In 1988, Ronald Reagan was wrapping up his second
presidential term in Washington. Perestroika came to the Soviet Union. NASA reported accelerated breakdown of the ozone layer by
chlorofluorocarbons. Steve Jobs showed off his first NeXT Computer in
San Francisco. The term "brick" was used not just for building materials but
for mobile phones, those novelty devices that people like doctors, lawyers, and
drug dealers carried.
A Toshiba T1200H personal computer (10MHz Intel 80c86 CPU,
1MB RAM, and a 20MB drive) cost $4098 ($7182 adjusted for inflation). Twenty
years later, in 2008, a Lenovo ThinkPad cost $1700 and put under your
fingertips 2GHz processing power, 2GB of RAM, and a 160GB hard drive (as well as
features that didn't even exist in the 1980s). And what Toshiba packed into its
T1200H in '88 you now hold in your hand in your smartphone...and then some.
In 1988, the cost per megabyte was $58 in adjusted dollars;
today it's $0.000363. In 1988, ASICs were electronics design's big daddy, and FPGAs
had on training wheels.
EDA was breathing the oxygen of emancipation from vertically
integrated electronics companies and thriving in its own little Wild West. And
a new company was born.
June 1, 1988 was the official launch date of Cadence Design
Systems, Inc., a merger of ECAD (founded by Glen Antle and Paul Huang in 1982)
and SDA Systems, which Jim Solomon founded in 1983.
The rest is history--history at Cadence, history within the
larger EDA sector--and we'll highlight some of that history with posts, videos,
and events in the next 12 months around the world. This week, to kick it off, we'll
hoist a 25th anniversary banner on one of the headquarters buildings
on Montague Expressway in San Jose, so if you're driving through the valley this
year, give a wave.
Twenty-five years is a long time in human terms, an eternity
in electronics. And while we'll look back in the coming months with interviews
with EDA industry titans and design gurus, it's intriguing to think what our
world will look like in 2038, a quarter-century from now.
In most ways, it's impossible to predict. Would you have
said in 1988 that you'd some day slip far more computing power into your pocket
than sat on your desk? I would never have thought then that in 2012 I'd be
driving across the California desert at 80 miles an hour one morning
watching--on my cell phone--a crazy Austrian in a space suit leap from the edge
of space and free fall to earth. Watching. In a car. In the desert. On my cell
It's really impossible to predict what the electronics world
will look like in 2038 because we couldn't predict today back then.
The Road Ahead
That said, we do a very good job of predicting technology
advancements in 5-10-year increments, argues EDA guru Gary Smith of Gary Smith
EDA. Take the ITRS
roadmap to bed tonight and you'll see a lot of the future going out to 2027.
The vast majority of that will come true, within a few years of prediction.
But what's the world going to look like?
Considering that what our industry looks and feels like
today is not fundamentally different than it was in '88, it'll look familiar in
2038. Back then, the industry had thrown off the vertically integrated model by
the end of the 1980s, and we're still operating in that specialization mode.
But we are entering a fascinating new era that will be the
foundation for technological innovation going forward. That is the disruption
of the semiconductor space, where companies are embarking on one of two paths.
On one path, a traditional semi company
focuses on components (in many cases, acquiring digital or analog businesses to
broaden its product lines). The other path finds companies focusing on systems.
(Frank Schirrmeister wrote
about this this week, and I added
a perspective as well).
Smith puts a finer point on the trend:
"What you're seeing in
the semi business--and this didn't become clear until a few years ago--ARM has
become the new prototype for an ‘integrated circuit company.' We might not be
using silicon in the future; we build ones and zeros and it goes into
something. In ARM's case it doesn't go into silicon until somebody else does something
with it. That's the pure form of the new concept."
As part of this transformation, the central processing unit,
for example, may actually fade in importance with the rise of special-purpose
processors, Smith believes.
"When we end up there
will be six to eight processing units optimized for certain sets of algorithms. We have
to specialize in different algorithmic patterns to optimize the processing so
we can minimize the power problem. A CPU by definition is not optimal."
For the EDA industry, the next 10 years will reveal a
transformation--already underway--from a tools business to a solutions business
as the vendors gobble up IP startups to enable those evolving semiconductor
companies moving up the system chain to fill up their billion-gate designs
productively and efficiently.
The better near-term news, according to Smith, is the EDA
industry has been acquiring semiconductor R&D know-how--for example
emulation technology or DFM--as the semiconductor companies move up the chain.
That, he insists, will push industry average margins north very soon.
He believes that will happen because of Ross Perot. Perot's
insight, according to Smith, was to outsource the data center's software. Then
those software providers sold it back to him at 40 percent margins, which
became the talisman for Wall Street. When EDA spun out of the semiconductor
industry, the semiconductor industry kept the ‘R' in the R&D, so the EDA
margins were 25 percent, Smith said. Now that the ‘R' is coming back in, those
margins should increase.
So what does 2038 look like? The best we can do at this
point is consider what 2028 looks like, given that's how far the ITRS roadmap
Fifteen years from now, we'll be making SoCs (or whatever
we'll call them) with 65 billion transistors at the high end. A hardware design
engineer will make $499,000, while a software engineer will pull in $265,000.
(By contrast the ITRS notes that a good hardware engineer made $228,000 last
year, while his software brothers and sisters made $182,000.)
In 2000, it took 18 engineers to design 1 million gates; in
2028 it will be 0.01. In 2000, the total cost to produce that 35 million-gate
design was $22.2 million; in 2028 it will cost $53.3 million to produce a
device with 16 billion transistors.
In 1988, there was no Internet as we know it, no Google, no
Facebook, no iTunes or Android. Cars had airbags but no side radar, no
autonomy. The only drones in the sky were space rockets.
So here's to 25 years. Let's raise a glass and then get back
to work. There's lots to do.