Error Factor Estimation Using Feature Contribution Under Data Drift
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data drift caused by changes in recipes, device components, or inspection targets makes it difficult for existing classification models to accurately estimate error factors in semiconductor inspection results, as the trend of past inspection results no longer suits new inspection results.
Innovation Solution
An error factor estimation device that processes inspection data to generate feature quantities, trains models to relate errors to these features, calculates contribution degrees, and acquires error factors based on selected feature quantities or their combinations, enabling effective error factor estimation even with continuous or discontinuous data drift.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a classification model is trained on past inspection results to estimate error factors, then estimation accuracy is improved for historical data, but the model becomes ineffective when data drift occurs due to recipe changes, device component updates, or inspection target changes
Solution Approach 1:
The patent implements dynamic model updating by continuously acquiring new inspection results and retraining the classification model. Instead of using a static model trained on historical data, the system periodically updates the model with new data to adapt to data drift caused by recipe changes, device component updates, or inspection target changes, thereby maintaining estimation accuracy over time
Solution Approach 2:
The system employs feedback mechanisms by acquiring actual error information from inspection results and using this feedback to retrain and improve the classification model. The model learns from actual error patterns in new data, allowing it to adapt to changing conditions while maintaining accurate error factor estimation
2Measurement precision
If extensive re-labeling processes are performed to maintain model accuracy during data drift, then estimation accuracy is preserved, but processing time and operational complexity increase significantly
Solution Approach 1:
The system performs self-service by automatically acquiring new inspection results and retraining the classification model without requiring extensive manual re-labeling processes. The automated workflow reduces human intervention and processing time while maintaining model accuracy through continuous learning from new data
3Measurement precision
If all feature quantities are used for model training to ensure comprehensive error factor estimation, then estimation coverage is improved, but processing complexity and computational resources increase
Solution Approach 1:
The system extracts and uses only the necessary feature quantities from inspection data for model training. By selectively processing relevant features rather than all possible features, the system maintains comprehensive error factor estimation coverage while reducing processing complexity and computational resource requirements
Data Source
AI summary
This error factor estimation device 100 is a device for estimating the error factor of errors that occur, and comprises: a feature-quantity-group-generating unit A2a that processes data including inspection results collected from an inspection device and generates a plurality of feature quantities; a model-generating unit 4 that generates a model A5a for learning the relationship between the plurality of feature quantities generated by the feature-quantity-group-generating unit A2a and errors; a contribution-degree-calculating unit 11 that calculates a contribution degree indicating the degree of contribution to the output of the model A5a for at least one of the plurality of feature quantities used for the model A5a learning; and an error factor acquisition unit 15 that acquires error factors labeled with feature quantities selected on the basis of the usefulness calculated from the contribution degree calculated by the contribution-degree-calculating unit 11.