No abstract

The dynamics of high-volume discrete-part semiconductor manufacturing supply-chain systems can be described using a combination of Discrete EVent System Specification (DEVS) and model predictive control (MPC) modeling approaches. To formulate the interactions between the discrete process model and its controller, another model called Knowledge Interchange Broker (KIB) is used. A robust and scalable testbed supporting DEVS-based manufacturing process modeling, MPC-based controller design, and the KIB DEVS MPC interaction model is developed. A suite of experiments have been devised and simulated using this testbed. The flexibility of this approach for modeling, simulating, and evaluating stochastic discrete process models under alternative control schemes is detailed. The testbed illustrates the benefits and challenges associated with developing and using realistic manufacturing process models and process control policies. The simulation environment demonstrates the importance of explicitly defining and exposing the interactions between the manufacturing and control subsystems of complex semiconductor supply-chain systems.Index Terms-Discrete-event system specification, hybrid simulation testbed, Knowledge Interchange Broker (KIB), model composability, model predictive control, optimization, semiconductor manufacturing, supply-chain management.

Large, fine-grain data collected from an actual semiconductor supply-demand system can help automated generation of its integrated simulation and optimization models. We describe how instances of Parallel DEVS and Linear Programming (LP) models can be semi-automatically generated from industry-scale relational databases. Despite requiring the atomic simulation models and the objective functions/constraints in the LP model to be available, it is advantageous to generate system-wide supply-demand models from actual data. Since the network changes over time, it is important for the data contained in the LP model to be automatically updated at execution intervals. Furthermore, as changes occur in the models, the interactions in the Knowledge Interchange Broker (KIB) model, which composes simulation and optimization models, are adjusted at run-time. INTRODUCTIONSimulation methods are used for modeling and simulation of supply-demand networks for predicting operations schedules and costs. For example, optimizing large-scale semiconductor supply-chain systems is very complex and the number of configurations of their processes and parameters are many. Therefore, there has been interest in using optimization modules for optimizing the operation of simulation. In this paper, we consider an actual semiconductor manufacturing supply-demand system (aka supply-chain system). Efficiency in the management of such systems can reduce costs by many millions of dollars each year (Wu, Erkoc and Karabuk 2005, Kempf 2004). Semiconductor supply-demand networks are a collection of suppliers, inventories, processes, transportations, and customers (Kempf 2006). Raw material is processed in sequential and parallel stages to produce many different kinds of products to customers. This enterprise can be divided to manufacturing processes and decision planning parts. The key variables of interest for discrete processes and logistics include stochastic processing times in building products in multiple stages and in different geographies. Another important variable is inventory holdings across manufacturing plants and logistics stages. Both the scale and complexity in manufacturing requires complex decision making, often supported with linear programming where reduced cost and just-in-time (raw material, semi-finished goods, packaging, etc.) delivery across the supply-demand is highly sought after. To simulate the operation of such dynamic enterprises, we can use discrete event modeling specifications such as Discrete Event System Specification (DEVS) and optimization modeling such as Linear Programming (LP) methods. Two important factors in creating realistic models of such enterprise systems are scale and model integration. Manufacturing has many tens of processes and inventories with alternative source to target routes and shipping modes. Similarly, logistics has many inventories and hubs for transportation and delivery to customers in different geographies. Simplifying generation of these simulation models is attractive. Considering the optim...

We present a simulation model constructed in collaboration with Intel Corporation to measure and gauge the interaction of non-linear supply chain phenomena (such as waste, uncertainty, congestion, bullwhip, and vulnerability). A representative model that mimics part of Intel's supply chain from fabrication to delivery is modeled using discrete-event simulation in ARENA. A "phenomena evaluation" framework is proposed to link model inputs and supply chain phenomena in order to evaluate supply chain configurations. Using a sample supply chain decision (safety stock level determination) we follow the "phenomena evaluation" framework to illustrate a final recommendation. Results show that our supply chain phenomena evaluation approach helps better illustrate some trade-offs than an evaluation approach based only on the traditional metrics (cost, service, assets etc.).

No abstract