One of the strangest realities of the past few decades has been the reliability of Moore’s Law. Gordon Moore in 1965 said that the number of transistors in an integrated circuit would double every year. Webopedia says that the prediction of the cycle soon stretched to 18 months.
What always seemed odd about it, and the fact that it has more or less kept pace for so long, is that the rule is a general overview that makes predictions relying upon highly technical developments that Moore had no way of predicting. Other laws, the theory of relativity, for instance, are based on the understanding at a very deep level of the physical world as it exists. Moore’s Law seems to be a statement of what will happen without really having much of an idea of how.
The future of Moore’s Law, which is a keystone of telecommunications networks in general and the explosion of mobility in particular, has always been seen to have one foot on an engineering banana peel. EE Times’ Rick Merritt offers an “advance” story on the Semicon West conference next week in San Francisco that discusses what in essence is the statute of limitations on Moore’s Law:
I wrote my first story about the death of Moore's Law back in the mid-1990s, when I had my first and only interview with Gordon Moore. He said his observation about doubling transistors in a given area every 18 to 24 months was bound to end -- it will hit atomic limits. Before it ends, he added, it will get really hard to do and slow down.
The thinking is that the end days are near. Another potential change to the equation is more subtle but just as potentially meaningful. Last week, I wrote about the impact of the Internet of Things (IoT). A high percentage of the mountain of wireless messages generated by the IoT will be short and just as well accommodated by older network technologies that are slower than LTE.
A Quartz magazine story suggests the same reality for chips. Much of the job that chips will be called on to perform will involve the IoT. Writer Leo Mirani suggests that the focus is moving from an endless cycle of developing faster and smaller chips to those that are less functional – but can get the job done – and perhaps are based on older designs. In other words, the army of IoT chips can move on horses and fire muskets. Writes Mirani:
Companies may do well to focus their energies on incremental improvements such as making chips slightly smaller, rather than the next revolutionary, Moore’s Law-extending processor. Such efforts aren’t particularly glamorous, but they fulfill a real need and perhaps broaden the ways that computing power can improve human life. Indeed, by broadening the functions of these processors and making them cheaper, rather than focusing on the rat-race of building one shiny new super-powerful processor after another when PCs are in decline, there’s a greater—and broader—impact to be made.
He saves an important point for the end: The horsepower-hungry PCs and the rise of devices that focus on a narrower and computationally simpler set of tasks create an environment in which the latest super chip isn’t necessary.
Research on faster chips continues nonetheless. The MIT Technology Review posted a story about what may be the next step for chips. IBM expects to switch the substance with which it makes transistors from silicon to carbon nanotubes by 2020. This will reduce the minimum size of features on transistors from 14 nanometers to 5 nanometers, which translates to faster performance.
This initiative, and other ones like it, will be interesting to watch over the long term. Big Blue’s success or failure will further attest, one way or another, to the durability of Moore’s Law. In the broader view, the changing nature of networking may reduce the importance of the work.