Fifty years ago, Gordon Moore, the co-founder of the chip manufacturer Intel described a regularity he had observed that would one day make him a household name. What he had noticed was that the number of transistors that could be fitted on a given area of silicon doubled roughly every two years. And since transistor density is correlated with computing power, that meant that computing power doubled every two years. Thus was born Moore’s law.
At the beginning, few outside of the computer industry appreciated the significance of this. Humanity, it turns out, is not good at understanding the power of doubling – until it’s too late. Remember the fable about the emperor and the man who invented chess. When asked to name his reward, the inventor requested that one grain or rice be placed on the first square of the board, two on the second, four on the third and so on. The Emperor readily agrees, not realizing that when you get to the 64th square the pile of rice required would be bigger than Mount Everest.
If you apply the metaphor to computing power, then we’re on the 25th square of the board – which explains why your smartphone has more processing power than even the big mainframe computers of yesteryear. It also explains why some writers (this columnist included) have often recycled a confident mantra: if all your problem requires for its solution is more computing power, then consider it solved – not tomorrow, perhaps, but in a few years at most. Moore’s law will take care of it.
And indeed, up to now it did. The semiconductor industry continued to stuff more and more transistors into smaller and smaller areas. But looming over this fabulous progress there was always a shadow, in the shape of the laws of physics. As the components on processor chips got smaller and were packed at increasing densities, and electrons had to move faster and faster through silicon circuits that got smaller and smaller, the chips got too hot. And they consumed too much power. At the moment, circuit features on advanced microprocessors are about 14 nanometres (one billionth of a metre) across, and it’s conceivable that the industry could get them down to 2-3 nanometres. But at that point we really will hit a wall, because then quantum effects will kick in that will make the devices chronically unreliable. Moore’s law will cease to apply.
We knew the wall was coming, of course, but most people didn’t worry about it because they assumed we’d find another processor technology that would enable us to slip the surly bonds of silicon. No such magic bullet has materialized, however. So why are senior semiconductor executives not queuing up to throw themselves off skyscrapers? After all, they run companies which are in thrall to Wall Street’s insatiable appetite for growth.
The answer suggested by a terrific article by M Mitchell Waldrop in this week’s Nature is that industry leaders have decided to sidestep the problem. (It’s a bit like the old joke about how many Microsoft engineers it takes to change a light bulb. The answer is none: Microsoft simply defines darkness as the new standard.) So – the industry’s leaders muse – maybe raw processing power isn’t all it’s cracked up to be. After all, the computers most people will use in the future will be smartphones, not clunky desktop machines. If they want raw computing power, they can turn to cloud computing like that provided by Amazon’s “elastic cloud”, EC2 or Microsoft’s Azure platform. But for normal people, the processing power that matters is that needed to run their smartphone apps, GPS, cameras and web browsers. And the things they worry about are weight, heat, battery life, portability, size and, of course, cost.
If ever a switch deserved that much‑abused term “paradigm shift”, this is it. And Waldrop produces evidence that it is indeed under way. It turns out that “since the 1990s, the semiconductor industry has released a research road map every two years to coordinate what its hundreds of manufacturers and suppliers are doing to stay in step with [Moore’s] law — a strategy sometimes called More Moore.” But that’s about to become history, he says. The industry road map to be released next month will be a research and development plan that, for the first time ever, will not be based on Moore’s law but on looking at applications that people actually use and trying to infer what kinds of processors will be needed to run them. It all makes good sense, of course. But it’s also the end of an era.