Soft simulation crucial for new automated fab decisions
06/01/1998
Case Study Soft simulation crucial for new automated fab decisions
Theron D. Colvin, Frederick P. Lawrence, PRI Automation Inc., Mesa, Arizona
Gerald T. Mackulak, Arizona State University, Tempe, Arizona
Software-driven simulation is a valuable tool for those evaluating and choosing automated material handling systems (AMHS) for new wafer fabrication facilities. While this approach is just a simulation, it provides more valuable data on possible real performance than just making decisions from comparative lists of specifications. With AMHS adding $75-100 million in hardware, software, and infrastructure to the cost of a new fab, but being essential for emerging 300-mm fabs, use of simulation is crucially important to help sort through management, engineering, and production questions. Simulation provides the ability to compare fab automation designs through detailed analyses of system component layouts, system performance, capacity constraints, wafer run rates, operational requirements, downtime parameters, automation needs, and the integration of all these elements.
NOTE: This is a hypothetical case study - the people and companies are not real, but the scenario is typical for new wafer fabrication facilities today.
Bill Maxwell recognized both a challenge and magnificent opportunity. As the wafer fab engineer in charge of AMHS, Bill had begun work to select a system for his company`s new fab (Fig. 1). He knew he could have real impact on the efficiency and long-term productivity of the fab.
The problem lay in choosing the right AMHS. When Hiroshi, Bill`s manager, gave him the assignment, Bill knew right away that the choice would be between two possible suppliers, X Corp., the current supplier, and Z Corp. Bill had already read the programming study completed by the architectural and engineering design firm. This study suggested that Z Corp. might be able to deliver better flexibility and productivity with its material handling equipment. At Hiroshi`s request, Bill had exploratory meetings with engineers and sales people from both companies. But Bill felt he knew practically nothing about either company`s equipment or system capabilities.
Thinking through the technical points that he had stressed during his exploratory meetings, Bill believed the automation should give the new fab facility better delivery times and more flexibility under changing factory conditions, compared to what his company was currently receiving. The long-term cost of ownership would apparently be lower with X Corp.`s system because they had support services in place. But he had no hard information from Z Corp. to compare these claims.
 |
Figure 1. The proposed new wafer fab.
Bill contacted a recognized leader in the field of system performance evaluation, Dr. Gary Compton, from the local university`s industrial engineering department. Gary was familiar with fab logistics systems design and knew how to determine the capabilities of these complex systems. Gary suggested a set of attributes to characterize competing systems performance:
work in process (WIP) capacity of the proposed systems,
required capacity measured against floor space,
simulation of system average and maximum transport times,
simulation of system average and maximum delivery times,
sensitivity analysis of system throughput potential to variability,
system capacity to compensate for shift start and start variability,
robot and vehicle utilization, and
maximum system design component thresholds (i.e., crossings, accumulations, and limitation values from simulation models).
In addition, Bill had lists from each company of supplier-provided data on system attributes:
system and components cleanliness,
system and components reliability,
flexibility to increasing level of fab change,
equipment shock and vibration control,
speed and ease of installation,
ease of maintenance,
control systems and ease-of-programming change capabilities,
ergonomics associated with load size and cumulative trauma disorders, and
density and change of capacity in transport and storage.
Gary cautioned Bill to make comparisons based on projected systems performance, not individual components. He advised that similar system characteristics can be organized into competing scenarios and compared quantitatively. Bill would then be armed with data that would facilitate an effective decision.
Deciding to bring all of his information together, Bill spent one Friday evening going over his own notes, sales literature, spec sheets, AMHS layouts, and everything else he had from the two suppliers. He made lists of critical parameters for the new fab`s AMHS, although its design was still undergoing changes. He made comparative lists of system features from both X Corp. and Z Corp., even though information about system features from the two companies was not always strictly comparable. By midnight, he realized that he didn`t have enough of the right information to make solid judgments. He was getting nowhere.
Bill put all his information aside and sat back to look at the whole problem. He wasn`t being asked to make a theoretical decision about the new fab, which would come into existence in a year or two. Either X Corp. or Z Corp. could probably supply an AMHS that would work, but that wasn`t enough. His real job, he realized, was to visualize in advance the operation of the new fab, and to visualize the actual operation of the two competing AMH systems. Most of the information allowed him to form impressions, but not to visualize the way the AMHS would actually work. He needed to see each system, including advantages and disadvantages, before he could make the decision that would affect the real operation of the new fab for a decade to come.
Z Corp. had included an on-disk simulation output report and animation of its AMHS running in the most recent version of the new fab`s layout. Bill began working with the simulation data. He calculated the cost of ownership and weighted averages of attributes and features. He learned that the reduced delivery time shown in Z Corp.`s simulation had the potential to slice three days from his company`s current 60-day average cycle time.
He saw that he could reconfigure the AMHS design at any time in the future. He could double the throughput or change the "from-to" routing matrix without major system downtime. The simulation data showed that he had headroom in the design for peak loads and for loads at the beginning and end of shifts, and that the system would perform in these time periods without delays. The system in his company`s current fab suffered from delays up to an hour long. Suddenly, Bill began to understand why these delays occurred and how they could be avoided.
By 3:00 am, Bill was making progress. He was beginning to see how the Z Corp. system might work in the new fab, but he would have to see a parallel simulation from X Corp. Maybe X Corp.`s simulation would also show a 3-day reduction in cycle time; it might even show more. Certainly it would differ from the Z Corp. simulation.
His recommendation of either company would have to survive the very close scrutiny that Hiroshi would give it, and Bill knew that Hiroshi would accept nothing without very solid evidence and documentation. Bill was on the right track. The pile of literature on the table in front of him gave impressions and scattered facts, but not the clear visualization he needed to make an intelligent and successful decision. Simulations, he recognized, were his best tools.
The analysis
A week later, Bill had the simulation from X Corp. He worked with it in the same ways and for the same amount of time as the Z Corp. simulation. Somehow, the X Corp. model didn`t feel as real as the Z Corp. model, but Bill wasn`t sure why. Certainly the lack of animation made it more difficult to visualize the system, but there was something else.
Bill called Gary Compton again. Gary asked Bill some questions that led him to a simulation modeling comparison. Gary suggested that maybe it was some of the model`s "simplifying assumptions" that reduced its capability to show system performance parameters, or maybe it was the lack of back-up data to indicate attributes that had been analyzed.
Bill was still apprehensive, but it was time for his initial report to Hiroshi. Before scheduling a meeting, he listed the key items he knew Hiroshi would probe closely, and organized this information into a database:
the simulation,
equipment downtime,
cleanliness, and
operational data.
When they met, Bill and Hiroshi went through the database point by point. As Bill had anticipated, his manager understood that Bill had done very thorough research on the systems offered by both companies. Not surprisingly, a number of issues were identified on which no agreement could yet be made. Hiroshi wasn`t endorsing either X Corp. or Z Corp., and neither was Bill. If he had had to make a final decision at that point, though, Bill would have chosen Z Corp. Their simulation tended to conform more closely to the current plans for the fab.
 |
Figure 2. EAI FactoryFlow intensity diagram for a) Z Corp. and b) X Corp.
With Hiroshi, he made visits to both companies to discuss the outstanding issues with respective engineers. In a meeting after both trips, Bill and Hiroshi found they disagreed on details, but they agreed they needed more information.
Bill turned his attention to delivery time and simulation statistics. While X Corp.`s simulation was based on an eight-hour simulated time, Z Corp.`s simulation covered 10 days. And the Z Corp. simulation integrated interbay vehicles or "cars" and off-line transfers, and made extensive use of industrial engineering techniques. Bill and Hiroshi put together a list of questions and issues that needed further clarification from Z Corp., and Bill called its design engineer.
There were several key questions: Since Z Corp.`s simulation included turntables and horizontal transfers, what was their precise purpose? The engineer had previously mentioned several other layout-related flow issues. Why were those issues important?
The engineer explained how Z Corp. determined flow patterns using EAI`s FactoryFlow software that generates color-coded flow intensity diagrams for AMHS track systems (Fig. 2a & 2b). Since Z Corp.`s "cars" could be flexibly routed through turntables in the track system, one could be sure to place the track in the correct places, just as one would do with a freeway, piping, or data system. And, as with a freeway system, undersizing at critical areas could cause problems that must later be solved by increasing system capacity. The engineer explained that Z Corp. used the new software, giving the company an advantage in determining factory flow prior to simulation. That, in turn, reduced the number of simulation iterations and reruns.
Z Corp. also needed to understand the goals of the proposed AMHS. It needed to know what moves it would handle, and what the peak number of starts and nonuniformity factors would be. Then the company could arrange its equipment based on the potential utilization of stocker robots.
After that, the Z Corp. engineer continued, they used software to identify high and low intensity flow areas and to determine locations for tracks, turntables, horizontal transfers, and vertical or elevation switches. Once collected, they imported this data from CAD right into the simulation software package. Then they performed short runs to determine car quantities, and long runs to determine system capabilities.
Bill called Gary Compton once again, to ask about the long, ten-day runs in the Z Corp. simulation, compared to the eight-hour runs in the X Corp. simulation. Ten days, Gary explained, was about the bare minimum required to characterize any complex system. Gary was surprised that any company would run less. He felt that with only an eight-hour run, X Corp. couldn`t accurately determine the 3s potential for all delivery and transport times, and other simulation outputs.
Z Corp. followed up by sending Bill an animation that showed its system operating in the latest design of the new fab. The Z Corp. engineers explained that they anticipated further changes in Bill`s AMHS design, but it would only take Z Corp. about a week to respond to each fab layout change with a new simulation and animation.
Bill also obtained, although not quite as rapidly, a simulation from X Corp. How X Corp. would respond to future design changes, though, wasn`t exactly clear, and Bill remembered that the company had been slow to respond to previous changes. He was concerned that X Corp.`s modeling effort would continue to lag his design process, leaving him to question the ability of the company`s AMHS to adapt to the latest layout changes.
Bill now went to work with simulation data from both companies. He analyzed both models by varying inputs and comparing results with the simulations. Now he could see real differences between the two systems. When he increased throughput, for example, the differences became very apparent.
Building a consensus
Gary Compton, now a consultant to Hiroshi`s group, helped Bill make further progress with the analysis. He and Bill went over the comparison of quantitative information and, based on the simulation output, concluded that the Z Corp. system had better delivery time and transport performance. When lot delivery times were compared for the optimal number of cars, Z Corp. had higher percentages of lower times, and lower percentages of higher times (Fig. 3), and hence lower average and maximum times (Table 1), than X Corp. A similar result was obtained for lot transport times. This held true for numbers of cars above and below the optimal (Fig. 4), which is only an approximation to expected AMHS behavior.
Bill, as he explained to Gary, was having a problem ordering project attributes so he could make a final decision; he just didn`t know how to quantitatively compare qualitative features. It seemed that no matter how he reviewed the material, there was always room for speculation if he picked one or the other competitor. Gary helped Bill develop a matrix that identified the key attributes in descending order and calculated weighted-attribute scores for the two competitors (Table 2). When the qualitative features were viewed, the choice was clear.
 |
Figure 3. Lot delivery time comparison.
 |
Figure 4. Delivery time versus transport time comparison.
The day after Bill completed his weighted equipment selection matrix, he was ready to proceed with his recommendation. He noted, for example, that if he selected Z Corp., operational overhead would be doubled, but the overhead would be offset by Z Corp.`s consigned parts and 7-day, 24-hr service coverage. He ran through the other categories, including delivery time, flexibility to change, power consumption, and support after sale. Other significant differences were apparent, and, in the end, the overall advantage was with Z Corp.
The next day he gave the raw matrix to Hiroshi, who did his own independent calculations. The two then compared their results. Their elements were different, but both gave the advantage to Z Corp.
In the ensuing days, Bill and Hiroshi made several management presentations, summarizing the results. Each time they were asked to validate, verify, and confirm all of their conclusions. As weeks went on, the fab design changed several times. Each time, Z Corp. responded to the change within a week. X Corp. actually failed to respond to more than half of the changes.
Conclusion
In 1513, Niccolo Machiavelli stated: "It must be remembered that there is nothing more difficult to plan, more uncertain of success, nor more dangerous to manage than the creation of a new order of things. For the initiator has the enmity of all those who would profit by the preservation of the old institutions and merely lukewarm defenders in those who would gain by the new ones." This is especially true of a semiconductor fab today, and when a new system must be introduced.
What tools can be used to assist in making these decisions? The answer: software-driven simulation. Simulation analyzes a system before it is built. It provides information on system capacities and the means to measure the system`s sensitivity to changes and to abnormal operation. Simulation-driven animation provides an additional benefit in the ability to see the system, to pan or fly through its design, and actually visualize the operation.
Keep in mind, however, that simulation is not an expert system. Rather, it is a tool used by competent system designers and modelers to develop and compare design alternatives. The skills and experience of the designers and modelers are of paramount importance. A bad design can be masked by a bad modeling effort. Managers must have the ability to view the results and ask questions that validate models - does the operation of a model make logical sense? Can designers and modelers reproduce the results that are experienced today in existing fabs?
Also keep in mind that software-driven simulation is not reality. Simulation is an art wherein the modeler must identify the key elements and use his or her experience to determine factors most likely to make the model resemble reality. This craft requires knowledge of statistics, mathematics, operation research techniques, and often just plain common sense. Making a decision for a new or existing supplier is stressful, but accurate simulation can make the decision-making more rational and less threatening.
Bibliography
F. Robertson, T. Nohara, Global Joint Guidance for 300-mm Semiconductor Factories, I300I and J300, July 1997.
M. Weiss, "Evaluating 300-mm Fab Automation Technology Options and Selection Criteria," MICRO, June 1997, pp. 65-78.
R. Muther, Systematic Layout Planning, Management & Industrial Research Publications, 1973.
G. Cardarelli, et al., "Analyzing Automated Interbay Handling and Storage," Semiconductor International, Sept. 1995, pp. 113-120.
D. Pillai, "Designing Automated Material Handling Systems for Large-scale Wafer Fabrication Automation," SME Technical Paper, MS89-785.
T. Colvin, G. Mackulak, "Designed-in Flexibility," European Semiconductor, Nov. 1996, pp. 33-35.
Theron D. Colvin received his BS degree in M.E. from the U. of Arizona, Tucson. In 1994, he joined PRI, where he directs the design and analysis of AMHS for semiconductor manufacturing fabs. He is the director of automation planning and design for PRI Automation`s Arizona office. He has more than 26 years experience as a manufacturing systems engineer and design engineer. PRI Automation, 1250 S. Clearview #104, Mesa, AZ 85208; ph 602/807-4747, fax 602/807-4441, email [email protected].
Frederick P. Lawrence received his PhD degree in industrial engineering from Arizona State U., Tempe. He joined PRI in 1997, and conducts operational requirement analyses for AMHS in semiconductor manufacturing fabs. He is the manager of operations planning for PRI Automation`s Arizona office. Contact him at email [email protected].
Gerald T. Mackulak received his PhD degree in industrial engineering from Purdue U., Lafayette, IN. He is an associate professor of engineering and the co-director of the Systems Simulation Laboratory in the ASU Department of Industrial and Management Systems Engineering. Arizona State University, P.O. Box 875906, Tempe, AZ 85287, ph 602/965-6904, email [email protected].