From Always On
GregBlonder [Morgenthaler] | POSTED: 02.15.04 @21:09
Now is time
for a few good researchers and entrepreneurs to come to the aid of the
semiconductor industry. The ever-evolving system of battery, integrated
circuit, and CMOS (complementary metal oxide semiconductor) transistor,
which for 20 years has lived in harmony with Moore's Law, is reaching
The doubling of transistor density every 18 months, which Gordon Moore
predicted, is coming smack up against nearly impenetrable heat and cost
barriers. Unless something radically new intervenes, the industry will
begin a gradual descent into slow growth and diminished profitability.
ability to deliver ever more intelligent portable devices will be the
first victim; such devices will start falling noticeably off the downward
cost curve around 2010. But ultimately other forms of computing will sink
into an ever-widening malaise. Given current trends, that could begin
about an emerging problem ten years out represent no idle wringing of
hands. That is because most conceivable solutions will take that long
to develop. The time to start building an entrepreneurial business around
the best of those solutions is now!
do we start?
depend on more powerful chemical bonds, but we've effectively run out
of atoms on the periodic table that could create them. Perhaps an improvement
factor of two or three is possible. The integrated circuit offers potential
for lots of short-term engineering fixesnew algorithms that reduce
current leakage, new rules for running wires, and so on. But potential
improvements are becoming increasingly incremental and will almost certainly
exhaust themselves over the next decade.
point of potential innovation, then, looks like it will have to be the
transistor. There's no getting around it: we need a replacement for the
is doable. Unlike batteries, transistors present no fundamental scientific
barriers to dramatic improvement. The laws of physics permit us to work
with far fewer electrons than the CMOS design currently demands. Fewer
electrons, in turn, would allow us to crank down the power necessary to
drive next-generation systems by a factor of hundreds. And less power
would mean, of course, less heat and more design breathing room to create
ever more intelligent and responsive portable devices.
replacement scenarios present themselves, in decreasing levels of desirability.
of dreams: Someone could come up with a CMOS design that, while radically
different, fits within basic silicon design experience. That would be
ideal, because it preserves the trillions of sunk dollars invested in
semiconductor infrastructure. But because no new suitable design has appeared
in over 20 years of trying, we can be reasonably confident that it won't.
2. The good: Researchers could solve the problem with new materials that lead to a
better insulator, a better dielectric, or an absolutely uniform thin film,
giving CMOS a new lease on life. In combination, such improvements might
both solve the heat problem and require changes in just a hundred or so
system processing steps. These improvements would offer the great advantage
of leaving another thousand steps in place and preserving infrastructure.
If no other alternatives emerge, a solution based on new materials might
win solid industry support.
3. The bad: We could invent a whole new semiconductor transistor and replace the CMOS
design entirely. Magnetic switching technology, adopted from magnetic
memories, is sometimes mentioned as a viable alternative approach. Almost
every step in the fab would require modification, and previously hallowed
design rules and prejudices would need to be reset, but necessity may
be the mother of invention.
4. The ugly: Finally, we could leave the cocoon of the silicon industry and head out
into wholly new territory like organic molecular switching or nanotubes.
These solutions would come genuinely out of left field and may have been
ignored by today's major corporate players. Of course, it's also the least
desirable alternative because we would be abandoning every last bit of
Of the alternatives,
only number 1 offers the potential for fairly rapid turnaround. Alternatives
2 and 3 will almost certainly require ten years of development and commercialization
before they become true mass products. It may already be too late to start
on 4 and still expect a smooth handoff with no diminution of growth.
times raise inevitable questions about financial viabilityespecially
given the standard venture capital time frame of five to seven years until
exit. Yet the stakes are high enough that a truly innovative approach
is likely to find serious, patient venture capital backing. The new company
would have to be very clever about finding government research dollars,
partnering, and watching its burn rate so that it could continuously fund
development of its technology. Only in the fifth or sixth year might it
break through with a couple applications where low power consumption is
absolutely critical and customers are willing to pay a premium.
You bet. But the right combination of technology, entrepreneur, and venture
capitalist could make it happen. No conceivable undertaking would do more
for the industry or the broader world economy.
is a general partner with Morgenthaler Ventures.