Product Model Reconstruction From Low-Resolution Fabrication Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing imaging tools for fabrication processes, such as brightfield imaging, provide low resolution images that are inadequate for detecting fine defects and measuring structural dimensions, making it difficult to assess the quality of manufactured products accurately.
Innovation Solution
A method and system that uses multiple low-resolution images to construct a product model with higher spatial resolution by minimizing a loss function based on imaging models, incorporating convolution and regularization terms, and applying a pixel resolution reduction model to enhance image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If brightfield imaging tools are used to capture images, then large areas can be measured quickly, but the resolution of the resulting images is low making it difficult to detect fine defects
Solution Approach 1:
The patent replaces the direct optical imaging mechanism with a computational approach. Instead of relying solely on optical resolution to achieve high-quality images, the system uses a neural network model that processes multiple low-resolution images to generate a high-resolution product model. This substitution of mechanical/optical systems with computational processing resolves the contradiction between measurement speed (using fast brightfield imaging) and measurement precision (achieving high resolution for defect detection).
Solution Approach 2:
The patent transitions from the spatial domain to the frequency domain through Fourier transformation. By transforming images to the frequency domain, processing to remove artifacts, and then transforming back, the system extracts high-resolution information that wasn't directly visible in the original low-resolution images. This dimensional transformation in the mathematical representation of images enables resolution enhancement without requiring higher optical power.
2Loss of information
If multiple imaging processes are used to capture different images, then more information can be obtained, but the complexity of processing and combining these images increases
Solution Approach 1:
The patent introduces a neural network model as an intermediary between the multiple low-resolution images and the final high-resolution product model. This intermediary learns the mapping relationship from training data and automatically handles the complex task of integrating information from multiple imaging processes. The neural network serves as a mediator that simplifies the processing complexity while maximizing information extraction from the input images.
Solution Approach 2:
The patent performs preliminary training of the neural network model using training images and corresponding product models before actual defect detection. This preliminary action prepares the system by pre-learning the relationship between low-resolution images and high-resolution structures. When actual images are processed, the pre-trained model can quickly generate high-resolution product models without requiring complex real-time processing, thus reducing processing complexity while maintaining information completeness.
Data Source
AI summary
A method is proposed of using low-resolution images of at least one product produced by one or more imaging processes, and imaging models characterizing the imaging processes, to determine values for plurality of numerical parameters which collectively define a product model of the at least one product. The determination of the values is performed by forming a loss function based on the acquired images, the imaging models, and the numerical parameters of the model, and performing a minimization algorithm to minimize the loss function with respect to the numerical parameters. Due to prior knowledge of the product encoded in the loss function, the product model may comprise reconstructed images which have a higher resolution than the low-resolution images.