Reinforcement Learning Dispatching for High-Yield Tool Allocation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional substrate processing facilities face a difficult tradeoff between on-time delivery and maximizing processing on higher-yield tools, leading to inefficiencies and increased costs due to the need to wait for high-yield tools to become available.
Innovation Solution
A reinforcement learning agent is trained using state and reward data to optimize lot processing on higher-yield tools while meeting threshold production values, automatically managing the tradeoff between on-time delivery and tool utilization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If the facility waits for higher-yield tools to become available to maximize lot processing, then manufacturing precision is improved, but loss of time increases
Solution Approach 1:
The patent implements dynamic dispatching rules that adapt in real-time based on tool availability, workload, and yield characteristics. The system continuously adjusts lot routing decisions rather than using static rules, allowing it to dynamically balance between waiting for high-yield tools and meeting delivery deadlines. This dynamic adaptation resolves the contradiction by making the system flexible enough to optimize for yield when tools are available while minimizing waiting time when they are not.
Solution Approach 2:
The system employs feedback mechanisms where actual tool performance, yield data, and delivery metrics are continuously monitored and fed back into the dispatching decision-making process. This feedback loop allows the system to learn from past decisions and adjust future routing choices to better balance yield optimization with time constraints, resolving the contradiction through data-driven adaptive control.
2Manufacturing precision
If the facility prioritizes processing on higher-yield tools, then manufacturing precision is improved, but productivity decreases due to reduced on-time delivery
Solution Approach 1:
The dispatching system dynamically adjusts its priorities based on real-time conditions, switching between yield-optimization mode and delivery-optimization mode as needed. When high-yield tools are available and workload permits, the system prioritizes routing lots to these tools. When delivery deadlines are approaching or tool availability is constrained, it automatically adjusts to prioritize on-time delivery, thus resolving the contradiction between manufacturing precision and productivity.
Solution Approach 2:
The system changes operational parameters such as dispatching priorities, routing preferences, and tool selection criteria based on current facility state. By adjusting these parameters dynamically according to yield targets, delivery deadlines, and tool utilization metrics, the system can shift its behavior to optimize for either yield or productivity as conditions require, resolving the contradiction between these two objectives.
3Manufacturing precision
If conventional dispatching rules are used to manage tool allocation, then device complexity is reduced, but manufacturing precision deteriorates due to inability to optimize yield
Solution Approach 1:
The patent implements a universal dispatching framework that can handle multiple objectives (yield optimization, on-time delivery, tool utilization) and various tool types within a single integrated system. This multi-functional approach consolidates what would otherwise require multiple separate dispatching systems into one unified platform, managing complexity while enabling sophisticated yield optimization across the entire facility.
Solution Approach 2:
The system uses configurable parameters and adjustable weights for different optimization objectives, allowing the dispatching logic to be tuned without changing the underlying system architecture. This parameter-based control enables sophisticated yield optimization while maintaining manageable complexity through a flexible, configurable framework rather than hard-coded complex rules.
Data Source
AI summary
A method includes identifying current state data associated with a substrate processing facility including one or more higher-yield tools and one or more lower-yield tools that have a lower yield than the one or more higher-yield tools. The method further includes providing the current state data as input to a trained reinforcement learning agent. The method further includes receiving, from the trained reinforcement learning agent, output associated with parameters. The method further includes causing, based on the parameters, maximizing of lot processing on the one or more higher-yield tools while meeting one or more threshold production values.