Cycle time, 300mm in spotlight at automation conference
08/01/2001
 |
Case studies and the migration to 300mm in the automation segment took center stage at the first Worldwide Automation & Performance Symposium, held in Salt Lake City, UT, and sponsored by Brooks Automation, Chelmsford, MA.
The symposium concentrated on customer case studies about wafer-fab tool and factory automation, factory interfaces, automated material handling, and tool control, especially from companies transitioning to 300mm. A key aspect of evolving technology for 300mm is the insistence by many fabs for equipment-centric solutions in the form of equipment engineering systems (EES), the natural extension of e-diagnostics, said John Biasi, director of strategic marketing for the software division at Brooks. With the complexity of today's automation on new tools, users have become very concerned about tool performance, availability, maintenance, etc. But more issues need to be addressed, referring to the current acceleration of EES standards and guidelines that are rolling out as Semi standards in 2001.
"Today, EES is a key concept of e-manufacturing and high efficiency factories, and a key concept in the transition to 300mm, all under international collaboration that only started about one-and-a-half years ago," said Biasi. "300mm is primarily about cost — a cost driven conversion — but there is a wake up call that it has to be applied all the way back to the equipment."
Keynote speaker Clayton M. Christensen (see photo), professor of business administration at Harvard Business School, gave symposium attendees a view of the future of semiconductor manufacturing based on his business model from his best-selling book, The Innovator's Dilemma.
Christensen's model builds from the premise that in every industry there is a trajectory of improvement that customers can use, and the pace of technology outstrips the ability of customers to use it, eventually giving rise to more disruptive technologies.
 Clayton M. Christensen of Harvard Business School sees big changes ahead in chipmaking. |
In Christensen's model, money stops when current products are not quite good enough compared to prior products to satisfy the needs of immediate customers in the value chain. "Because it is not good enough, the architecture is nonstandard, proprietary in nature, and so the scale of economies is skewed. Market leaders have to disintegrate by selling off the subassembly part of their businesses that makes money."
Christensen sees a potential for this scenario possibly happening with Intel and the microprocessor industry, for example.
Now that microprocessors have become more than just good enough, we are likely to see emerge a low end to the microprocessor market for different applications; these will be very modular and the competition will be different from the way Intel has competed on the high end of the market. At this low end, speed to market and the ability to customize is going to be the way these companies do business.
Christensen foresees that the semi industry's current three-month throughput factor won't suit this market, and that wafers will have to move through fabs in three weeks, and then, perhaps, in three days. "The implications for the kinds of equipment that KLA-Tencor, Applied Materials, etc. make, and the kinds of software tools that Brooks has to make, put great value on integration that a company like Intel has historically enjoyed. The world is likely to change as a result of these things happening," concluded Christensen.
Rob Leachman, director of the Competitive Semiconductor Manufacturing (CSM) Program at the University of California at Berkeley, added credence to the future importance of cycle time, but for other reasons, with phenomenal bottomline facts from a DRAM manufacturing case study. The study came from the work of the CSM Program, Leachman's consulting to IC manufacturers, and research commissioned by Sematech to update cost resource modeling approaches for evaluating industry roadmap improvements.
"Most manufacturing improvements are traditionally evaluated in terms of costs. For example, reduced working capital decreasing the time between when we spend money and when we are paid for selling that product, or lower WIP and inventory levels, means you can have smaller cleanroom and warehouse spaces. But in a technology industry, the impacts of what we do to revenue streams can be much greater. And, in particular, for technology goods where prices are falling over time, the impact of manufacturing speed on revenue is very great; this is the real key to value in cycle time reduction," said Leachman.
Leachman admits that with shorter and more controllable cycle time a company has a more attractive product and service. But his work has concentrated on the juxtaposition of cycle time improvement with selling price.
From manufacturing benchmarks collected in the CSM Program, Leachman's calculations show that extending IC development time an extra day, to get a stepper or process qualified, is like paying $3.44 for every wafer that the factory will ever make. In addition, if it takes one more day to reach mature die yield, it is like paying $1.35 for every wafer that will be made, or if cycle time is one day longer, it is like paying $3.04/wafer.
For a real world example, Leachman outlined the results of a five-year project at Samsung that began with two fab lines and was eventually expanded to 12, plus die sort and assembly. This project involved methods for calculating target cycle times for individual process steps, online heuristic scheduling of all fab processing equipment, situation-based planning of maintenance and engineering work, and training over 3000 people about how to think about cycle time.
"We saw the cycle time gradually come down from 80 days or more to about 30 days to make DRAM, a manufacturing cycle where the theoretical cycle time is 19-21 days. In the end, we computed DRAM revenue if the cycle times had stayed at 80 days and found that cycle time reduction increased sales revenue by $954 million, in an operation with annual revenue of $22 billion. When we included non-DRAM production in the fabs the gain was over $1.1 billion," said Leachman. For this period, Samsung's DRAM market share rising from 18 to 22%.
"Altogether sales revenue went up 4% and market share 4% without increasing production volume, capital investment or people. We simply increased the overall selling price by 4% — all profit," Leadman said. "Being in the market early before the price goes down made all the difference."