The hardware versus software debate..............
eetimes.com
ISSCC: Hardware vs. software debate draws range of supporters
By Peter Clarke
EE Times
(02/18/99, 3:27 p.m. EDT)
SAN FRANCISCO — "Has hardware become a second-class citizen to
software?" That was the knotty question posed at one of the evening panel
sessions at this week's International Solid State Circuits Conference.
While the panelists took a range of positions in the hardware vs. software
debate, it was Robert Brodersen, a professor at the University of
California, Berkeley, who took up the banner most vehemently for
hardware solutions.
Brodersen's primary argument was that the energy-efficiency advantage of
hardwired logic solutions over software running on a general-purpose
processor was at least four orders of magnitude, and that was too big a
deficit to be "left on the table." He also challenged the software model,
saying that it was a legacy of the pioneers of computing in the middle of this
century for whom cost, size and power consumption were not an issue.
Brodersen also attacked the inefficiencies in the practice of taking parallel
algorithms, coding them sequentially in a high-level software language and
then struggling to rediscover the parallelism and running it on
general-purpose processor architectures. Purposefully built hardware is
much better able to reflect the parallelism of the original algorithms, he said.
The flexibility of software, most often touted by the panelists as its key
advantage, was not universally true, Brodersen argued. "Software is
becoming harder than hardware. People are afraid to change the software
in case it breaks the system. Hardware is more flexible and is being
designed to be backwards-compatible with software," he said.
Taking an opposing position, Dan Dobberpuhl, chief executive officer of
SiByte Inc. (Menlo Park, Calif.) and formerly a leader of several
microprocessor development projects at Digital Equipment, argued that
software plus an appropriate CPU is king.
Some of the arguments he marshalled were that while power efficiency and
die size may favor a hardware solution, such solutions naturally require
smaller volumes than more general CPUs, and that volume drives price.
Software also allows late customization of hardware, bringing
time-to-market advantages, which can be extended beyond a product's
delivery through field upgrades or over-the-Internet fixes.
Dobberpuhl's final argument was that software would be the more fertile
area of development because there were almost no barriers to entry. "The
design cost of any ASIC must be $3 or $4 million, and it needs high
volume. How many 12-year-olds can design a VLSI circuit?" he
challenged.
Patrick Bosshart, an engineering fellow at Texas Instruments Inc. (Dallas,
Texas) was firmly in agreement with Dobberpuhl. "Anything that can be
done in software will [be done]. ASIC development cost will go up while
available processor performance, 50 to 2,000 Mops today, will increase to
10 to 200 Gops in five year's time," he said.
Raul Campasano, chief technology officer at Synopsys Inc. (Mountain
View, Calif.) took a more moderate position, reflecting that hardware and
software are both in play in most systems and should be developed in
tandem. He also observed that while both pure software and pure
hardware solutions can be viable in different applications, there is also a
middle road. "FPGAs are the fastest growth area. They can be faster than a
processor but slower than dedicated hardware. There are a host of startups
that favor a combination approach."
Professor Takayasu Sakurai from the University of Tokyo put up a slide
that showed just such a solution, but showed his hardware colors when he
said, "Without hardware, software is nothing, but without software,
hardware is something."
It was also observed that hardware design now often starts with high-level
language descriptions so that, in a sense, software is the root of both
hardware and software-architected solutions.
A member of the audience observed that the hardware vs. software debate
was similar to the analog vs. digital debate held many years previously and
now largely dropped in the light of digital circuitry's dominance of the
landscape.
The audience member said that things that could not be done digitally were
done in analog circuits. In like fashion, digital hardware solutions would be
used only for those things that couldn't be done in software.
Brodersen seemed to draw some comfort from the assertion that software
would be used for the "easy" things, while much of the challenging work in
new applications would continue to be pioneered in digital circuits. |
