Magnetic media, in the form of disk and tape drives, has been the dominant way of storing bits. But the speed and low power of flash memory has been displacing it from consumer systems, and various forms of long-term memory are in development that are even faster. But a new paper suggests that magnetic media may still be competitive—you just have to stop reading and writing it with magnets.
Using a specific form of garnet and some ultrafast laser pulses, a Dutch-Polish team of researchers performed what they suspect is the fastest read/write of magnetic media ever. And, for good measure, the process was extremely energy efficient.
Heat is actually a problem for both hard drives and flash. Although it doesn’t create a problem in most consumer systems, dealing with excess heat is a major issue in data centers. The problem, according to the authors of the new paper, is one of scale. While we can calculate the minimum energy needed to flip a magnetic bit, we use much more than that to ensure that every bit gets written as intended. Eight orders of magnitude more, in fact. Most of that excess energy ends up dissipating into the environment, where it ends up as heat.
The authors decided to find a way around using electromagnetic switching of bits. Instead, bits are read and written by shining light at a largely transparent medium. Presumably, a couple of well-designed mirrors could carry any waste energy away.
The magnetic medium that makes this all work is an yttrium-iron garnet crystal doped with cobalt. The garnet itself is largely transparent to light, but the cobalt atoms provide it with interesting features. They end up oriented by the crystal structure, which places their electron’s orbitals in a specific geometry. With the right wavelength of polarized light, however, the orientation can be shifted to any of three others. Photons that power this shift are absorbed, and any that aren’t simply pass through the transparent garnet crystal.
Changing the cobalt’s magnetic orientation has a dramatic effect on the propagation of light at another wavelength. The authors showed that they could switch patches of the garnet from fully reflective to fully absorbing, based on the precise wavelength and polarization of a laser pulse. Thus, the initial pulse of laser can act to write a bit that can then be read by a second pulse. The authors showed that, once written, bits remain stable for days, but the entire substrate can be erased with an external magnetic field.
(Note that this is different from the technology in magneto-optical drives, which would use lasers to heat a part of a disk in order to make it susceptible to change by an external magnetic field. No external magnet is ever involved here, and the lasers aren’t doing any heating.)
The striking thing is that both the read and write laser’s functions were performed with femtosecond lasers: “for the recording and reading out, we used just two femtosecond laser pulses; to the best of our knowledge this experiment is the fastest-ever write–read magnetic recording event.”
The authors estimate that the energy lost when writing a bit 20 nanometers square would only dissipate 22 attojoules (that’s 10-18 joules). In flash memory, by contrast, each written bit dissipates 10 nanojoules into the hardware—that’s nine orders of magnitude higher.
In principle, all this is fantastically good. But it leaves out something important: that femtosecond laser. Even the most compact femtosecond is typically the size of a laptop hard drive. And they’re only capable of firing pulses at about 100MHz, which means we’d need to divide up each pulse and/or have a number of lasers firing in parallel to get really good bandwidth. And, finally, they’re not especially energy efficient. So, the energy advantage you get from reading and writing with light might be swallowed entirely by the process of generating the light.
Still, it’s pretty neat tech, and someone might one day find a way of working around some of these problems. And, if there’s any indication that this could scale beyond scientific curiosity, you can bet the authors of this paper are trying them out.
Nature, 2016. DOI: 10.1038/nature20807 (About DOIs).