Inside the memristor

This is the sexy part. The very sexy part.

Like a capacitor, a memristor has a ‘memory’, but unlike a capacitor the memory is a function of the fluctuation in current that changes its resistance. Said another way, a memristor will ‘remember’ the last charge that passed through it. Which just happens to be a feature of non-volatile RAM, one of the perceived benefits to the new technology.

Chua had postulated that something like the memristor should exist based on the four electromagnetic qualities of current, voltage, charge, and magnetic flux. Electrical engineering teaches that resistors relate current to voltage, capacitors relate voltage to charge, and inductors relate current to magnetic flux. So where was the piece of the pie that related charge to magnetic flux? This would be one of its defining properties: becoming more or less resistive depending on the charge that flowed through it.

According to Williams, this works through virtue of hysteresis, where the rate of change of the memristor speeds up as it moves from one state to another. This is one of the ‘anomalies’ that was previously explained away by current circuit theory, but which is a core feature of the memristor.

An example of this is the way titanium dioxide changes its resistance in the presence of oxygen, an effect which electrical engineers haven’t been able to explain but currently exploit to create oxygen sensors. But this mysterious effect had always been a pointer to the existence of the memristor. Indeed, Williams and his team used titanium dioxide in construction of the first memristor.

Williams’ proof comes in the form of a thin pair of layers of titanium dioxide inserted between a pair of platinum electrodes. One layer of titanium dioxide has depleted oxygen atoms which, when a charge is applied, shift between layers, in turn changing the resistance of the film. When a positive charge is applied, the oxygen atoms spread out to the second layer increasing the resistance; according to Williams, by a thousand-fold or more. When the charge is reversed, the atoms return to the first layer, decreasing resistance. Most importantly of all, the atoms remain in whatever state they are in when the charge is removed, and with each ‘memristor’ capable of represeting ‘1’ or ‘0’ through its level of resistance, you have the building blocks for non-volatile memory. While obviously not tested, the physics model suggests a memristor should be able to hold its state for years.

It’s predicted that the energy use to switch a nanoscale memristor compares to that of flash memory, but with lower energy requirements to read it. HP has already applied its prototype memristors in its ‘ultra-high-density crossbar switches’ using nanowires to compress 100 gigabits into the same space that current flash memory can use to store only 16 gigabits. That’s six times the space efficiency, while using less power; and it’s still only in the labs testing phase. The implications for SSDs (solid-state drives) could be immense – right now SSDs are exceedingly fast and use very little power compared to hard drives, but can’t match hard drive densities. Memristors could change that.

Aside from memory, the memristor, given its non-volatile properties, could also lend itself well to FPGA (Field Programmable Gate Array) designs. Williams sees a combination of transistor and memristor packages in the future that could allow high-efficiency, low-power and low-heat devices perfect for mobile applications.

Indeed, the addition of memristors could revolutionise traditional transistor heavy designs. “Instead of increasing the number of transistors on a circuit, we could create a hybrid circuit with fewer transistors but the addition of memristors – and more functionality,” Williams says.

As some are saying – not the least Williams and Chua – the memristor is a revolution. Getting it to manufacturing, however, could be a way off yet – especially if, as the pair suggest, memristors are throwing current theory out the window and textbooks need to be rewritten. Established technology vendors like Intel, AMD, and NVIDIA have their entire business models built around the transistor; re-inventing that will take time, if indeed they are inclined to do so at all.