Focus on Taiwan: UMC on the Move

Industry observers could be forgiven for thinking that Taiwan's UMC and TSMC are now the unquestioned leading players in foundry. They are. But that doesn't mean anything is taken for granted at UMC. The company is on the move, driving the world's first 300 mm production silicon at Trecenti Technologies, its joint venture with Hitachi, and pioneering 0.13 µm process technology in partnership with IBM and Infineon.

The Trecenti joint venture has had significant success with its first SRAM products, and the Trecenti 300 mm operation is now running customer products. Meanwhile, industry analysts have been quick to point out that the move to 300 mm will prove a costly one for any company. The high cost of 300 mm wafers could offset gains in overall productivity. UMC Chairman John Hsuan is undaunted, and he expects at least a 30% cost advantage in the move to 300 mm.

"Previously, the expectation was that a single 12 inch raw wafer would cost approximately $1000, but already we see that figure down to $400, so we believe that when we start 300 mm production the cost of a raw wafer will be down to $300 or less," Hsuan said. "If that proves to be the case, then we expect to break even by the time we are managing 8000-10,000 wafers per month."

Liou's expectation is that 300 mm production will unquestionably improve UMC's competitiveness. While 300 mm presents formidable challenges — "It is commonly agreed that automation is the major challenge for 300 mm fab manufacturing" — UMC's partners, Hitachi and Infineon, both have considerable experience in 300 mm production and, Liou said, "The team as a whole will benefit from each other's background in the technology." Hitachi, he pointed out, was one of the original members of the Selete (Semiconductor Leading Edge Technology) program in Japan, and Infineon gained a background in 300 mm in work with Motorola.

1. Now producing customer products on 300 mm wafers, UMC hopes to break even on cost by the time it is running 8000-10,000 wafers a month. (Source: UMC)

That "next line" will be Fab 12A, at the new science-based Industrial Park in Tainan, southern Taiwan. UMC's Tainan site, at 40 hectares, is large enough for five 300 mm fabs, Hsuan pointed out. "The original idea was to add one new fab each year, but in practice this will depend on the learning curve for 300 mm, and on market conditions."

UMC broke ground on a third 300 mm wafer fab in April. Located in Singapore, UMCi Pte. Ltd. is a joint venture with Infineon and Singapore's Economic Development Board.

The total number of electronic devices sold may continue to increase, but the growth rate of the traditional PC will be low or flat because the original pattern of consumption is being replaced by a new one. "For that reason, I think high-end, high-performance processors will also see much slower growth, while the numbers for the embedded RISC type of processors will grow strongly — although embedded RISC will not need a huge amount of system DRAM," Hsuan said.

2. UMC recently broke ground on a third 300 mm wafer fab — a joint venture — in Singapore. Pictured is an artist’s rendering of the facility. (Source: UMC)

For Liou, the changing marketplace is resulting in a general long-term trend to SoC (system-on-a-chip) solutions. The issue here is whether foundry is suitable for SoC manufacture, with some industry observers declaring it is not. Liou is affirmative on this question. "I also have seen reports saying foundry is not suitable for SoC manufacturing, but I do not agree. Rather, I believe that foundry technology is now very advanced, and that the portfolio has undergone a very significant expansion from its first involvement with logic products. As well, chip scaling continues to advance."

Liou detailed how UMC's foundry business has advanced from logic, through mixed signal, SRAM, rf, flash and DRAM to BiCMOS, while simultaneously the technology has scaled from 0.25 through 0.15 to 0.13 µm, with 0.10 µm process technology on the horizon. As the technology advances, he noted, more and more devices are integrated on the chip, constantly adding to the technology portfolio, both in density and speed. "When you look at all this, UMC is clearly in a position to offer the best platform for SoC manufacturing, with a common platform and IP verification through programs such as our Silicon Shuttle, which enables IP reuse by customers."

When you avoid redesign through IP reuse, you immediately gain in time-to-market in an industry where chip complexity is increasing as market cycle time is decreasing, Liou emphasized. "That's why we have this common platform, to advance the portfolio in the interests of our customers. Foundry will be able to link with the IP companies to provide these technologies."

Liou concedes that IDMs will continue to have some advantages in SoC manufacturing, and for that reason he thinks SoC technology will be strongest in Europe and Japan, simply because there are more system companies there. That means UMC must work closely with the IDMs to create better opportunities for SoC products, according to Liou. "UMC has partnered with Japanese and European companies quite heavily. UMC is already well prepared for the SoC era," he said, citing UMC's relations with Hitachi, Infineon, STMicroelectronics and Alcatel as examples in which UMC has already been involved in the manufacture of system chips.

With at least two new companies from Taiwan having recently declared their ambitions to become operational foundries (on the Chinese mainland), the industry will now be watching to see what happens. Will companies such as UMC and TSMC feel the heat as world demand for foundry services continues to accelerate, or will the established players maintain an effortless technology and market leadership while the newcomers find the barriers to entry simply too high?

3. Technicians check systems at UMC’s new Fab 8C. (Source: UMC)

"The critical mass is already there for UMC," Hsuan said. "The only way a new foundry might start up is by trying to build up its legacy or niche business, whereas UMC is trying to exit those areas. Right now, the SoC concept, and with it the building of IP blocks, is the trend in design. But when people migrate to SoC production, they won't want a niche platform. Everything will need to be on the same technology platform."

Hsuan also feels that, if existing IDMs want to try to transform themselves into foundries, they will find it difficult. "It was certainly a painful process for UMC, during the transition years of 1995-2000, and that was in a good IC demand cycle, accompanied by successful joint ventures."

Also, foundries must earn the trust of their customers, and that can take time. "In our case, UMC customers are free to visit our offices and fabs, and when they do, they see we have no preference among companies," Hsuan said. "In this way we get their trust."

Liou also thinks that, although foundries have proved highly successful in meeting the needs of an increasingly complex industry, the barriers to entry will remain high. The foundry business is heavily oriented to advanced technology, he noted, as well as customer service. High technology imposes formidable entry barriers, while successful customer relations are all about meeting the demands of the productivity cycle time. It's a difficult equation to get right. "The technology demands that you make a process shrink with each generation, and customers will need to see this," Liou noted. "If they don't see it, they will seek other solutions. This will be the challenge for new entrants to foundry."

Hsuan pointed out that, beyond the formidable technology barriers, a successful foundry business is all about service, efficiency and driving down costs for customers. UMC has managed to achieve these goals by a variety of means — establishing a common design platform for 0.13 and 0.10 µm processes, for example. The company aligned with IBM and Infineon to create a standard. With third-party IP houses choosing to build upon the standardized process, companies will be able to achieve an early time-to-market, Hsuan said. "If they have capacity issues, they can go to either IBM or Infineon."

Then there is the ever-increasing cost of masks. "In the past, there has been no second source in the foundry business, and companies always needed to rebuild masks," Hsuan noted. "With our Silicon Shuttle system, customers will have the opportunity to improve their design and IP and jointly share engineering and mask costs. Under this system customers are allowed to keep their masks; they do not need to make a new one [each time]. A 300 mm wafer costs a lot. Using the Silicon Shuttle, each customer can share 5% of the total costs."

Hsuan also pointed out that UMC is now offering e-services. Normally, a customer order must be checked with production control and then confirmed with the customer, which takes time. But through UMC's Web site, my.umc.com , customers can have their orders confirmed immediately. This on-line service uses ATP (Ability To Promise) software, developed in-house. "It works rather like an airline reservation. You click in your requirement, and it is confirmed in about 30 seconds. The Internet is a very efficient tool," Hsuan said.

"We hope that one day we can offer some form of virtual review process, via the Internet," he added. "When customers do layout and design review, we can participate through the Internet." But this would not be simply a conferencing procedure. Rather, the reviews would stimulate some kind of "think tank" on the UMC side and provide knowledge not simply for an individual but for a whole group.

The complete version of this article originally appeared in Electronic Business Asia , February 2001. Check it out at www.eb-asia.com.