Future unclear for specialty DRAMs
Chipmakers may opt to outsource as business improves
By Jack Robertson
EBN
(07/19/02 02:39 p.m. EST)
Makers of specialty DRAMs for communications and networking applications could end up outsourcing the chips rather than producing them in-house, according to some participants at the Platform Conference held here this week.
The specialty DRAMs, including fast-cycle and reduced-latency memory, generate only low-volume orders, so while vendors are happy enough to push them in today's slow market, they may have second thoughts if they need capacity to meet the needs of key customers when the DRAM business rebounds.
Sherry Garber, an analyst at Semico Research Corp., Phoenix, said when the market improves, many DRAM makers will shift production of the specialty chips to foundries.
“It makes sense with the limited volume of communications and networking DRAMs to move that output to foundries,” Garber said. She estimated that specialty chips in 2002 will constitute no more than 3% of the total DRAM market, and will only grow to 5% to 8% by 2006.
Bob Eminian, former Samsung Semiconductor Inc. marketing executive and now president of consultant Creor Group in Campbell, Calif., concurred that specialty DRAMs are likely to migrate to foundries.
“When the DRAM market is down, vendors will sell any product. When the market turns up, companies will devote full resources to their main lines. The staying power of specialty DRAMs may rest with the foundries.”
Jim Cantore, an analyst at iSuppli Corp. in San Jose, believes that second-tier DRAM producers will pick up on specialty chips to fill any void created by major DRAM makers pulling out of them.
But for now, at least, most of the Big Five DRAM players vow they won't back off the new memory versions.
“Samsung firmly believes that the special needs of many different kinds of users require different types of DRAMs,” said Mian Quddus, director of technical marketing at Samsung Semiconductor in San Jose.
Switches and routers, for example, need special DRAMs to provide more packets per second-up to 125 next year, from 31 last year, Quddus said. Mobile phones need low power and high data rates, and digital TV and set-top boxes need low memory density but very high performance, he added.
“No longer will one size of DRAM fit all applications, and Samsung will make each type the market needs,” Quddus said.
Jim Sogas, vice president of sales at Elpida Memory (USA) Inc., Santa Clara, Calif., doubts that memory makers will turn to foundries to make the specialty parts.
“We can make them as cheaply as foundries can. I don't see any advantage in foundries,” Sogas said.
Chee Ho, Infineon Technologies North American Corp. director of product marketing, San Jose, said, “Volumes are very small. You have to be able to make a business selling only a few million pieces a year.”
The dilemma for makers of low-volume specialty chips is how fast they can recover their investment, according to Lane Mason, an analyst at Denali Software Inc., Palo Alto, Calif. The devices require different designs than mainstream DRAMs, even though they use the same cell structure. On-die controllers are different, and often the specialty chips are larger than conventional DRAMs.
“If any of these chips fail to generate sufficient return for a vendor, they will end up a one-generation product,” Mason said. “Or you will start to see some consolidation of different specialty chips as a few survive.”
Memory makers had hoped that OEMs would pay premium prices for the lower-volume specialty parts. But this isn't always the case.
Elpida's Sogas said communications OEMs are taking advantage of the slow DRAM market to pressure for the lowest price possible, even though order volumes are low. EMS companies, he said, are attempting to get similar discounts for specialty chips that they or their OEM partners obtain on mass-volume PC DRAMs.
Yet another issue facing makers of specialty DRAMs is the need for an upgrade path. Infineon's Ho called for chipmakers to come up with roadmaps for two or three generations of specialty chips. “Otherwise, you end up re-inventing the wheel over and over again. This can't just be a specific product that goes away after one generation.”
Other conference panelists, however, argued that the six- to seven-year lifecycle of communications systems was long enough for memory companies to prepare follow-on devices.
“This isn't the PC mentality on product cycles. In fact, the long life of specialty products in these markets should lead to greater stability in DRAMs,” said Steve Chen, vice president of sales at Rambus Inc., Los Altos, Calif.
Sogas questioned whether communications customers were locking themselves into proprietary specialty DRAMs now on the market. Elpida is considering a possible XDDR-II design that would meet the special needs of communications and networking customers, but in an open-architecture device. He said the company has not yet decided whether it will go ahead with the DDR-II variant. |
