by Bob Haavind, Editorial Director, Solid State Technology
Capital and infrastructure needs will far outweigh labor as drivers for chip fabs, so the major industrial nations should retain chipmaking for the foreseeable future, concluded a distinguished panel at the International SEMATECH Manufacturing Initiative conference in Austin, TX, on Oct. 23.
“I never believed that much in China as a place to make chips,” commented Brett Hodess, Merrill Lynch analyst, adding that he did not see it as a good location to lower costs. Labor costs are a small factor in wafer fabs, and capital costs are lower elsewhere in Asia, such as Japan, he said.
“In 2007, only 7% of equipment sales were to China, and only half of those were to Chinese companies,” Hodess pointed out. And now China is slowing down, since the Chinese government wants to provide incentives for industries that create more jobs. There is little job creation for huge investments in wafer fabs — unlike solar, for example, where lots of workers are needed to screw panels together.
L-R: Michael J. Cadigan/IBM, Harvey Frye/TEL America, Kenneth S. Flamm/UT-Austin, Brett Hodess/Merrill Lynch. (Image courtesy of SEMATECH)
Dan Hutcheson, moderator and CEO of VLSI Research, stated that China did not have a coherent business strategy for chips. SMIC, for example, tried to lure Texas Instruments to its foundry, but did not have the advanced technology required. China is a technology power, he explained, and has not differentiated its products — Samsung, by contrast, makes memory chips cheaper than ProMOS, so it is differentiated by cost.
Ken Flamm of the U. Texas/Austin agreed that cost of capital dominates decisions about where to put chip fabs, and labor plays a small role. “There is no cost advantage in going to China, and the infrastructure is not strong,” he commented, adding that Oregon would be a better choice. SMIC’s financials show orders-of-magnitude difference between depreciation/amortization and sales, so huge subsidies are needed. Unless regional governments step up, he does not see much chipmaking expansion in China soon.
New fabs are going up in New York State, however, because of good infrastructure but also incentives. He sees this as a more likely future model for placing chip plants, so G7 locations may continue to be attractive.
There will be turbulent waters ahead in the semiconductor business, and there are likely to be fewer players, suggested Michael Cadigan, GM of semiconductor solutions in IBM’s global engineering solutions unit. It is not only becoming hard to get financing for new fabs, but it is also difficult to get cash for semiconductor development. That’s why IBM has stressed joint ventures, partnerships, and alliances, he said. There are now eight alliance partners for the Common Platform, and recently IBM has entered into a partnership with ARM for embedded devices.
Cadigan illustrated the benefits of this approach by pointing out that in the past, IBM has had a reputation for innovation, but without much regard for cost. Samsung, by contrast, makes consumer products, so it is always seeking out cost savings in processes. Samsung might suggest a step using a lower dose, for example, and this might be adopted. Chartered also makes useful contributions, he added.
The partners feel they will get the products they need designed, and as far as manufacturing goes, Cadigan called them “location agnostic.” IBM is a technology enabler and the partners are free to put plants where they want. Development work is done in Albany, NY, and the initial launch facility for new technology is in Fishkill, he explained.
To illustrate the high cost of developing technology for a new node, Cadigan said that spending could go above $1B. IBM has not reached that level, but he explained that costs are spread out through partner contributions. He feels they have industry leadership at 32nm and believes this can be maintained at 22nm. Cadigan granted that things did not work so well at the initial stages of the Common Platform approach. The partners did not recognize the difficulties of fab synchronization as a vehicle for simultaneous release. Also, they found they needed to build up ancillary capabilities, such as libraries of IP blocks and EDA tools.
He sees a rough ride ahead as the semiconductor industry slims down to fewer players. He cited his experience in the disk drive industry, which shrunk from 20 to five players as competitors drove each other out of business.
“It will be the same situation for chips,” he said.
Hodess of Merrill Lynch pointed out that every 2-3 years some 40% of the installed base becomes outmoded. “No other industry has this problem,” he said.
In the 2005 period, all sectors were having success: NAND, DRAM, microprocessors, and everything else. Fabs were built up to meet demand, and now memory makers have too much capacity. They were changing technology every year, doing half-node advances. Now they realize this was a mistake and they are slowing down. This will make fabs last longer, giving them a chance to get a return over time. Meanwhile, 200mm memory fabs are either shutting down or doing trailing-edge work.
Harvey Frye, president of TEL America, discussed how the strategy for locating factories has evolved. He said that TEL has become a global company to be close to customers. First, he explained, you look at the big picture and narrow it down. Whereas 200 countries may be users, this must be focused to perhaps 20 key locales. So he sees the G7 as remaining strong except for some like Canada.
He does see government as a driver, however, so that sovereign money might play a part in the future in places such as the Arab emirates. Technology and risk factors must also be considered, however, he explained. Links in the supply chain have become much more seamless over the years. To build tracks, for example, integrated metrology is needed and materials are critical. In the old days, the Rubik’s Cube for materials might have had three elements, but now there might be eight or more as the technology becomes increasingly complex.
Because of this added complexity, Frye noted, while TEL continues to want to be close to customers, quality and efficiency may now sometimes trump this. Where you do development may be the overriding factor. It may not matter where you make something; the more important consideration may be where you do development and partnering. As examples, he cited IMEC and Albany as well as Intel.
Hodess pointed out that equipment companies are becoming less capital-intensive as they expand outsourcing. In good times, they are great cash generators as a result, and there has been pressure to pass this on as dividends or stock buybacks. But, he said, they resist this because they know how vital it is to retain key technical people during downturns.
Flamm from UT/Austin suggested that the semiconductor industry is breaking down into four business models. One is design firms that do proprietary, high-performance devices, and they need advanced manufacturing plants to enable them to keep ahead. Others design trailing-edge devices where performance is not critical, so manufacturing is not a differentiator. Instead, the design is the key.
Then there are pure foundry manufacturers aiming for economy-of-scale. Only TSMC is thriving in this sector, he suggested.
Finally, there are commodity products where vendors try to stay ahead in technology. Samsung has been the only company successful at this, he said, but it is so difficult even they have tried to expand into logic.
Hodess suggested that tension was building between high-performance fabless designers and the foundries. TSMC has recognized that while its return-on-equity (ROE) is improving, its stock isn’t. So management believes it must get a better return-on-invested-capital (ROIC). Therefore, it is ramping up 45nm slower than it has moved to new nodes in the past, which is frustrating to companies like NVIDIA, he concluded.
Packaging is gaining in importance, Hodess added. TI [not National Semiconductor: ed.] is building a plant in Singapore to do trailing-edge products, for example. Flip-chip is moving to lower-technology devices, and LCD display drivers are being put directly on the back of flat-panel screens. While the panel agreed that chipmaking will remain in the G7 countries, it did not rule out continued movement into other areas where big markets emerge, including China. — B.H.