Semiconductor R2R Control With Dynamic Grouping for Low Runners
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Run-to-run controllers in semiconductor manufacturing struggle with limited feedback for 'low runner' products, leading to suboptimal learning and control due to static grouping, where less frequently produced items receive insufficient data for effective recipe adjustments.
Innovation Solution
Implementing a dynamic grouping method that uses a higher-level scheduler to determine and adjust links between control contexts based on measurement data, allowing for continuous learning and adaptation across all products, including 'low runners', by recalculating a link matrix that weights feedback from frequently and infrequently produced items.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If static grouping is used to organize control contexts, then the R2R controller can manage frequently produced products effectively, but low runner products receive insufficient feedback and learning opportunities
Solution Approach 1:
The patent merges control contexts by introducing a similarity measure that identifies relationships between different product groups. Low runner products are merged with similar volume products based on their parameter proximity, allowing them to share feedback and learning opportunities while maintaining their own control context identity.
Solution Approach 2:
The patent transforms the static grouping into a dynamic structure by calculating similarity measures between control contexts and determining links based on these similarities. The grouping structure can adapt and change as new products are processed and more data becomes available, allowing the system to dynamically reassign low runners to appropriate parent contexts.
2Adaptability or versatility
If separate control contexts are maintained for all products, then each product can have dedicated control parameters, but the learning capacity for low runner products remains limited due to insufficient data
Solution Approach 1:
The patent implements a nested structure where control contexts are organized hierarchically. Low runner products are nested within or linked to parent control contexts that represent volume products. This allows the low runners to inherit and contribute to the learning of their parent contexts while maintaining their own specific control parameters, effectively creating a multi-level control hierarchy.
Solution Approach 2:
The patent makes control contexts universal by allowing them to serve multiple functions simultaneously. A single control context can represent both a specific low runner product and a broader category of similar products. This multi-functionality enables the same control context to provide product-specific control while also accumulating data from multiple similar products to improve learning reliability.
3Productivity
If the R2R controller focuses on volume products, then optimal control is achieved for frequently produced items, but by-products and low runners cannot be processed effectively
Solution Approach 1:
The patent segments the control system into different levels: volume products that maintain their own dedicated control contexts and low runner products that are linked to similar volume products. This segmentation allows the system to provide focused optimization for volume products while simultaneously enabling by-products to benefit from the established control patterns of their counterparts.
Data Source
AI summary
A method for adjusting a process controller in order to manufacture one or a plurality of components in production is disclosed. The method includes (i) obtaining input data characterizing one or a plurality of components, (ii) determining clusters in the input data, (iii) grouping the components into the determined cluster, and (iv) determining a parameterization of a specified process step for the one or plurality of components or subsequent components by way of a R2R controller, depending on the measurement results and the grouping.