Users must define comm-specific DRAMs, panel says
By Craig Matsumoto, EE Times
Mar 13, 2001 (7:32 AM)
URL: eetimes.com
PHOENIX — DRAM vendors will need to work more closely with communications systems OEMs to define new comms-specific memories, according to panelists speaking Monday (March 12) at the Semico Summit. Tomorrow's DRAMs will not be as homogeneous as they were in the past, nor will they follow vendor-defined standards such as the Rambus DRAM interface, the speakers said. Rather, "we need to have the application-specific DRAM," said Tokumasa Yasui, executive vice president of Elpida Memory Inc. (Tokyo). "Not only the DRAM vendor, but also the user [the systems OEM], must be involved."
DRAM vendors have already been forced to make changes as demand from the PC sector falls while rising in the communications sector. Projections by Semico Research Corp. show that communications applications will make up 29 percent of the DRAM market in 2005, up from up from 13 percent in 2000. Consumer electronics will run a close second, at 25 percent of the DRAM market in 2005.
In contrast, the desktop PC market will shrink to 12 percent of the total DRAM market in 2005, down from 36 percent in 2000, Semico said.
The shift away from PCs changes the balance of power in DRAM requirements, according to the Semico Summit panelists. While PC standards were essentially dictated by Intel Corp., communications standards have no centralized point of origin. Moreover, the needs of communications OEMs vary: handheld devices require low-power components, while routers and switches strive for the largest and fastest memories possible.
As a result, DRAMs vendors no longer pursue a one-size-fits-all model, said Sanjay Srivastava, chief executive of Denali Software Inc. In this way, it tracks the specialization seen in the flash and SRAM markets, which have spawned niche parts such as CompactFlash and quad-data-rate SRAMs, he said. Similar specialty DRAMs, such as the Windows RAM or the Rambus DRAM, haven't met the same success as those niche parts, Srivastava said.
Panelists agreed that specialized DRAMs need to be developed with the cooperation of systems vendors. In that regard, vendor-driven collaborations such as the JEDEC Solid State Technology Association or the Advanced DRAM Technology consortium have been lacking, because "in all of these, users are not playing a central role," Srivastava said.
Speaking separately with EE Times, Yasui outlined his ideal solution to the problem: a collaboration including the major DRAM vendors and all the major OEMs of a particular sector — something not possible under Intel's PC regime.
"I don't want to work with [only] Intel," Yasui said. "I want to have direct talks, all the end users and the DRAM vendors in a room together, to find the logical solutions." Citing 3-D graphics accelerators as an example, Yasui said he would collect the major vendors, such as ATI Technologies Inc. and Nvidia Corp., together with the major DRAM players to discuss the road map for future graphics memory.
A single company such as Elpida could individually contact systems vendors and develop a proprietary DRAM. But such a part would be a niche product, Yasui said, whereas a part developed with other DRAM makers could become a high-volume part that drives future industry growth.
Separately, panelists agreed that the large DRAM makers of today are the likely candidates to produce the more specialized parts of the future. This won't be a venue for small players, because the requirements to survive in the DRAM market include technology, a design team that can turn products quickly, and "large financial resources," said Jan du Preez, president of Infineon Technologies North America Corp. (San Jose, Calif.).
"It's not just saying you can manufacture 300 mm and go to 0.15-micron. You have to be able to design a product and meet customer needs," du Preez said. "I don't think you'll see changes in the top five now, or five years from now." |
