After he wrote his seminal paper in 1971 on the four fundamental classes of electronic devices, Professor Leon Chua had to wait almost 40 years for someone to come up with a device that had the right properties to complete the set: a resistor with a memory.

Once a team at HP Labs led by Stan Williams had demonstrated memristor behaviour with devices they made, old research began to pop up out of the woodwork showing that, long before the device had a name, the behaviour was all over nature.

Chua described in his talk at the recent Design Automation and Test in Europe conference in Dresden how a model of brain cell behaviour developed in the 1950s, which puzzled electronics designers for more than a decade because it seemed to require inductors the size of a kettle to function using 'classical' devices, showed memristor-style operation - a telltale pinched hysteresis loop on an IV graph.

The archive dates back even further, it seems. In the latest issue of Nature Materials, Chua together with Professor Chris Tomazou and Themistoklis Prodromakis from Imperial College, London claim experiments on electric arcs and thermistors from the 19th Century show the same kind of hysteresis. Memristors, it seems, are everywhere and apparently predate the other three devices in order of discovery - they just didn't become recognised for close to two centuries.

The one place where the memristor is not - although that is open to debate given the way in which the hysteresis curves keeping popping up - is in our computers. But that could change over the next decade. Williams claimed at DATE that the obstacles to using memristors in computers for memory as well as logic are being overcome.

The regular structure of memristor arrays makes the technology a good candidate for replacing transistors in future machines. The increasing problems with defining transistors in sub-50nm processes tip the balance in favour of technologies that can use regular arrays and self-assembly techniques. The work on memristors so far ticks both of those boxes. The question is how the memristor will begin to displace the transistor. Memory is probably first but logic may not be that far behind, especially for low-cost systems that do not need very high performance.