Semiconductor Chamber State Prediction Using Cross-Correlated Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies have not effectively addressed the challenge of accurately predicting the performance of semiconductor manufacturing devices by considering the uniformity of features between acceptable and unacceptable chamber states, which leads to inefficiencies in existing technologies.
Innovation Solution
A state prediction device, including a data acquisition unit, a feature management unit, a correlation calculation unit, a ranking unit, and a state prediction unit, and a correlation calculation unit are used to predict the performance of semiconductor manufacturing devices by considering the uniformity of features between acceptable and unacceptable chamber states.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If more features are extracted from time-series sensor data for state prediction, then prediction accuracy is improved, but computational resources and processing time increase
Solution Approach 1:
The patent extracts only the most relevant features from the complete set of sensor data features. By identifying and removing redundant or less important features, the system maintains prediction accuracy while reducing the computational burden associated with processing all features equally.
Solution Approach 2:
The patent applies different treatment to different features based on their individual importance and characteristics. Rather than uniformly processing all features, the system identifies specific features that contribute most to prediction accuracy and focuses computational resources on those, while simplifying or eliminating processing of less critical features.
2Reliability
If more features are analyzed for state prediction, then prediction reliability is improved, but training data requirements and processing time increase
Solution Approach 1:
The patent extracts and retains only the essential features that contribute most to reliable predictions. By removing redundant features from the analysis set, the system reduces the time required to process and analyze features while maintaining the reliability needed for accurate state prediction.
Solution Approach 2:
The patent applies partial action by analyzing only a subset of available features rather than all features. This selective approach provides sufficient information for reliable predictions without the excessive processing time that would result from analyzing every available feature.
3Measurement precision
If features are selected based on deviation from normal state only, then abnormal state detection is achieved, but uniformity between acceptable and unacceptable states is not considered
Solution Approach 1:
The patent employs a feature selection criterion that serves multiple functions simultaneously: it identifies features that deviate from normal states (abnormal detection) and also evaluates the uniformity of these features across both acceptable and unacceptable chamber states. This multi-functional approach ensures comprehensive state prediction capability.
Solution Approach 2:
The patent incorporates feedback by considering the uniformity of features across different chamber states. The feature selection process uses information about how features behave across the full range of states (both acceptable and unacceptable) to refine the selection, ensuring that chosen features provide consistent and reliable prediction signals throughout.
Data Source
AI summary
Aspects relate to generating a highly-accurate state prediction result for a semiconductor manufacturing device. A state prediction device for a semiconductor manufacturing device includes a data acquisition unit for acquiring a first set of operation data for a first processing chamber and a second set of operation data for a second processing chamber; a feature management unit for generating first and second feature maps; a correlation calculation unit for calculating a normalized cross-correlation result that indicates a uniformity level of a target feature between the first and second feature maps; a ranking unit for ranking target features based on the normalized cross-correlation result and selecting a subset of target features that achieve a ranking threshold; and a state prediction unit for generating a state prediction result that characterizes a performance difference of the second processing chamber with respect to the first processing chamber based on the subset of target features.