Some years ago, the once-admired BBC science documentary slot Horizon ran a program on nanotechnology. This was preposterous in many ways, but one sequence stands out in my mind. Michio Kaku appeared in front of scenes of rioting and mayhem, opining that “the end of Moore’s Law is perhaps the single greatest economic threat to modern society, and unless we deal with it we could be facing economic ruin.” Moore’s law, of course, is the observation, or rather the self-fulfilling prophecy, that the number of transistors on an integrated circuit doubles about every two years, with corresponding exponential growth in computing power.
As Gordon Moore himself observes in a presentation linked from the Intel site, “No Exponential is Forever … but We can Delay Forever“ (2 MB PDF). Many people have prematurely written off the semiconductor industry’s ability to maintain, over forty years, a record of delivering a nearly constant, year on year, percentage shrinking in circuits and increase in computing power. Nonetheless, there will be limits to how far the current CMOS-based technology can be pushed. These limits could arise from fundamental constraints of physics or materials science, or from engineering problems like the difficulties of managing the increasingly problematic heat output of densely packed components, or simply from the economic difficulties of finding business models that can make money in the face of the exponentially increasing cost of plant. The question, then, is not if Moore’s law, for conventional CMOS devices, will run out, but when.
What has underpinned Moore’s law is the International Technology Roadmap for Semiconductors, a document which effectively choreographs the research and development required to deliver the continual incremental improvements on our current technology that are needed to keep Moore’s law on track. It’s a document that outlines the requirements for an increasingly demanding series of linked technological breakthroughs as time marches on; somewhere between 2015 and 2020 a crunch comes, with many problems for which solutions look very elusive. Beyond this time, then, there are three possible outcomes. It could be that these problems, intractable though they look now, will indeed be solved, and Moore’s law will continue through further incremental developments. The history of the semiconductor industry tells us that this possibility should not be lightly dismissed; Moore’s law has already been written off a number of times, only for the creativity and ingenuity of engineers and scientists to overcome what seemed like insuperable problems. The second possibility is that a fundamentally new architecture, quite different from CMOS, will be developed, giving Moore’s law a new lease of life, or even permitting a new jump in computer power. This, of course, is the motivation for a number of fields of nanotechnology. Perhaps spintronics, quantum computing, molecular electronics, or new carbon-based electronics using graphene or nanotubes will be developed to the point of commercialisation in time to save Moore’s law. For the first time, the most recent version of the semiconductor roadmap did raise this possibility, so it deserves to be taken seriously. There is much interesting physics coming out of laboratories around the world in this area. But none of these developments are very close to making it out of the lab into a process or a product, so we need to at least consider the possibility that it won’t arrive in time to save Moore’s law. So what happens if, for the sake of argument, Moore’s law peters out in about ten years time, leaving us with computers perhaps one hundred times more powerful than the ones we have now that take more than a few years to become obsolete. Will our economies collapse and our streets fill with rioters?
It seems unlikely. Undoubtedly, innovation is a major driver of economic growth, and the relentless pace of innovation in the semiconductor industry has contibuted greatly to the growth we’ve seen in the last twenty years. But it’s a mistake to suppose that innovation is synonymous with invention; new ways of using existing inventions can be as great a source of innovation as new inventions themselves. We shouldn’t expect that a period of relatively slow innovation in hardware would mean that there would be no developments in software; on the contrary, as raw computing power gets less superabundant we’d expect ingenuity in making the most of available power to be greatly rewarded. The economics of the industry would change dramatically, of course. As the development cycle lengthened the time needed to amortise the huge capital cost of plant would stretch out and the business would become increasingly commoditised. Even as the performance of chips plateaued, their cost would drop, possibly quite precipitously; these would be the circumstances in which ubiquitous computing truly would take off.
For an analogy, one might want to look a century earlier. Vaclav Smil has argued, in his two-volume history of technology of the late nineteenth and twentieth century (Creating the Twentieth Century: Technical Innovations of 1867-1914 and Their Lasting Impact