HP Labs has made a nano-scale device which stores data, explains previous anomalies in nano-device characteristics, and may be able to act as a synapse in analogue neural networks. It is also the first physical implementation of the memristor, claims HP, a theoretical partner to resistors, capacitors and inductors invented in 1971.
“Transistors are slightly faster at switching, but this is fast,” HP Labs spokesman David Harrah told Electronics Weekly. “Stan [research head Stan Williams] is talking in terms of taking it down to 14nm and possibly single digits, so it has the potential to go to 1Tbit/cm2.”
Possible in several ionic-electronic systems, the metal-insulator-metal device demonstrated has a stack of two titanium oxide layers its insulator. One is doped with oxygen, the other slightly starved of oxygen.
Electrons flow first through one layer and then the other. “The oxygen creates vacancies and these vacancies migrate to the other layer with current flow,” said Harrah.
That is: the oxygen ions move, causing the boundary between oxygen-rich and oxygen-starved TiO2 to move towards one electrode or the other changing the overall resistance of the insulator layer.
Depending on current direction, resistance increases with time or decreases with time, with the magnitude of resistance change dependent on charge (current x time). When current flow stops, resistance freezes. When current is reversed, resistance change reverses.
Plotted, current and voltage form a hysteresis loop which can be frozen at any point by stopping current flow and the device can therefore store data.
According to research head Williams, this behaviour is only noticeable in nano-scale dielectric layers as these offer the huge potential gradients required to shift ions at reasonable voltages.
Realising that ion drift in insulators can cause hysteresis may explain previously nano-scale electrical anomalies.
“The rich hysteretic I-V characteristics detected in many thin film two-terminal devices can now be understood,” Williams told the journal Nature.
HP also sees the junction as possible component of artificial brains. “It is very similar to a brain synapse,” said spokesman Harrah. “You could simulate a brain with transistors as neurons and a bunch of these as synapses.”
37 years ago, scientist Leon Chua, now at the University of California at Berkeley, mused theoretically that with: resistors linking current and voltage, capacitors linking voltage and charge, and inductors linking current and flux there should be a component that linked charge and flux.
He named it a memristor – where memristance(M) links the two like this: d(flux)=Md(charge).
“In the trivial case of linear elements in which M is a constant, memristance is identical to resistance, and thus of no special interest,” Williams told the journal Nature. “However, if M is itself a function of charge to yield a nonlinear circuit element, then things become quite interesting.”
For limited excursions, Williams’ device is exactly a memristor, said HP.
“This is an amazing development,” Chua said. “It took someone like Stan Williams with a multi-disciplinary background and deep insights to conceive of such a tiny memristor only a few atoms in thickness.”
Atomic force microscope images of 17 HP Labs non-linear devices in a row, each a pair of oxide layers between the single bottom wire and one of the top wires. “The devices act as ‘memory resistors’ [memristors], with the resistance of each device depending on the amount of charge that has moved through each one,” Stanley Williams director of HP’s quantum systems lab, told Electronics Weekly. The wires are 50nm – about 150 atoms – wide.