IEEE Spectrum November, 2010 - 58

up to the vice president of engineering, who eventually
agreed to use the new circuits. They worked, and the controller ended up in Hewlett-Packard drives used by a wide
variety of manufacturers.
Looking for more challenges, Sehat joined Integrated
Information Technology in Santa Clara, Calif., where he
designed circuits for digital video compression and decompression. The technology was novel but ended up in a dog of
a product: the AT&T VideoPhone. It was time to move on,
but Sehat didn't know what to do next. His wife did, though.
"Engineers always talk about technology," she says.
"They figure out a way to design something, to make a product better." But Sehat was a little different, she says, for he

also thought about how the product would fare in the market. When they were still in school, she had told him, "After
your Ph.D., you get some real-world experience. But whenever you're ready, we're going to do a company of our own."
In 1995, he was ready. He asked his brother Pantas, who
had recently left IBM's Almaden Research Center, to join
the start-up. With some savings and money from Dai's parents, on 10 February 1995 the three incorporated Marvell-
so named because they were planning to do marvelous things,
and, they had noticed, a lot of successful companies' names
ended with "el" or "ell," like Intel, Novell, and Nortel.

SEhAT ALREAdY hAd plans for the first product: a better
read channel for disk drives. It sounds
incredibly specialized and it is, but it's
also one of the drive's key components.
The read channel takes the analog signal coming from the magnetic head as
it scans the disk, converts the noisy signal to digital, and puts that information
out onto the bus that will take it to the
computer. Existing read channels used
a bipolar transistor on a complementarymetal-oxide semiconductor substrate
(BiCMOS), but Sehat planned to use only
CMOS. That way the channels could be
manufactured by a chip foundry like the
Taiwan Semiconductor Manufacturing
Co., so Marvell wouldn't have to build
its own fab. Using CMOS also meant
that the device would consume less
power. This would, however, present an
engineering challenge: Existing CMOS
read-channel designs were much slower
than BiCMOS.
However, Pantas had worked on hard
drive technology at IBM, and Sehat had
experience in mixed signal chips-those
that require combining analog and digital technology. Between them they
thought they could design a read channel
that could handle 180 megabits per second, putting them well ahead of existing
100 Mb/s BiCMOS products and, they
A low student-to-faculty
projected, equal to the next generation
ratio, world-class
of BiCMOS chips that would be coming
researchers, collaborative
from
the established companies.
laboratory facilities,
The
traditional method used in read
and industry partnerships
channels
involved looking for peaks in
equip Tufts engineers
the analog waves to define the digital
with the skills to improve
bits. But as the bit density increased, it
people's lives through
got harder to distinguish the peaks. The
a passion for innovation
Sutardja brothers instead took multiand a commitment
ple samples of the signal and then used
to excellence.
digital signal processing circuitry to sort
them out. This approach reduced noise
and also sped up the bit rate.
"The Marvell founders would come
to conferences, say they could do it in
CMOS, but wouldn't reveal their secret
sauce," says Stephan Ohr, an analyst with Gartner who covered analog

Electrical &
Computer Engineering
Graduate Program in

Apply now:

http://gradstudy.tufts.edu
58

NA * iEEE SpEctrum * NovEmbEr 2010

spectrum.ieee.org

http://gradstudy.tufts.edu http://spectrum.ieee.org

Table of Contents for the Digital Edition of IEEE Spectrum November, 2010