TAIPEI, Taiwan In a push to diversify, foundry United Microelectronics Corp. is gearing up for a drive into CPUs and NAND flash memory production on its 65-nanometer and 45-nanometer technology nodes. The company said it is in talks on a CPU production deal, but it is being circumspect on how it will attack the high-volume flash memory market.
Saying UMC "cannot ignore" the rapidly growing commodity flash market, UMC chief executive Jackson Hu plans to push UMC to pursue deals in both NAND, NOR and SRAM, but not DRAM. Hu was tight-lipped on details, but hinted that UMC doesn't intend to wade too deeply into the turbulent waters that typify competition in the commodity flash market.
"NAND and NOR have huge fluctuations in price, so that is something we have to consider very carefully," he said in an interview with EE Times at the company's Hsinchu headquarters. "Clearly, the high-end, high density area is a very challenging board game. It takes significant investment and the technology moves much faster."
Hu acknowledged that opportunities in the commodity NOR market may be limited. That may result in the foundry expanding its business in embedded NOR, similar to a deal it struck with Cypress Semiconductor in March. Cypress plans to shift its 0.13-micron S8 embedded-flash technology and two future generations of embedded-flash technology to UMC. The foundry will also make advanced SRAMs for Cypress, the first of which will be made on a 65-nanometer process later this year.
In CPUs, it looks like UMC will also take a step-wise approach and avoid the significant resource investment needed for production of bleeding edge CPUs. "Honestly speaking, if you try to aim for high-end silicon on insulator for desktop, that goal may be too high to reach," he said. "So we will do bulk CMOS and that automatically translates into low-power processors."
UMC is currently developing a silicon-on-insulator process, which should be introduced later this year on the 65-nm technology node. "We believe that down the road, maybe at 32 nanometer, that SOI will become more important, and therefore from a technology development point of view we need to have that knowledge and experience," he said.
UMC also plans to have several customer tape-outs at the 45-nm technology node during the second half. Typical of advanced nodes at foundries, FPGA makers will be the early adopters, as well as graphics chip designers. In particular, Hu noted cell phone chip designers looking to cut costs will also quickly move on to 45-nm. "You will have the side benefit of higher performance and lower power for sure, but the primary reason is for cost reduction," he said.
Like the transition to 130-nm, Hu said the shift to 45-nm is fraught with challenges because of the new materials and equipment being introduced. UMC struggled more than its main rival at 130-nanometer, but it recovered well at the 90-nm node. "Because of the bad experience at 130nm, the industry is more cautious. But based on my understanding, the chance of success at 45nm is much higher than before," he said.
Despite growing participation in group process development, UMC will continue to go it alone. "Technology development is not a team sport," Hu said. "Group development has its own limitations because each company has its own priority."
Yet he said it is inevitable that IDMs and foundries will continue to work more closely together because of the rising costs of process development and fab ownership. That doesn't mean co-process development to Hu, but rather co-development of process and product specifications to ensure that devices can be made in high volumes at a general-purpose foundry.
Hu also touched on the topic of consolidation in the foundry industry, saying that it is sorely needed. He said UMC is watching a wide range of seemingly unrelated developments, such as Samsung potentially increasing its foundry business, India showing interest in chip making, and the introduction of a "rent-to-own" or "managed foundry" by Semiconductor Manufacturing International Corp.
"Those are spotty events and whether they will produce significant results is to be watched," he said. "But I do feel that there is too much capacity on the mature technology because in the past several years companies have rushed to build in places like China and now they clearly realize that building the fab is the easier step. Filling them is the tougher part."
Hu said UMC is hoping to increase its footprint in China by asking the Taiwan government to let it work more closely with affiliate HeJian Technology. UMC has a 15 percent stake in the Suzhou, China-based fab but it has not been able to put it on its books because the fab was producing wafers at 0.18 micron, considered illegal until recently by the Taiwan government.
Although UMC will ask the government once again for permission, Hu did not seem optimistic that approval would be quickly granted. However, once UMC is able to officially work with HeJian, he said UMC would consider acquiring the foundry.