Clustered Anomaly Detection for Multi-Recipe Semiconductor Tools
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current diagnostic models for semiconductor fabrication processes require significant computing resources and time to create and manage, and struggle with detecting defects in new or under-data recipes, leading to inefficiencies in operation and management.
Innovation Solution
An anomaly detection method using a classifier with an encoder and decoder that learns from training data subsets, extracts features, clusters data, and relearns classifiers to efficiently detect abnormalities in input data, reducing the need for multiple diagnostic models and minimizing computing resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If separate diagnostic models are created for each recipe, then detection accuracy for specific recipes is improved, but computing resource consumption and time increase significantly
Solution Approach 1:
The patent creates a single universal diagnostic model that can handle multiple recipes through data clustering and selective relearning. Instead of maintaining separate models for each recipe, the system clusters training data by recipe characteristics and relearns the universal model only when needed, allowing one model to serve multiple functions across different recipes while consuming fewer computing resources
Solution Approach 2:
The system dynamically adjusts the diagnostic model by relearning only when recipe parameters change significantly. By monitoring recipe characteristics and triggering relearning only when necessary, the system maintains high detection accuracy for specific recipes while avoiding continuous full retraining, thus reducing overall computing resource consumption
2Measurement precision
If separate diagnostic models are created for each recipe, then detection accuracy for specific recipes is improved, but operation and management complexity increase
Solution Approach 1:
The patent creates a single universal diagnostic model that can handle multiple recipes through data clustering and selective relearning. Instead of maintaining separate models for each recipe, the system clusters training data by recipe characteristics and relearns the universal model only when needed, allowing one model to serve multiple functions across different recipes while consuming fewer computing resources
Solution Approach 2:
The system segments the training data by recipe characteristics and uses clustering to organize data into manageable groups. This segmentation allows the universal model to process different recipe types efficiently without requiring separate models, simplifying management while maintaining specialized detection capabilities for each recipe category
3Measurement precision
If diagnostic models are relearned whenever recipes change, then detection accuracy is maintained, but time consumption increases
Solution Approach 1:
The system dynamically adjusts the diagnostic model by relearning only when recipe parameters change significantly. By monitoring recipe characteristics and triggering relearning only when necessary, the system maintains high detection accuracy for specific recipes while avoiding continuous full retraining, thus reducing overall computing resource consumption
Solution Approach 2:
The system performs preliminary clustering of training data by recipe characteristics during the learning phase. This preliminary organization allows the model to quickly adapt to new recipes by referencing pre-clustered data patterns, reducing the time required for relearning when recipes change without sacrificing detection accuracy
4Measurement precision
If sufficient training data is accumulated for each recipe, then detection accuracy is improved, but the time to accumulate data increases
Solution Approach 1:
The system dynamically adjusts the diagnostic model by relearning only when recipe parameters change significantly. By monitoring recipe characteristics and triggering relearning only when necessary, the system maintains high detection accuracy for specific recipes while avoiding continuous full retraining, thus reducing overall computing resource consumption
Solution Approach 2:
The system performs preliminary clustering of training data by recipe characteristics during the learning phase. This preliminary organization allows the model to quickly adapt to new recipes by referencing pre-clustered data patterns, reducing the time required for relearning when recipes change without sacrificing detection accuracy
Data Source
AI summary
An anomaly detection method capable of minimizing the consumption of computing resources and time is provided. The anomaly detection method includes: learning a first classifier, which includes an encoder and a decoder, using a plurality of training data, which are classified into a plurality of first subsets; extracting features from the plurality of training data by computing the plurality of training data with the encoder of the learned first classifier; reconstructing the plurality of training data into a plurality of second subsets by clustering the plurality of training data based on the extracted features; learning a plurality of second classifiers, which correspond to the plurality of second subsets, using the second subsets; and detecting any abnormality in input data using the plurality of second classifiers.