Relentless Doubling of Power Was Great While It Lasted
Smaller, faster and cheaper. The global audience for computers has become completely accustomed to the current rhythm of improvement. It’s been a constant and incredibly reliable pace. In 1965, Intel co-founder Gordon Moore made his eponymous prediction that the number of components in an integrated circuit — essentially the computing power of processors — would double every year for the next 10 years. In 1975, he revised the rate to a doubling every two years. This revision has stayed remarkably accurate for almost 40 years.
But the laws of physics still affect the laws of Moore. As transistors have become harder and harder to shrink, computing firms are starting to look at making better use of the transistors they already have. As Mr Moore admitted last year, it looks like the rate of progress is about to change.
In a recent regulatory filing, Intel hinted at an adjustment to the microprocessor development strategy that to date has been driven by a rigid application of Moore’s Law. It is still sticking with “a regular and predictable upgrade cycle” and introducing a new generation of silicon process technology every two to three years. However, it appears to be moving away from the two-phase “tick tock” strategy in use since 2007, replacing it with a three-phase chip development model known as “process, architecture and optimization”.
Under the “tick tock” model, Intel would create chips based on a new, more-efficient manufacturing technology every two years. This is the “tick”. In the intervening years, the company would introduce new microarchitectures for these chips, giving further improvements in performance. This is the “tock”.
But in 2014, when its first 14-nanometre chips, code-named Broadwell, ticked their way to market, they were nearly a year behind schedule. The “tick” to a 10-nanometre process that was meant to follow the “tock” of 2015’s Skylake microarchitecture has slipped, with Intel predicting that the product is not expected to arrive until 2017.
The expansion of the “tick tock” model is perhaps not surprising. It applied to traditional computing CPUs during a period of predictable advances in technology. However, after a few years of being wrong-footed by the rise of the mobile phone, Intel is reshaping its business to address opportunities in a post-PC world. The company that provides processors for data centres and servers now also offers chipsets for the Internet of things, drones and wearables among many other things.
Moore’s Law was never meant to apply for all time, and all chipmakers will have been making plans for the protracted slowdown in its application. Its end phases are expected to bring changes to an industry that used to rely on steady and sure improvements. As Intel and its rivals move to focus on new computing applications, the next generation of processors will have to be inexpensive and highly energy-efficient. There’s no end of innovation to help create such products, but they won’t follow an immutable rule forever. It seems that even laws are made to be broken.