Slashdot Mirror
Self-Assembling Nanocomputers
A Semi-Anonymous Coward writes: "According to this article a researcher at Harvard University has developed techniques for self assembly of nanoscale wires that operate without resistance due to a property called ballistic conductivity. He hopes the research will provide an 'end run' around convential top-down circuit designs, allowing much smaller, faster and more energy efficient computers."
But since this is a Harvard researcher being written up in the Harvard press, my hype-o-meter is on the alert. Then I read this:
Lieber has "philosophical differences" with the industry's "top-down" approach to nanotechnology--taking big things and making them smaller. "The way to truly revolutionize the future," he says, "is to take a completely different approach: build things from the bottom up."
Pardon me, but have these philosophical differences yielded even a working flip-flop yet? The world is littered with "proofs of concept" that are too difficult to implement. I'll admit that this technology is extremely promising, but at this highly experimental stage of development it's hardly time to go bashing the accomplishments of the semiconductor industry. Unless, of course, you're trying to drum up press for yourself.
That said, sounds pretty cool. I'll be even more interested when they can form some basic logic circuits with it.
If guns kill people, then CmdrTaco's keyboard misspells words.
I believe people once thought that about what I am using currently to post this message.
just a thought . . .
With reproduction added to the mix, it can be argued that 3 of 4 of these benchmarks are covered. Whose to say that the fourth, evolution, wouldn't follow naturally?
ps: Once these nano-machines develop opposable thumbs, I think we could be in trouble.
Ballistic conductivity refers, and I'm sure I'm not getting this quite right, but that when you coearse an electron to travel down a carbon nanotube, literally, the nanotube forms an effective electron shell that makes the electron fire down the tube with little other interaction with the carbon lattice.
The Israelis came up with a dna-based nanowire a couple years back. There's some talk on nanotech mailing lists about using ribosomes (the things inside cells that assemble proteins from instructions encoded in RNA) as organic nano-assemlers. Theroretically (once someone figured out how to code RNA to produce the right molecules), the ribosomes could be used as self-assemblers to churn out miles of organic nanowire. You could even code robosomes to assemble other ribosomes, thus exponentially increasing output. The only costly part would be the (gold) electrodes.
There's certainly a lot to be said for the 'bottom-up' approach to nanotechnology. Cost for starters! One issue though is, how does one address these very tiny devices?
The problem with a whole bunch of identical tiny circuits is of course that they're all identical - there's no way to differentiate between them. There will have to be some way of distinguishing and interacting with these units.
A couple of ideas spring to mind though. One is to encode the position of one of these units in the unit itself as it is being assembled, by interacting with some sort of precisely engineered field. What would work (if anything) depends very much on the chemistry, but it could be something as simple as a gradient in an electrostatic field, to aligning with a very fine grid of polarized light. There are options, but it all sounds Hard. Schemes like this could attack the problem of differentiation, but there's still interaction and addressing.
One way to solve the addressing problem is to bypass it almost entirely. If these structures are sufficiently small, and can be engineered to act as a giant grid of finite-state automata with evolution rules based on neighbouring states, one can simulate a computational device with a version of Conway's Life on speed. Input and output can be done at the edges of the constructed array, which is probably going to be more simple than trying to address the middle of the structure. The problem lies in initialising the state of the array - clearing it is probably easy enough, depending on how state is stored, but priming it with a state that admits the computational task desired seems to be almost as hard as addressing the cells in the first place.
Another approach might be to give each cell some random state as it is constructed (and there should be plenty of sources of randomness at the molecular level to draw on.) Imagine that this state corresponds to an "activation key": when an appropriately modulated high frequency EM signal hits the cell, it pushes it over into an active state. Before this, it's effectively off (perhaps an off cell would simply propogate signals from its neighbours and do no computation). Give each cell some way of indicating that it has been activated (eg, it emits some light upon activation), and then fire random keys at the cells. This solves the addressing problem, and the interaction problem (one could use the same key for changing the cell's state) - but then one has no easy way of telling how the newly identified cell connects to the other addressable cells.
Do any slashdotters have any ideas? Or can point to literature where these problems are (ahem) addressed?
Self assembly is how the body builds a lot of its internal structures. I did a bunch of work on this in my doctorate - basically you can get some reasonably complex structures (e.g. a virus shell) from a small set of repeating sub-units.
One of the common structures found in all cells are 'micro-tubules' - long cylinders made of repeating tiles of a protein called, imaginatively, 'tubulin'. They look a bit like a coil of rope; technically it's most common form is a '4-start, 13 unit helix'.
Now the place these protein structures are found *most* commonly is in neurons, which are crammed to the gills with these things. And there is a (way-out, whacky, widely discredited, completely batshit, but still very cool) theory that the way our brains actually work is not just at the synapse level, but at the sub-cell level using these microtubules. (This would add maybe another 5 orders of magnitude to the available computing power of the brain if it were true; these suckers are small and there's *heaps* of 'em!).
The idea (and it keeps cropping up in papers 'cause it's just so appealing :-) ) is that
computations can be done using a 'game of life'
like system of electical charges on the outside
of the microtubule, where each unit adops an
electric polarity, and then 'flips' it's
neighbours depending on a simple set of rules.
It's a very cute idea, completely lacking in
anything so crass as experimental evidence.
These days of course no one believes a word of it.
<false modesty>For some dodgy work on nanoscale self-assembly, and for some half decent pictures of microtubules, check out my thesis at nanoscale simulation </false modesty>
Wer mit Ungeheuern kämpft, mag zusehn, dass er nicht dabei zum Ungeheuer wird.