|Making memories: This colorized|
atomic-force microscopy image shows
17 memristors. The circuit elements,
shown in green, are formed at the
crossroads of metal nanowires.
Credit: StanWilliams, HP Labs
Memristors are nanoscale electronic switches that have a variable resistance, and can retain their resistance even when the power is switched off. This makes them similar to the transistors used to store data in flash memory. But memristors are considerably smaller--as small as three nanometers. In contrast, manufacturers are experimenting with flash memory components that are 20 nanometers in size.
"The goal is to be at least double whatever flash memory is in three years--we know we'll beat flash in speed, power, and endurance, and we want to beat it in density, too," says Stanley Williams, a senior fellow at HP who has been developing memristors in his lab for about five years.
HP makes memristors by laying down parallel metal nanowires onto a substrate, coating them with a layer of titanium dioxide, and placing a second layer of nanowires perpendicular to the first layer. Where the wires cross, a memristor is formed. HP expects the first devices containing memristors to offer about 20 gigabytes of storage per square centimeter, twice the projected capacity of flash at this time. The company has dubbed memristor-based data storage "ReRAM", which stands for Resistive Random Access Memory.
Like other silicon technologies, flash memory is approaching the physical limits of what's possible in miniaturization. Flash memory also wears out after about 100,000 read-write cycles (longer than the lifetime of most gadgets), while lab tests have shown that memristors can withstand up to about a million read-write cycles.
Under the terms of the new agreement, HP will maintain the intellectual property related to memristors. Hynix will make and sell memristor memory to HP and other customers. Williams says the company's goal is to encourage the industry to adopt memristor memory. "The economic benefit to HP will be as the first mover," says Jim McGregor, chief technology strategist at industry analyst firm InStat.
Displacing flash could still take years and billions of dollars, and the industry has other experimental kinds of memory to consider, notes McGregor. Researchers are working on phase-change and ferroelectric materials that can make new forms of memory. McGregor believes that, given the likelihood of speed bumps in manufacturing, it's unlikely a commercial memristor product will be available in 2013, as HP and Hynix predict.
But Williams does not foresee any major manufacturing hurdles. He says HP has been working for the past year on prototype devices with an undisclosed semiconductor manufacturer. Williams adds that memristors can be made with materials and machinery already present in semiconductor factories.
Dan Olds, a consultant with Gabriel Consulting Group in Beaverton, Oregon, is optimistic about the technology. "The sky's the limit if they can deliver on the promise of memristors--the question is at what price, and how fast prices will come down," he says. "Any new technology is a crapshoot, but if it's a matter of engineering and not basic research, then you feel more confident betting on it."