There’s a new law in tech town, guiding how we plan our computational future. If you think mobile is interesting now, you ain’t seen nothin’ yet.

Last month Stanford engineering professor Jonathan Koomey Ph.D. gave a brief presentation at VERGE, a sustainability conference in Washington DC. Koomey was not listed as one of the top speakers; that honor was reserved for various government officials and philanthropists.  But it was Koomey’s presentation that got them talking.  He explained in 10 minutes how computing efficiency—the number of computations completed per kilowatt-hour of electricity used—has doubled about every 18 months, going all the way back to Eniac and Univac I in the 1940s. Such an increase translates into a 100-fold increase in the electrical efficiency of computing every decade.

The first article on this rate of change was just published today (in MIT Technology Review), but already people are referring to Koomey’s Law as the energy equivalent of Moore’s Law.

Koomey’s Law identifies a remarkable rate of change, but what does it mean on a practical level? First of all, the current generation of smartphones and tablets owe their existence to this trend. It was inevitable they would develop. As the energy efficiency of microprocessors increased, they were able to shrink and still perform. Looking forward the implication is for even more powerful mobile computing, as well as sensors and controls that use very little energy and perform a wide array of tasks. As Koomey writes:

“The most important future effect is that the power needed to perform a task requiring a fixed number of computations will continue to fall by half every 1.5 years (or a factor of 100 every decade). As a result, even smaller and less power-intensive computing devices will proliferate, paving the way for new mobile computing and communications applications that vastly increase our ability to collect and use data in real time.”

The new chips and devices Koomey envisions are already in the lab.  They will scavenge ambient energy flows from the environment, from light, heat, motion, radio waves, solar radiation, and other existing sources. They will run for decades without need for refueling or a change of batteries.

Koomey sees deeper implications for the next wave of computing. Customized data collection and more precise control over processing opens the door to real-time analysis, enabling an “internet of things” making smartphones seem quaint. The idea of ‘big data’ is popular today, but Koomey says it is time to start thinking about nanodata, data about individual transactions collected on a massive scale. “When analyzed in the right way [nanodata] can lead to great insight,” says Koomey.

Another benefit is more direct control over processes; always improving the ability to do exactly what we want to accomplish with computer processing using a minimal amount of energy.

The challenge is huge. Software engineers will have to provide new ways to pull insights from the massive amounts of data that will become available. Koomey was speaking at a conference on sustainability, and he pointed out the revolution in computational efficiency means we will have new tools at our disposal to more effectively manage energy supply and demand.

What’s the run rate?

In recent years some have questions the continued validity of Moore’s Law, although recent reductions in CPU size are keeping pace with the basic premise. Koomey sees plenty of room for continued improvement in the increase of computational energy efficiency. In 1985 famous physicist Richard Feynman envisioned a theoretical limit of a 3-atom transistor, and that if such a transistor was built it would have energy efficiency 100 billion times greater than the transistors on the market in 1985.

If Koomey’s trend line continues, we will hit the Feynman line in 2041. Koomey says that from 1985 to 2012 there has been a 40,000-fold increase in energy efficiency, so he sees a rosy future for his new law, at least for the next three decades. After that, as Feynman speculated, continued advances in computational energy efficiency will require single-atom transistors or transistors based on individual nuclei inside an atom. It just so happens researchers at Purdue University and the University of New South Wales have recently created a reliable one-atom transistor.

Below is the video from Koomey’s 10-minute presentation at the VERGE conference in March, which ends with a challenge to scientists and engineers to follow Alan Kay’s notion that “the best way to predict the future is to invent it.”