Deep Convolutional Neural Network for Semiconductor Image Super-Resolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor inspection and metrology processes face challenges in generating high resolution images from low resolution images, leading to time-consuming defect review and metrology processes that require physical specimens and additional tools, limiting throughput and accuracy.
Innovation Solution
A system utilizing a deep convolutional neural network with sub-pixel convolution layers to transform low resolution images into high resolution images, enabling the generation of high quality, de-noised, and super-resolved images without the need for physical specimens or additional tools, applicable to various imaging platforms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If inspection processes are used to detect defects on semiconductor specimens, then defect detection capability is improved, but image resolution remains limited requiring additional defect review processes
Solution Approach 1:
The patent creates a virtual copy of the physical specimen through image processing. A neural network generates a virtual high-resolution image from the low-resolution inspection image, eliminating the need to physically revisit the specimen for defect review. This virtual copying approach preserves all specimen information while providing enhanced resolution for defect analysis.
Solution Approach 2:
The patent replaces the mechanical/physical defect review process with an computational image processing system. Instead of physically examining specimens under high-magnification microscopes, the system uses neural networks to computationally enhance image resolution, substituting mechanical inspection with intelligent algorithms.
2Measurement precision
If physical specimens are used for defect review and metrology processes, then high resolution measurements can be obtained, but process time increases and throughput decreases
Solution Approach 1:
The patent performs preliminary image enhancement during the initial inspection process. By generating high-resolution virtual images from low-resolution inspection images at the time of inspection, the system eliminates the need for separate defect review and metrology processes, thereby maintaining high measurement precision while improving overall throughput.
Solution Approach 2:
The patent merges multiple separate processes (inspection, defect review, and metrology) into a single integrated workflow. By generating virtual high-resolution images during inspection, the system combines the functions of defect detection and high-resolution analysis, eliminating redundant processing steps and improving productivity.
3Measurement precision
If additional tools are used for defect review and metrology, then high resolution image generation capability is improved, but device complexity and cost increase
Solution Approach 1:
The patent makes the inspection system multi-functional by enabling it to perform both low-resolution inspection and high-resolution defect review/metrology through computational image processing. The same inspection system that captures low-resolution images can generate virtual high-resolution images, eliminating the need for separate high-magnification microscopes and reducing overall device complexity.
4Measurement precision
If high magnification optical systems or SEMs are used for defect review, then image resolution is improved, but processing time and specimen requirements increase
Solution Approach 1:
The patent replaces mechanical high-magnification optical systems and SEMs with computational image processing using neural networks. This substitution eliminates the need for time-consuming physical examination processes while maintaining or improving image resolution through intelligent algorithms that enhance virtual images.
Data Source
AI summary
Methods and systems for generating a high resolution image for a specimen from a low resolution image of the specimen are provided. One system includes one or more computer subsystems configured for acquiring a low resolution image of a specimen. The system also includes one or more components executed by the one or more computer subsystems. The one or more components include a deep convolutional neural network that includes one or more first layers configured for generating a representation of the low resolution image. The deep convolutional neural network also includes one or more second layers configured for generating a high resolution image of the specimen from the representation of the low resolution image. The second layer(s) include a final layer configured to output the high resolution image and configured as a sub-pixel convolutional layer.