Physics-Informed Neural Operators for Limited-Data Wafer Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge in semiconductor manufacturing lies in acquiring high-quality, diverse, and representative data for training AI-based image processing models due to the sensitive nature of semiconductor wafer data, which is crucial for defect inspection and quality enhancement.
Innovation Solution
A Physics Informed Neural Operator Based Learning (PINOBL) model that integrates physics-based mathematical solvers with neural networks to process images, using functional units like uniform grids and probability distributions, and computes residual loss to generate intermediate images, ultimately producing final outputs that comply with physical laws.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional AI training methods are used with extensive training data, then model accuracy improves, but data acquisition difficulty and cost increase due to sensitive semiconductor wafer data
Solution Approach 1:
The patent introduces physics-informed mathematical solvers as an intermediary between the neural network and the image processing task. These solvers encode domain knowledge and physical laws (such as diffusion equations for denoising or sharpness enhancement equations for super-resolution) to guide the learning process, enabling the model to achieve high accuracy without requiring extensive sensitive training data from semiconductor wafer inspections
Solution Approach 2:
The patent transforms the training approach by changing the parameter optimization objective. Instead of purely data-driven loss functions, the system incorporates physics-based constraints and mathematical solvers that define the optimization landscape, allowing the model to learn effective representations with fewer training samples while maintaining accuracy
2Reliability
If more training data is collected to improve model robustness, then processing reliability improves, but processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary action by pre-defining physics-informed mathematical solvers and domain-specific constraints before the training process begins. These pre-encoded physical laws and mathematical relationships (such as diffusion equations or sharpness enhancement models) provide a structured framework that guides learning, enabling the model to achieve robustness with reduced training time and fewer data samples
3Measurement precision
If complex physics-based mathematical solvers are integrated into the neural network, then image processing accuracy improves, but device complexity increases
Solution Approach 1:
The patent segments the image processing system into distinct functional components: a neural network for feature extraction and pattern recognition, and separate physics-informed mathematical solvers for specific processing tasks (denoising, super-resolution, sharpness enhancement). This segmentation allows each component to specialize in what it does best while maintaining overall system manageability and interpretability
Data Source
AI summary
Disclosed is a method for advanced image processing using physics-informed learning for artificial intelligence (AI) based applications using a Physics Informed Neural Operator Based Learning (PINOBL) model in semiconductor manufacturing. The method includes determining a attributes based on functional units associated with input images. The method also includes selecting, based on the set of attributes, a set of physics-based mathematical solvers corresponding to an image-processing task. The method also includes generating according to the set of physics-based mathematical solvers, a set of intermediate images corresponding to the set of input images.