While we’re at it at fundamental change in computing, how about this one…
Thanks for the memristor
Feature Could transform computing
By Clive Akass
AN INDUSTRY as hype-ridden as IT is bound to treat all claims of major breakthroughs with a certain scepticism. But the fabrication of the memristor by researchers at HP Labs really could turn out to be the most important development in computer electronics since the invention of the integrated circuit. It is certainly one of the most intriguing.
Memristors hit the headlines again last week when HP announced that it had demonstrated they can be used to create logic circuits that can retain their state without power, which means they can be used simultaneously for computation and fast non-volatile memory.
The company says memristors could bring us superdense non-volatile memory with ten times the speed of flash memory within three years, followed by computers that boot instantly to the state they were in when switched off. And that is just the near-term prospect.
So what is a memristor? For nearly a century and a half, capacitance, resistance and inductance were considered to be the only three factors governing the electrical behaviour of a passive circuit, and the capacitor, resistor and inductor were seen as the three fundamental components.
Then in 1971 a University of California engineer called Leon Chua revisited the basic mathematics of circuits and postulated a fourth factor called memristance, promising a corresponding component called a memristor.
The crucial properties of a memristor would be that its resistance would change with how much charge had flowed into it – or, put another way, how long a voltage was applied. And that when you removed the voltage, it would “remember” that resistance until a voltage was reapplied.
Chua’s work was equivalent in electrical engineering terms to suggesting to mathemeticians that there was another basic arithmetic operation besides addition, subtraction, multiplication and division, and it was widely dismissed as a mathematical game, particularly as nobody could find any memresistive effects. Or rather, as it turned out much later, they did but did not recognise them.
Chua’s paper was largely forgotten for 30 years until it was picked up by Greg Snider, part of an HP Labs team investigating nanoscale electronics. He thought it might explain some odd results the team had been getting. Even then it took years to establish the fact that memristance exists, and that the reason it had gone unnoticed for so long was because it becomes significant only at nanoscale.
Not until 2008 did the HP team leader, Stan Williams, announce that they had developed a memristor, which took the form of a silicon dioxide nanowire doped to increase its conductivity.
Transistor silicon, which has four outer electrons, is similarly doped with elements having either three or five outer electrons which add either positive ‘holes’ or loosely-bound negative electrons to facilitate charge flow.
Dopant atoms in well behaved transistors stay in place within the crystalline structure as charges flow. However in a memristor the atoms themselves move, driven by the enormous electric fields at that occur at nanoscales (one volt across a nanometre would create an electrical charge of a billion volts per metre). The dopant atoms remain in place when the voltage is removed because there is nothing to drive them back, hence the memory effect.
The effect is bi-directional. Shove charge in one direction, and the resistance goes up. Reverse the polarity of the applied voltage and the resistance goes down.
Memristors can be made relatively easily using existing semiconductor fabrication technology but Williams says he expects them to complement, rather than supersede, transistors. And the fact that they can be used to create traditional logic gates does not necessarily mean they will be used that way.
For one thing, chip designers have an enormous investment in their existing circuits, and producing the developer tools and designs for memristors would cost a lot more because the entire architecture would have to be rethought. So the pace and direction of memristor technology development is a business issue as much as a technological one.
And, long before an actual memristor was developed, Chua pointed out that it was very similar to a brain synapse with its ability to sum inputs and store the result. This makes it very interesting for the emerging field of neuromorphic engineering, the development of computer system than mimic the brain.
Rahul Sarpeshkar, associate professor of electrical engineering at MIT, tells in the IEEE’s Spectrum magazine how the inner ear uses 14 microwatts of energy, an amount so small that a single AA battery could supply it for 50 years, and it does what would take a games processor 50 watts and more than a billion floating point operations a second to do.
It does so by doing a lot of analogue preprocessing, by using the structure of the inner ear, before in effect digitising the filtered and compressed result for processing by the brain. This mix of analogue and digital conceivably could be performed by a semiconductor processor chip having a mix of memristors and transistor gates.
Brain-like neural nets are already used for tasks like pattern recognition but currently they are simulated using digital circuits. Synapse-like memristors should be able to do the job more efficiently and cheaply.
It took two decades for transistors to evolve into the integrated circuit, and almost as long for them to supersede vacuum tubes, even though both were used for more or less identical purposes. The memristor is utterly and subtly different from any other component and researchers are well aware that there might be ways of using it that no one has thought of yet. µ