 |
In recent years, due to the shift towards customized production models, rapid product changes, and market fluctuations, manufacturing systems have faced immense competitive pressure. Traditional centralized system models have become less suitable for next-generation manufacturing systems as they fail to efficiently handle high-mix production, complex manufacturing processes, and frequent system reconfigurations.
Our laboratory's research focuses on manufacturing systems built upon distributed components to meet the demands of modern manufacturing systems. These demands include modularity, reconfigurability, flexibility, scalability, object-oriented modeling and planning, decentralization, fault tolerance, load balancing, network penetration, and integration of new and legacy systems. We have proposed the theory of Distributed Colored Timed Petri Net (DCTPN). This theory is an extension of the traditional Petri Net (PN) theory, allowing simultaneous consideration of system timing and attribute issues. Additionally, this paper defines, transforms, and establishes theorems for Distributed Colored Timed Petri Nets. Since DCTPN is constructed in a decentralized and modular manner, it can achieve various designed combination architectures and is applicable to frequently reconfigured systems. Notably, the newly proposed Component Object Model Server Place (COM-Server Place) can be used to simulate manufacturing execution systems. As a result, DCTPN not only enables horizontal integration of production processes but also vertically integrates manufacturing execution systems.
Our laboratory has also developed a DCTPN inference engine, a distributed statistical process control system, and a distributed order management system to demonstrate this manufacturing system architecture. All developed applications utilize DTMS to achieve distributed penetration, fault tolerance, and load balancing in a distributed environment. Finally, we use DCTPN to model a semiconductor factory section and a reconfigurable cluster tool system. These models can be tested using different scheduling methods.
To enhance production capacity and market competitiveness, IC foundries must manufacture and utilize larger wafers, an inevitable development trend. Therefore, the next-generation 12-inch automated IC foundry will become the mainstream of future semiconductor manufacturing. However, as all current 12-inch foundries are still in the design and preparation stages, with no existing factories for reference, a simulation environment is essential to model possible system behaviors and verify design feasibility.
Generally, in a manufacturing system, the time spent transporting and waiting for workpieces is significantly longer than the actual processing time on machines. Therefore, the performance of an automated material handling system (AMHS) critically affects the overall factory efficiency. In recent years, wafer foundries have faced significant challenges. Most semiconductor manufacturers plan to increase wafer size to boost production capacity. However, the increase in size makes manual handling difficult, making automated material handling systems a crucial component in 300mm wafer fabs.
Our laboratory utilizes DCTPN to model the entire AMHS. Within this AMHS model, we implement "Push Vehicle," "Send to Stocker," "Zone Control," and related dispatching rules to improve system performance. The completed AMHS model is integrated with the Process Flow Model of wafer fabrication. Additionally, the concept of virtual machine clusters is developed and applied to address tool coupling issues in simulations. The integrated model serves as a complete virtual semiconductor factory, allowing for more precise evaluation of key performance indicators such as Work In Progress (WIP), Vehicle Utilization, and Product Cycle Time.
Our laboratory also researches the architecture and development of remote diagnosis and maintenance for semiconductor cluster tools. Statistical process control and batch control modules are used to detect and eliminate process variations. The diagnosis module applies neural networks and case-based reasoning for machine prediction and fault diagnosis, while the maintenance module provides repair time estimation and strategy selection for machine maintenance. All critical information can be transmitted to remote users via the web and a message notification platform (GMPP) to monitor the real-time status of machines.
Our laboratory has also developed a two-tier process time prediction system based on the Decomposed Queueing Model. This system consists of two main modules: the System Analyzer Module and the Performance Measures Predictor Module. The System Analyzer Module employs Empirical Distribution Analysis to analyze wafer arrival patterns and service time distributions for different machine groups, as well as the Overall Equipment Effectiveness (OEE) of individual machines. The Performance Measures Predictor Module then selects appropriate queuing models based on the analysis results to compute key performance indicators, such as Cycle Time, Queue Length, and Move Rate, with cycle time estimation being particularly crucial. This research adopts a decomposed queuing model, where each machine group is modeled independently, making the approach adaptable to various semiconductor factory configurations without being constrained by the diversity of machines, machine groups, and product types. This method also facilitates the integration of priority queue models. The system implementation follows a two-tier modular architecture, providing advantages in maintainability and scalability.
|
