ML Wafer Fab Dispatching for HMLV Lot Scheduling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current dispatching systems in wafer fabrication facilities rely on common priority selection or human experience, failing to optimize production efficiency and adaptability, especially in complex production environments like high mix low volume (HMLV) and batch job processes, which hinders overall factory performance and resource efficiency.
Innovation Solution
A machine learning-based intelligent dispatching system that simulates human learning behavior by analyzing history and basic data to predict runtimes and switching times, optimizing schedule results, and automatically dispatching lots, using algorithms like decision trees, random forests, artificial neural networks, and Bayesian networks to improve production efficiency and adaptability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If common priority selection based dispatching rules are used, then the dispatching system is easy to implement, but it cannot take into account the substantial benefits of entire Fab and fails to optimize production efficiency
Solution Approach 1:
The patent replaces traditional mechanical dispatching rules (priority selection, idle tool-based selection) with an intelligent system using machine learning algorithms and neural networks. This substitution enables the system to learn from historical data and optimize dispatching decisions automatically, resolving the contradiction between ease of implementation and production efficiency optimization.
Solution Approach 2:
The dispatching system performs self-learning through machine learning algorithms that automatically analyze historical production data, tool performance, and recipe information. The system improves its dispatching decisions over time without manual intervention, enabling it to capture substantial benefits for the entire Fab while maintaining automated operation.
2Adaptability or versatility
If human experience based dispatching is used, then the system is simple to operate, but it cannot adapt to complex production environments like HMLV and batch job processes
Solution Approach 1:
The patent implements a feedback mechanism where the system continuously monitors actual production results and compares them with predicted outcomes. This feedback loop enables the machine learning algorithms to learn from real-world performance and adapt to complex production environments like HMLV and batch job processes, while the automated nature maintains ease of operation.
Solution Approach 2:
The system performs preliminary analysis of historical data, tool capabilities, and recipe requirements before making dispatching decisions. By pre-processing and learning from historical patterns, the system adapts to complex production environments in advance, eliminating the need for manual adjustment while maintaining operational simplicity.
3Productivity
If traditional dispatching rules are used, then the implementation is straightforward, but cycle time cannot be reduced and production capacity cannot be improved
Solution Approach 1:
The patent implements a dynamic dispatching system that continuously adjusts scheduling decisions based on real-time tool status, recipe requirements, and historical performance data. This dynamic approach enables the system to optimize production capacity and reduce cycle time by making adaptive decisions rather than following static rules, while the automated machine learning processes keep implementation straightforward.
Data Source
AI summary
A machine learning intelligent dispatching system, including a history information module storing various history data, a basic information module storing various basic data, an algorithm module working out predicted runtimes and switching times of recipe groups based on the history data and basic data through machining learning, a robot module working out an optimized schedule result based on the history data and basic data and the predicted times, and a dispatching module dispatching lots according to the optimized schedule result to obtain an actual production result, and the actual production result is fed back to the robot module as a basis for the machine learning.