Slashdot Mirror
Nanotube Transistors
orn writes: "Reuters is reporting that IBM has made a step toward using carbon nanotubes as the next best thing since sliced silicon wafers. They created a process that could potentially be turned into a manufacturing method to make extremely small transistors. Yahoo has a copy of the article. We all know that small transistors means higher speeds! Yum!" The NY Times has another article about the same technology, which we've mentioned before.
in my dream world it would be built into the OS, making the programming yokel easy.
Unfortunately, this doesn't make it easy.
To run efficiently in parallel, a program must be written to use a parallel algorithm. The best parallelized OS and hardware in the world won't save you if the algorithm is serial. Writing parallel algorithms is more difficult than writing serial algorithms (usually), and turning a serial program into a parallel program automatically is extremely difficult (some optimizing compilers try to do this when building for parallel platforms, and usually don't do very well).
To make matters even worse, many problems are intrinsically serial - impossible to reduce to parallel form. As long as even part of your program is irreducibly serial, there will be a limit to how much you can boost performance (Amdahl's Law).
In summary, trying to use a massively parallel machine to boost performance doesn't work as well as you're assuming (though it's good for many special cases).
The transistor itself is smaller, but again the capacitance isn't necessarily decreasing but the ability of the transistor to drive current is.
:).
I'm afraid this isn't the case. As you make the gate of a transister shorter, its ability to drive current increases. Shrink the entire transistor, and as the W/L ratio of the gate stays the same, the transconductance of the transistor stays (roughly) the same.
Yes, process differences and short-channel effects muck this up somewhat, but the effect is still there. This is why we're bothering to move to finer linewidths at all.
Until recently, shrinking circuits like this produced a *big* speed boost, because most of the parasitic capacitances scaled directly with device area. Same drive current with 1/x^2 the capacitance means a speedup of x^2 for a linewidth shrink of factor x. Recently, as you point out, wire delays have become relevant, but while that does limit signal propagation speed, it doesn't affect a transistor's ability to drive current.
I've been studying this in far too much detail recently
What I would like to see is cooler chips so I don't toast my lap on the train. If new chips can do that, well, then it may be time to pick up some Big Blue stock...
You mean like Transmeta chips? They don't even need CPU fans. You *DO* own some Transmeta stock, don't you? Surely you're not just shooting off at the mouth about buying IBM's stock?
Sorry to be grumpy about this, but I can't believe how far Transmeta has fallen, and how much geeks like us have let them down. They built exactly what they said they were going to build - a cooler, lower-power CPU that's 100% Intel-compatible, and nobody seems to be buying the product that we all got so excited about. Everybody says they want cooler chips, lower-power chips, but then it's the ridiculously hot, power-sucking Athlons that get all the press. Wonder why that is?
What's your damage, Heather?
I've been reading about all sorts of uses for these carbon nanotubes, the most recently being hyperdense packaging of gases for energy use.
A good summary of some of those uses, and other resources can be found here.
my understanding is that the REAL importance of nanotubes is that they can be built in 3d - not constrained to the 2d of current chip tech.
-k
IMB's Press Release, as well as a neat article at USA Today (believe it or not) and, finally, a very informative article at TechReview.com.
Do something about world hunger. Click here
You know, I disagree.
.. I wish things were built more like my PalmPilot - the damn thing is so stingy, it's wonderful!
Unless you've got a real use for that 2GHz processor, anything else is just wasted electricity/heat.
Lets look at my house for a minute - a fairly modest setup. The old 486 running Linux to protect my Windows machines from the 3l33t h4x0rz is idle most of the time. Like as in less than 2% load. I probably should replace it with a little LinkSys box, but I need Linux access every once in a while at home. That would be a great use for a processor whose rate could be increased/decreased dynamically.
Look at my Pentium 233. It's idle most of the time I spend composing email, browsing, and programming. It's only when I want to do a brain sucking compile or image manipulation that I need more juice.
Here at work it's even more wasteful. We've got tons of machines that idle all night, spinning hard drives and running stupid screen savers. Power management is often set incorrectly, and the screen savers can burn quite a bit of needless CPU. Wouldn't it be great if they could slow down and save some juice?
Energy is too cheap
Yes you're describing branch prediction. This is universally used in all modern processors. It gives probably 2x speedup or so in general. What you're describing is a way to follow multiple branch paths when you reach a branch that is hard to predict. Actually, my PhD advisor did a paper on this before http://www-cse.ucsd.edu/users/tullsen/ISCA98.ps I don't remember the exact numbers, but they probably saw speedups of 15% or less. As you increase the number of paths you follow simultaneously (in your massively parallel processor) the gains speedup. Hell, if you give perfect branch prediction (rendering your optimization unnecessary) you probably will only get 50% speedup maximum. That's a hell of a lot less than the 1000x you'd get from a perfect speedup. Actually, my current research involves ways to use processors similar to what you describe (but much less parallel--8 way at the most) to improve the performance of programs. But trust me, it's not as simple as just waving your hands and saying 'the xxxxx will take care of it'--we've been waiting for compilers to automatically take care of it for more than 10 years and they still aren't very good at it.
I must burn in hell, suffer and pay for my sins
But Gods the one who's losing, Satan always wins!
I know you're being silly, but there was talk a few years back about actually using microscopic vacuum tubes for certain very high speed application. IIRC, the tubes were supposed to be less succeptable to interferance, hence allowing them to be driven faster. Of course, this was about 2-3 years ago, and I haven't heard a lick about it since, so it was probably just some nostalgic hacker's pipe dream.
If god had intended you to be naked, you would have been born that way.
And look, Intel is already moving all of its lines to .13 micron, and their existing low-heat chips are already in the same ballpark as Transmeta's in term of cooling characteristics (albeit still with a thin fan package).
Meanwhile, Transmeta's still struggling to break through to >700Mhz speeds (necessary to compete with even slower Intel chips considering that the instructions are emulated, Transmeta propoganda aside). So net-net, Intel is going to roll over Transmeta like a Mack truck.
Which is sad, Transmeta has a lot of promise, but results need to live up to hype. Brento asks why geeks have abandoned Transmeta, didn't we want cooler chips? Yes, but first we want them to be fast.
Invisible Agent
Invisible Agent
This post is a mirror; when a monkey stares in, no hacker gazes out.
"IBM researchers said they've found a process by which they can form batches of nanotube transistors, which are as small as 10 atoms across. Until now, nanotubes had to be positioned one at a time or by random chance, IBM said."
batches != individually maufactured
Stop Continental Drift! Reunite Gondwanaland!
b/c the chips are just not fast enough. Low power devices are all well and good, but most of us want to compete w/the other geeks and have a system that just fucking screams!
I love the work that they did, but I have no use for it. I don't want a 700mhz processor that isn't running at 700 all the time. I want 2ghz that runs on low power and low heat and all the time.
I am sure that most agree w/me.
Carbon structures can withstand way higher temperatures than silicon. Probably carbon nanotube will generate less heat for the same performance, because it moves lower charges via better conductors on a shorter distance. Hoowever since your processor is only part of your lapop and your processing requirements will increase, and your laptop will only get thinner, not smaller (for the screen size), it will dissipate perhaps only half of current energy. The carbon nanotubes will get much hotter, just because they can. Probably the CPU core will easily be able to reach betwern 400 and 600 dec C. That can be efectively distributed using carbon nanotube heatsink, so there will be no burrning hot spots on your laptop. BTW, that is another applicatin of carbon nanotubes - very effective heatsinks. What conducts electricity efficiently often conduct also heat wery well. Heatsinks were will be probably the first carbon nanotube techology in your laptop. Petrus Vectorius Petrus Vectorius.
--
A feeling of having made the same mistake before: Deja Foobar
The researchers they are referring to are, amongst others, from the "Delft Univeristy of Technology" (Delft, The Netherlands) better know as "de TU-Delft"
I know they did a Science article on this subject, but since www.sciencemag.org doesn't even have abstracts on-line, I can't verify that this is the one:
Kouwenhoven L, Science Vol. 275 (28 March 1997) pages 1896-1897. Title: "Single-Molecule Transistors"
There are also a few other articles about Nanowires in Science by the same guy. There's also a (very) short piece on the "TU-Delft" homepage.
So would a chip made with nanotube transistors run cooler than current chips do? While I'm all for smaller chips, my laptop really can't get much smaller than it is now due to keyboard size and the fact that I like bigger, not smaller screens.
What I would like to see is cooler chips so I don't toast my lap on the train. If new chips can do that, well, then it may be time to pick up some Big Blue stock...
By themselves, perhaps, nanotubes are not going to make computers faster. But if one can pack the current equivalent of 8 CPUs onto the space currently occupied by one, and run them SMP, or even (perhaps!) MMP, that would in fact be a speed increase in terms of end use.
What I think makes this even more exciting is the article I read via /. yesterday about a 1x1x1 cm glass cube holding 1TB of data. Add together the space of a current IDE drive (say, 8x10x21 cm, or 1680 cm^3/1680 TB) and a nanotube CPU running the equivalent of 8 CPUs as SMP, and all in the space of a current small desktop machine.
Only two things are infinite, the universe and human stupidity, and I'm not sure about the former. (Einstein)
Where are the massively parallel machines? And I'm not talking about piddly 2/4/8 way or even 64 way SMP machines. I'm talking thousands to millions of independent CPUs all fitting inside a desktop sized PC.
Speed isn't the end all answer to computing. Look at the human brain and how insanely slow large ion synapses are yet look what the brain can do. It's not the speed. It's the parallelism.
Your understanding is basically correct - buckballs and carbon nanotubes are generated in what amounts to a low pressure carbon arc plasma. This helps to prevent normal chemistry from interfering and pulling out carbons before it is "too late" and a stable structure has been produced.
However, this doesn't require that the transistors be individually manufactured. Many molecules are produced in the plasma, and the next step of the work is to separate out the ones that are desired.
Right now the hard part of the process is moving the molecule into the right place to make the transistor. That is what, so far, must be done individually. However, ataching a type of molecule to a particular place is not something that must necessarily be done by hand. If you think about it, lithography is all about getting the etchant (or modern equivalent) to deliver itself to the right place. Also, recent 'designer' medicines have useful properties of self-delivery. So although this is a tricky problem, there is every reason to believe the engineers will find a solution.
-posting anonymously due to unjustified "bitchslap" thanks to the fascist editors
Certainly right now it's cost ineffective to make large batches of these, the article says as much. Remember though that the first transistors were hand soldered devices..
It's just a matter of time.
Trolling is a art,
One of the largest factors inhibiting speed of microelectronics is parasitic reactances (inductive and capacative), which you've casually mentioned, but not fully.
Every length of wire has some finite inductance, and the longer the wire, the more inductance it has. This is one of the hardest parasitic effects to kill. To see how this limits speed, picture how an inductor works. It basically stores energy in a magnetic field around itself, dependent on the current through the device. Magnetic fields don't like to change instantly, and when the current does change too quickly, it takes time for hte magnetic field to collapse (or create). This causes these parasitic inductances to act as virtual low-pass filters all over the circuit. Thus large inductances prevent one from quickly-changing currents, which limits quickly-changing voltages, which ultimately limits quickly-changing logic states of the device.
This is only talked about self-inductance, this inductance really depends on the location of all components in the circuit, as their magnetic fields all interact and they'll exhibit mutual inductances too.
The other reactive effect is parasitic capacitance, in which an electric field is created between varying components, usually between circuit-board traces and ground-planes, or between layers within the lithographed silicon devices. You mention capacitance, but the largest effect comes from the surface-area in common, not just the distance between traces. These parasitic capacitances don't like it when the electric field between them changes too quickly, so they also act as virtual low-pass filters all over the place, and also limit quickly-changing voltages. Ie, with a single-pole low-pass filter, it takes time for the voltage to ramp up, which means to ramp up to the threshold for a high/low transition will be limited.
Sometimes, the parasitic inductances and capacitances work together in nasty ways that they cause oscillations (or ringing), which is a real pain in the butt to try to kill off.
This is probably the primary reason the engineers want to shrink circuits - to kill the parasitics. That's why lead lengths are as short as possible and as narrow as possible on boards.
I always hear of overclocker enthusiasts talk about cooling their CPU's to lower and lower temperatures as if they can find no limit to their clock speeds. But they don't seem to acknowledge these inherent low-pass filters all over the place in terms of the parasitics. .V / _` (_-<_-<
.\_/\_/\__,_/__/__/
__ __ ____ _ ______
\ V
make world, not war
Which is one reason why smaller == faster. Smaller transistors means less distance between them, which means shorter signal paths.
The other big factor is the energy involved in switching the transformer, which must be dispersed as heat. Smaller transistors also means less capacitance (or less charge to alter the state of the transistor), which means less current and less waste heat.
Unfortunately, these devices are getting to the point where they are triggered by only a few electrons, which means they will be easily affected by thermal noise. This will make designing a processor extremely difficult.
Unfortunately, we're getting to the point that the actual speed of the transistor/gates is nearing the speed of the traces between the different gates. Simply having faster transistors won't help bring faster computers.
We've begun to reach the point where the distance between the gates is the limiting factor; the amount of time it takes a signal to traverse the distance between gates being the critical part.
Which is the biggest reason why the description, at least, is inaccurate. Faster transistors can help to a point - but by making the entire chip die smaller, and reducing the distance between the logic gates - that is where the speed gain is to be found.
-- Sometimes you have to turn the lights off in order to see.
I thought the whole point of transistors was to get away from tubes! And isn't it going to be a pain to find and replace those tiny little tubes when they burn out?
Fore more deep background check out the Nanotube website at: http://www.pa.msu.edu/cmp/csc/nanotube.html
---
--Got Lists? | Top 95 Star Wars Line
"Intel has technical operations in China, Russia, India and so forth, and we have them there frankly to a significant extent because that's where there are trained people." (See! It has nothing to do with money!)
I'll do my best to not flame you but come on, use your head. Even if the amount of technically competent people were the same here (in the US) as in India, China, Russia and so forth they'd STILL have all those plants and research centers abroad.
It's all about the money cause they don't have to pay as much in those places. Take GM or whoever it was that moved a whole car manufacturing plant to Mexico. Are American plant workers any more skilled at manning a production line then their Mexican counterparts (after being trained)? In Intels case, I'm sure there aren't as many American engineers as Russia, China, etc but you have to admit saying it's not about the money (esp in Intel's case) is kinda ridiculous.
BOSTON SUCKS!
"It will take several years and additional breakthroughs before nanochips become as prevalent as silicon chips are today," said Harvard Professor Charles Lieber.
A possible shortcut, Lieber said, is the development of hybrid silicon-nanotechnology chips. Hybrid chips could be ready in about five years, he said, while pure nanochips will likely come within a decade.
Hybrid chips would be extremely dense, allowing substantial gains in processor speeds or the amount of data a memory chip can hold.
Looks like a substantial achievement from all this.
There's always sufficient, but not always at the right place nor for the right folks.
Wasn't there an article just yesterday about a new litho technology that allowed chipmakers to etch on the molocule-size scale?
Faster transistors may give you a few more FpS in the FPS of your choice (I made a funny!) but Nanotube-transistors have a much greater application than speeding up existing technology. This kind of logic device can be used in small and nano-sized robots, increasing the range of tasks they can do and making them more controllable.
Ultimately, this is going to be the kind of advance we site as making the artery-scraper nanites and nerve-tissue replacements possible. Don't limit yourselves to thinking *solely* about data processing when new logic devices come along.
The next Slashdot story will be ready soon, but subscribers can beat the rush and slashdot the links early!
My understanding of the process is that the nanotubes form in very hot furnaces by passing lasers and high current electric arcs through the carbon vapour in the furnace. This is not an easy process, and essentially requires that these transistors are individually manufactured.
This is by no means a commertially viable process.... yet. Don't hold your breath.
.. if only.
Check out:
Chip Evolution: IBM Scientists Develop Breakthrough Transistor Technology with Carbon Nanotubes
They've got more dirt on this at
http://www.research.ibm.com/nanoscience/
http://www.research.ibm.com/atomic/
The Register's got it too...
http://www.theregister.co.uk/content/3/18556.html
Cool stuff.
Si.
This is way cool. More deep science stuff about nanotubes is at the homepages at http://www.research.ibm.com/nanoscience/nanotubes. html and http://www.research.ibm.com/nanoscience/index1.htm l
Can I get these with filters?