Chromatic .......
By Rick Boyd-Merritt
SUNNYVALE, Calif. - Five long years ago, a Silicon Valley startup called Chromatic
Research Inc. burst on the scene in a media blitz, trumpeting its plans to build a new
kind of PC component - a media processor - based on an emerging technique known
as VLIW and sold on a new business model it dubbed the "chipless" chip company.
Partners LG Semicon, Toshiba and STMicroelectronics would make the Mpact device
and Chromatic would recoup its investment in designing the part by selling software to
run PC graphics, audio, modem and other functions on it simultaneously. With a handful
of seasoned executives in tow, the operation leapt onto the charts of hot companies to
watch.
Today, the leadership of a radically scaled-back Chromatic says the idea of a media
processor for PCs lives on, although they have discontinued work on the Mpact line,
abandoned the very-long-instruction-word techniques, the chipless business model and
what has become a dog-eat-dog computer-audio and graphics market. Instead they will
pursue new arenas such as digital-TV decoding, with a completely different architecture
in the works. Although Chromatic is on the verge of losing at least one major
backer - Toshiba - what may be the most surprising thing about this high-flying startup
in the process of completely rebuilding itself, is that some people still believe in it.
The company recently concluded a new round of investments that president Dave Holt
says will take it "a long way down the path toward our next generation." And customers
such as Compaq Computer, Gateway and STB Systems, which were among the first to
conclude that Mpact could not deliver on the company's vaunted promises, give
Chromatic a fair shot at competing for slots in PCs that will sell next year.
"They have a chance," said a senior engineer at Compaq, which took an equity stake in
Chromatic earlier this year. "The question, as always, will be about their execution and
timing. Strategically they are headed in the right direction, and I think they will be
competitive because they are approaching the problems from a different angle."
All about parallelism
In the wake of the hype behind its foundered Mpact, the company refuses to divulge full
details of its next-generation plans. But it is clear the new part will abandon the complex
VLIW-like engine that could issue as many as three instructions per cycle, in favor of a
design that combines simpler processor cores working in parallel on an "SMP on a
chip." The part would handle tasks such as DTV and possibly other functions such as an
ADSL modem or home-network controller. And the company will act as a more
traditional, fabless semiconductor maker designing and selling its own chip and
associated hardware.
"I'm not convinced what we were doing with Mpact was really VLIW, but it was our
marketing stance," said Chromatic founder and lead engineer, Mike Farmwald. "In any
case, the road we are heading down now is even less VLIW.
"The future is all about parallelism, and there are easier ways to get parallelism than to
build a complex [VLIW] machine that issues five to 10 instructions per cycle,"
Farmwald added. "Even if you don't fill all those instruction slots, you still take a hit in
terms of power consumption and having all those transistors. In short, it's better to have
two simple machines than one complex one. I think it's absolutely clear we will move to
SMP on a chip."
Chromatic's VLIW-like approach apparently was as inefficient as its new-age chipless
business model, under which both Chromatic and its semiconductor partners called on
customers together, then debated how to split the profits from each sale. These
discussions became even more "intense" when Chromatic's sales failed to meet
expectations.
"The chipless model ends with Mpact," said Holt. "That model was developed at a time
when capacity was tight and it gave us the capacity and helped net the interest of large
semiconductor partners," he added. "Now there's a glut of capacity, so a fabless model
makes more sense. You have to be flexible to change your models as the business
conditions change."
But more than business conditions changed to create a situation in which the Chromatic
master plan crumbled. The company's first-generation Mpact 1, released in September
1996, failed to meet its performance goals. Its software-intensive approach demanded
hard work from highly skilled software engineers who were hard to come by in Silicon
Valley. And the graphics and MPEG-2 applications areas where Chromatic's Mpact 2
found some success are expected to dry up next year, a prediction that forced the
company to cancel plans for a third-generation part.
"We wound up designing it into a dedicated DVD decoder board," said Jim Hopkins,
vice president of strategic marketing at STB Systems Inc. (Richardson, Texas), which
still sells a few thousand Mpact adapter cards each month. "You needed more
horsepower if you were going to be able to do two or three things on the chip, and
unfortunately that's where the media processor was supposed to get the bang for the
buck."
A board based on the Mpact 2 handles 2-D and 3-D graphics as well as DVD decode
and sells for about $70 to OEMs. However, it tends to drop video frames when
handling graphics jobs, and its 3-D performance is as much as 50 percent below boards
offering separate 3-D graphics and DVD decode chips, and cost just $25 more,
Hopkins said. "It's not up to snuff with what you can do with dedicated processors even
at a lower cost," he added.
Despite the problems, the Mpact boards sold as DVD decoder cards in volumes as
high as 25,000 per month at their peak, Hopkins said. One of the main users was
Gateway, which launched one of the first PCs to offer DVD drives, thanks to the
Mpact's software architecture.
Just as the first DVD-equipped systems were set to ship last year, a requirement for
content-scrambling software came to the fore. Chromatic revved its software for the
new algorithms and systems shipped within two months, while others took longer to
re-spin silicon, according to Mike Grubbs, director of global convergence products at
Gateway (North Sioux City, S.D.).
But software was also Chromatics' Achilles' heel. The Mpact's compilers were difficult
to use and good programmers can be hard to find. Said Hopkins of STB, "Software
represents the biggest risk for delay and costs because it's all manpower."
To make matters worse, the applications space Chromatic made a dent in last year -
combo DVD and graphics boards - is under attack from two sides.
"We've got a couple quarters of hardware DVD left until we cycle through to the point
where OEMs are shipping mostly 300-MHz-and-higher systems," said Hopkins. "Once
we get there, DVD will be a software function on the Pentium II."
In 3-D, a handful of companies compete for high-end sockets, while Intel is likely to
sweep the low end next year with its Whitney chip, which will integrate an i740 3-D
controller and a north bridge chip. "On the high end people are leapfrogging each other
every three months, and at the low end Intel will eat everyone's lunch," said Holt.
Bob Brown, president of Toshiba America Electronic Components Inc. (Irvine,
Calif.), said Chromatic and the media processor are effectively dead in the PC
space. "Mpact was a very good product, but every time we got close to doing
something, there was a glitch in the software," said Brown. "We couldn't make
any money on it." |
