It's always better to make mistakes on a model than to screw up in production. That's why the solar industry's first clean room started life as a simulation.
John R. Vacca
If at first you don't succeed, you probably didn't run enough simulations. Take Siemens Solar Industries. It succeeded. How? Before committing its photocell production to a new-to-the-industry clean room, SSI took advantage of simulation software and a nearby university to fine-tune the processes and work-flow procedures it would use. The results were dramatic. The Camarillo, California, company saved over $10,000 in consulting fees. And because it could clear up production bottlenecks in advance, the new clean room is saving the cell-fabrication department roughly $7.5 million annually.
The key was simulation. No longer ju
st computer games for the intellectually curious or exercises in the computer-science curriculum, today's simulation software has come down from the world of FORTRAN subroutines and is appearing on desktop computers.
Now simulation is making its way into the world of business and industry. It was important in an earlier business-process reengineering effort at SSI, and during the past year simulation helped fine-tune the planning and operation of a new production process and fabrication facility.
Making a Clean Start
SSI is the world's largest volume maker of solar electric products. Everything from power supplies for telecommunications systems to utility-scale solar electric plants uses its cells and modules. Your lawn might even be an SSI customer: The company also produces solar panels and solar-powered outdoor lighting. It's part of the worldwide Siemens Solar Group, which has locations in Germany, Japan, Singapore, and the U.S.
SSI decided to introduce clean-roo
m contamination-control technology into its photocell fabrication. That spells money -- to upgrade equipment and to create a new facility.
A
clean room
provides a controlled environment that filters incoming air and water to high standards of purity. It also controls temperature, humidity, and air pressure, but the key element is air filtration. Clean-room personnel wear special coveralls, hair covers, shoe covers, face masks, and gloves. Clean rooms are standard practice for fabricators in the semiconductor business -- you couldn't make a CPU without them -- but the solar industry had never used them before SSI decided to take the plunge.
There was a lot at stake. And there were high hopes. "This new facility would enable Siemens to attain higher levels of quality in the cell-fabrication process," says
Mike Fahner
, the SSI project manager responsible for improving information accuracy and implementing logistics systems to enhance supply-chain
performance. "This was also an opportunity to improve productivity by redesigning the facility layout and material flows to increase throughput, decrease the work-in-process levels, and reduce cycle times," he adds.
Efficiency Is the Name of the Game
World-class manufacturing organizations strive to minimize the level of what is called
work-in-process
-- that is, the inventory of unfinished goods on the shop floor that are waiting for the next processing step. Work in process represents a capital asset that's tied up -- money that could be better invested in things such as plant upgrades.
High inventory levels can be costly. If things move slowly enough, there can even be a cost of obsolescence. If the products are technology-driven where changes occur rapidly, then the administrative costs of managing engineering changes for partially finished goods can be cumbersome.
"At SSI, we considered that computer simulation was an ideal tool to help improve the pro
ductivity and quality of the cell-fabrication process," explains Fahner. "Computer simulation provides a virtual laboratory where an engineer can experiment with effects such as adding machines or changing their configuration without the time and expense of making the physical changes."
Computer simulation also allowed SSI to compare numerous alternatives quickly. "We believed that simulation models could predict the effects of proposed equipment investments or modifications with minimal risk," Fahner adds. "We also believed that the primary benefit of the simulation-modeling process was the knowledge and insight we would gain in understanding the interactions within the systems we were designing and studying."
According to Fahner, SSI's primary goals were to model the proposed production process, concentrating on the wafer-diffusion, oxidation, and plasma-etch processes for the new cell-fabrication room. The company also wanted to identify and remove system constraints and evaluate alternative s
cheduling, delivery rules, and material flow with respect to queue levels, throughput, cycle time, machine utilization, and work-in-progress levels.
Partnering with Poly
Dr. Sema Alptekin, professor of industrial and manufacturing engineering at California Polytechnic State University in San Luis Obispo, California (generally known as Cal Poly), wanted to get her students involved in a real industrial problem using simulation techniques. She approached Fahner after a mutual contact suggested she consider SSI.
Fahner was interested in a partnership with Cal Poly because he's always looking for efficient ways to solve problems with limited resources. Siemens keeps a core group of highly skilled and focused individuals who manage projects using external resources on an ad hoc basis. This approach has allowed Siemens greater flexibility by accessing a specialized knowledge base only when needed, and without carrying costly overhead. (see the sidebar "Learn by Doing".)
Th
e project began on September 23, 1994. Alptekin visited SSI to tour the plant and discuss the project's objectives. Fahner visited Cal Poly twice during the 10-week project, defining project objectives for the students, providing data, and answering questions whenever industrial expertise was needed.
Alptekin's students were excited and willing to visit SSI to learn more about the problem. They also felt greater-than-normal pressure, since the projects weren't hypothetical exercises created by an instructor and ultimately destined for the trash can; their work would be evaluated by industry, and their recommendations might actually be implemented within a few months.
Picking the Right Simulation Tool
Alptekin notes that simulation software has been around for a long time. "It was part of my Ph.D. dissertation in 1981 to write code in FORTRAN as a simple simulation tool to solve a simple problem. With advances in computer technology and software, solving the same problem, w
hich took several years in the 1970s, could be accomplished in several days today," she observes.
After numerous meetings and discussions, the Cal Poly students, with Alptekin's help, selected ProModel from ProModel Corp. (Orem, UT) as the software they would use for the SSI computer simulations. "ProModel is just one of the simulation tools we use," says Alptekin. "We picked ProModel because it was the only available simulation product that was developed as a Windows application," she adds. "It offers the advantages of a true Windows-based product." (see the sidebar "What ProModel Can Do".)
The package's relatively gentle learning curve was critical, according to Alptekin. "Since this was a class project, the simulation tool had to be easy enough to learn and apply in a short time," she explains. "Anybody who's using Windows could learn to use ProModel in a short time." The user does not have to learn a simulation language. It's a point-and-click development tool through its library of tools, ico
ns, and constructs. The package's on-line help and on-line training were also important factors.
In addition, ProModel was well suited for this project-workgroup situation because different members of the group could develop submodels and integrate them into a final model. "It allows the user to start with a small model and develop it to a larger model through the use of submodel and model-merge capabilities," Alptekin continues. "ProModel is also well suited for fast what-if analysis because of its flexibility to accommodate changes and its fast running speed."
The Windows underpinning of ProModel offered other advantages. The students found ProModel's output charts and graphs superior to those of other simulation products, and they could easily paste them into other Windows applications for reporting purposes. Finally, ProModel's animation capabilities were a definite plus. "Animation helped SSI managers validate the baseline model and provided them with a better understanding of their proposed
system," says Alptekin.
How the Students Used ProModel
The students used ProModel to develop a baseline that defined the model's scope and objectives based on real data. They also used it to make assumptions to simplify the model, as well as to create a model of the network of resources in the system by defining the machines, their capacities, and the probability-distribution functions of their stochastic service times and interarrival rates of incoming parts. Some of the resources in this model included chemical-etching baths, spin-rinse dryers, wafer-transfer stations, diffusion tubes, oxidation tubes, and plasma-etch ovens.
Also, the students used ProModel to define the routines used for producing wafers. There were two product lines in the Siemens model, plus a sequence of operations, batch sizes, and logical rules that determined which machines could be used under specific conditions.
Furthermore, they used ProModel to validate the baseline model, which handled
iterative simulation runs, comparison of simulation runs to real performance, and baseline-model validation when the simulation-generated statistics matched the actual performance to the given assumptions. Finally, they used ProModel to develop alternative scenarios to design enhanced systems that met the modeling objectives.
Simulating the Processes
The actual project modeled the wafer-diffusion, oxidation, and plasma-etch processes. The simulation involved the machines, equipment, workstations, storage-and-handling devices, operators, and material and information flows necessary to support the process. SSI and the students developed a baseline model of the proposed system and validated the baseline results from the simulation with actual process data.
Also, SSI and the students developed and evaluated alternative scenarios. Many involved the problems of producing two different models of solar cells that called for different machine setups and process times. The students
considered adding extra spin-rinse dryers, diffusion and oxidation tubes, and/or plasma-etch ovens to the system; modifying machine-setup times; and using a different type of dryer that required no setup or changeover. Other potential changes included a variety of scheduling options for both production and parts arrival, modifying batch sizes according to cell model, and changing the distribution of diffusion and oxidation tubes within the facility.
Recommendations
The Cal Poly students delivered their final report and presentation on December 2, 1994. They proposed specific recommendations on given criteria and objectives, as well as on new operations and organizations. As a result of this project, SSI was able to make significant changes that improved the efficiency of the clean-room facility and plan for further changes that would allow even more productivity.
For example, SSI's solar-cell line includes two types of photocells, called the M-line (which accounts for 62 p
ercent of production) and the PC-line (38 percent), that require different machine setups and production times. The facility's four spin dryers were initially set up so that three handled M-line cells and one handled PC-line cells. Simulation showed that changing the setup to two machines for each model would improve throughput and cut-queuing times.
Another change that grew out of the spin-dryer study involved scheduling dedicated shifts for each model type. By running M-line product for 13 consecutive 8-hour shifts and then changing over to PC-line product for eight consecutive shifts, SSI could realize a significant increase in throughput, a reduction in waiting times, and a minimization of time devoted to equipment setup and changeover.
But SSI and the students acknowledged that a radical change in scheduling like this could significantly impact -- either positively or negatively -- other SSI operations that weren't studied in the simulation project. Thus, SSI needed to do further investigatio
n before implementation.
Simulation showed that, once the spin-dryer constraints were removed, the next bottleneck in the system was diffusion. The students recommended an increase in diffusion capacity by converting a bank of four oxidation tubes to diffusion or, budget permitting, adding another bank of four tubes. They determined that converting tubes would lead to an increase in the overall system throughput and utilization as compared to the baseline model. This finding, coupled with the importance of flexibility in balancing the diffusion/oxidation mix, pointed up the need to find simple and expeditious techniques for tube conversion.
In Production
In February, SSI's Cell Fab I clean room commenced operations. The processes of diffusion, oxidation, and plasma etching for silicon wafers are all done within the 2500-square-foot clean room. Approximately 50,000 wafers per day can be processed through the new $1 million clean-room facility.
The improved capabilitie
s of the new facility, fine-tuned by simulation, have enabled SSI to realize significant production improvements. Also, the simulation project validated the capabilities of the proposed system, and this helped SSI meet its cost and capacity expectations. Finally, the simulations gave SSI a much better insight into the interactions among various elements within the production process, which has led directly to better-defined scheduling rules and operating procedures.
Product yield,
which is defined as the ratio of quantity output to quantity input in the cell-fabrication process, has improved from a baseline of 83 percent to its current level of nearly 98 percent. The reduction in scrap loss is primarily the result of better material flow and contamination control in the process.
In addition, process variability, as determined by the wafers' surface resistivity (called
sheet rho
), has been reduced. As the variation in surface resistivity decreases, the fabrication process becomes
more repeatable and yields more consistent consistent electrical properties.
As a result of the clean-room implementation, the bottleneck in the process has moved from the diffusion and oxidation processes (which are done in the clean room) to downstream operations. Most of the improvement was due to the installation of additional equipment to increase capacity. However, SSI designed the facility to get more throughput with the same number of operators, and thus the cost per unit has been reduced. Also, improved material flow has resulted in lower inventory levels and reduced cycle times. This allows SSI to be more flexible in responding to customer demands.
Lessons Learned
While SSI's new fabrication process operates effectively, thanks in part to the simulations that went into the planning, there's still room for improvement. In particular, Fahner says that if SSI were to do the simulation project all over again, he would make some important changes.
First, he wo
uld involve manufacturing personnel to a greater degree during the what-if discussions when deciding what options and alternatives to explore. Second, he would work harder to develop a stronger in-house simulation capability. This would allow SSI to take better advantage of the model and to use it as a tool for continuing improvement. Finally, he would develop run-time models so that users not familiar with programming could still run the simulations and explore new alternatives with new data.
At the moment, primarily due to cost and time constraints, SSI has no further major simulation projects on the docket, although SSI engineers use in-house simulation tools to create simple models as the need arises. While simulation languages and software applications are improving at an accelerated rate, Fahner acknowledges that simulating a complex process still requires highly skilled engineers, a clear focus, and dedicated time for modeling.
Joint projects, like this one with Cal Poly, can be an effectiv
e way of leveraging resources, but this type of cooperative effort requires a significant amount of project management effort. It's an effort Fahner thinks was worth making.
"Overall, it went fairly well," he says. "The Cal Poly students did an excellent job in performing simulation modeling of Siemens's new cell-fabrication clean room. Their analysis was thorough and insight-ful. In addition to learning about Siemens's processes, they were concurrently learning how to use ProModel simulation software. It was evident that they quickly grasped the fundamentals of analyzing system constraints and applied these methods to Siemens's case study."
In sum, he says, "Their conclusions and recommendations were sound and reasonable, having direct applications to improving Siemens's processes. The presentations were very professional, and the interactive discussions were even more helpful. The real value of this project came from their fresh perspective and insight. The knowledge that they have gained can gr
eatly improve our understanding of the cell-fabrication clean-room process."
WHERE TO FIND
ProModel Corp.
Orem, UT
(801) 223-4600
fax: (801) 226-6046
http://www.promodel.com
THE PROBLEMS
-- Poor
material flow.
-- Unbalanced
resource utilization.
-- Bottlenecks
in throughput.
-- Schedule
delays.
THE SOLUTION
-- Model
alternative scenarios and select the best attributes from
each.
-- Study
the interactions between machines to eliminate bottlenecks.
-- Improve
material flow and reduce work-in-process levels by
implementing a "pull system" based on co
nsumption replenishment.
THE BENEFITS
-- Validated
process capabilities prior to implementation and met
budget and capacities.
-- Modified
scheduling rules and operating procedures to improve
process performance.
-- Gained
a better understanding of the process, reducing the risks
involved in decision-making.
-- Developed
an analytical tool that can be used for continuous
process improvement.
photo_link (46 Kbytes)
Mike Fahner, project manager at Siemens Solar Industries, in the new clean-r
oom fabrication facility.
John R. Vacca is a freelance information-technology and air-and-space contract writer based in Houston, Texas. He can be reached on the Internet at
74044.164@compuserve.com
or on BIX c/o "editors."
Faking It, Then Making It
