AI Mask Defect Classification With Data Augmentation for EUV Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing defect classification methods for masks, particularly in the context of Extreme Ultraviolet Lithography, are time-consuming and prone to inconsistent analysis due to reliance on visual inspection and algorithms like CNN or K-NN, necessitating a more reliable and efficient automated process.
Innovation Solution
An electronic device and method utilizing a communication circuit, memory, and processor to receive defect data, perform data augmentation, and train an AI model to classify mask defects, incorporating SEM and GDS images, and enhance classification accuracy through data augmentation and transfer learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If visual analysis by operators is used to detect mask defects, then defect detection can be performed, but the process is time-consuming and produces inconsistent analysis results
Solution Approach 1:
The patent replaces the manual visual inspection system with an automated defect classification system using artificial intelligence models. The processor automatically classifies defect types based on defect data without human intervention, eliminating the time-consuming and inconsistent nature of operator-based visual analysis while maintaining high reliability through consistent algorithmic processing
Solution Approach 2:
The defect classification system performs self-service by automatically analyzing and classifying defect data without requiring operator intervention. The AI model independently processes defect images and data, generating classification results that are both time-efficient and consistently reliable, freeing operators from manual inspection tasks
2Extent of automation
If traditional algorithms like CNN or K-NN are used for automatic defect classification, then automation is achieved, but the classification reliability is insufficient for sophisticated EUV processes
Solution Approach 1:
The patent changes the parameters and architecture of the AI model from traditional CNN or K-NN algorithms to a more sophisticated neural network structure with multiple layers and advanced feature extraction capabilities. This parameter change in the model complexity and architecture enables reliable classification for sophisticated EUV processes while maintaining full automation
Solution Approach 2:
The patent employs a composite approach by combining multiple defect data types (defect images, process data, and other relevant parameters) into a comprehensive training dataset. This composite data structure, processed through an advanced AI model, creates a robust classification system that achieves both automation and high reliability for EUV mask defects
3Measurement precision
If data augmentation is performed to secure sufficient training data, then classification accuracy is improved, but the processing complexity increases
Solution Approach 1:
The patent performs data augmentation in advance during the dataset preparation phase, before the actual defect classification process. By pre-processing and augmenting the training data to include various defect types, conditions, and scenarios, the system secures sufficient diverse training data without adding complexity to the real-time classification operation. The AI model is trained on this pre-augmented comprehensive dataset, achieving high classification accuracy while maintaining efficient operational simplicity
Data Source
AI summary
An electronic device includes: a memory configure to store a first training dataset labeled with a defect shape and a mask pattern of a mask and at least one artificial intelligence model; and at least one processor configured to receive defect data including first defect data, second defect data, and third defect data; by using the second defect data, obtain a second training dataset by performing data augmentation on a defect shape for which an amount of labeled data in the first training dataset is determined to be less than a threshold value; train a first artificial intelligence model generated to classify a defect type of the mask based on the first training dataset and the second training dataset; and classify a defect type of the mask based on at least one of the first defect data, the third defect data, or the first artificial intelligence model.