TOKYO Japan's top five chip makers may form a giant foundry consortium and build a 300-millimeter fab capable of producing 40,000 wafers a month as early as late 2003, if they can reach agreement on the project this year.
The plan, promoted by Nippon Foundry Inc., the Japanese subsidiary of UMC Group, has been circulating among the top management of NEC Corp., Mitsubishi Electric Corp., Toshiba Corp., Fujitsu Ltd. and Hitachi Ltd. since the beginning of this year, NFI president Yukio Sakamoto told EE Times.
"You need annual semiconductor revenues of $7 billion or $8 billion to build a 12-inch line and no one company can do it," Sakamoto said. "I'm asking the top five Japanese semiconductor companies for 30 billion yen $261 million each. The IDM-foundry relationship is changing and we need a new model."
If an agreement can be forged to raise 150 billion yen from the five companies, NFI will match that funding with 150 billion yen of its own, giving the consortium the funds it needs to build a cutting-edge plant. The fab would be able to produce SRAM, flash, DRAM, and system-level chips on UMC's 0.13-micron process in 2003. About half the fab's projected monthly capacity of 40,000 wafers would switch to 0.10-micron processing as early as 2004, according to an outline of the plan given to EE Times.
The proposal calls for NFI to conduct day-to-day operation of the plant. Decisions on production by individual chip makers would be made on a case-by-case basis under an aegis of avoiding a cumbersome layer of management. Individual capacity allowances would be based on an individual company's rate of investment. Production schedules would be fixed for six months, after which members would be allowed to decrease production by up to 50 percent, according to the outline supplied by NFI.
"We want members to use UMC's process, but if they don't want to, we'll be able to use theirs," Sakamoto said.
But the telling point, he said, would be cost savings, with each company risking only a tiny investment compared to the cost of building an individual 300-mm fab, while gaining access to 12-inch production, especially if the companies concentrated on larger, 100 cm2 devices such as CPUs and DRAM.
"The most important thing is the cost reduction," Sakamoto said. "Members can totally ignore the initial investment" and concentrate their scarce resources on engineering and product development, two areas that Sakamoto said badly needed shaking up. "Resources for systems-on-silicon need so many applications engineers," he said. "Unless they Japan's semiconductor industry create products and markets together, they can't succeed."
Further cost savings would result from building out the shell of an older plant, taking advantage of its ready-built infrastructure, he said.
"We haven't looked at the candidate sites yet and we may use phased-out sites," he said. "Japan has many 4-, 5-, and 6-inch fabs, for example, that can be demolished. If we could get a site like that, it would also help that company's restructuring costs."
Earlier this month, NEC refurbished an old fab and opened it as a new LCD driver circuit line at a cost of $200 million, and made the change in only six-and-a-half months. Beyond that, Japan's semiconductor makers have shown recent willingness to carry out some restructuring and teamwork. The creation of Elpida Memory Inc. from the combined DRAM operations of NEC and Hitachi is one example, and Elpida is now building its own 12-inch fab. But such moves miss the point, according to critics such as Sakamoto and Sony Corp. president Nobuyuki Idei, who contend that Japan's electronics industry must first solve deep-set structural problems and form more strategic alliances to rebuild the market share they lost in the 1990s.
While Sakamoto presented the 300-mm foundry project as flexible and workable, he said meetings with executives from Japan's big five, plus Sharp, Matsushita and Sony, led him to estimated it had a "50-50" chance going through.
"Everyone says that it is interesting, but that now is not the time to invest in new capacity. But now is the time to invest," he said, arguing that UMC's ambitious capacity buildup was correct even in the face of a slump in demand that some already characterize as a recession.
Sakamoto's own predictions were dire. While the consensus opinion of Japan's memory makers holds that bit demand will rise over 60 percent in 2001 and that memory prices will recover in the second half of this year, Sakamoto said he expects PC sales will fall 10-to-15 percent from last year. With PCs absorbing about 40 percent of all semiconductors, this will prove a crushing blow, limiting semiconductor growth to 0-to-3 percent overall for 2001. Further, Sakamoto said the real levels of inventory for cellular phones had built to 50 million units.
"Last year there were 370 million cellular phones bought but there were 400 million made, while parts procurement was for 420 million. Even if the cell phone market is up 25-to-30 percent this year, it's not enough" to burn off inventory, he said.
"Semiconductors for cell phones and PCs make up 60 percent of the market, and networking is really cooling. It's going to be very, very tough."
Sakamoto also confirmed reports in the Taiwanese press stating that UMC's two biggest customers, Xilinx Inc. and Altera Corp., have heavily slashed their foundry orders buy up to two-thirds.
Against this backdrop, UMC is believed to be scaling back initial production at its $3.5 billion 300-mm plant in Singapore, which UMC is building with Infineon Technologies. While Sakamoto would not confirm the exact reduction or UMC's initial production plan for the fab, the cut is believed to be as high as 20 percent.
But UMC is sticking to its guns in every other area, Sakamoto said. There are no plans to push out the Singapore fab, which will start groundbreaking this spring, with volume production on a 0.13-micron process coming in 2003.