Article

AutoSimulations, Inc., a Brooks Automation Companynductor manufacturers have highly complex scheduling challenges. National Semiconductor's Arlington, Texas facility faced especially rigorous hurdles. The plant runs high numbers of high-volume products. "It's like a big chess game," says Jacques Mercier, corporate manager, manufacturing decision tools and methodology. "You have to plan hundreds of moves ahead, but you don't have 16 chessmen. You have 500 to 1000 pieces of equipment."

The facility was already using Consilium's (Mountain View, Calif.) WorkStream MES product to track and manage production. The company recently discovered that a scheduling approach developed by <%=company%> (Bountiful, Utah) could leverage the Workstream data by providing real-time dispatching capabilities. However, to generate complex schedule optimizations in real time, National Semiconductor required powerful, reliable, open platforms.

National Semiconductor has implemented the AutoSimulations' scheduling solution on a two-way Sun Enterprise 3000 server. Following an intense, multiple-front cycle time minimization effort, which also included AutoSimulations' scheduling system, the plant has cut cycle times in half, which in turn has lead to improved product yields and better response to customer requests.

The Nature of the Challenge
Scheduling semiconductor fabs is a difficult task. Ideally, the most expensive equipment should be highly utilized. But there are complications. In any manufacturing process, one or two pieces of equipment—called "bottleneck" equipment—dominate plant throughput. Ideally, bottlenecks should be highly utilized in order not to hold up the rest of the line.

Then there is the nonlinear nature of process routings. Semiconductor wafers visit the same pieces of equipment numerous times throughout the manufacturing process, resulting in highly complex, interwoven routings.

Once each of these factors is balanced, planners must decide whether to optimize for high volume or high product mix. At National's Arlington facility, the answer wasn't so simple, because the plant runs high volumes and large numbers of products.

For such a complex problem, the goal was surprisingly simple and straightforward. "We needed to increase equipment utilization and cut our cycle times," said Mercier.

Consulting a Grand Master
AutoSimulations, a leading provider of factory simulation and capacity analysis products for designing and sizing production facilities, was working on extending its solution from planning to scheduling. "The challenge was to provide a solution that enabled production analysis, off-line simulation and real-time dispatching," said Peter Elleby, chief technologist at AutoSimulations. "It all had to be seamlessly integrated with WorkStream."

AutoSimulations solved the problem by intercepting WorkStream database updates in real time and using object-oriented database technology to map the updates into a model stored on the Sun platform. This object database is the central point of communication between AutoSimulations' analysis, simulation and scheduling tools, and more importantly the link with WorkStream.

"We provide the tools to allow production managers to constantly tweak the schedule optimizations because generic results are never adequate. In the semiconductor industry, no two plants run the same processes, and naturally, different products share common production lines or combinations of equipment," said Elleby.

Benefits as Big as Texas
"Simulation had always been an off-line approach, but we needed optimal solutions on-the-fly," noted Mercier. "While the AutoSimulations' approach solved this former limitation, management was nervous about how on-line scheduling would affect plant floor operators. Operators simply need lists of tasks to perform, and in what sequence. They can't afford to be caught up with complex analyses."

Mercier reassured management that there would be no impact on operators, because the new scheduling engine would be embedded under the hood. The new optimized schedules, dispatched in real-time by the AutoSimulations Real-Time Dispatcher, would look just like the old ones. The only difference is that the schedules would be better. "AutoSimulations even made sure the screen was identical to the old system," added Mercier.

The scheduling model involved tens of thousands of variables, including nearly 1,000 pieces of equipment; hundreds of products; and between 50 and 100 product flows, each of which comprise hundreds of steps. To build the models, and determine the necessary level of detail for variables such as cycle time, setup time, and utilization, National conducted validations comparing simulation models to actual conditions over the previous two months. It learned that it would have to recalibrate the model monthly to reflect the constant flow of equipment changes that are made to the lines.

The project began in the summer of 1997, and went live with dispatching in the first quarter of 1998. "The results from real-time dispatching and other cycle time reducing measures have been outstanding," said Mercier. Cycle times dropped by 50 percent. Significantly, the variations in cycle time have also been reduced 50 percent, meaning the facility performs more consistently.

The Sun Infrastructure
Running such complex scheduling models in real time requires powerful, robust platforms.

National Semiconductor runs AutoSimulations' models on two-way Sun Enterprise 3000 servers, using Solaris 2.6. "Sun was a very reliable hardware partner," says Mercier. "The AutoSimulations' package was so well-tuned for Sun that we had no implementation issues that were hardware-related."

Simulation time depends on the model size and the length of the simulated period in terms of computing time, but typically took five to fifteen minutes per actual month simulated. Input and output files are handled on PCs using Microsoft Windows NT. "We just update all the data on Windows, export everything to Solaris to run the simulation engine, then the output is sent back to [Microsoft] Windows-based PCs," Mercier notes. "Users get their simulation output in Microsoft Excel spreadsheets and graphs."

Thanks to the combination of AutoSimulations' modeling tools and the Sun Solaris platform, National Semiconductor can deliver improved dispatching plans to operators quickly and reliably, using the systems that they already know. As a result of an all out attack on cycle time, National Semiconductor's Texas plant decreased production cycle times by 50 percent and improved utilization of expensive manufacturing equipment. Reduced cycle times improved learning cycles, product yields and response to customer requests.

Free Logistics Online Newsletter