating a denoised inspection image comprises acquiring an inspection image; generating a first denoised image by executing a first type denoising algorithm on the inspection image; and generating a second denoised image by executing a second type denoising algorithm on the first denoised image.