300 mm fabs are changing the way the semiconductor industry does business
by Allan Richter
A few years ago, tools that could process 300 mm silicon wafers were under development, but they didn't quite achieve critical mass or significant adoption. A cyclical business lull ended suddenly in 2000, yielding enormous orders that dropped like a winning lottery ticket into the laps of both the IC makers and the equipment manufacturers that supply them. Alas, 300 mm technology was not yet mature enough for most manufacturers to put their trust in it, and 200 mm continued to rule the day.
That was then. As the industry slowly emerges from its worst downturn, many expect chip makers to invest enough in 300 mm fabs and the tools to fill them to call the shift a trend. If not next year, they say, then 300 mm will be here in 2004. In any case, 300 mm is coming. As of early last month, equipment makers said they were tracking 16 active 300 mm fabs, most in production.
"That's beginning to show a commitment to it, and it seems to be working, so I expect you'll see more and more companies ramp and more companies include 300 mm in their strategies," said James C. Morgan, chairman and chief executive officer of Applied Materials. The equipment giant is expected to be one of the key beneficiaries of the shift because 300 mm favors large tool companies that have strong R&D budgets. And on the chip side, only the biggest integrated device manufacturers can afford to build 300 mm fabs.
If this is indeed the upturn for 300 mm, then the roughly $3 billion needed to build a fab for the larger wafers, versus roughly $1.8 billion for a 200 mm facility, will indeed force unprecedented manufacturing alliances. That's the known part of the equation. The unknown factors – and it still may be too early to tell – are the strategies of those alliances.
However they take shape, some industry participants expect the transition to the larger wafer size, combined with other moves to advanced technology nodes, to create new opportunities for suppliers, in effect, wiping the slate clean for the coming upturn.
Michael J. Luttati, executive vice president and chief operating officer of Axcelis Technologies, an ion implantation equipment supplier, said he expects customers to proactively look for tweaks and upgrades as they make their second and third rounds of 300 mm equipment purchases. That can give incumbents that have kept up with improvements a huge advantage. But suppliers that haven't kept pace, Luttati says, create what he termed another industry "discontinuity" – an opportunity to change vendors.
One likely scenario that could help create those opportunities in the longer term is that chip makers will roll out the same toolsets from fab to fab to help offset their intense investments. Intel calls this equipment deployment practice "copy-exact" – a term the chip giant coined when it applied a cookie cutter approach to several of its fabs.
With huge capital investments in 300 mm fabs requiring the deep pockets of not one IC supplier but several partners, chip makers are now likely to make a more determined effort to adopt the copy-exact approach both internally and with alliance partners that co-invest in the fabs. "It's going to force you to deal with capital availability issues to a much greater extent than it did at 200 mm," says Steve Newberry, president and chief operating officer of Lam Research.
IS COPY-EXACT THE WAY TO GO?
Many equipment makers expect copy-exact to become more popular at 300 mm because the manufacturing strategy minimizes the risk between product development and full production. Indeed, key links in the supply chain stand to benefit. After completing a four-year review of more than 130 300 mm tools, the organization Semiconductor Leading Edge Technologies (SELETE) concluded two years ago that suppliers and chip makers alike could reduce product development costs, according to a SalomonSmithBarney report.
The qualifier: BIG. It will take bigger investments by bigger chip makers to build the fabs for bigger wafers, with a bias toward bigger equipment companies with bigger R&D budgets. That provides a great deal of upside to the copy-exact trend. Imagine the reward for spreading development costs across customers with the bulk of Taiwan Semiconductor Manufacturing Company (TSMC) or United Microelectronics Corp. (UMC) if the chip makers apply the strategy.
"It's clear that if you develop the technology on one toolset and then put a different toolset in the production line, there is significant risk in achieving the same results in a short period of time," says Wilbert van den Hoek, Novellus' chief technical officer.
But the copy-exact practice is as risky as it can be rewarding. Just as it can yield a highly lucrative win from several customers on one roll of the dice, it can also mean a devastating loss if the lead chip maker in the partnership rejects a supplier's tools. The leader decides – a move that could favor that company's incumbent tool suppliers – and the other partners have to follow suit because of the cost. Thus copy-exact raises the stakes: Win or lose, the opportunities at 300 mm are bigger.
"It certainly makes it easier for us to go into a new fab and get the equipment started up," said Lam's Newberry of the copy-exact model. "But it's a double-edged sword. It increases the need for an equipment supplier to understand who those lead companies are and who are the ones that in essence will make the decision for equipment selection. Winning at IBM is not just winning at IBM; it's also winning wherever IBM sells the technology. If you don't win the application tool of record at IBM or TSMC or any of these other copy-exact exporters, you've lost the business at that other place."
But Mihir Parikh, chairman and chief executive of Asyst Technologies, an automation tool supplier, believes that suppliers to the joint ventures now forming to leverage 300 mm costs will be simpler to manage than the variety of technology and development joint ventures he has observed over the years. The simple reason: The newer ventures are production alliances in which one partner is the obvious leader. "We have fewer players to go to," Parikh says.
Some equipment executives, however, don't believe that is quite how business will be won at multiple chip makers in an alliance.
Don Mitchell, chairman and chief executive at FSI International, which supplies wafer cleaning and resist processing equipment, said he rejected the term "copy-exact" in the context of the 300 mm rollout. The way FSI sees it, if the technology and all other sales and service conditions are right, newcomers have as much chance at winning contracts as the incumbent equipment supplier at whichever chip maker is perceived as the dominant partner.
"In a joint venture, the implementation of copy-exact is one partner dictating the entire toolset to all the other venture partners," Mitchell said. "We are learning that that is not happening, but rather they are cross-pollinating their engineering teams to determine who has the best process performance for different parts of the manufacturing sequence. Then they use a more collaborative approach toward equipment selection for the joint venture investments. What emerges from that is a common toolset" for all the partners.
So far this year, 62 percent of FSI's orders are for 300 mm solutions, pacing well ahead of the 40 percent of capital expenditures that analysts expect to be earmarked for 300 mm throughout the industry this year, according to the company.
Whatever its name, the strategy of using like toolsets can be too inflexible despite the potential early cost benefits and the maximum return on development it can provide. Ultimately, it can cause chip makers to fall behind the technology roadmap if they adhere to outmoded tools. That can cost equipment vendors future business. Without accounting for later advances in technology, copy-exact can break long-established links in the supply chain.
COPY, BUT WISELY
Applied Materials' Morgan says it is still too early to tell how wedded chip makers will be to copy-exact. "I think it also depends on what the technology requirements are," Morgan says. "You're really concentrating on the best process capability, and so each company has to judge for its process and for its particular application what its strategy will be. Having a blanket strategy may hurt them in maintaining technology leadership. You'd like to not have to change any more than you have to, but if you don't have the best process steps you're in danger of losing your competitive position."
Novellus' van den Hoek agrees, saying chip makers should execute a variation of copy-exact known as copy-smart, an approach that embraces flexibility and allows more current technology enhancements to the replicated tools. After all, he points out, equipment is typically chosen for development about three years before volume production.
The SELETE study cited by the equity research company SalomonSmithBarney in its recent report on 300 mm supports that idea. While 300 mm tools have nearly the same performance levels today as 200 mm tools, the report said, continuous improvement would be needed to meet cost-benefit goals of 30 percent to 40 percent.
The roll-out strategy adopted by Infineon Technologies, the first chip maker to build a 300 mm facility, beginning with 64 megabit DRAMs in Dresden, Germany, three years ago, may shed some light on how copy-exact and copy-smart could be applied in similar projects. Two of Infineon's 300 mm fabs have very similar tools that follow the copy-exact model, said Peter Schaefer, vice president of memory products at Infineon Technologies North America, but only because they ramped up within a few months.
The fabs, both of which run 200 mm and 300 mm production lines, are the facility in Dresden and a joint venture, ProMos Technologies, in Hsinchu, Taiwan, with Mosel Vitelic. Instead of copy-exact, Schaefer said the dominant model at work is a "fab cluster" framework that encompasses the two facilities and others. Those include a 200 mm fab in Richmond, Va.; a 300 mm joint venture with UMC and the Singapore Economic Development Board in Singapore; and a 300 mm joint venture with Nanya Technology planned for Taoyeun, Taiwan.
The "fab cluster" model embraces copy-smart, in large measure because of sensitive human resources issues. "If you say you have only one factory that is providing the process and nobody can vary from the process, you have a problem motivating workers in the other factories," he said. "If everyone can contribute to process development, you get a better result."
FINDING BALANCE
With the copy-smart strategy, however, the stew that ends up on the table may hardly resemble the original recipe at all. "The problem with copy-smart is that it opens the door for ignoring copy-exact completely, allowing engineers to pick new tools for whatever reason," van den Hoek says. "Copy-exact is black and white, which means you don't have any decision to make. Copy-smart has shades of gray, and managing shades of gray is much more difficult."
Often, the problem isn't that customers look for too many changes, but too few. Customers are not as proactive about accepting the kinds of improvements that equipment vendors want customers to accept to keep competing tool suppliers at bay. Van den Hoek says Novellus tries to get customers to adapt as many improvements as possible, but customers typically accept two out of three vendor-proposed changes at best.
One example of a tool modification more readily accepted at 200 mm versus 300 mm is a change eliminating the need to scrub a wafer of backside silicon dioxide particles, van den Hoek said. Those particles could deform a wafer slightly, but even a slight distortion can hamper a lithography tool's ability to print small features because the tool's depth of focus is only a few thousand angstroms.
At a 200 mm fab, a chip maker would likely accept such a change as late as six months before full volume production, van den Hoek says, but nowhere near that time frame on 300 mm equipment. Rather than risk endangering the whole process with a modification, a chip maker might be forced to invest $10 million in capital equipment that could scrub the wafer separately.
Here's how the numbers play out. Buying that $10 million of extra equipment ends up a better choice than accepting even a 1 percent lower yield. A wafer from a state-of-the-art 200 mm fab producing devices in a 0.13-micron copper process costs roughly $5,000, so a 300 mm equivalent would cost $10,000. Churning out 30,000 wafers a month, for instance, a 1 percent loss in yield would cost the fab $18 million in just six months.
The conundrum: Another vendor poses a challenge by offering the scrubbing function built into its tools. "Copy-smart certainly gives you the highest return with minimum risk but is the most difficult to implement," van den Hoek says. "There are a lot of decisions."
WRITING ON A CLEAN SLATE
Semiconductor capital equipment makers recognize that the fledgling 300 mm market represents a "clean slate" in the industry. The larger wafer sizes present an opportunity for chip makers to buy tools from new suppliers or remain with their incumbent vendors if their offerings are strong enough. Equipment makers say they are working to produce highly robust 300 mm tools while trying to make the transition from 200 mm to 300 mm as steady as possible for customers.
Unlike the move to 300 mm wafers, the shift to 200 mm involved less stark transitions between linewidth geometry requirements, said Scott Becker, FSI's vice president of product management for surface conditioning products. As a result, the transition to 300 mm wafers has let the company start fresh when designing tools to accommodate the larger wafers.
"We were able to implement all the knowledge and technology improvements that we've been wanting to incorporate in our products," Becker said, "but it's fairly prohibitive to do that in a lot of older products. In 300 mm, we started with a clean sheet of paper." Leveraging the 300 mm advances, FSI has retrofitted 200 mm tools with innovations it applied to tools for the larger wafer size. (For more on 200 mm, see the sidebar, "Okay, 300 mm is Here. Whither 200 mm Tools and Fabs?" )
Some strategies for helping customers straddle both 200 mm and 300 mm production until they get a firmer footing in 300 mm are more subtle than bringing to market an array of classically defined hybrid, or bridge, tools.
Axcelis' Luttati said his company is trying to ease its customers' transitions to 300 mm tools by combining 200 mm features that are less of a culture shock with more enticing advances. On its high-current, low-energy ion implant product line, for example, Axcelis scaled the end-station, wafer-handling technology from its 200 mm version and applied it to 300 mm. Newer to the very low-energy implanters in the 300 mm product line, Luttati said, is advanced beam line technology needed for forming shallow-junction transistors.
Another supplier, the SEZ Group in Villach, Austria, has been deploying a strategy to help buttress the market – and its position in it – with several support braces. On the one hand, the company has been trying to help customers make a relatively easy transition to 300 mm with a new support lab, which it opened in Phoenix, Ariz., in late April, where customers can "kick the tires" on its new equipment.
At the same time, the company has been positioning its single-wafer surface-preparation tools for greater acceptance by chip makers deploying 300 mm, but it also rounded out its play in the market by acquiring a small German wet bench manufacturer, HMReinraum-Technik Ltd., earlier this year.
"We didn't want to buy a large wet bench company because we felt the direction was definitely single wafer, but we needed to address the whole market," said Jim Mello, vice president of technical operations for SEZ America in Phoenix.
OPPORTUNITIES, HIDDEN AND OBVIOUS
With a 300 mm fab depreciating at $56,000 per hour, according to SalomonSmithBarney, more inspection and metrology will be needed to maintain uptime, making that segment another area the supply chain will have to carefully navigate. The equity research company estimates that the capital outlay for inspection and metrology accounts for up to 5 percent at 200 mm; it is expected to grow to up to 9 percent at 300 mm. But suppliers and chip makers will have to tackle some new issues, like determining the optimum sampling size for the larger wafers.
To pinpoint threats to yield, fabs will have to sample more than usual. But at 300 mm, figuring out precisely how much more has not been fine-tuned. Equipment suppliers who help customers make "data-driven" buying decisions, says Peter Nunan, vice president of KLA-Tencor's Yield Technology Services Group, will more likely win their confidence than by wantonly selling tools.
"If you're going to build a 30,000-wafer-a-month fab, you don't put in all that capacity on day one. You put it in chunks," Nunan advises. "So if you do your over-sampling when you put your first 10,000-wafer chunk in, and you determine what your excursion frequency is, when you put the next chunk in, you purchase the additional inspection and metrology capacity based on the initial sampling. If you've done this right, you'll over-sample at 10,000 wafers, so you absorb that capacity when you go to 20,000."
One of the more apparent opportunities for suppliers is helping fabs automate their 300 mm production lines, where carriers will have to deal with 20-pound containers – more than twice the weight of a carrier in a 200 mm line. Safety regulations could set weight limits, and security-minded fab managers may be more prone to limit access to the carriers, whose value rises considerably at 300 mm.
Dave Huntley, president of Kinesys, a 10-year-old Netherlands-based automation software supplier that has focused on the back-end of the fab for the past several years, says the transition to 300 mm is also turning up some hidden opportunities. A Kinesys focus is in inkless assembly, that is, transferring test results from a probe on the assembly side without marking failed devices with ink dots, considered a contaminant.
Huntley says many chip makers are seizing the opportunity to go inkless along with the other sweeping changes that come with rolling out a 300 mm fab. "It's the window of opportunity. The investment is going in anyway. Why not do it all?"
At 300 mm, it seems, everything is bigger – the wafer size and the stakes. Incumbent suppliers have a good shot at extending their contracts with sizable increases – if they can keep pace with technology advances and work through the whims of their customers. But competitors are sure to be watching much more closely to see if the incumbents blink.
semi.org |
