Moore's Law states, in brief, that the total transistor count
on a semiconductor device of a fixed size will double every 18 months. In essence,
this means the power of a device will double every 18 months.
But Moore's Law has more than one axis, or component, and I wonder how many
of the power improvements we see are artificially stimulated by a belief in
Moore's law itself. I also wonder whether Moore's law can be violated permanently.
In this column I'm going to suggest that IBM may do just that.
Let's start with the notion that Moore's law is a self-fulfilling prophecy.
Even Gordon Moore has said more than once that he was surprised Moore's law
has kept going for so long. It was never a real law, but an observation turned
into a law. It was merely interesting. But as time progressed and the law seemed
to stick and continue, did it become a self-fulfilling prophecy? You expect
processing power to double in 18 months, so it does because you expect it and
ramp up for it. In the semiconductor business, if you assume the competition
is investing heavily in R&D to develop process technology that will add to this
growth, then you do the same.
Everyone in R&D then pinpoints targets that match Moore's law, thus the law
sticks. Is it possible that if they pinpointed a different target the results
would hit that new target instead of the 18-month target? Is 18 months just
artificial? A convenience?
A gentleman's agreement?
I mention this because of a topic I often harp on: the rate of growth in hard
drive performance, which switched from roughly following the Moore's law cycle
of 18 months, to a new 12-month cycle. The new cycle came about as a result
of IBM R&D willpower. Is it possible that this same kind of adjustment can happen
in the semiconductor industry?
Maybe. IBM microprocessor R&D has also shown that its development cycles can
be way ahead of Moore's law. And we now see the same kind of energy coming from
IBM in this arena that we saw in the company's hard drive business a while back.
As you may recall, IBM was making serious inroads into the market for hard
drive subsystems with announcements of things like advanced MR heads and then
GMR heads and other technologies. Then the company buys the nifty hard drive
business (and technologies) from Digital. Shortly thereafter all hell breaks
IBM has been making similar noise in the microprocessor arena lately, first
with a very early gigahertz chip, then with copper interconnects and other technologies.
What is the point of all this? Fun?
Having already seen what happened to the hard disk business and Moore's law
under pressure from IBM, there is no reason to think the company will stop with
hard disks. IBM has been more than dabbling in microprocessors, with an early
investment in Intel, the codevelopment of the PowerPC chip with Motorola and
Apple, and a strange partnership with Cyrix.
I have to assume IBM just needs a moment in time when it can disrupt the microprocessor
business the way it did the small-hard-drive business. The company has to ask
itself, though, if the buying public can absorb more powerful technologies any
faster. If so, then this could become one wild ride.
The stopper that might prevent this from happening lies deeper within Moore's
law. In fact, there are three inter-related effects of Moore's law: performance,
size, and price. The performance aspect is what people tend to focus on regarding
Moore's law. But the same general law means that as transistor count increases
the cost per chip should decrease for a chip of given power. The law should
also mean that because transistor counts increase more integration can take
place and the size of a given device will decrease. These three factors -- size,
price, and performance -- interact, and are pushed around by market demand.
We've seen a lessening of demand for computer power lately, so price drops
have grabbed the spotlight. People are settling for less performance to get
a better price. The other area of big demand is for laptops. Here you have the
two axes of performance fighting with size. In this case the performance side
of the equation suffers as the size of the equipment shrinks. When the price
axis is tapped to make a cheaper laptop, performance suffers or the laptop has
to be bigger.
These three axes of Moore's law are doing more of a balancing act now than
ever before. The complexity of the balancing act could easily deter IBM or anyone
else from messing with Moore's law, but let's see what happens.