Slashdot Mirror
'Millipede' Prototype Shown at CeBIT
neutron_p writes "It was a subject of much controversy for last 5 - 7 years, but it's finally got protyped. At CeBIT, IBM for the first time shows the prototype of "Millipede" - nanomechanical data storage device. Using revolutionary nanotechnology, scientists at the IBM Zurich R&D Lab, Switzerland, have made it to the millionths of a millimeter range, achieving data storage densities of more than one terabit per square inch, equivalent to storing the content of 25 DVDs on an area the size of a postage stamp. The principle of operation is comparable with the old punch cards, but now with structural dimensions in the nanometer scale and the ability to erase data and rewrite the medium."
This kind of device would be incredible for backup purposes, and the recording method seems to be fast as well, but would they accept almost-unlimited rewrites? In that case, this technology could finally replace magnetic devices. Solid state is always better, but so far, the existing alternatives don't offer the durability and flexibility of hard disks.
The article quotes 10,000 read/write cycles. Given that this number is probably a slight exagerration for PR purposes, it's a good start, but needs optimizing. Hopefully by the time this technology makes it to market, that will have increased that number enough that it will be competitive with magnetic drives. I think that this will definately be a viable replacement for flash drives.
The technology uses localized heating of a polymer past its glass transition. There is no reason that this should cause much material degredation if it is done properly (i.e.: avoiding temperature spikes, and engineering polymers that have an accessibly low glass-transition temperature while also being robust against thermal cycling). I think with enough engineering this could be done. There is alot of research on heating polymers past the glass-transition temperature, so they won't be reinventing the wheel or anything.
Holding enormous amounts of data becomes less and less useful in practical situations if you can't access a decent sized chunk of it quickly.
A high storage density will require a lot of error correction and redundancy built into the medium (much like ECC RAM), which may affect the data transfer rate and increase the cost. The manufacturer's will have a choice of either putting their own ECC chip onto the medium, rasing the cost
or leaving the error correction to the memory controller, lowering the data transfer rate.
I can't answer on behalf of IBM or the millipede project, but if you want my opinion (as an academic researcher who uses similar technology), then I'd guess:
1. Competitive to HDD, since the tips don't seek very far (100 microns max) and since data output from multiple tips can be done in parrallel (in principle, 4000 bits at once, depending on data contiguity, etc.). The time required to actually 'melt' the divots might be the limiting factor, but again that should be offset by the ability to write 4000 bits at once.
2. Room temperature is fine for piezos and cantilevers. Even cold temperatures should be fine. I imagine the material they use would stop responding properly if the device were too hot (above 70 C maybe), but if placed in a computer case away from the hottest components, it should be fine.
3. Even though each tip uses local heating, I don't think the device temperature would be very high. In read mode, the cantilevers are passive and the piezo doesn't generate much heat (I use AFMs at work, and they don't generate heat the way a magnetic HDD does).
4. As I describe in another post, each array in principle alloys thousands of tips to read/write together, at the same time. Stacking a bunch of arrays in a real device is straight-forward.
5. Failure rate might be a problem, and needs consideration. In the lab, sometimes I can use a tip for a long time without damage, but sometimes they can snap off. If the device is properly designed I would guess failure rates for each tip would be okay. Polymer degredation or aging is a very real problem. Presumably they are optimizing that as best they can. I think initial devices will probably have extensive error correction, so that if one tip dies, it can recover the data from that region and write it somewhere else.
6. The current cost for MEMS tips batch-processed like this can be from 1$ per tip to as much as 50$ per tip, depending what you want. So an array might cost thousands of dollars. Of course, the tips are use are for a small market (academic research). It is easier to use lithography to make a bunch of chips than to make a Pentium chip, though, so I imagine if it went into mass production, it wouldn't cost more than 100$ per array. So competitive with HDD.
7. My guess: initial devices to hit the market will have 10 redundant arrays with tons of error-checking. The storage will be competitive with magnetic drives and transfer rates will be too. Cost will be a bit higher, but after being in production for about 5 years, most figures of merit will be better than HDD, and cost will be down to what we're currently used to paying for storage.
But these are, of course, just my (hopefully educated) guesses.
Modern magnetic HDD stores on the order of 100 Gb in a 3.5 inch platter, which is ~0.4 Gb/cm^2.
If each surface atom on a material encodes one bit of data, then your storage density depends on the density of your material. For example, let's say that the atoms are on a square grid, and are spaced by 0.15 nm (i.e.: 1.5E-10 m, the length of a typical carbon-carbon bond). That means that you have about 4E15 atoms per cm^2. So if each atom one holds a bit, that means about 600,000 Gb/cm^2.
Of course, actually using each atom to store a single bit may or may not be feasible. On the other hand, using the entire volume of a material (which you seem to think won't happen) may be possible. Various (far-fetched) nanotech proposals exist. Assuming we're allowed to use the entire volume of the material, and conservatively estimating that it requires ~6 atoms to store a single bit, Drexler calculates you should be able to get ~5 bits/nm^3, or 5E21 bits/cm^3 (refer to Drexler, Nanosystems, p. 366).
This is all quite far off. It will require alot of work for us to get anywhere near these values. But in terms of fundamental limits, we have quite a way to go!
Actually, not too far off the bat... :)
Nanoimprint lithography has been demonstrated to reliably produce replicas in curable polymers on the order of around 10 nanometers.
Basically, you start with a "hard" patterned surface (e.g. SiO2, quartz) press it into a polymer (e.g. PDMS-polydimethyl silizane), heat it up to the glass transition temperature of the polymer (so that it flows and conforms around the master) and then proceed to cool and/or cure the polymer. You're left with a rubbery mold that can be subsequently used to "cast" replicas of the original.
It seems, from all I've read about this millipede technology, that the real bugaboo is re-writing bits. I'm wondering just how important that really is. While I would preserve the ability to destroy data (easily implemented by writing pits at every location) I think that 99% of the uses of this massive storage could be done without re-writing.
Let me think of a couple of scenarios for these chips:
1) Music storage and playback, as in an Ipod.
This is a perfect example of something that you never need erase. You very rarely want to replace the previous version of a song with a newer one -- mostly you just want to add to your collection. In the very odd case that I never want to hear a song ever again, I could destroy it.
2) My own business -- visual effects.
We scan and create a few terabytes a year of images. Perhaps surprisingly, we throw almost none of them away during production, keeping old versions of images as reference. Disks are cheap enough that there's no need to erase frames during a project, and these millipede devices promise to be rugged and permanent enough to act as their own long-term backup. We'd just disconnect the drives and store them on a shelf forever.
Clearly, we'd want to change the way that filesystems work -- maybe the directory structure would be kept in flash memory where just the data bytes are on the millipede surface until it's time to inter the disk in the archive.
I think that IBM, and others, should really consider the possibility of non-rewritable millipedes, especially because abandoning that capacity would appear to make everything else much much simpler and cheaper. They might make it into production sooner too.
Thad Beier
I love Mondays. On a Monday, anything is possible.