News
04/01/2002
Korea puts the worst behind; hopes for an early spring
With DRAM-dependent Korea hit by the rout in memory prices over the past year, it's no surprise that chipmakers there have cut capital spending to the bone.
But what may be surprising is that for Korea, the worst year in the history of the semiconductor industry turned out to be a better one than 1998, when the country nearly went bankrupt and had to be bailed out by the IMF.
Korea's capital equipment spending plummeted 53% year-on-year to $1.2 billion in 1998 — the height of the Asian Financial Crisis — but the fall in 2001 was 42% year-on-year to an estimated $2.1 billion, according to Semi figures recently released at Semicon Korea. Korea was also able to maintain its relative share of worldwide capital expenditures at 8% over the past two years, whereas Taiwan saw its share almost halve from 20% in 2000 to 12% last year.
 |
For 2002, semiconductor capex in Korea will fall 10% to $3.5 billion, according to the Korean Semiconductor Industry Association. [KSIA's figures, which are higher than Semi's because they include expenditures on R&D and other items, put 2001 spending at $3.9 billion.]
While that's more bad news for equipment and materials suppliers, Samsung Electronics Co. is providing some relief. Although the company cut its overall capital spending by 40% this year, the memory chip operations have been allocated about $1.15 billion for 2002 — a respectable amount given the slimmed down capex budgets of other major IDMs.
"We see some opportunity for Samsung's 300mm pilot line this year," said Ben Chen, VP and GM of Asia for PRI Automation. "We also expect a 300mm fab from Samsung, but maybe not until next year."
By comparison, financially strapped Hynix Semiconductor is at the bottom of the DRAM ladder when it comes to capital expenditures. But if the company is to be believed, it gets a lot more bang for the buck.
"With one third of our competitors' capex, we successfully operated 0.18μm fabs to 0.15μm and full volume production came up with [good] manufacturing yields," noted Youm Huh, executive VP of the nonmemory division at Hynix. In spite of what he called "a really tough year for Hynix," Huh said the company "demonstrated its engineering and manufacturing capability in achieving the most competitive manufacturing cost for 0.15μm-based DRAM."
In any case, what little money Hynix has to spend has been on hold because of the prolonged negotiations with DRAM maker Micron Technology over possible mergers and joint venture agreements. Another casualty of the ongoing negotiations was a proposal announced in December for Hynix to sell some of its nonmemory fabs to a consortium of fabless Korean IC design houses. That's now on hold pending an outcome of the Hynix-Micron talks, according to industry sources.
With capital expenditure budgets under pressure, the name of the game this year will be making the most of what's already installed.
"Korean chipmakers are very cautious and not confident about recovery," said Hye-Bum Choi, a director of the KSIA. "The focus is on boosting productivity, not building new fabs."
For tool suppliers, that means keeping the communication lines open with their Korean customers to ensure they are positioned to benefit from the upturn when it comes.
"We are focused on positioning ourselves with regard to the technology for 300mm transition, for copper transition ... as well as advanced process control," said Dean Duffy, VP of sales for FSI International. "The downturn has given us the opportunity to work with these customers, to support them with the recipes, the equipment configurations, and the integration activities necessary so we are in a position [to benefit] during the upturn."
While Korea is known for its DRAM device prowess, the country has also developed a large number of manufacturers of components and systems for fabs and backend operations. In fact, two-thirds of the exhibitors at Semicon Korea over the past two years have been local companies.
An example of a local toolmaker is A-Tech Ltd., which manufactures wet stations and sputtering systems. Besides selling to Korean customers, "we will focus on China, Taiwan, and Malaysia," said company president S.K. Suh. Still, despite Korea's fierce nationalism, the country's leading chipmakers prefer to buy known-brand equipment. "For major equipment in the wafer fab, they buy 99% from outside Korea," noted Suh. "Noncritical equipment they will buy from Korea."
While Korea's chipmakers have felt the icy winds of the DRAM market for several months, the same can't be said of Seoul, which has had a particularly mild winter this year. "We expect an early spring this year so I hope to see an earlier recovery coming from Seoul, too," said Huh of Hynix. Even his competitors would agree with that.
— Craig Addison, Asian Correspondent
Downturn pushes China foundries toward smaller geometries, more emphasis on service
Even with China's low costs and vast potential market, it looks like the only way to turn a profit making chips is with newer technologies and more value-added service. So Grace Semiconductor Manufacturing Corp. and Semiconductor Manufacturing International Corp. (SMIC) are planning to make 0.18μm chips, looking toward 0.13μm, and emphasizing customer service.
Grace plans to build only two fabs, but will continually upgrade those to new-generation technology and contract everything else out to other foundries. It aims to differentiate itself by serving as a coordinator of many partners to supply customers with a total solution. More than 90% of Grace's initial production will be exported, though many of the systems that the chips go into may then be re-imported back to China. Customers will likely include overseas electronics systems producers, and assembly and design houses.
"Our strategy is not expanding by building fabs, but by upgrading," explains Nasa Tsai, Grace's president and COO. Beyond that capacity, "We'll contract to other foundries," he says, "like ASMC for bipolar." Another difference between Grace and the other foundries, says the Stanford PHd and Mosel founder, is that it offers a wider range of services to the customer.
He notes Grace plans more deals like its recent agreement with Avant! to provide different IP for customers, so it can coordinate the entire production process for the systems house, from design through production, test, and assembly, even including sales. "The difference [between us and] other foundries is this kind of total solution," says Tsai.
 |
Grace plans to complete its cleanroom this summer, install tools in the fall, and start pilot production by year's end with mainstream 0.25-0.18μm technology. Workers will go to Oki fabs in Japan this fall to learn to run the equipment, to ramp in March.
Meanwhile Grace is working on developing 0.15μm and 0.13μm technology. "We'll move ahead in step with the industry," asserts Tsai. "In three to four years we'll be on par with the most advanced technology in the world. When 0.13 is ready for mass production, we want to be ready." The two shells now under construction have enough space for capacity of up to 100,000 wafers/month.
The company's $1.6 billion financing is to come from $700 million paid in capital, $700 million in long-term bank loans, and $200 million in short-term loans. Some $200 million is already paid, another $400 million committed. Meanwhile, SMIC president and CEO Richard Chang said his company's yields were up to 90.4% for mask ROMs, 93.6% for 0.18μm 4M SRAMs, and 95.6% for 0.2μm FCRAMs.
"We don't want to compete with other foundries on price," Chang told Japanese institutional investors at a Mizoho Securities forum. "That's the major differentiation from other foundries. And we'll support the IDM company more."
SMIC's first round $1.5 billion investment is committed for building capacity up to 42,500 wafers/month of 0.25μm and 0.18μm standard CMOS. The second round of financing, not yet committed, will fund expansion to 85,000 wafers/month by June 2004, and to 0.13μm and bi-CMOS processes. The company is talking to IMEC, STMicroelectronics, and Infineon about acquiring 0.13μm technology.
Both SMIC and Grace have recruited flocks of overseas Chinese workers with experience at big name chipmakers to Shanghai. Some 430 of SMIC's 1300 total employees are from abroad, 300 of those from Taiwan, 100 from the US, and 30 or so from Japan, Korea, and Europe. Subsidized housing and stock options help, with the company's IPO planned for 4Q03. Grace has recruited about 150 engineers from overseas, mostly from Taiwan, and an equal number of Chinese. By year's end, the company aims to have 400 foreign and 400 local Chinese engineers.
Still, it takes more than just money and people for a new foundry to succeed, says Bruno Guilmart, Chartered Semiconductor's president for Asia Pacific and Japan. "The biggest challenges are ramping the fab and building trust. You have to build experience and expertise with time. You can't underestimate the effort required." And, he notes, "The jump to 0.13μm is not an easy one for anyone."
Main customers so far for the foundries in China are the multinational systems and chipmakers, starting now to design in China for the local market, and looking for some local manufacturing to bypass the 17% value-added tax on imports and meet the fuzzy Chinese requirements for local content. Those customers prompted Chartered Semiconductor to license its 0.18μm technology to SMIC in exchange for some capacity.
"We wanted to offer more flexibility to our customers, who tell us one day in the future they may need capacity in China," explains Guilmart. "We thought of selling an old fab to China, or building one there, though it may well be more cost-effective to produce in Singapore with a fully depreciated fab. We decided to partner, and transferred some technology for access to some capacity in case customers have need for local content, without making a big investment."
China's own fabless design houses are still a small market, perhaps $100 million estimates Guilmart, but are fast moving up the ladder to more advanced technology, like Tsinghua Chip Crystal Microelectronics' recent 0.18μm RF CMOS chip, prototyped on one of Chartered's multiproject wafers and made by Chartered. "This is definitely a first for us," says Guilmart, "And it may well be a first in China." Likely to give the sector a boost is the recent arrival in Shanghai of some US design houses run by overseas Chinese, like Silicon Storage Technology, Trident, and ESS. Toshiba also recently announced plans to transfer several hundred of its chip designers there.
Most of the $10-12 billion chip demand in China, however, still comes from the makers of communications gear, white goods, and brown goods, who buy chips off the shelf for things like refrigerators, TVs, DVDs, or wireless phones. But they're expected to move increasingly, like companies elsewhere in the world, to designing their own chips, and bringing more business to the foundries. "Big changes are coming in this area over the next three to five years," says Guilmart.
— Paula Doe, Contributing Editor
Made in China: Huge assembly site for world's electronic systems
Aside from the potential demand from China's own vast population for cell phones and all the other silicon-rich gadgets of modern life, the country is already a huge assembly site for a big chunk of the world's electronic systems. It produces more mobile phones and desktop personal computers than Japan, about a quarter of the world's televisions and VCRs, and more DVD players than anyone in the world. The China Center for Information Industry Development figures about 15% of those chips are made in China, while Shanghai Hua Hong NEC Electronics executive VP Toshio Ohta says it's only 6%.
— Paula Doe, Contributing Editor
 |
Japan's struggle opens door for foreign tool vendors
Ten years ago, Japan boasted more than its fair share of leading chipmakers, but convincing them to even consider buying any non-Japanese equipment was difficult.
What a difference a decade makes. Not only has the Japanese stronghold in chipmaking deteriorated, but the "not invented here" syndrome has all but gone away. At Semicon Japan 2001, a number of foreign-based equipment suppliers reported success or strong interest in their products, something that would have been a rare feat in the early 1990s.
"It's more of a global manufacturing environment now. Ten years ago, a company like Applied Materials was not very strong in Japan; now they are very strong [here]," noted Robert Owen, CEO of start-up nLine Corp., which unveiled a wafer defect inspection system. Owen expects to sell "increasing quantities" of the new tool in Japan in 2003.
 |
Ask ASML how hard it is to crack the Japanese market. The Dutch lithography supplier all but gave up during the 1990s, preferring to focus on markets like Taiwan and Korea. At Semicon Japan this year, it was able to show — for the first time ever — a chart with a Japan breakout of sales. For 1H01, ASML's Japan sales represented 5% of its global lithography tool sales and 13% of track and thermal products, for a companywide total of 6% in Japanese sales. "Last year, I could not show this chart because the number was zero," said Dave Chavoustie, the company's executive VP of sales.
Chavoustie said ASML had "multiple systems installed in Seiko Epson and many other customers in Japan." The company also announced its first technology training center in Japan, a 36,000 ft2 facility in Kawasaki.
In other lithography news, Japan's Canon claimed to have toppled Nikon from its perennial No. 1 spot in the Japanese stepper market. Of the 280 steppers Canon shipped in 2001, about half of them were for the domestic Japanese market, according to Hiroshi Shibuya, head of the semiconductor equipment sales group of Canon Sales Co. "We could have the No. 1 share in Japan this year [2001]," he said.
Other things considered unlikely a few years ago are also happening, such as mergers and acquisitions between US and Japanese companies. Last month, Advanced Energy Industries Inc., Fort Collins, CO, entered into an agreement to acquire Aera Japan Ltd., a Japanese supplier of mass flow controllers. The president of Aera, Peter Aoyama, noted that attitudes in Japan are changing, allowing such collaborations to take place, and permitting Japan to become more open.
That view was supported by Mihir Parikh, CEO of Asyst Technologies Inc., Fremont, CA, who doesn't rule out future Japanese acquisitions. "The issue for Japan is that it hasn't fully developed a global reach," he said. "Japanese companies have been Japan-oriented. One of the ways [they] can really get global reach is through a partnership or M&A with a US or non-Japanese company."
Semicon Japan was also the venue for the announcement of the Advanced Packaging and Interconnect Alliance, an effort by seven companies to commercially develop advanced interconnects. Member companies include Casio, Dainippon Screen, Ebara (from Japan), August Technology, the Kulicke & Soffa Flip Chip Division, Ultratech Stepper from the US, and Unaxis Balzers from Europe. "Flip chip technology is going to revolutionize the packaging environment going forward," said Arthur Zafiropoulo, CEO of Ultratech. "Our objective is to provide a solution to the industry with minimum risk."
The group plans to build two 300mm pilot lines for process development, one in the US and another in Asia. The first line is expected to be operational by 3Q02 and the second by 1H03, with the precise locations yet to be determined.
The focus on 300mm fabs also raised questions about Japan's role as a technology leader going forward. Figures presented by Applied Materials, based on a forecast from the Semi Fab Database, showed that in 2002, Japan's share of leading-edge capacity (≤0.25μm) will be 18%, compared with 42.5% for Asia Pacific, and 27.7% for North America.
"A few Japanese guys will invest in 300mm, but the majority of them will have to go to the 300mm foundries," said Parikh of Asyst.
One bright spot for Japan has been the consumer products segment. That includes new computer games from Sony and Nintendo, DVD players, digital still cameras, and the 3G-cell phone service introduced in October by NTT DoCoMo.
— Craig Addison, Asian Correspondent
Maybe not so far out after all: Tools for the minifab hit the market
Toshiba Semiconductor's novel plan to rethink the fab to make small runs of systems chips without big capital investment may not be so far out after all.
Tools developed by Toshiba and its partners for such a minifab started showing up on the floor at Semicon Japan last December. Ulvac showed a direct-write ion implanter to cut out resist steps. Tokyo Electron had a scan coater to replace spin coating and CVD, and a flexible thermal processing system to shorten process time to about one hour for most processes.
Such tools will be integrated with near real-time lot management software in a demonstration line at the new Japanese government cleanroom in Tsukuba opening this month. The goal is a low-cost little fab that can quickly turn out small volumes of ever-changing systems chips at a profit. Toshiba figures such a minifab could run 100 lots/ month using a total of only about 35 specialized production tools, compared to the 70-some tools it now takes to run the same volume using conventional equipment.
By doing ion implantation directly through a stencil mask, Ulvac and Toshiba have said they figure they can eliminate up to 10 rounds of resist coating, exposure, development, and removal for all the different p-n junctions, potentially cutting process time from four hours down to one, and costs in half. All that other lithography and resist equipment would no longer be needed for ion implantation. Ulvac has worked with its maskmaker to improve mask resolution beyond what it reported at IEDM in the past, and has worked with Nippon Seiko to improve the accuracy of stage positioning, to get the tool ready for the market.
TEL's newly announced TELFORMULA thermal processing system has been Toshiba's minifab poster child, a model of all sorts of the properties needed for fast and flexible small lot production. It runs small lots in about an hour, instead of large batches in four or five. It can change the load from 1-25 wafers in a batch without changing the recipe or running dummy wafers. It also uses in situ dry chamber-cleaning technology, eliminating the need for off-line wet cleaning, and could potentially be used for multiple processes in succession. The companies have said the furnace heats up at 200°C/min compared to 15°C for standard batch tools. TEL says shipments will begin in the spring of this year.
TEL also has a new approach to coating low-k interlayer dielectric at lower cost with its demonstration scan coater, also developed with Toshiba Semiconductor. By controlling the pressure of the liquid coming through a very fine spray nozzle and maintaining a stable speed of swing back and forth across the advancing wafer, the coater can make a surface having only 3% variation with the same throughput as spin coating but using only 10% as much material. By eliminating the spun-off materials waste, it also reduces impurities in the chamber and cuts cleaning solvents and waste liquids. TEL will mount the scan coater on its existing Cleantrack ACT 8 SOD system. It plans sales by the end of this year.
Toshiba Semiconductor has also been working with Ebara on developing a low-cost immersion copper-plating system, to make a thinner and more uniform copper layer while using only a fraction of the usual plating solution. The tool uses a porous ceramic plate with high resistance, which allows the electrodes above and below it to be brought very close together, so the plating solution bath can be reduced to only about 1mm deep. Cleaning and drying are also done in the same module to reduce process time.
The Japanese government will kick in $33 million (¥4 billion) for integrating these tools into a demonstration line with flexible and almost real-time lot management software under the HALCA project. Consortium members will contribute another $21 million (¥2.5 billion). Participants besides Toshiba, TEL, Ulvac, and Ebara include Dainippon Screen and potential users Sharp, Seiko Epson, Sony, and Rohm.
— Paula Doe, Contributing Editor