Alan Boyle writes:Will memristors save Moore's Law? The answer appears to be yes … that is, if you redefine Moore's Law, which has fueled the growth of the computer industry for four decades. Research groups say that memristors, a new type of memory device that's on the verge of going commercial, will dramatically enhance the storage capacity and usability of computers.
HP, the world's top PC manufacturer, today announced a collaboration with memory-chip maker Hynix to get the first memristors to market in three years. One of the first goals will be to create a computer you can “turn on and off like a light bulb,” said Stan Williams, founding director of HP Lab's Information and Quantum Systems Laboratory. But that's just the beginning.
HP isn't the only company joining the memristor revolution: IBM and Samsung have also looked into the technology, and in the journal Nano Letters, Rice University researchers today report the development of silicon-based memristors that they say will extend the limits of circuit miniaturization for years or even decades to come.
Memristors, or "memory resistors," take advantage of the fact that passing electrical current through particular types of material will change the molecular structure of that material so that it "remembers" which way the current was running, and at what voltage, even when the power is turned off.
Memristors are said to represent a "fourth class" of basic electrical circuit, alongside resistors, capacitors and inductors. The concept behind memristors was first proposed in 1971 by circuit theorist Leon Chua, but for decades it was nothing more than a concept.
"It was only two years that we essentially announced that memristors were real, that they're more than a theoretical prediction," Williams said. "To me, it's so amazing that this concept lay dormant for nearly 40 years."
Memristors are built up from tiny sandwiches of thin-film circuitry. HP's experimental devices, for example, use a layer of titanium dioxide with wires that are about 50 nanometers wide. The silicon-oxide circuitry being developed at Rice contains nanocrystal wires as small as 5 nanometers. Layers of nanocircuit sandwiches can be stacked up to create three-dimensional memory arrays.
The result is that huge amounts of data can be retained inside your computer in an instant-on, instant-off mode, with much less energy required for operation.
Once memristors really take hold, consumers may well think of their computers in a completely different way, HP's Williams said. "We do see terabyte thumb drives — multiple terabytes — as being possible," he told me. "You can think about storing lots and lots of high-def video, you can think about storing 3-D video. That's something that's coming down the pipe."
Even processing 3-D video will require quick access to huge amounts of data, he noted. "You have to have something [for memory storage] that's very dense but also very fast," he said. "You have to be able to write to it very, very fast, and you need to be able to do it without the thing burning up."
Thanks to the research being done at HP and elsewhere, memristors could someday do the processing as well.
"I think that the memristor is the gift that's going to keep on giving," Williams said. "I do believe that about 10 years from now we will see memristors used in some type of logic. Either standard logic ... or there's another type of logic that memristors are capable of, and that's what we call synaptic logic. The type of logic that brains use."
Sounds interesting, esp. the part about synaptic logic, which is what neural networks use.
I love when news people fine a fun phrase to use like "Moore's law" and then use it wrong and sound stupid.
Moore's law was that every 18 months you would be able to fit twice as many transistors onto a chip of a certain size, if you arent using transistors then his law isnt applicable at all. It has nothing to do with the speed or capabilities of a chip, just the amount of transistors you can fit on a chip.
Memristors still have a while to go before they become useful for large scale systems, we need to be able to mass produce them cheaply and arange them in a logic that will be compatible with current systems, i dont expect to see them anywhere in a desktop for 10 years.