Shahen and Thread - Article poses possibility of $10B fabs [wow].
supersite.net
How real is the $10 billion fab?
To some experts, it is just as sure to come as the $1 billion IC plant was a decade ago
By J. Robert Lineback
If history is any teacher, the semiconductor industry is headed straight for the $10 billion wafer fab.
No one can say exactly when this milestone might occur, but some experts already are raising warning flags that major changes in chip processing could easily jump the cost of a new production fab to what can only be described as an astronomical figure within the next 10 years.
For now, production equipment suppliers, chip industry executives, and analysts disagree widely over whether this will happen. But some of these same experts also are quick to note that it was just as difficult back in the late 1980s to believe that the cost of building a wafer-processing plant would hit $1 billion in the mid-1990s.
"When we started predicting [the cost of a new fab] would cross $1 billion, I remember one Intel executive telling Sematech that his company would never build a $1 billion plant," recalls G. Dan Hutcheson, president of VLSI Research Inc. in San Jose. "If things continue [today] the way they have been, I would say yes to the $10 billion fab," acknowledges the veteran analyst. But, he adds, "my own sense is that production yields are getting so high now that the pay back could be diminishing."
"More-expensive fabs have always yielded higher," Hutcheson notes, "making it possible to put more transistors on a chip." But, he says, "when Moore's law runs out of steam and you don't get the higher yields, the rate of increase [in fab cost] will slow down. I don't know if that will happen tomorrow or in the next 10 years, but we are heading into that realm."
Other experts also expect to see the rate of increase in fab costs slowing - but for different reasons. They predict that the cost models in high-volume fabs will change due to the advances in plant automation, higher tool productivity, larger wafer diameters, and greater uses of mini-environments to isolate silicon from contaminates.
Prior to 1995, the cost of fabs had been growing 10 times every 10 years, says Russ Lange, director of silicon technology strategy at IBM Corp.'s Microelectronics Division in Fishkill, N.Y.
"That was a big problem," he says, "because industry revenues were not going up at that rate. Sooner or later, the cost of fabs would [have grown] substantially larger than the revenues being produced," Lange notes. "But in the past decade, the increase in total fab costs has slowed to about 4 times in 10 years - which is good news."
The reduced rate of increase can be linked to the spread of wafer-isolation systems that lower the cost of cleanrooms, Lange says. Reusable designs in modular tool platforms are also helping, he adds.
"The tool industry has gotten big enough to where it can reuse components to help drive down costs," IBM's Lange says. "In the 1980s, we were all using custom tool designs and everything was unique."
When the chip industry reached the point where a production fab was costing $1 billion, alarms went off everwhere. This milestone caused tool suppliers and device makers alike to begin reexamining the way that equipment was being designed, used, and not being used.
"Ten years ago, a fab manager wouldn't immediately be able to tell you which [of his] tools were constraining throughput in the factory," says Jim Doran, vice president of operations at Advanced Micro Devices Inc. in Austin, Tex. "But with equipment costs rising, that has changed."
"We now drive [to fix] those 'constraint tools' to reach 75% utilization in 168 hours of operation in a week," reports the AMD vice president. "That would have been unheard-of 10 years ago. It is all being driven by the need to use these expensive tools at high utilization rates."
The struggle to make increasingly expensive tools worth the price is as old as chip making itself, notes Daniel Queyssac, chief operating officer for the frontend tool operation at ASM International in Phoenix. A senior chip executive for two decades, Queyssac recalls a heated debate over the growing cost of lithography during a management meeting at Motorola Inc. back in the 1970s.
"There was much trepidation in going from 2-inch to 3-inch wafers," he recalls. "Our boss was completely beside himself after learning we had to replace the proximity aligners with $110,000 steppers. No one thought we could make a profit if we had to spend $100,000 on a machine."
The ever-rising costs of fab gear has continued to haunt management ever since. The hot spot now is in lithography, where the new generation of 193-nanometer exposure tools is pushing dangerously close to the $10 million mark.
Even this awesome figure may be eclipsed during the next decade when wafers make their biggest shift yet in production technology - moving to the post-optical exposure systems now in the early stages of development.
Current lithography costs also are being driven higher by other factors. They include the need for more expensive photoresists and the growing use of phase-shifting mask technologies which make it possible to extend the use of exposure tools and print feature sizes that are smaller than the wavelength of light.
But it's not only lithography. Other fab tools are getting to be much more expensive as the need for greater control in process steps increases with each device shrink and new semiconductor materials - such as copper and low-k dielectrics - are introduced.
Plants also are beginning to add new layers of systems and software to their wafer-processing lines as the need for real-time inspection and metrology grows. More closed-loop process control systems and wafer transporting robotics and carriers are also being installed. And automation software is growing in use as the value of a wafer keeps rising and the need to reduce tool downtimes gets more critical.
With the price of tools exploding, chip makers are being forced to focus on making their plants run more efficiently. Fab managers keep looking for new ways to squeeze costs out of everything they can - from utility bills, starting wafer substrates, consumable materials, payrolls, and miscellaneous building costs.
But there's another problem here. Chip makers often don't know just how well they are doing in wafer processing as compared to manufacturers in other industries, claims Rudolf Simon, executive vice president at M+W Zander Facility Engineering GmbH. The German firm studies and builds fabs.
Speaking before an executive summit of chip and equipment managers in Dresden in April, Simon warned that a new wafer fab could reach $10 billion by 2005 if costs were not curbed. He also criticized the chip industry for failing to benchmark itself across companies on a range of manufacturing issues.
"Compared to other industries, the semiconductor industry has done very little benchmarking outside of yields and a few other areas," Simon pointed out. "There is no benchmarking on the cost of ownership," Simon said. "Even inside companies, benchmarking is very sporadic."
As a result, industry managers often are surprised by what's going on within their fabs. "At a Sematech meeting several years ago, we took a very rough cut at the so-called $1 billion wafer fab," recalls C. Scott Kulicke, CEO of Kulicke & Soffa Industries Inc. "We concluded that the theoretical $1 billion fab [would have] only $500-to-$600-million worth of equipment in it." Another $250 million went into bricks, mortar, cleanroom, and systems for air-handling and DI water and about $50 million went for odds and ends, he says. "But the real shocker was the cost of installing and qualifying equipment, which was another $100 million," says the head of the Pennsylvania chip-assembly systems supplier.
But one of the few positive developments coming out of higher equipment prices is that the trend is driving chip makers closer to their equipment suppliers. "We work with all of them to get as much as we can out of the equipment," declares Lewis McMahan, worldwide facilities manager at Texas Instruments Inc. in Dallas. "One of the keys is to get the designs of IC products as small as possible and simple [in terms of process steps]. Whether that's big enough to make a difference, I don't know, but it's the only 'big knob' we've got to turn right now."
The equipment suppliers also are trying to get closer to their customers. "Barring a revolutionary way to process semiconductors, we are forced to continue to use standards, cooperation between customers and suppliers, and new system designs that are upgradable," comments Mike Parodi, president of Tegal Corp. "A tool cannot stand alone for one process generation any more," comments the etch gear supplier based in Petaluma, Calif.
Parodi and other equipment executives are becoming increasingly worried that the next round of wafer fabs will push the total cost of a plant to more than $2 billion, meaning that it becomes a bet-the-company decision for chip makers.
"I think we're already running into the economic wall," warns Brad Mattson, CEO of Mattson Technology Inc. The Fremont, Calif, supplier sells tools for diffusion, thermal processing, photoresis stripping, and epitaxial silicon. "You already see the problems in DRAM factories," he says. "They cannot make a profit, and clearly the whole foundry movement is a response to the economic wall. All of these mega-changes" he believes, "started when the industry hit the $1 billion fab."
With the cost of manufacturing growing and industry revenues flat for the past three years, concerns are growing within the chip industry regarding the impact of technology acceleration on industry health. Sematech, the Austin-based industry consortium, is now pursuing an economic model that it hopes will aid chip makers and their capital equipment suppliers in determining the tradeoffs in launching such new efforts as a new process generation, a larger wafer size, or new types of factory systems (see story in the June publication).
"A lot of people feel the technology roadmap [sponsored by the Semiconductor Industry Association] has done a great job identifying technology trends and challenges, but there are other issues that must be considered," points out Mark Melliar-Smith, CEO of Sematech. He was referring to the growing need for economic data in chip manufacturing.
Until recently, Melliar-Smith was one of the believers who was certain the industry was heading toward the $10 billion fab. Today, he is less certain that chip makers would be able to build these huge factories and operate them with acceptable profits. The Sematech CEO, as well as some chip makers, now predict that the industry's persistent boom-and-bust cycles will most likely douse any interest in $10 billion factories during the next decade.
Without a doubt, though, next-generation fabs will be far more expensive than today's $1.2-to-$1.8 billion plants. They will perhaps rise to around $4 billion in 2009, suggests semiconductor analyst Bill McClean, president of IC Insights Inc. in Scottsdale, Ariz.
"If [the pricetag] goes up from $1 billion to $4 billion, it won't make a lot of difference in the number of companies being able to afford fabs," McClean says. This is because "so many [chip makers] are dropping out [and going to foundries] as we move to the $2 billion fab."
The prospect of fewer chip companies being able to afford manufacturing plants is a daunting prospect for the capital equipment industry. More silicon foundries are expected to become major customers of fab tools, but major equipment vendors are looking for ways to change the rules. Richard S. Hill, CEO of Novellus Systems Inc., suggested at the April executive summit in Dresden that the industry needed to develop a business model for the reuse of fab equipment by chip makers.
"A more efficient way to utilize the entire asset base of the industry would be a huge potential," Hill explains. "To do this, we need to better understand who needs the performance level of different types of equipment and develop an open market for equipment reuse. It's done now on a limited scale but not strategically by chip manufacturers," he says.
Hill figures that capital equipment suppliers still will be in a good position to grow in the next 10 years because either $10 billion fabs will be built with more expensive tools - assuming circuit shrinks continue at their current pace - or many more lower-cost fabs will be built if device scaling begins to slow.
Executives at Applied Materials Inc. also see an opportunity to shift the focus of equipment designs to provide "cost-driven solutions," according to Dan Maydan, president of the giant Santa Clara, Calif., equipment company.
"Instead of just supplying technology for etch machines or position machines, we will focus on modules, which would have somewhere between one and four different kinds of systems," the Applied president figures. "Instead of doing one type of technology, [these modules] would perform the types of activities for complete microstructures," he says. "All of a sudden - if you do it that way - the number of steps required in a fab line will be reduced dramatically.
The big challenge for this kind of strategy, however, is to guarantee the module approach will produce the performance of individual tools," Maydan says.
But no matter how much the chip industry complains about higher equipment prices, Maydan insists the latest production equipment has been well worth its price. Today's 8-inch wafer fabs can handle 30,000 wafer starts a month, he points out, which is double that of the 15,000 four-inch wafers turned out in the early 1980s.
Figure in such improvements as device shrinks and volume runs, and Maydan, ever the salesman, maintains that the chip industry has improved its ability to turn out good transistors by 2,000 times, while costs increased just 16 times. He sees no slowdown ahead.
This issue features the first of a two-part series on the rising costs of manufacturing in the semiconductor industry. The following stories cover the soaring costs of equipping and operating future fabs: |
