SANTA CLARA, Calif. Discussion of China's emerging CMOS foundries has centered on their perceived role as low-cost suppliers of last-generation capacity. But to Elpida Memory Inc., the DRAM company forged from the memory divisions of Hitachi and NEC, Chinese chip lines are part of a global strategy based on a radical segmentation of the world DRAM market. And the strategy holds lessons that may prove important to the survival of large semiconductor manufacturers in other markets as well.
Elpida announced earlier this month that it will fab its 0.13-micron, stacked-capacitor DRAM product at foundry partner Semiconductor Manufacturing International Corp. (SMIC; Shanghai, China). In turning out Elpida DRAMs, SMIC will join an NEC fab in Japan, a Hitachi/Nippon Steel fab in Singapore and an Elpida-owned fab in Japan.
The plan, said Elpida's new chief, Yukio Sakamoto, is to move Elpida's 0.13-micron process from its Japanese fab to SMIC and to begin production before year's end. Sakamoto described it as a pure foundry relationship, not a technology exchange, since the contract stipulates that SMIC will use the Elpida process only for Elpida DRAMs.
Sakamoto Elpida's president, CEO and representative director said that the cost structure on the SMIC-built DRAMs will be substantially lower than for Japanese-built parts. But this is not the primary reason for the relationship, Sakamoto said in an interview here. In fact, Elpida does not plan to ship SMIC-built parts out of China. Rather, the deal is part of a more strategic view of the global DRAM market.
As a DRAM specialty house, Elpida depends on foundry and jointly owned capacity for much of its product and is beholden to one of the most volatile price structures in the industrial economy. But, Sakamoto pointed out, it's not the entire DRAM market in which prices are cyclic. The volatility is primarily in the high-volume personal-computer segment of the DRAM business. Specialty and high-performance DRAMs tend to have higher, much more stable margins.
Elpida's strategy hinges on the observation that the most volatile segment of the market is almost entirely contained now within China. Much of the PC motherboard and memory add-in business has moved from Taiwan to the mainland and, given current cost structures, the rest will likely follow. So it can make a great deal of sense to put a DRAM fab in China, where it would provide local content to help manufacturers avoid a very stiff tax, where the logistics of serving Chinese assembly facilities is vastly easier and where it would benefit from the same cost structure as its customers.
This cost advantage, Sakamoto said, has little to do with low labor costs. Labor is a relatively small component of wafer cost. But the Chinese government has been aggressive in attracting foundry business, and that translates to a lower cost structure overall, through means that are not easy to trace.
Very different customers
Within China, Elpida will employ two means of distribution. "There are two very different kinds of customers in China: branches of multinational companies, and locally owned companies," Sakamoto said. "It is relatively easy to serve the multinationals with our direct sales organization. We get the orders from headquarters and ship DRAMs directly from SMIC to the plant. For locally owned companies, we believe that local distributors will be much more effective. So we will use both direct sales and distribution."
Sakamoto said that working with executives at SMIC had proved satisfactory, unlike the protracted negotiations traditionally required when doing business with Chinese companies. "All of the executives with whom I was dealing have been in major companies in the U.S. or Taiwan. My direct counterpart at SMIC was originally at Texas Instruments," he said. "The negotiating process is Westernized; in some ways, it is easier than doing business with a Japanese company."
The executive sees the new face of Chinese business as part of a trend. "If you are a process engineer in Taiwan, you can't get rich. You have to move to a fabless startup in Taiwan or move to a new foundry in China to make a lot of money. There is a lot of flow of experienced people people who have worked in the U.S. and Taiwan to the Chinese companies."
Sakamoto's strategy overall is to keep the differentiated, leading-edge DRAMs in Elpida's captive fabs, and to serve the volatile PC segment from its China source. That will isolate the company from the excessive swings in commodity DRAM pricing, he said, while giving it the freedom to concentrate engineering resources on the profitable leading edge of the business.
He is betting a lot. For one, Sakamoto has said publicly that he will resign if he hasn't made Elpida profitable by the fourth quarter. But a larger issue is at stake as well. Sakamoto, 55, feels strongly that Japanese industry is in a malaise brought on by the inability of an aging, deeply conservative generation of Japanese managers to adapt to the times. "People want to blame Japanese deflation on the banking system. They say if we fix the banks, things will get back to where they were," he said. "But I don't agree. Deflation isn't from the banks. With globalization of industry, prices are heavily influenced by developing countries. As long as they are willing to sell product for less to get into the world market, prices will go down.
"If you look at it that way, deflation isn't necessarily negative. It's good for the countries that are growing in the world market. It's good for people who have fixed incomes: they can buy more. Japanese people used to shop wherever else in the world they went. Now Taiwanese people come to Japan to buy clothing made in China.
"The point is that you can't just wait for deflation to go away. You have to understand how the global economy is working, have to learn to make money in the new situation."
Certainly the markets will watch Elpida's results. But so, too, will the ranks of conservative Japanese executives, one suspects. For them, Sakamoto may have important news.