Case Study: APF Fusion Integrates Dispatch with Simulation
Rapidly test dispatching rules without impacting production
Factory-wide continuous improvement programs often include efforts to improve dispatching rules for meeting cycle-time commitments and increase throughput for bottleneck tools.
Factory simulation software is used in today’s semiconductor environments to manage the risks of changing dispatching rules. Current simulation models require manual coding of dispatching rules into C++ extensions—an expensive, error-prone process because it’s difficult to recreate a complex and dynamic production environment with complete fidelity. The alternative—analyzing and validating changes to dispatching rules and policies directly on the factory floor—is disruptive, risky and costly.
APF Fusion, the latest addition to Applied Materials’ Advanced Productivity Family (APF) software suite, is designed to solve this problem. A dispatching and capacity-analysis solution, APF Fusion provides a bridge between Applied’s AutoSched AP simulation and APF real-time dispatching (RTD) software. It enhances users’ ability to perform the “what if” simulations needed for more realistic dispatching/scheduling policy changes and short-term capacity planning.
APF Fusion allows users to port existing dispatching rules from production APF RTD into simulation automatically, enabling the same dispatching rules to be used both on the factory floor and in simulations. This eliminates the need to create and maintain duplicate sets of rules for test and production environments, and dramatically reduces simulation times (see figure 1).
“APF Fusion makes it possible to quantify the effects of proposed rule changes before they are put into production,” said David Hanny, marketing manager for automation products at Applied. “When companies have one set of rules on their factory floor and a different set of rules in their planning and simulation systems, it becomes difficult to predict, for example, how many wafers to start or how much more equipment to buy.”
Figure 1. A single set of production and simulation dispatching rules enables faster testing and validation cycles.
SEAGATE TECHNOLOGY REDUCES TIME, COSTS
Seagate Technology, a leader in data storage technology, is an early APF Fusion adopter, using it in their 200mm fabs in Bloomington, Minnesota, and Londonderry, Northern Ireland, which produce disk drive recording heads. Both sites have online factory simulation and RTD capabilities. The company’s dispatch team wanted to improve its ability to examine trade-offs and then validate dispatching and scheduling policies.
Brian Gowling, senior manager of Industrial Engineering for Seagate’s RHO-Wafer Manufacturing operation, said the dispatch team sought to develop rules without adversely affecting production schedules and R&D cycle times. Working in an offline model, the team looked at how new rules might affect product cycle times, equipment capacity, utilization and throughput. They also sought to record factory events, so they could develop a more efficient response to unforeseen events such as down tools, work-in-process (WIP) bubbles, and changes in demand.
Prior to installing APF Fusion, Seagate ran automated online factory simulations that updated actual factory conditions every 30 minutes and provided forecasts ranging from one shift ahead to one week ahead. The main applications for these simulations were to define the goals for factory shifts, prioritize bottleneck equipment, schedule preventive maintenance (PM), prioritize process “No Path” and “Limited Path” efforts, and set weekly factory velocity goals.
However, the dispatch rules in the simulation model were simplistic compared to those in the factory dispatcher (see figure 2), resulting in significant simulation inaccuracies (see figure 3).
Figure 2. Differences between the detailed dispatch rules in the factory dispatcher at a Seagate Technology 200mm fab (left column) versus the more simplistic rules used in the fab’s simulation model before the installation of APF Fusion software.
Figure 3. Before APF Fusion software was installed at a Seagate Technology 200mm fab, more than 40% of the inaccuracies in lot-by-lot dispatching simulations were caused by major differences in wafer queue times. These inaccuracies occurred, in part, because specific dispatch rules could not be adequately modeled.
“Because APF Fusion enables us to maintain the same dispatching rules for both production and simulation model environments, in one case we implemented and adjusted a batching parameter from the dispatch rules from within the online simulation,” Gowling said. “Previously, we didn’t have a method of modeling this rule to view the impact.” (See figure 4.)
With APF Fusion, Seagate was able to integrate an RTD rule into simulation, validating the effects of five different dispatching parameters on key performance indicators (KPIs) at both its manufacturing sites in just one week instead of the more than six months it would have taken otherwise. That modeling ability directly translated into about $100,000 in labor savings.
Figure 4. The graph shows the impacts of implementing and adjusting a batching parameter from within the online simulation provided by APF Fusion. The data comes from three simulation model runs where a WIP bubble event was recorded at a constrained tool. Seagate previously had no way to model this dispatch rule to view its impact.
BETTER ALLOCATION OF METROLOGY RESOURCES
APF Fusion has also helped Seagate better allocate its metrology resources. Out of eight metrology tools at a fab, only two can perform specific functions Seagate calls “CUT” and “MEAS; the remaining six tools only perform “MEAS.”
Under normal operating conditions, the CUT function is primarily used for engineering and R&D WIP. Specific engineering and R&D teams are given a daily allocation of CUT metrology capacity. Some teams are allowed to run one or two wafers per day; others can run as many as 10 per day (see figure 5).
When WIP is balanced, the six dedicated MEAS tools can keep up with all the MEAS-related WIP. In such instances, the R&D teams are allowed to exceed their allocation. However, if a WIP bubble begins to grow, either because of too few available tools or because of a WIP influx, CUT-related WIP is deprioritized and the two “dual-capability” tools process MEAS-related WIP instead.
The factories’ dispatching rules control the balance of the R&D and production lots run on these metrology tools. Seagate engineers can adjust that balance dynamically to smooth out the WIP bubbles.
Figure 5. At Seagate, factory dispatching rules control the balance of the R&D production lots run on eight metrology tools. If a WIP bubble occurs, CUT-related WIP is deprioritized, and the two metrology tools performing this function will process MEAS-related WIP instead.
While some manufacturers rely on spreadsheets for capacity planning, Madhav Kidambi, Applied APF Fusion product manager, said discrete event simulation is far more accurate, and factory-wide simulations enabled by Applied’s APF suite, which includes Fusion, will enable more accurate “what-if“ analyses.
“Engineers will be able to see the impact of dispatching rules not only in their specific area of interest, but in other areas of the factory as well. APF Fusion makes it possible to see the impact of changes everywhere else in the factory and to view overall factory performance.”
For example, many companies pay particular attention to the dispatching/scheduling algorithms in the photolithography area because those tools consume a large part of the budget. However, Kidambi said companies may find that over time lithography is not the true constraint in the fab.
“It could be etch, it could be some other area. If you just do a local optimization in the litho area, it could very well lead to big WIP bubbles in other areas. Then, at the end of the day, you’ll find the overall improvement is not there, and all you have done is created another bottleneck.”
APF Fusion best fits into a simulation environment used for tactical-level planning; that is, relatively short-term capacity planning of one week to three months.
“For longer-term simulation, you are better off running a steady-state simulation analysis,” Kidambi said. “But Fusion can support simulations which indicate where there are bottleneck tools that are keeping the fab from meeting customer commitments in short-term scenarios.”
To get the most benefit from APF Fusion, the latest versions of AutoSched AP and APF RTD and Reporter software are required.
THE HUMAN FACTOR
According to Hanny, one challenge facing the industry is the difficulty of coordinating the efforts of engineers working separately in dispatching, on the shop floor, and in planning.
“In many cases their work is siloed. APF Fusion will both enable and require engineers to work more collaboratively, so that faster, better operational decision-making is achieved.”
Kidambi said several semiconductor companies have expressed a desire to integrate their homegrown scheduling/planning tools using APF Fusion.
“We have heard from some companies that they want this kind of capability so they can use our SmartSched [short interval real-time scheduling] software, and also integrate with homegrown or commercial scheduling products. That would give them a great capability to integrate external scheduling algorithms. At this point, however, these are just internal development projects.”
For additional information, contact firstname.lastname@example.org.