Apparently AMD sees the end of Moore’s Law approaching. While the law may cease to be true in the strictest sense, I believe like many futurists that the spirit of the law will continue. The future of Moore’s law is about a continuing growth in the bang for buck that computing delivers.
Gordon Moore, co-founder of Intel, never intended to create a ‘law’ in his own name, he merely observed that the number of transistors on a silicon chip was doubling approximately every two years. This was way back in 1965, and the trend he had observed stretched back to 1958. Amazingly this statement became a ‘law’ because it remained true and continues to be so right up to the present day.
Now we are constructing chips with such tiny components — working on a process at 22 nanometres, the width of just 220 helium atoms — that we are coming up against the limits of the laws of physics. AMD is struggling to shrink its transistors to this scale and beyond, though its main competitor Intel seems confident it will get down to 14nm and even 10nm in the next few years. The slower transition to 22nm is effectively breaking Moore’s law, at least for AMD.
Moore’s Law was always going to expire eventually. There are only so many transistors you can cram on to a chip however incredible your technology. But to understand the spirit of the law you have to go back to what Moore originally said: he wasn’t talking about what was technically feasible, but what was economically feasible. Now if you accept that transistor count used to be a reasonable analogue for computing power, what Moore’s Law really represents is that computing power per pound increases exponentially.
Of course the measure of computing cost is not only the unit’s acquisition: you need to consider its power consumption. This too though has been falling exponentially. We can now deliver more computing power per pound and per watt than ever before. Continuing the trend is a challenge but you only need to look at initiatives like HP’s Moonshot to see the gusto with which it is being attacked (think of Moonshot like Southampton’s Raspberry Pi ‘supercomputer’ on a grand scale). After all, no-one would bet against there being a big market in the future for the infrastructure behind all internet services.
Future of Moore’s Law: Continuing exponential
Ultimately we will reach the outer limits of what can be achieved with silicon. Physicist Michio Kaku predicted some time ago that we will reach that limit by the middle of the next decade. Though this may slow our ability to increase absolute computing power for a short time, it is unlikely to diminish our ability to refine the production processes and hence further diminish the financial and energy cost of each unit. And beyond this there are a number of candidates on the horizon to replace today’s silicon chips.
Many people believe that Moore’s Law or its successors simply cannot continue: exponential growth would ultimately end up consuming everything in the pursuit of computing resource were it to continue. But some don’t rule out even this extreme future. In Charles Stross’ visionary novel Accelerando, humans (or our information-based descendants) begin to demolish whole planets and asteroids to support the manufacture of ever greater volumes of ‘computronium’ — smart matter — the substrate on which their pure-intelligence life form exists.
That today though is the stuff of science fiction, outside of the 20 year window in which I choose to operate. Suffice to say, within that time frame I certainly don’t expect the rate of growth of computing power per watt and per pound to diminish. The future of Moore’s law is safe, in spirit if not in specifics.