Clairvoyant's View: Not Too Clear -- Some prognosticators see a single architecture in the DRAM industry's future, while others expect market fragmentation.
Aug. 28, 1998 (Electronic Buyers News -) -- It's hard
to predict the future if you're looking into a mud-spattered crystal
ball. Such is the problem people face today in trying to figure out in
which direction the troubled DRAM market is likely to go.
Many DRAM suppliers and analysts believe the high-volume PC market
will remain essentially anchored to one specific DRAM architecture,
reprising the roles played in the past by such technologies as EDO and,
more recently, by SDRAM.
But a growing number have a different take. They believe the DRAM
market will fragment, and in fact is already doing so. These observers
expect the market to be composed of at least two major DRAMs, possibly
containing different cores and each grabbing a substantial share of PC
demand.
Which, if either, of these prognostications turns out to be correct
will depend on the development of several factors.
Right now, as everyone is painfully aware, there is too much
production capacity, and DRAM prices are in the cellar for the third
year in a row. DRAM suppliers are just hoping to make it through
another week or month without rupturing a major profit artery.
The Asian financial crisis has worsened matters, forcing severe
curtailment in capital spending, particularly by Japanese and South
Korean companies.
In response, some DRAM suppliers are doing what they can to shrink
die in order to eke out a small profit in the face of decreasing
available funds for future-generation research.
And given these current market misfortunes, there is uncertainty
about which type-or types-of DRAM will dominate the next generation, as
well as when demand for that technology will kick in.
Who's on first?
Depending on whom you ask, the dominant next-generation DRAM will be
double data rate (DDR) DRAM, Rambus DRAM (RDRAM), SLDRAM, or at least
two of the above. As for what succeeding generations might look like,
some predict that question will be pushed to the back burner or
relegated to low-level research until the market improves.
Intel Corp. also has a say-particularly about architecture and
chipsets.
Most observers expect that Direct RDRAM (DRDRAM) will be ready by
mid-1999, along with an enabling chip- set from Intel. But no one seems
certain whether availability of and demand for DRDRAM will be in proper
sync.
Some analysts claim DDR will displace the RDRAM after SDRAM loses its
popularity; they suggest this period could last from six months to two
years.
Others expect devices designed to meet the PC-100 specification to
extend the SDRAM's life. This camp projects that DDR will take a small
share-less than 20%-of the total DRAM market.
But most prognosticators expect that DRDRAM will grab a good portion
of the market. When, to what extent, and whether it will share the
spotlight with other technologies remain unanswered.
The push to develop low-latency DRAMs indicates to some that demand
will diverge along two lines: A large number of buyers will demand
DRDRAMs, while an equally large number will require low-latency
designs, depending on the application. Others see niche applications
such as embedded DRAM continuing to grow and capture market share. (See
sidebar on page 50.)
Major wild cards are capacity and capital investment. Typically, a
down cycle has been a time for suppliers to add capacity and equipment,
and to set up new processes to meet anticipated demand for
higher-density and higher-performance devices.
But in the past three years, most companies haven't increased
capacity or added equipment. Some, like Texas Instruments Inc., have
left the market altogether. Other large suppliers are idling capacity,
and some have shut plants or fab lines.
However, when market conditions improve and demand strengthens, DRAM
makers may have to take costly measures to ramp up production of new
devices. In such an environment, economies of scale will be hard to
achieve, and prices may climb.
Changes afoot
Historically, only one type of DRAM has dominated, with niche
applications filling in the remainder of the market in bites of less
than 10% or 15% each. Typically, only one device, such as EDO or SDRAM,
most suits PC design engineers at any given time.
Recently, however, that has begun to change. Many high-end PC,
workstation, and mainframe design engineers, for example, have found
that EDO works as well for them as does SDRAM. It has the performance
they require and is less expensive. But this situation could change as
suppliers continue to scale down production of older DRAMs.
Other changes are afoot. DRAM suppliers previously designed the
memory devices and let OEMs work out a way to integrate them into their
systems. But today's system designers dictate to or work closely with
DRAM suppliers to ensure that their needs and specifications are
designed into newer memory-chip architectures.
Early SDRAMs were not compatible or drop-in replacements for each
other. Intel and OEMs found this unacceptable.
The current solutions for next-generation DRAM designs all are
compatible. So compatible, in fact, that Intel, Rambus, and several
suppliers expect DRDRAM to provide memory solutions for the next five
years.
Intel chose the interface design to meet a series of criteria based
on performance, cost, flexibility, and various other factors, according
to a spokesman for Intel, Santa Clara, Calif. "Direct RDRAM meets those
criteria for the next five years. We chose it because we found that
those requirements provided flexibility and headroom for the future,"
he said.
Currently, Intel is focusing on the PC-100 specification, "but we're
definitely keeping an eye on the landscape. We're constantly working
with the entire industry to support new and emerging technology," the
spokesman said.
Rambus Inc., like Intel, takes the view that one major technology
will dominate in the next five years. "DRAMs are a commodity product,
and most DRAMs go into PCs; 55% of all DRAMs go into PCs, and the
dominant architecture will be the one that wins the PC market," said
Billy Garrett, product planning manager at Rambus, Mountain View,
Calif.
He's confident that the Rambus interface will be the one, and that
chip suppliers will release parts using this technology through the
1-Gbit density. "We have one part providing all the bandwidth, and we
go to more channels to increase bandwidth; it's evolution in memory
controllers with DRAM arrays."
Rambus obviously hopes to keep subsequent generations of devices
using its architecture in the demand mainstream for more than five
years.
"Rambus is optimized to [PC] systems that will debut in the year
2000, and it will coincide with the dominant market share," said
Michael Sporer, technical marketing manager for memory at LG Semicon
America Inc., San Jose. "It will have a long life.
"In two or three years, we'll start thinking about the next
generation of architecture. Whether it's proprietary or open-standard
depends on the DRAM industry," Sporer said. "Rambus has certain
technical and performance advantages, but it's the compatibility of
these DRAMs that will make it wildly successful."
New rule makers
"[In the old days], customers could tell us what they needed, when
they needed it, and why," said Jim Sogas, director of product marketing
for memory at Hitachi Semiconductor (America) Inc., Brisbane, Calif.
"What's changed is [that] the big driver for DRAM is the PC, and the
control of design decisions has gotten into a third party's hands.
Intel is making decisions in place of the clear decisions we used to
receive from our customers."
The PC market has changed in that there is little differentiation in
the hardware.
"Open standards allow OEMs to buy parts from anyone and assemble
their PCs," Sogas said. "It creates some problems for DRAM makers; it's
caused problems and will cause more before it stabilizes."
Sogas would like to see DRAM makers more involved in the design
process, but he doesn't contemplate any chance of returning to the
older methods.
"Instead of system designers and box builders, some of the leading PC
companies today characterize themselves as the sales and marketing arm
of the PC industry," added an industry source who asked to remain
anonymous. "Too often these companies are moving more into the
logistics and fulfillment arenas instead of offering customers
technical value," he said.
Other DRAM suppliers describe the situation differently. They claim
to be part of a relationship that began years ago when customers wanted
ideas about improving system performance.
"[In the past, our customers] asked us questions about cost,
performance, and so forth two to three years down the road," said Kevin
Kilbuck, memory engineering manager at Toshiba America Electronic
Components (TAEC) Inc., Irvine, Calif. "We certainly have input, but we
don't decide what they will use in their system; we have to work
together."
But, today, chipset support is more critical than memory. "Our number
one priority is to provide solutions for that market," said Jeff
Mailloux, DRAM marketing manager at Micron Technology Inc., Boise,
Idaho. "How it dictates what those requirements are is something we all
have to work with. It's driven by the chipset approach, and that's
driven by Intel."
This situation, however, is likely to reverse itself, said Sam Chen,
director of engineering at Mitsubishi Electronics America Inc.,
Sunnyvale, Calif. "Intel is the technical provider for the PC industry.
When the industry has grown to a certain extent, IP will dominate
system technology, and that will not be welcomed by system houses.
"There's a potential conflict," Chen said. "If this continues and
Intel has all the system IP and PC makers don't innovate,hey will lose all the IP to Intel. I don't see this happening
to major system houses, like IBM, in the future," Chen said.
Is compatibility the answer?
Within today's PC industry, different platforms are starting to
require different memories. This explains the current preference for
PC-66 and PC-100 SDRAMs, which some analysts expect to remain popular
for at least 18 to 24 months.
Others expect DDR, a direct evolution from SDRAMs, to reap the next
demand sweep; still others predict that DRDRAMs will come on next year
with demand ramping quickly. DRDRAMs will meet the need for PC
applications requiring higher bandwidths, but demand won't take off
before 2001, according to Sherry Garber, senior vice president at
Semico Research Corp., Phoenix.
But Garber believes that the lack of supplier capital investments
over the past three years will have a more certain impact.
Suppliers concurred and noted that R&D, either for shrinking die or
developing processes, has pre-empted capacity expansion and, in many
cases, new-equipment additions.
Samsung gives R&D top prior said Avo
Kanadjian, vice president of memory marketing at Samsung Semiconductor
Inc., San Jose. "In a company like ours, cutbacks are implemented in
nonsemiconductor areas first, and when it's applied to semiconductors,
it's more applicable to capital expansion."
Die shrinks have been a priority recently for Hyundai Electronics
Inc. in an effort to bring costs down and increase profits, said Farhad
Tabrizi, director of strategic marketing at the San Jose company.
The combination of the Asian financial crisis and low DRAM prices has
shrunk investment monies significantly, reducing Hyundai's ability to
invest in new assembly lines, packaging, or architecture.
"There's not enough money to invest in new architectures and
infrastructure modifications; [this has delayed] our move to new,
revolutionary products," Tabrizi said.
And, if the market continues to decline, there will be further
investment cutbacks next year, according to Jamie Stitt, TAEC's DRAM
marketing manager.
Observers expect some market consolidation next year. Even
Taiwan-based companies that have publicized relatively aggressive
investment plans are starting to cut back on those commitments.
"New capacity is dependent on market conditions. When the market
warrants it, we'll add fab capacity," Micron's Mailloux said. "Until
then, we won't move forward. In the past, DRAM investment was distorted
because it wasn't financed off the DRAM business, it had government or
conglomerate backing. That's changed too; the DRAM business has to pay
for itself, there has to be enough profit. People [right now] can't
afford to invest in new technology and new factories." |
