More than half the number of people now alive on earth weren't even born in 1965 so they will
have little appreciation of the progress that has been made in computing over the last 50 years.
Back then in the “swinging sixties” in order to access a computer you would have to use a terminal
connected to a mainframe computer, and that mainframe filled a large room and needed air
conditioning to keep the computer at a stable temperature. It seems an incredible comparison but
nowadays, the smartphone you may have in your pocket has over 3,500 times more computing
power than one of those gigantic 1960s mainframes.
In 1964 Gordon Moore, who was Head of Research at Fairchild Semiconductor, was asked by
Electronics Magazine to submit an article for their 35
annual edition, predicting what was likely to
happen in the next ten years of semiconductor component technology. Moore explains in his own
words how he went about writing this now famous article: “And I looked at the few chips we had
made and noticed we went from a single transistor on a chip to a chip with about eight elements –
transistors and resistors – on it. The new chips coming out had about twice the number of
elements, about 16. And in the laboratory, we were creating chips with about 30 elements and we
were looking at the possibility of making chips with twice that many, around 60 elements on a chip.
Well, I plotted these on a piece of semi-log paper starting with the planar transistor in 1959 and
noticed that, essentially, we were doubling every year. So I took a wild extrapolation and said we’re
going to continue doubling every year and go from about 60 elements at the time to 60,000 in 10
years.” This concept later became known as “Moore's Law” and that rate of progress in the
electronic component industry has continued to the present day. If you own a laptop bought in the
last couple of years its microprocessor is likely to contain around a billion or more transistors.
Moore's Law didn't just continue for ten years, it has proved to have been consistent for the last 50
years, although unless there is a sudden radical technological breakthrough it will probably fail in
around eight more years. A copy of Gordon Moore's original illustration from that historic 1965
article is available as free PowerPoint slide, just click on any of the free download links. To
understand how prescient Gordon Moore was in his article fifty years ago you can read the original
(courtesy of Intel) by clicking on this link.
This exponential development of the microprocessor, and associated integrated components
produced from silicon, has shaped much of the modern world. Moore's Law is often boiled down
to the phrase: “that microprocessors double in complexity every 24 months.” But that
simplification ignores the fact that the acuteness of Gordon Moore's article was the observation
that electronic components were going to get cheaper at the same time as they doubled in
complexity. This is the salient economic aspect of Moore's article, if complex chips were expensive
to produce in volume, their development would have been stymied and their customers would
have been just a handful of rich corporations. Instead complexity and low cost went hand in hand
and enabled the mass production of the electronic goods that we see all around us now.
Gordon Moore went on to found the company Intel that also grew into a colossus on the back of
the personal computer revolution. Those increasingly lower costs enabled computing to expand
from large companies to the down to computers in nearly every home. On the 50
the creation of Moore's Law it is time to take stock, just as Gordon Moore did, and consider what is
likely to happen to electronic component production in the course of the next decade. Over half
the chips that Intel is shipping at the moment are based on a 14-nanometre process, this is a
degree of miniaturisation that equates to a size comparable to a virus. Just to make this clear, one
nanometre is one billionth of a metre so microprocessor production now occurs at an atomic scale
of miniaturisation. As if that isn’t extraordinary enough, Intel expects to produce 10-nanometre
chips in the next year or so and has already taken “an early look” at 7-nanometre chips which may
possibly ship before the end of 2020. After this there may conceivably be a 5-nanometre process
but by then the physical limits of the miniaturisation of chip production will probably have been
reached according to our present knowledge.
You probably know that the function of a transistor is to act as a switch or gate that is either open
or closed. Microprocessors rely on the quantum mechanics of how atoms respond to an electrical
charge to determine whether the gate is open or closed. The diameter of one atom of silicon is
0.27-nanometre and in the 14-nanometre chips that Intel is now shipping only a few dozen atoms
are involved in each gate. In the near future, developing 7 or 5-nanometres per chip, fewer and
fewer atoms will be involved and there’s a problem: they don’t all respond the same way. At the
moment an open or closed state can be determined by what the majority of atoms do, but this
becomes progressively harder when fewer and fewer atoms are involved. The physical limits are
reached when there are so few atoms that their behaviour to an electrical charge becomes
uncertain. If 70% of a dozen or so atoms in a transistor behave in one way we can calculate the
probability of the gate being either open or closed, but when only 51% of a handful of atoms
behave uniformly no amount of error correction can determine whether the gate is open or closed.
It is quite likely that we will not see any further reduction in the size of transistors beyond 5-
One possible way round this problem that Intel is already using is to build the microprocessor up in
three dimensions, layering one level of gates on top of another, yet this approach has its own
limitations which hark back to Gordon Moore's original economic observation. Rapid
microprocessor miniaturisation was predicated as long as costs also continued to shrink. But the
costs of microprocessor production are now so great that the global industry has minimised just
like the size of microprocessors themselves. Last year Intel spent $11.5 billion on research and
development, and another $10 billion on capital expenditure, to finance chip production. The
company had spent a similar amount the preceding year. Few businesses can sustain this amount
of continual expenditure: 10 years ago the number of companies with “state of the art” chip
manufacturing round the globe numbered 18, today there are only four – Intel, Samsung, Global
Foundries and the Taiwan Semiconductor Manufacturing Corporation. In purely economic terms a
limit of four players in a market leads to a lack of competition which means prices are almost
certainly more likely to go up than down. Additionally, these remaining four corporations rely on
other highly specialist companies to manufacture the equipment they need to produce silicon chips
at atomic scales of miniaturisation. And, of course, the problem for the enterprises producing this
extremely technical equipment is that only four companies do not make a market. So we can
determine that overall manufacturing costs are set to increase and, now that the global market is
an oligopoly, the retail prices of all computer products are likely to rise.
Some analysts think that Moore's Law will break down in the next eight to ten years because of the
limitations of the physical size of atoms of silicon, others reckon that the vast sums of money being
spent on research and development by the four remaining chip manufacturers will produce a
technical breakthrough and Moore's Law will continue. Moore's Law is not an immutable “law” of
nature, the term was coined by a Caltech professor and caught everyone's imagination as it seemed
to capture the rapid progress being made in chip production at the time. Putting on my scientific
hat I wouldn't like to bet that a technical breakthrough will allow the fast rate of progress in
microprocessor production to continue. However, in economic terms for the reasons I've already
stated: oligopoly and high costs, I would argue that Moore's Law is already breaking. But not
perhaps in a way that he could have foreseen. The financial ceiling has been reached, Intel's
current generation of 14-nanometre and next generation 10-nanometre microprocessors are
already costing more than $20 billion a year in annual investment to design and produce. Such
massive up-front costs dictate that millions of chips will have to be sold over increasingly longer
time periods to get a return for the amount of investment involved.
In the last 50 years because of the outstanding developments in microprocessor production,
computers have gone from being room-sized to a powerful, super computer in a watch you can
wear on your wrist or, or course, a smart phone. It’s interesting to note that there are other
technologies involved in a smart phone that haven't developed nearly as fast as microprocessors,
which is why your smartphone battery irritatingly needs charging every night, and doesn't last a
month or even a year.
...with analysis & insight...
Is Moore’s Law already breaking?