Slashdot Mirror
IBM Shrinks Bit Size To 12 Atoms
Lucas123 writes "IBM researchers say they've been able to shrink the number of iron atoms it takes to store a bit of data from about one million to 12, which could pave the way for storage devices with capacities that are orders of magnitude greater than today's devices. Andreas Heinrich, who led the IBM Research team on the project for five years, said the team used the tip of a scanning tunneling microscope and unconventional antiferromagnetism to change the bits from zeros to ones. By combining 96 of the atoms, the researchers were able to create bytes — spelling out the word THINK. That solved a theoretical problem of how few atoms it could take to store a bit; now comes the engineering challenge: how to make a mass storage device perform the same feat as scanning tunneling microscope."
...once they have these new mass-storage devices, how can I turn it into a homebrew tunnel scanning microscope?
There are theoretical limits to how much information can be stored in a molecule -- this given by the molar entropy, typically expressed in J/(K*mol). But it can also be expressed, more intuitively, as bits per molecule.
(Yes, you can convert between J/K and bits -- they measure the same thing, degrees of freedom.)
Per this table, iron has a molar entropy of 27.3 J/K*mol, or 4.73 bits/molecule.
IBM is claiming an information density of (1/12) bits/molecule, which is reasonable -- the thermodynamic limit is ~57x greater.
Information theory is life. The rest is just the KL divergence.
There's a better article here which includes some more information on the experiment. In particular the temperature was 0.5K.
Also the computerworld article claims that using an antiferromagnetic arrangement of atoms is an advantage because it pulls the atoms more tightly together. I'm not convinced that this is true but even if it is the effect would be completely negligible. The interesting aspect of this arrangement is that each atom cancels out the magnetic field of the atoms either side of it which should help with data stability (a similar effect is seen in perpendicular recording today).
Unrelatedly: have they/will they publish a paper on this? I can't find anything mentioning a paper in the press releases.
I'm a materials science graduate student, and my research is on semiconductors. While I don't work with materials for data storage, I have a pretty good background in electronic properties of materials so maybe I can shed some light on the situation.
Basically, I suppose this would be hypothetically possible but the problems you'd face would be very, very difficult to solve. The big problem here is that in order to keep something ionized, you would have to completely isolate it from any other atoms that might donate/steal an electron. Again it's hypothetically possible, but impractical considering most of those are noble gasses. Not to mention, storing data as ionized/unionized atoms is fundamentally different from the way we store data now (magnetic domains). I think the more reasonable idea would be to shrink magnetic domains, as well as the number of magnetic domains required to form a bit. If I remember correctly, currently each magnetic domain consists of several hundred atoms and each bit consists of around 100 magnetic domains. As the article states, the best we could get is one atom representing one bit, and the probability of using magnetism over changing to ionization as the mechanism for differentiation between ones and zeroes is very high.
An actual STM instrument is pretty big. About the size of, say, a mini-fridge. But the majority of that is the computer to drive the system, the readout electronics, and the enclosure (to dampen out vibrations, establish vacuum, etc.). The actual readout tip is pretty small: a nano-sized tip attached to ~100 micron 'diving board' assembly.
A related problem with STM is that it's a serial process: you have a small tip that you're scanning over a surface. This makes readout slow. However in a separate project, IBM (and others) has been working on how to solve that: the idea is to use a huge array of tips that scan the surface in parallel (IBM calls it millipede memory). This makes access faster since you can basically stripe the data and read/write in parallel, and it makes random seeks faster since you don't have to move the tip array as far to get to the data you want. It increases complexity, of course, but modern nano-lithography is certainly up to the task of creating arrays of hundreds of thousands of micron-sized tips with associated electronics.
Using tip arrays would make the read/write parts more compact (as compared to having separate parallel STMs, I mean). The enclosure and driving electronics could certainly be miniaturized if there were economic incentive to do so. There's no physical barrier preventing these kinds of machines from being substantially micronized. As others have pointed out, the first magnetic disk read/write systems were rather bulk, and now hard drives can fit in your pocket. It's possible the same thing could happen here. Having said that, current data storage techniques have a huge head-start, so for something like this to catch up to the point where consumers will want to buy it may take some time.