NeoMagic Running On All Cylinders In New Direction
Oct. 06, 2000 (Electronic Buyers News - CMP via COMTEX) -- Now that NeoMagic
Corp. is engaged in a major refocus from notebook-PC graphics to the wild and
woolly mobile-Internet-appliance arena, new embedded-DRAM products from the
company that pioneered the technology will arrive no sooner than the middle of
next year.
But the Santa Clara, Calif., company is betting that the latest eDRAM wizardry
it's got tucked up its sleeves will be worth the wait for Web-appliance OEMs,
their potentially enormous customer base, and, of course, NeoMagic shareholders.
The company is slated to ramp production of its first complete system-on-a-chip
(SoC) in the second half of 2001. If NeoMagic's widely acknowledged edge in
eDRAM manufacturability holds true and that schedule doesn't slip, the device
could blaze a trail as the first eDRAM-based SoC for Internet appliances to
emerge in volume.
The product will feature a MIPS32 4K processor core from MIPS Technologies Inc.,
NeoMagic's signature graphics-processing, audio-processing capabilities, and 2
to 8 Mbytes of on-chip DRAM, according to NeoMagic president and chief executive
Prakash Agarwal.
NeoMagic wouldn't name development partners or prospective customers and their
product lines for its eDRAM SoC.
Agarwal noted, however, that the device "can address many different Internet
appliances-including handheld multimedia terminals, handheld PDAs, and combined
cellular phones and PDAs-of the type that trendsetting companies like Ericsson,
Nokia, and Sony are turning out."
Follow-on products will be targeted at future generations of mobile Web
appliances by layering eDRAM-enabled Internet video processing and
wireless-communications connectivity atop a graphics and audio-processing
foundation, he said. To deliver that much functionality in a single chip, these
planned SoCs will incorporate new and, in some cases, unproven technologies that
NeoMagic is feverishly developing, licensing, or both.
Examples include a new spin on MPEG-4 compression and decompression that
leverages the company's recently acquired intellectual property in associative
processor array (APA) processing, which the company is wedding to embedded DRAM.
The idea behind the marriage is to boost MPEG-4 performance significantly.
"That would move Internet imaging and video playback considerably closer to the
full-motion, broadcast quality consumers have been awaiting in vain," Agarwal
said. "It's part of our new company strategy, which is to bring rich multimedia
capabilities plus 'any time, anywhere' mobility to the Internet experience so it
becomes that much more satisfying."
Fueling NeoMagic's overhaul is its belief in a potentially huge payoff if it
gets to market first, said Bob Merritt, director of emerging markets for Semico
Research Corp. in Redwood City, Calif.
"If the high-performance, small-size, low-weight, and low-power-consumption
benefits provided by eDRAM technology made sense for notebook and desktop PCs,
they'll make a whole lot more sense for the coming waves of handheld Internet
appliances," Merritt said. "That's because these products-at least as they are
currently envisioned-will have a whole lot more stringent size, weight, and
power requirements compared with PCs, along with similarly demanding performance
needs."
The arrival of an SoC that can meet those requirements, plus deliver a judicious
array of multimedia and mobile-communications capabilities, "could play a key
role in spurring Internet-appliance market development," Merritt said. "Once a
technology platform is defined, products based on that platform can also start
to emerge at an accelerated clip."
And the size of the Internet-appliance market could fulfill predictions right on
time. IDC, Framingham, Mass., for example, believes information-appliance sales
will soar to $17.8 billion, or 89 million units, in 2004, compared with $2.4
billion, or 11 million units, in 1999.
On the other side of the Atlantic, Datamonitor, a U.K.-based market research
firm, is even more bullish: By 2005, it foresees 200 million users of wireless
Internet gaming devices alone.
Given those projections, NeoMagic stands to gain a lot if it can replicate its
success in notebook PCs by pioneering an eDRAM platform for Web appliances that
ends up being the one to copy or beat. In the company's corner is the sheer
complexity of integrating sizable amounts of DRAM, complex logic, and analog
functions into a single, manufacturable device. It's an art form that only
NeoMagic and a handful of other ASIC vendors-notably LSI Logic Corp.-have so far
mastered.
So it's little wonder that no other company has announced a strategy for eDRAM
SoCs for Web appliances. Not yet, at least.
"Certainly, NeoMagic faces plenty of potential competition in the
Internet-appliance arena, since the advantages of on-chip DRAM for appliances
are so apparent," Semico Research's Merritt said.
Merritt said the contenders will come from multiple camps. Besides eDRAM
heavyweights such as LSI Logic, they could include established suppliers of SoCs
for first- and second-generation Internet appliances, such as Cirrus Logic Inc.,
the legions of Linux processor suppliers for cellular phones, and former
competitors in graphics accelerators for PCs, such as S3 Inc.
S3, however, is no longer hot on eDRAM. The company recently shut down its
add-in graphics-board business and transferred its graphics-chip operations to a
new joint venture with Via Technologies Inc., called S3 Graphics, in Santa
Clara.
Contrary to some early reports, S3 is still using the new venture to aim
integrated graphics products at the volume OEM PC market, according to Rick
Bergman, chief operating officer at S3 Graphics. But it has abandoned eDRAM
technology in favor of multichip modules to achieve its performance and cost
goals.
"Notebook PCs in particular need the latest in logic processes and fast cycle
times as well as low cost from graphics solutions, and that doesn't work so well
with eDRAM," Bergman said.
It didn't for S3, at least, which appears just as likely to eschew eDRAM
technology for silicon targeted at Internet appliances. But even for chip
companies that echo Semico Research's Merritt on the benefits of eDRAM for the
appliance arena, the difficulty of developing or acquiring a competency in eDRAM
production could slow their emergence as major eDRAM contenders. There is also
the uncertainty about just how Internet appliances will evolve.
"At this stage, no one knows precisely what Internet appliances will look like,
what range of multimedia and communications capabilities they will have to
deliver, and what types of user interface they'll need to make millions upon
millions of consumers reach for their wallets," Merritt said.
"If NeoMagic's idea of those requirements proves both mainstream and flexible
enough, then it should be able to capture a leading position [in SoC sales for
Web appliances] and run with it," he said. "But market dynamics [affecting eDRAM
suppliers] could march off in an entirely unanticipated direction, as they did
in notebook and desktop PCs."
Then NeoMagic could find itself playing catch-up. But that's a risk Agarwal and
his recently reshuffled management team are willing to take.
What finally propelled NeoMagic, S3, and several other eDRAM suppliers out of PC
graphics was a sudden and substantial drop in discrete-DRAM prices. DRAM became
so cheap that PC OEMs couldn't resist the opportunity to use more discrete DRAM,
despite eDRAM's considerable value proposition for notebook and desktop
computers.
By contrast, in the handheld-Internet-appliance sector, high levels of circuit
integration will be less a luxury than a necessity-the current uncertainty about
market evolution notwithstanding. So lightning, if it comes, won't strike twice
in the same way for NeoMagic.
"It's more likely we'd be somewhat off-base as far as functional requirements
go," NeoMagic's Agarwal said. "But if anyone has a feel for what those
requirements will be, it's today's trendsetting Internet-appliance OEMs. And
that's why a major component of our new company strategy is partnering with such
companies as early as possible in our product design cycle. It's how we're
pinning down the right product vision."
With the vision in place, the next big challenge is executing it, Agarwal said.
For NeoMagic, execution so far means shedding operations that dilute the
company's focus on enabling a mobile and multimedia-rich "Internet experience"
and bringing on board the requisite people, technologies, and manufacturing
partners "to hit the market window with a high-yield SoC that is low cost,
highly reliable, and on spec with die-size requirements," he said.
Last spring, NeoMagic began restructuring from two separate product divisions to
one functional organization with centralized engineering. It also sold its
DVD-controller product line to LSI Logic. These changes reduced the company's
head count by more than 35% and included six departures from the executive
staff, most notably its chief financial officer and vice president of
manufacturing.
NeoMagic quickly added several new members to its executive staff. Sudhir
Chandratreya joined as vice president of technology, and Yogendra Shah was named
vice president of wireless engineering.
Even so, Agarwal described the restructuring part of NeoMagic's execution
strategy as "quite painful," and said he is relieved it's nearly done.
Nowhere near done-but a lot more fun, according to corporate vice president of
marketing Mark Singer-are the company's ongoing technology- development efforts.
On the Internet-video front, NeoMagic is busy building on the competency in APA
processing it obtained last year by acquiring Associative Computing Ltd. from
Robomatics Inc. Long used in Cray supercomputers, for example, APA processing
can crunch vast amounts of data at a dizzying clip. That makes it an excellent
candidate for boosting MPEG-4 encoding/decoding speeds and helping Internet
video go mainstream, Singer said.
"By contrast, embedded DRAM loads very fast," he said. "Plus, there's no
inherent limit to DRAM interface size since communications all occur on-chip. So
if we can pull high-bandwidth eDRAM and APA processing into a single chip, we'll
be able to beat Rambus Inc.'s DRAM speeds [the fastest on the market], since the
Rambus interface doesn't scale."
The only downside to this approach is "no one has ever done it before, so it's
risky," Agarwal said.
That's not the case with NeoMagic's choice of foundry suppliers. NeoMagic has
been working with Mitsubishi Electric, Toshiba Semiconductor, and Infineon
Technologies for seven years and will continue to do so, Agarwal said.
Semico Research's Merritt said those relationships translate into a big
competitive edge for the company.
"When you embed DRAMs with logic, you typically have to use a manufacturing
process that is a compromise between the ideal process for either," he said. "So
you usually lose some performance out of your chip. NeoMagic had an advantage in
the PC-graphics-accelerator business because it really worked with its foundries
on developing processes that delivered the best possible trade-offs among
performance, yield, and die size. Now, it can leverage that expertise in the
Internet-appliance sector."
By the same token, the lack of such foundry relationships and process expertise
for competitors amounts to a big barrier to entry into the world of embedded
DRAM. "That's not something we're complaining about," Agarwal said. |
