The future of computer memory is largely an unsettled place. This is true for volatile RAM-type system memory, but also non-volatile long-term memory, like hard drives, flash memories, and the like. This unsettled nature makes for a fertile climate for new and unlikely-seeming ideas, like, say, using sound waves to manipulate stored data.
This particular scheme comes courtesy of engineers at the Universities of Sheffield and Leads in the UK. In a paper published in the Applied Physics Letters, they describe a new approach to an experimental technology known as "racetrack memory" utilizing surface acoustic waves, a variety of sound wave existing in only two-dimensions, to manipulate bits of information. The bits move in different desired directions as determined by variations in the sound's pitch, so, in a near-literal sense, the memory is manipulated by "singing."
We all know that conventional hard drive memory, implemented as spinning magnetic platters that can be written and rewritten with data many times over, isn't cutting it anymore. Once a pretty slick technology, hard drives haven't really been keeping up with the rest of the computer in terms of ever-increasing speeds, and they also happen to be pretty unreliable. Something like 15 percent (a rough average) of all hard drives will fail eventually.
What comes after spinning hard drives are solid-state drives (SSD). Flash memory is one variety of SSD, but it's also not ideal—flash drives are fast-ish, but they also become unreliable even more quickly than conventional hard drives and are expensive.
Flash memory stores data electronically rather than magnetically, which it accomplishes thanks to a certain sort of transistor called a floating gate transistor. (In the most basic sense, a transistor is a semiconductor device used to amplify and-or switch electrical signals.) A hard drive, by contrast, stores data by changing the magnetic properties of some material. In a sense, racetrack memory combines the two, offering a solid-state alternative that's based on changing magnetic rather than electrical properties.
This is where the "racetrack" comes in. We take a very, very fine wire—like a thousandth the diameter of a human hair—and twist it into a loop (the racetrack). Different regions of the loop are then magnetised either parallel to the direction of the wire or perpendicular to it (polarized south-north or north-south, in other words). These orientations, or the "domain walls" separating the two, are what represent bits and they can be moved around the loop using an electric current or applied magnetic field.
The MIT video above (unrelated to the current study) explains it better.
Anyhow, we can imagine billions of these racetracks all packed together onto a silicon chip. The catch is that to read and write data off of them, we need to provide a current, which creates heat and costs energy. This is where the new research comes in.
Here, surface acoustic waves (SAWs) generated by a pair of transducers are made to propagate across a surface of racetrack memory units from opposite directions. Where they come together, a standing acoustic wave is formed and this can be used to isolate arrays of domain wall-represented bits. By changing the frequencies of the waves—changing the sounds' pitches—it becomes possible to move the bit arrays in either direction.
"Our proposed approach has a number of attractive features," the study explains. "First, in materials such as lithium niobate, SAWs can propagate distances [a centimeter scales] with little power loss. Potentially, this would allow very large numbers of [domain walls]/devices to be controlled by a single transducer pair, making the approach attractive from the perspective of power efficiency."
Given the nanometer scales involved in racetrack memory loops, a centimeter is a vast distance. It's an impressive effect.
"Because of this, we think a single sound wave could be used to 'sing' to large numbers of nanowires simultaneously, enabling us to move a lot of data using very little power," explains Tom Hayward, a study co-author and engineering professor at the University of Sheffield. "We're now aiming to create prototype devices in which this concept can be fully tested."