Deep Learning Denoising for Semiconductor Specimen Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current noise mitigation strategies in semiconductor inspection, such as computing linear combinations of reference frames and using difference filters, are inadequate as they do not effectively denoise test images and can leave residual noise, which affects the accuracy of defect detection.
Innovation Solution
A deep learning model, specifically a convolutional neural network (CNN), is configured for denoising images of specimens, allowing for the reduction of noise and improvement of signal-to-noise ratio (SNR) in images used for defect detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional noise mitigation strategies (linear combination of reference frames, difference filters) are used, then processing speed is maintained, but noise reduction effectiveness deteriorates leaving residual noise that affects defect detection accuracy
Solution Approach 1:
The patent replaces conventional mechanical/image-processing-based noise mitigation methods (linear combinations, difference filters) with a deep learning-based neural network system. The neural network is trained to automatically learn and apply complex denoising transformations, substituting traditional algorithmic approaches with an intelligent system that adapts to different noise patterns and achieves superior noise reduction while preserving defect information.
Solution Approach 2:
The patent transforms the noise mitigation approach by changing from fixed, predetermined filter operations to dynamic, learned transformations. The neural network adjusts its processing parameters based on the input image characteristics, enabling adaptive noise reduction that responds to varying noise levels and patterns rather than applying uniform filtering across all images.
2Measurement precision
If deep learning denoising is applied to test images, then noise reduction effectiveness improves, but processing time and computational complexity increase
Solution Approach 1:
The patent performs the computationally intensive work of training the neural network in advance, before actual defect detection is needed. During production inspection, the pre-trained network processes images much faster than conventional methods would require for equivalent noise reduction. This separates the heavy computational burden (training) from the time-critical operation (defect detection), achieving both high noise reduction and fast processing speeds.
3Adaptability or versatility
If standard difference image filters are used, then ease of implementation is maintained, but adaptability to different noise distributions deteriorates requiring special expertise to define custom filters
Solution Approach 1:
The patent enables the system to automatically adapt to different noise distributions without requiring expert intervention. The neural network is trained on diverse image data representing various noise patterns, allowing it to self-adjust its denoising behavior based on the specific characteristics of each input image. This eliminates the need for specialists to manually design and tune filters for different noise conditions.
Data Source
AI summary
Methods and systems for determining information for a specimen are provided. One system includes a computer subsystem and one or more components executed by the computer subsystem. The one or more components include a deep learning model configured for denoising an image of a specimen generated by an imaging subsystem. The computer subsystem is configured for determining information for the specimen from the denoised image.