IBM Announces Breakthrough in Nanotech

[Darth Raptor]

IBM researchers this week announced they've made major strides in nanotechnology by studying how to build storage and other computing devices out of components no bigger than a few atoms or molecules.

Researchers at the company's Almaden Research Center in San Jose, Calif., report in Science that magnetic anisotropy could eventually be used to store information in individual atoms, paving the way to pack as much as 150 trillion bits of data per square inch, 1,000 times more than current data storage densities. In other words, the ability to store data in individual atoms could lead to devices capable of storing the equivalent of 30,000 movies in a device the size of an iPod.

Anisotropy measures how long a magnet has a pull in any single direction. "Every atom has a magnet inside," says Cyrus Hirjibehedin, a researcher at the Almaden lab, noting that the magnetic orientation of an atom is called its "spin." "We want to understand the properties of an atom and were able to measure the anisotropy for a single atom in a particular environment."

Almaden researchers used IBM's scanning tunneling microscope to manipulate individual iron atoms and arrange them with atomic precision on a specially prepared copper surface; scientists previously were unable to measure the magnetic anisotropy of a single atom. IBM used the microscope to determine the orientation and strength of the magnetic anisotropy of each iron atom.

"Now we have a means for understanding anisotropy," says Andreas Heinrich, manager of Almaden's Scanning Tunneling Microscopy lab. The next step, he says, is fashioning a system in which the atom's spin is stable enough to be used for data storage--something that scientists may achieve in several years or, says Heinrich, may not even be possible. "Our job is to jump ahead," he says. "We hope to make a drastic change rather than incremental improvements."

Another Science report describes research by scientists at IBM's Zurich Research Lab in Switzerland on ways to use a single molecule to perform many of the same functions now carried out by silicon. The study indicates that it's possible to turn a single molecule into a switch without disrupting its outer shell--a significant step toward building computing elements at the molecular scale that are vastly smaller, faster and use less energy than today's computer chips and memory devices.

Switches inside computer chips turn the flow of electrons on and off and, when put together, form the logic gates that make up the electrical circuits of the computer processors. Having ever-smaller switches allows the circuits to be shrunk to ever-tinier sizes, making it possible to crowd more circuits into a processor, boosting speed and performance.

Researchers at IBM and elsewhere previously demonstrated switching within single molecules, but the molecules would change their shape when switched, making them unsuitable for building logic gates for computer chips or memory elements.

Next up for the Zurich research team: building a series of these molecules into a circuit, and then figuring out how to link them to make a molecular chip.

Um, wow.

So I guess this would pretty much be the physical limit of computing using binary electrical signals, huh?

[Spin Echo]

Um, wow.

So I guess this would pretty much be the physical limit of computing using binary electrical signals, huh?

Well, I imagine if you went from spin 1/2 to higher spin molecules you could move past the binary limitations.

What I don't get is how they expect to keep the spins from relaxing and interacting. Need to track down the original article...

[Sikon]

So I guess this would pretty much be the physical limit of computing using binary electrical signals, huh?

Even the awesome 150 trillion bits of data per square inch figure in the article wouldn't be remotely close to the absolute physical limit. Even that is mainly "2D" as opposed to "3D" data storage.

Here's a random example for silicon:

One mole of silicon is 28.09 grams (considering the atomic mass), while its typical density is around 2.3 g/cm^3, so one cubic centimeter of silicon corresponds to ~ 0.083 moles of silicon. A mole is 6.022 * 10^23 atoms (Avogradro's number). A cubic centimeter of silicon contains about 5 * 10^22 atoms.

That's 50 billion trillion atoms per cubic centimeter in the case of silicon.

Silicon is just a random example. One could have calculated for carbon, iron, or any of a variety of substances, or a mixture of substances, and similarly obtained a figure at least tens of billions of trillions of atoms per cubic centimeter of solid material.

For the theoretical physical limit for sufficiently advanced technology, consider if the ratio of the number of bits stored to the number of atoms in the computer's storage device could be up to X.

The number of atoms in the computer is tens of billions of trillions per cubic centimeter. And X wouldn't be many orders of magnitude less than unity or better for the ultimate technology possible within physical laws. (There have been theoretical techniques imagined of storing even a number of bits per atom, although, of course, not all the atoms in any storage device would be storing data). So, the conclusion is that physical laws permit having many billions of trillions of bits of data storage per cubic centimeter.

If technology ever someday approached the true physical limit of computing, for 3D data storage, performance would not be describable as 150 trillion bits per square inch but rather literally many billions of trillions of bits per cubic inch.

For example, as awesome as it would be to store the equivalent of 30,000 movies in a device the size of an iPod as described in the opening post article, even that is many orders of magnitude less than the limit possible within the laws of physics.

The equivalent of at least many trillions of movies could be stored in a device the size of an iPod, if technology ever approached true physical limits for theoretical 3D data storage.

Of course, the point is not to literally suggest such being used for storing trillions of movies but just to observe how theoretical limits differ from current computing by so many orders of magnitude.

With that said, the performance described in the article is, obviously, still extreme at a thousand times current data storage density.

Techniques like the article's described manipulation of a small number of individual atoms with a STM electron microscope have been done for a number of years without that leading to a practical method of chip manufacture on that scale yet, since it would be so different of a challenge to have a manufacturing method affordably working with not just a few but sufficient trillions of atoms at once.

Still, it's definitely at least conceivable that a 1000x improvement could occur between now and later in this century, analogous to the history of such improvements over past decades. And, though Moore's law (e.g. >= ~ 1 order of magnitude performance increase per decade) must break down someday, it hasn't yet.

[Molyneux]

Every time I think I'm impressed with how much storage we can pack into a small space, for how cheaply, something new comes out that's even more impressive...I guess that's Moore's Law for you.

Good luck to them, and I do hope that this pans out quickly!

[Ariphaos]

Every time I think I'm impressed with how much storage we can pack into a small space, for how cheaply, something new comes out that's even more impressive...I guess that's Moore's Law for you.

Good luck to them, and I do hope that this pans out quickly!

There will, eventually, be a limit, though given that the transition to 65 nm is already underway (laying the groundwork for 45 nm, after 90 was supposed to be the 'limit' due to quantum tunneling according to some), I'd be hesitant to claim just where it is.

[His Divine Shadow]

How bump-resistant are these atomic sized thingies going be anyway?

[Lord of the Abyss]

How bump-resistant are these atomic sized thingies going be anyway?

I'm not an engineer, but I'd think they'd be pretty resistant. Structures on that scale have very little mass, and therefore very little inertia. If anything broke from a bump, I'd expect it to be whatever substrate they are a part of, not the microscopic components.

[Admiral Valdemar]

Indeed. Think of how tough cells can be, then imagine something even hardier. At those scales, things are quite difficult to mangle.

[Darth Wong]

How bump-resistant are these atomic sized thingies going be anyway?

When you're very small, physical accelerations and decelerations are not a problem. Heat, however, is a big problem.

[Admiral Valdemar]

This is where the likes of reversible computing come in. Using traditional computing techniques, the waste heat would be a big interference factor with the delicate calculations taking place and at those sizes, a bit of molecular bouncing can skew things quite a bit. Anything at that level will likely have thermal properties act as the limiting factor. Enzymes, for instance, have very specific optimal temperatures.

[Starglider]

This is where the likes of reversible computing come in. Using traditional computing techniques, the waste heat would be a big interference factor with the delicate calculations taking place and at those sizes, a bit of molecular bouncing can skew things quite a bit. Anything at that level will likely have thermal properties act as the limiting factor. Enzymes, for instance, have very specific optimal temperatures.

Storage like this is almost certainly going to be very slow compared to electronic solutions made with the same technology base, it isn't going to be used for main or cache memory. As such keeping processor heat down isn't really relevant to the viability of this tech unless its used in a very small device (which usually don't have enough computing power to pose a heat dissipation challenge anyway). However it is likely that a very large fraction of that storage will be devoted to error correction and redundancy. It may well make sense to use a highly unreliable medium with lots of redundancy as opposed to a more reliable lower density medium, if the differential in density is high enough; you can always use arbitrary amounts of error correction to bring the probability of failure up to the required level (though almost always with penalties to read and write speeds - whether that matters depends on how parallelised access is).

[Admiral Valdemar]

I was speaking in generally, I've not fully read the OP, but the fun things we could be doing with storage and processing of data in the future gives me a hard one. I just hope direct neural linkages appear soon enough to really give us something to enjoy.