Process Metric Prediction with Reinforcement-Led Model Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current prediction models in manufacturing processes, such as semiconductor manufacturing, become less accurate over time due to process drift and increasing data volumes, leading to suboptimal performance.
Innovation Solution
A computer-implemented method using a reinforcement learning framework to evaluate and select the most appropriate model configuration for predicting process metrics by analyzing performance indications based on process data, allowing for real-time updates and adaptability to changes in the manufacturing environment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a static prediction model is used initially, then the model can be implemented with simple configuration, but the prediction accuracy deteriorates over time due to process drift and increasing data volumes
Solution Approach 1:
The patent implements a dynamic model selection system that automatically transitions between different model configurations based on real-time process conditions. The system monitors process data characteristics and dynamically selects the most appropriate prediction model from multiple candidate models, allowing the system to adapt to process drift and changing data volumes without manual intervention.
Solution Approach 2:
The system changes the parameters of the prediction system by maintaining multiple model configurations with different characteristics (e.g., different data window sizes, different model complexities) and selecting among them based on current process conditions. This allows the system to optimize prediction accuracy for varying process states while managing computational resources effectively.
2Adaptability or versatility
If multiple model configurations are maintained for different process conditions, then adaptability to process changes improves, but the system complexity increases
Solution Approach 1:
The patent introduces an intermediary model selection layer that sits between the process data and multiple prediction models. This intermediary automatically evaluates current process conditions and selects the most appropriate model configuration, thereby managing the complexity of having multiple models while maintaining high adaptability to process changes.
Solution Approach 2:
The system implements feedback mechanisms where prediction performance is continuously monitored and used to inform model selection decisions. This feedback loop allows the system to automatically adjust which model configuration is used based on actual performance, reducing the need for complex manual configuration management while maintaining high adaptability.
3Productivity
If manual model selection and updates are performed, then the system remains simple to implement, but the productivity and responsiveness of the prediction process decreases
Solution Approach 1:
The patent implements a self-service prediction system where the model selection and updating processes are automatically performed by the system itself based on monitored process conditions. The system autonomously evaluates which model configuration is most appropriate and updates its prediction approach without requiring manual intervention, thereby increasing productivity while maintaining reasonable automation levels.
Solution Approach 2:
The system uses feedback from continuous monitoring of process data and prediction performance to automatically trigger model selection and updates. This feedback-driven automation increases prediction process efficiency by ensuring the right model is used at the right time, while the automation level is managed through intelligent thresholds and decision rules.
Data Source
Figure 1(a)~1(d)
Figure 2
Figure 3
AI summary
A method, the method comprising: obtaining one or more models configured for predicting a process metric of a manufacturing process based on inputting process data; and using a reinforcement learning framework to evaluate said one or more models and/or model configurations of said one more models based on inputting new process data to the one or more models and determining a performance indication of the one or more models and/or model configurations in predicting the process metric based on inputting the new process data.