'Electronic blood' for computer
IBM has unveiled a prototype of a new brain-inspired computer powered by what it calls "electronic blood".
The firm says it is learning from nature by building computers fuelled and cooled by a liquid, like our minds.
The human brain packs phenomenal computing power into a tiny space and uses only 20 watts of energy - an efficiency IBM is keen to match.
Its new "redox flow" system pumps an electrolyte "blood" through a computer, carrying power in and taking heat out.
A very basic model was demonstrated this week at the technology giant's Zurich lab by Dr Patrick Ruch and Dr Bruno Michel.
Their vision is that by 2060, a one petaflop computer that would fill half a football field today, will fit on your desktop.
"We want to fit a supercomputer inside a sugarcube. To do that, we need a paradigm shift in electronics - we need to be motivated by our brain," says Michel.
"The human brain is 10,000 times more dense and efficient than any computer today.
"That's possible because it uses only one - extremely efficient - network of capillaries and blood vessels to transport heat and energy - all at the same time."
IBM's brainiest computer to date is Watson, which famously trounced two champions of the US TV quiz show Jeopardy.
The victory was hailed as a landmark for cognitive computing - machine had surpassed man.
But the contest was unfair, says Michel. The brains of Ken Jennings and Brad Rutter ran on only 20 watts of energy, whereas Watson needed 85,000 watts.
Energy efficiency - not raw computing power - is the guiding principle for the next generation of computer chips, IBM believes.
Our current 2D silicon chips, which for half a century have doubled in power through Moore's Law, are approaching a physical limit where they cannot shrink further without overheating.
"The computer industry uses $30bn of energy and throws it out of the window. We're creating hot air for $30bn," says Michel.
"Ninety-nine percent of a computer's volume is devoted to cooling and powering. Only 1% is used to process information. And we think we've built a good computer?"
"The brain uses 40% of its volume for functional performance - and only 10% for energy and cooling."
Michel's vision is for a new "bionic" computing architecture, inspired by one of the laws of nature - allometric scaling - where an animal's metabolic power increases with its body size.
An elephant, for example, weighs as much as a million mice. But it consumes 30 times less energy, and can perform a task even a million mice cannot accomplish.
The same principle holds true in computing, says Michel, whose bionic vision has three core design features.
The first is 3D architecture, with chips stacked high, and memory storage units interwoven with processors.
"It's the difference between a low-rise building, where everything is spread out flat, and a high rise building. You shorten the connection distances," says Matthias Kaiserswerth, director of IBM Zurich.
But there is a very good reason today's chips are gridiron pancakes - exposure to the air is critical to dissipate the intense heat generated by ever-smaller transistors.
Piling chips on top of one another locks this heat inside - a major roadblock to 3D computing.
IBM's solution is integrated liquid cooling - where chips are interlayered with tiny water pipes.
The art of liquid cooling has been demonstrated by Aquasar and put to work inside the German supercomputer SuperMUC which - perversely - harnesses warm water to cool its circuits.
SuperMUC consumes 40% less electricity as a result.
But for IBM to truly match the marvels of the brain, there is a third evolutionary step it must achieve - simultaneous liquid fuelling and cooling.
Just as blood gives sugar in one hand and takes heat with another, IBM is looking for a fluid that can multitask.
Vanadium is the best performer in their current laboratory test system - a type of redox flow unit - similar to a simple battery.
First a liquid - the electrolyte - is charged via electrodes, then pumped into the computer, where it discharges energy to the chip.
Redox flow is far from a new technology, and neither is it especially complex.
But IBM is the first to stake its chips on this "electronic blood" as the food of future computers - and will attempt to optimise it over the coming decades to achieve zettascale computing.
"To power a zettascale computer today would take more electricity than is produced in the entire world," says Michel.
He is confident that the design hurdles in his bionic model can be surmounted - not least that a whole additional unit is needed to charge the liquid.
And while other labs are betting on spintronics, quantum computing, or photonics to take us beyond silicon, the Zurich team believes the real answer lies right behind our eyes.
"Just as computers help us understand our brains, if we understand our brains we'll make better computers," says director Matthias Kaiserswerth.
He would like to see a future Watson win Jeopardy on a level playing field.
Other experts in computing agree that IBM's 3D principles are sound. But as to whether bionic computing will be the breakthrough technology, the jury is out.
"The idea of using a fluid to both power and cool strikes me as very novel engineering - killing two birds with one stone," says Prof Alan Woodward, of the University of Surrey's computing department.
"But every form of future computing has its champions - whether it be quantum computing, DNA computing or neuromorphic computing.
"There is a long way to go from the lab to having one of these sitting under your desk."
Prof Steve Furber, leader of the SpiNNaker project agrees that "going into the third dimension" has more to offer than continually shrinking transistors.
"The big issue with 3D computing is getting the heat out - and liquid cooling could be very effective if integrated into 3D systems as proposed here," he told the BBC.
"But all of the above will not get electronics down to the energy-efficiency of the brain.
"That will require many more changes, including a move to analogue computation instead of digital.
"It will also involve breakthroughs in new non-Turing models of computation, for example based on an understanding of how the brain processes information."