Next-gen memory modules ready to roll
Jan. 29, 1999 (Electronic Engineering Times - CMP via COMTEX) -- The
coming year will be a watershed for memory technology as chip and
module makers start fielding the next generation in DRAMs.
Throughout the decade, DRAMs have steadily evolved through multiple
generations of higher-performing parts: from conventional DRAMs to
fast-page-mode (FPM) versions, then to EDO (extended-data-out) DRAMs
and finally to synchronous DRAMs (SDRAMs). We're now on the cusp of a
move to the next iteration, but which candidate technology will emerge
as the heir apparent has yet to be determined. Thus, memory-module
makers are positioning themselves to move rapidly in whichever
direction the market tells them.
With a majority of DRAM sales going into the PC market, the DRAM
arena has become, not surprisingly, very PC-centric. Noting the heavy
influence of Intel Corp. on PC evolution, some memory-module makers
have concluded that the Intel-sponsored Direct Rambus DRAM will be the
next big thing, and that's where they're focusing their efforts.
Yet other memory-module makers note the appeal of avoiding
revolutionary technology changes as long as possible, such as those
inherent in Rambus and Synclink (SL) DRAMs, and of pushing established
technology as far as possible. Just as 66-MHz SDRAMs passed the baton
to 100-MHz SDRAMs last year, they expect the next big thing to be a
move from 100-MHz SDRAMs to 133-MHz and faster SDRAMs. Beyond that,
they say, SDRAM technology still has a fair amount of headroom for the
future, with double-data-rate (DDR) SDRAMs about to get real.
No one can be sure how the coming year will pan out among the
unprecedented variety of DRAM contenders. "One year ago, EDO DRAM was
entrenched in the PC space and now it's essentially dead," said Bob
Johnston, vice president of marketing at Smart Modular Technologies
Inc.
Bosco Sun, president of Camintonn Corp., said that the SDRAM took the
market by storm last year. "The transition happened quite rapidly. In
the last six months or so, of all new PCs coming out, 90 percent or
more required SDRAM."
Whether SDRAMs will remain the dominant DRAM type for main memory
and, if so, for how long are open questions. Clearly, though, DRAM
evolution will depend on two interrelated issues: volume and price.
These in turn will depend on supply from chip makers and demand from
equipment makers, and not on module makers. "We really don't know what
will happen and what will happen first," said Sun. "All we can do is
monitor the trends, work on them all and then jump on whatever
bandwagon develops." Besides chips and system sockets for the new
memories, of course, infrastructure items also must be put in place for
a market to take off: logic and control chips, for example, appropriate
assembly and test equipment and completed specifications.
The pricing picture on the next wave of memory modules is also
unpredictable, the vendors said, citing the chaotic nature of memory
chips. "If I quoted you a price and you printed it next week, it might
no longer be accurate," said Walter Juras, Hitachi's DRAM product
marketing manager. "The market can fluctuate by up to 10 to 15 percent
at a time and it changes daily, so it's difficult to talk about any
kind of pricing about any particular type of module."
As a rule, though, pricing on the latest memory technologies will
follow a traditional path, said Bob Fusco, a product marketer at
Hitachi. "Initially, things may have a 5 to 10 percent premium and the
market [volume] will shape how fast that premium goes away," he said.
"In the past, premiums have been killed pretty quickly. The market
won't bear anything larger than 5 to 10 percent and, even then, only
for short periods of time. And, inevitably, the cheaper, higher-volume
alternative will come to dominate."
The 100-MHz-SDRAMs that joined 66-MHz SDRAMs in the market in 1998,
along with Intel's PC-100 specification, now command only "a very small
price premium in the end-user market and none for large OEMs," said Don
McCord, vice president of marketing at Tanisys Technology Inc.
Virtually all the memory-module makers, including NEC, Toshiba,
SiliconTech, Memory Card Technology and Dane-Elec Memory, started
fielding 100-MHz parts some time within the past year, but none as yet
has a handle on the market split between them and 66-MHz parts.
Boba Propobic, vice president of engineering at Camintonn, said
increasing the frequency from 66 MHz to 100 MHz on a conventional
64-bit memory bus translates into a bandwidth increase from 528
Mbytes/second up to 800 Mbytes/s, and 133 MHz pushes bandwidth out
another fraction.
A transition to DDR or Rambus, however, approximately doubles
bandwidth out of the chute: DDR by transferring data at both the rising
and falling edges of a clock signal; Rambus by jacking up the frequency
to 400 MHz-effectively 800 MHz since it also uses rising and falling
edges. Rambus shrinks the width of the data bus from 64 to 16 bits,
however, and Hitachi's Fusco said that Rambus and DDR SDRAM at a
100-MHz (effective 200-MHz) frequency have comparable 1.6-Mbyte/s
performance. "DDR SDRAM [at 64 bits] has a 4:1 advantage over Rambus in
the number of bits transferred per clock," Fusco said, "but Rambus
compensates by running at four times the frequency."
All the module makers expect both Rambus and DDR to have a place in
the market, but there's disagreement over how big each one's slice will
be, how things will evolve over time and even when the first production
products will become readily available. "You never know what the truth
level is till things show up on the street," said John Sutherland,
director of new-business development for Kingston Technology Co. "The
PR people are all saying summer but people in the labs are saying they
want to get things out by March."
Like many of the module makers, Smart Modular Technologies has a foot
in both the DDR and Rambus camps, and like many, Smart's Johnston sees
Rambus starting life as primarily a PC phenomenon, while DDR SDRAMs
address applications in high-end workstations and servers. "DDR will be
a little more expensive than SDR [single-data rate] initially but less
than Rambus," he said. "But as Rambus ramps, it'll come to cost about
the same."
As for Celestica, "We're on the Rambus bandwagon," said Shal Grant,
director of the memory-operations group. Said advisory engineer Thilo
Sack, "Rambus is inevitable because Intel will make it a de facto
standard, so it won't be too long before volume is high enough to drive
costs down, and there will be a fairly quick rampdown on pricing."
Tanisys expects to be shipping Rambus modules some time in the first
half of this year, said the company's McCord, but they're "not quite
sampling yet, the major reason being a major shortage of the chips."
Once Rambus chips and modules really start shipping, he predicted "a
significant premium" for them, "perhaps through the end of 1999 because
there will be more demand than supply. There are really only four or
five chip manufacturers with production-ready chips today and no one is
close to full production yet."
DDR SDRAMs are also on Tanisys' road map, "but we don't see the DDR
systems shipping as soon as the systems for Direct Rambus," he said.
"It's not clear to me that there will be many DDR systems shipping
until 2000."
McCord noted an important advantage for Rambus, with its narrow data
path. "With Rambus memories located so close to the core logic and with
such a small number of lines between them, that greatly simplifies
system-board manufacturing and simplifies complying with the regulatory
requirements of the country you're selling it in," he said.
As for Samsung, it is active in both Rambus and DDR, "but we're
moving more quickly with Rambus," said Nick Schwartzman, product
marketing manager. "Why? I think the mainstream is obviously behind
Rambus and DDR better fits niche markets. Server manufacturers like to
see more stable transitions from one technology to another, so DDR for
them would be the next logical step."
Schwartzman said it's too early to speculate on Rambus pricing. He
said that memory vendors hope to make DDR pricing "something palatable
for consumers, but it depends on the time frame. Our goal is to get as
close to SDR as possible, but that might take time. We're probably
looking at 2000 or 2001 for the crossover [when the DDR premium over
SDR disappears]."
For its part, Hitachi "is locked and loaded to do either architecture,
" Rambus or DDR, said Juras. When exactly a volume market will develop
for either technology "is really difficult to say because there are no
real shipments out there right now." And when that happens "depends on
how far PC-133 really travels."
Hitachi, however, clearly appears more bullish on SDRAM evolution
than on Rambus for the near term. The company has been shipping
engineering samples of DDR chips and modules for about a month. "DDR is
a very good cost-effective solution because you're essentially taking
current 64-Mbit SDRAM technology and tweaking it to squeeze as much
performance out of it as possible," Juras said. "It is a really solid
fit as a good evolutionary path from traditional SDRAM."
Fusco noted one other advantage of evolutionary SDRAMs: a hedge
against availability problems. "Voltages will be key," he said. "It's
easy to do a memory controller that can do both 2.5 V and 3.3 V and
talk to both SDR and DDR, so DDR provides a nice backup plan to stay
with older technology if that makes sense. If you break compatibility
and go to some new interface, you don't have that option: You have to
have that new memory; you can't switch back and forth with
previous-generation technology."
|
