Renewed growth should bring end to DRAM pricing free fall in ‘99
By Andrew MacLellan
Electronic Buyers' News
(12/18/98, 02:14:24 PM EDT)
The DRAM market appears headed for another wild ride in 1999.
Dwindling capital investments, rising equipment costs, and an unwieldy production mix will complicate the DRAM picture next year, just as OEMs begin to decide the fate of several new high-speed architectures.
After cresting at more than $40 billion in 1995, DRAM revenue fell below $15 billion this year, triggering a wave of industry consolidation, deferred investment, and ASP erosion.
For buyers, the fallout was manna. OEMs gorged themselves as DRAM tags dropped an average of more than 30% in each of the past three years. Since 1995, the average main-memory capacity of a PC has increased from 8 to 64 Mbytes, according to Semico Research Corp., Phoenix.
But the pricing free fall has lost its momentum, a number of DRAM makers and analysts now say. In a trend reversal, 1999 should bring renewed revenue growth of as much as 30%, according to Dataquest Inc., San Jose.
Whether stronger sales will translate into better earnings for cash-poor DRAM makers is uncertain. But one thing is clear: Pricing for new technology will present OEMs with a confusing landscape next year, given the number of DRAM options in the wings.
Having introduced just two DRAM versions since 1996, suppliers next year will ramp up three separate lines, letting OEMs choose between PC-133 SDRAM, double-data-rate SDRAM, and Direct Rambus DRAM. An informal poll of DRAM manufacturers garnered differing opinions as to how each type of memory will play out in the market.
“With a revolutionary architecture, forecasting becomes not a science but something of a mystical feat,” said Stephen D. Marlow, vice president of memory operations for the semiconductor group of Toshiba America Electronic Components Inc., Irvine, Calif.
In a general scenario, Direct RDRAM will target the high-performance end of the PC spectrum, while DDR SDRAM is slated for workstation and server applications.
PC-133 SDRAM, meanwhile, could be a wild card next year and into 2000. A natural speed extension achieved by shrinking PC-100 SDRAM wafers to sub-0.25-micron linewidths, PC-133 is expected to serve a growing percentage of the mainstream PC market.
PC-133 memory supports an emerging 133-MHz front-side bus, but is not found in Intel Corp.'s roadmap, which moves from PC-100 SDRAM straight to Direct RDRAM. In keeping with this strategy, Intel has disclosed no plans to support PC-133 with a separate chipset.
However, PC-133 support will soon be available from several third-party chipset vendors, which will ensure the longevity of the SDRAM family. LG Semicon Co. Ltd. confirmed that at least 20% of its SDRAM output next year will consist of PC-133 chips, while Hitachi Ltd. said that on an annualized basis as much as half of its 64-Mbit capacity will run at 133 MHz.
PC-133 “will find enormous penetration on the desktop,” said Jim Sogas, marketing manager for memory products at Hitachi Semiconductor (America) Inc., Brisbane, Calif. “It's very easy to use, and all of the DRAM makers who are shrinking [their wafer linewidths] will have 133 right out of the chute.”
NEC is also planning to ramp PC-133 SDRAM, and is looking at the higher-speed chips as an interface option for its low-cost Virtual Channel Memory (VCM) DRAM core.
“VCM will have a 133-MHz offering,” said Will Mulhern, product marketing manager for advanced DRAMs at NEC Electronics Inc., Santa Clara, Calif. “Our position with VCM is to target it to a market that's already using SDRAM and plans to use SDRAM through 2001.”
Combined with a low-latency core such as VCM, PC-133 SDRAM could enable system OEMs to design a single memory type into their notebook and desktop PCs, workstations, and servers, according to analyst Victor deDios of deDios & Associates, Newark, Calif. This differs from other new DRAM, which is aimed initially at specific performance segments.
And the chips have another potential application. If Direct RDRAM is in short supply, PC-133 could serve in the interim by operating in Rambus-enabled systems with the aid of a specially modified Rambus memory module known as a Synchronous RIMM.
“PC-133 is a natural fallout of our next-generation sub-0.25-micron process,” said Michael Sporer, technical marketing manager for LG Semicon America Inc., San Jose. “If S-RIMMs become necessary, I believe PC-133 would be a natural part to build that product.”
Following years of negative margins and increasing front- and back-end costs, many IC makers say DRAM oversupply is beginning to dry up despite an ongoing spate of die shrinks.
Toshiba, for example, expects a relatively stable supply/demand scenario to unfold in 1999, with the scales tipping from 5% oversupply to 5% undersupply as the year wears on. Toshiba projects that total demand for Rambus chips next year will be about 200 million 64-Mbit equivalents, but supply will amount to only about 120 million units.
DDR should also be in short supply, but will account for just a tiny fraction of next year's 2-billion-unit DRAM market. While companies such as Hitachi and Fujitsu Ltd. believe DDR will challenge Rambus for desktop sockets, others say these memory types will serve different segments.
“I don't think there will be a battle between DDR and Rambus,” said Avo Kanadjian, vice president of memory marketing for Samsung Semiconductor Inc., San Jose. “I think DDR has found its niche in applications where customers have the ability to build their own chipsets, such as in high-end servers.”
ebnews.com |
