Diversification rewrites DRAM maxims
By Brian Fuller
EE Times
(03/07/00, 1:35 p.m. EST)
PHOENIX ? The large number of DRAM flavors has led to a handful of suppliers able to sustain the intense capital and R&D demands of serving a few high-volume markets, while the rest of the old-line memory vendors fight for the peanuts of niche businesses.
That was the take of panel members at Semico Research's annual Semico Summit here on Monday (March 6).
"DRAM is no long a one-size-fits-all commodity," said Dean Klein, vice president of the Integrated Products Group at Micron Technology Inc. "Memory is likely to become . . . a strategic component of each design."
Klein, whose company surged to the lead in the DRAM market since it acquired Texas Instruments Inc.'s memory business, said it will take enormous resources to tackle any large market going forward. But second-tier companies can't handle those financial resources or won't handle them because the DRAM depression of the past two years has left them shell-shocked, he said.
Keith McDonald, president of Smart Modular Technologies Inc. (Fremont, Calif.), took a slightly different tack on the same issue. A module maker can only support so many variations of increasingly varied DRAM architectures and hope to make money, he said.
In the past, four module types accounted for 80 percent of the PC100 memory business. But the market now also includes 200-MHz and 266-MHz version of double-data-rate Rambus DRAMs, and various SDRAM flavors, which leaves a module maker with at least 22 variations to possibly sell.
The market is tricky "when you're in 5 percent supply-demand ratio in DRAMs," McDonald said. But trying to forecast supply and demand for three different DRAM types is almost impossible, he said.
For the first time over the past few years, panelists acknowledged that pricing isn't the only thing that keeps them awake ? although recent memory pricing has been shaky at best. Semico researchers estimate the worldwide DRAM module market this year will reach $27.7 billion and escalate to $41.4 billion next year ? just a fraction over the industry's record high of $40.8 billion in 1995.
Apart from pricing, technology issues are now a concern, participants said.
"Some applications will require blazing logic speed, others chunks of memory," said Micron's Klein. And embedded memory will always find a home, for instance, in ink-jet printers with 60-MHz processors and a need for 2 Mbytes of memory, he said. "You can integrate a small piece of DRAM in this application. It's a perfect marriage. You can save a lot of cost in that equation."
Diversification dilemma
DRAM diversification doesn't bode well for the push to 300-mm wafers, which were once forecast to ramp up in 1999 and are now just starting to garner semiconductor-gear production capability.
"This [diversification] does not look good on a 300-mm line right now," Klein said. "We can continue on eight-inch wafers by improving geometries, shrinking dice, squeezing the equipment and really not flood the market with any particular part type."
Panelists also agreed that there is a place for Rambus DRAMs in the drive for differentiated products.
"Today you take jumps at 64 megs at a clip," said Smart Modular's McDonald. "When you go forward, Rambus allows you granularity in 32-meg increments."
He noted that Sony's Playstation ? first seen as a great game platform and now seen in its latest iteration as the consumer's TV-based entry to the Web ? takes two pieces of Rambus DRAM.
"You'd have to use 8-to-16 pieces of other DRAMs in that system. Is Rambus a premium? I say no because you have fewer components," he said.
eet.com |
