————————————————————————————
Monthly Discussion
Moore’s Law
Gordon
E. Moore co-founded Intel in 1968 and directed the company for many years. In
1965 he made his famous observation that the press called "Moore's
Law". In his original paper, Moore observed an
exponential growth in the number of transistors per integrated circuit and
predicted that this trend would continue. Through Intel's relentless technology
advances, Moore's Law—the doubling of the number of transistors on a chip every
couple of years—has been maintained, and still holds true today. Intel expects
that it will continue at least through the end of this decade, reaching 1
billion transistors on a microchip.
But if Moore’s law is a natural
law, it should follow an S-curve rather than a simple exponential. So I decided
to graph the evolution of the number of transistors as a function of time from
the commercial appearance of the first integrated microchips. Exhibit 3 shows
the data as well as an S-curve that is well adapted to them.
Exhibit
3. Data and S-curve fit (purple line)
for the evolution of the number of transistors on a chip. The vertical scale is
logarithmic to accommodate the many-order-of-magnitude rise. In turn, the
logarithmic scale makes an exponential (and the early part of an S-curve)
appear as a straight line.
The
data line up on a straight line overlapping the early part of the S-curve,
which is very similar to an exponential. But as we approach the mid-point of
the S-curve (around 2012), the pattern begins deviating from an exponential. In
fact, after 2012 the rate of growth slows down and the S-curve begins to
approach a ceiling of about 2.7 million transistors. All and all Intel’s
expectations are reasonable; 1 billion transistors on a chip should appear
around 2010.
However, progress is better represented by the rate
of growth. The annual increments in the number of transistors crammed into a
chip are shown in Exhibit 4. Here we see a peak around 2012 followed by a
decline. The decline means that the chips appearing in the market from then
onward will be better by a progressively diminishing amount.
Such a decline does not signal bad news. It signals
the industry’s maturity. A mature industry, for example that of automobiles,
does not feature increases in engine power for new car models. This may
eliminate easy gains but does not prevent companies in the car industry from
doing well. Similarly, the semiconductor industry will reach maturity toward
the end of the decade and computer performance will stabilize in the 2020s. I
believe this will be conducive to the economic boom and the stability
anticipated for that period (see Kondratieff cycle in Predictions),
because rapidly increasing computer performance can have adverse effects on the
work ethic.*
Undoubtedly, new technologies will not cease to
appear. People are already talking about quantum computers, which themselves
may follow Moore’s law. But as airplanes replace car for intercity traveling
(see Predictions), new computer technologies will replace Intel’s
technology. But substitutions of this magnitude are slow and typically extend
beyond one Kondratieff’s cycle.
For what concerns Intel and the near future of
computers, the pot will soon stop boiling and begin simmering.
Exhibit
4. This graph is obtained from Exhibit
3 through successive subtractions of adjacent points, so as to depict
increments per years. We thus obtain a lifecycle for the growth process. The
peak of the lifecycle is around 2012. The vertical scale is again logarithmic.