Defect Image Prediction From Fast Scans for Wafer Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning models for defect detection in semiconductor substrates face challenges in differentiating between defects and nuisances, particularly when training data is insufficient, leading to decreased capture rates due to inadequate focus on defect differentiation and insufficient characterization features.
Innovation Solution
Enhancing defect signals and training models using customized loss functions based on defect distribution and classifier feature maps, along with selective image pair selection, to improve the conversion of low-resolution fast scan images to high-resolution slow scan images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution images are captured to distinguish defects from nuisances, then detection precision is improved, but image capture time increases significantly
Solution Approach 1:
The system performs preliminary defect localization using fast scan images to identify candidate regions, then applies slow scan imaging only to these specific regions. This preliminary action reduces the overall inspection time while maintaining high detection precision for critical defects.
Solution Approach 2:
The wafer inspection process is segmented into two stages: fast scan for broad defect localization and slow scan for detailed characterization. This segmentation allows the system to capture full-wafer images quickly and then focus high-resolution imaging only on regions containing potential defects.
2Measurement precision
If machine learning models are trained with sufficient defect differentiation focus and characterization features, then defect detection accuracy is improved, but training data requirements and model complexity increase
Solution Approach 1:
The loss function applies different weighting and optimization focus to different regions of the image and different types of features. Defect regions receive higher weighting in the loss calculation, and the model is trained to prioritize learning defect-specific features over general image features, reducing unnecessary model complexity.
Solution Approach 2:
The system uses a specialized loss function that provides targeted feedback during training by comparing predicted defect locations and characteristics against ground truth. This feedback mechanism guides the model to focus learning on defect differentiation without requiring excessive model complexity.
3Measurement precision
If conventional loss functions are used for training, then training speed is maintained, but defect differentiation capability is insufficient
Solution Approach 1:
The loss function parameters are specifically designed to emphasize defect-related features and characteristics. By changing the parameters of the loss function to weight defect differentiation more heavily, the model achieves superior defect detection capability without requiring a complete redesign of the training framework.
Data Source
AI summary
A method for training a prediction model to generate a high-resolution image representing defects on a substrate from a low-resolution image of the substrate. The method includes inputting a first image and a reference image of defects on a substrate, which are representative of images captured using different image capture conditions, to a neural network. The neural network is executed to generate a predicted image in response to the first image. A loss function that is indicative of a difference between a defect distribution in the predicted image and a defect distribution in the reference image is calculated and the neural network is modified based on the loss function. The neural network may be trained until the loss function is minimized.