Graphene Transistors 10x Faster Than Silicon 170
Asadullah Ahmad writes "IBM has created transistors made from carbon atoms, which operate at 100 gigahertz, while using a manufacturing process that is compatible with current semiconductor fabrication. With silicon close to its physical limits, graphene seems like a viable replacement until quantum computing gets to desktop. Quoting: 'Researchers have previously made graphene transistors using laborious mechanical methods, for example by flaking off sheets of graphene from graphite; the fastest transistors made this way have reached speeds of up to 26 gigahertz. Transistors made using similar methods have not equaled these speeds.'" The other day we discussed what sounds like similar research by a group of scientists at Tohoku University; that team did not produce transistors, however.
Didn't Produce Transistors? Oh Come On! (Score:5, Informative)
The other day we discussed what sounds like similar research [slashdot.org] by a group of scientists at Tohoku University; that team did not produce transistors, however.
Surely that is some sort of joke. From the summary of the Tokyo University article:
A new paper entitled Epitaxial Graphene on Silicon toward Graphene-Silicon Fusion Electronics published by a group of physicists at Tohoku University in Japan has demonstrated that they can grow graphene on a silicon substrate and pair that technique with conventional lithography to create a graphene-on-silicon field effect transistor.
Not to mention that article is a myriad of highly moderated comments admonishing the staleness of graphene on silicon transistors.
Re: (Score:2)
> Surely that is some sort of joke.
Too bad if it is, we have been waiting for this for a while now since silicon based chips kind of reached their frequency limits. Of course, there is quantum computing but it is not coming to your local store soon ;-))
It would be nice to be able to fit a 100 gigahertz chip in current hardware architectures...
Re:Didn't Produce Transistors? Oh Come On! (Score:5, Insightful)
But there is a limit, no mistake about it. Look at modems. They went through this same limit/new limit methodology for years before they were replaced outright. I think this definitely puts silicon in it's death throws, but I expect some last minute breakthroughs that will push it a bit farther than previously though possible. This is a good thing, in that it forces us to optimize current technologies in ways that we didn't previously consider (like compression did for modems) that in turn was applied to all sorts of communication technologies, and arguably to other technologies outside of communications.
I just see this as a necessary step before pushing off into the next big thing.
Re: (Score:3, Interesting)
Hard drives would be a more interesting example. There is an indus
Re:Didn't Produce Transistors? Oh Come On! (Score:5, Informative)
I think you're just too young to have seen the whole chain of "limits" on modem speeds. For a long time we were told that 9600 baud was the absolute maximum speed, limited by the fundamental physics of modem technology over phone wire.
See http://en.wikipedia.org/wiki/Modem#Breaking_the_9.6k_barrier [wikipedia.org]
Re: (Score:2)
That's the context of the discussion, what are 'the limits' of silicon physically, not wh
Re:Didn't Produce Transistors? Oh Come On! (Score:5, Insightful)
Again, that's very easy to say in retrospect. I believe this is an almost identical situation: we have a very complex set of interactions from which we derive one number: "transistor switch speed". We believe we understand those relations well enough that we can derive a fastest speed any possible silicon design can give.
This speed is far more similar to the "maximum" modem speed than it is to the melting point of some substance.
Before Ungerboeck's work, information theory seemed very clear about the fastest possible rate at which data could be reliably sent on the frequencies that would "stay on the wire" without bandwidth bleedover. Ungerboeck just demonstrated that there were artificial assumptions underlying the information coding theory on which that speed was based.
You're looking at documentation after-the-fact on modem speeds, which rightly enough talks about revolutions in theory. From the point of view of people before the revolution in the theory, you talk about physical limits. All limits we calculate are by definition theoretical limits, though.
To paraphrase Arthur C. Clarke: When a scientist or engineer states that something is possible, he is almost certainly right. When he states that something is impossible, he is very probably wrong.
Re: (Score:2)
There is a difference between paraphrasing and butchering. you butchered this quote:
"When a distinguished but elderly scientist states that something is possible, he is almost certainly right. When he states that something is impossible, he is very probably wrong."
The context is not science or engineers, but about how someone can become intrenched in a form of thinking.
Your version doesn't even make sense, and it implies anything is possible;which isn't true.
Modem is a horrible example because the limitatio
Re:Didn't Produce Transistors? Oh Come On! (Score:4, Insightful)
Re: (Score:3, Informative)
[quote]Again, that's very easy to say in retrospect. I believe this is an almost identical situation: we have a very complex set of interactions from which we derive one number: "transistor switch speed". We believe we understand those relations well enough that we can derive a fastest speed any possible silicon design can give.[/quote]
No. We know the fastest speed of a MOSFET made with current fabrication technologies. The problem is that MOSFET (specifically CMOS topologies) has very very good characteris
Re:Didn't Produce Transistors? Oh Come On! (Score:5, Informative)
I think you're just misunderstanding the problem.
The "baud rate" of telephone lines is pretty slow. Baud rate is the number of symbol transitions per second the media can support. Baud rate and bits/second have not been equivalent since Bell103a/V.21 frequency-shift-keyed modems, where 300 baud meant 300 bps, each state transition being a discrete tone that indicated a "mark" or "space" (0/1). From then on, Bell 212a/V.22 used phase-shift keying to get 1200 bps out of a 600 BAUD symbol rate, encoding two bits of information per symbol.
POTS lines are pretty pokey - the practical maximum BAUD rate is less than 3500 symbols/sec. Where speed advancements were made in later evolutions of POTS modems were in the number of bits that could be encoded per symbol, using QAM and Trellis Modulation. A 33.6 kbps modem is encoding 10 bits per symbol onto a 3429 baud carrier.
So, when you kept hearing "phone lines max out at less than 4800 baud", that was correct. The engineers kept wringing higher bit rates out of narrow-band POTS by putting more information on each of the symbols transmitted.
Then, with V.70 and V.90, the modulation schemes took advantage of certain characteristics of non-muxed POTS lines to use PCM digital encoding instead of an analog audio carrier. Unfortunately, if you were serviced through a SLC-96 ("Slick") muxed subscriber loop, which multiplexed the signal from your subscriber line to the central office, you could only connect with older analog modulation schemes such as v.32/v.32bis/v.34.
Re: (Score:2, Informative)
The link you provide speaks to the problems of bit-packing on the symbol states, and the solution of Trellis Modulation, which I mentioned. Trellis coding allowed for packing more than 4 bits to each symbol without increasing the error rate, leading to the development of the v.32bis standard and 14.4Kbps modems. Which is what I said - it wasn't high baud rates, but better bit packing that realized faster speeds.
And you're still saying "baud" when you mean "bits per second".
Re: (Score:2)
um, modems did hit a physical signal limit. We just found ways to get multiple pieces of data and a waveform.
Re: (Score:2)
Re: (Score:2)
You would have been 4 when I started using 9600 bps modems, and 5 when I owned my first. So yes, at 27, you missed most of the theoretical discussions over the maximum rate we could transmit over POTS lines. I remember back when they said 1200 was it. The absolute maximum speed the phone lines could handle. Of course it wasn't. And with each new breakthrough, the new absolute limit was raised just above it.
Re: (Score:2)
Re: (Score:2)
Modems have a fixed limit, because on the ISP side the audio is converted to digital, and goes over a 64Kbps link, of which a part is reserved for signalling, leaving 56K for the user.
It's not a question of being unable to make a better encoder, it's that the line is not able to transmit data any faster.
If you have a line that once a second measures the voltage and outputs a "1" or "0", it doesn't matter what fancy stuff you put on the sending end, the receiver still won't output more than a bit a second.
Re: (Score:3, Interesting)
I never mentioned the 56k limit. I'm referring to the fact that the same signal is used but tweaked each generation to allow greater speeds in ways that weren't even considered. For instance, from 300 baud modems to 56K modems. Frequency shifts, phase shifting, duplexing, echo cancellation, QAM, etc. All of these pieces allowed more data to be sent over the same old twisted pair in ways they never thought possible.
All of those advances were evolutionary rather than revolutionary, and they benefited all sort
Re: (Score:2)
Re: (Score:2)
Ultimately, home internet connectivity still required a new infrastructure (thick copper lines). There were fundamental physical limitations of phone wires; those wires were replaced.
The situation with silicon is similar. We've gotten to the point where FETs can't get any thinner (and therefore, faster). Changing the semiconductor material allows better current per area but it has its own complications as well.
Re: (Score:2)
Re: (Score:2)
You are obsolete. You just choose to look at a very tiny piece of the timeline in order to delude yourself otherwise.
Re: (Score:2)
> Silicon has been "about to reach it's limits" since the late 90's.
Maybe a little later if we trust this graph, granted it has been forecasted for longer than that although :
http://smoothspan.files.wordpress.com/2007/09/clockspeeds.jpg?w=805 [wordpress.com]
Re: (Score:2)
Re: (Score:2)
And have you noticed it's speed isn't ramping up like it used to any more?
Yes it is approaching a practical limit. Fabs are having a tough time creating an environment to make sub 20nm. We are talking about an environment where 1 part part billion isn't clean enough for a Fab, and the vibration from someone walking in th same room screws up calibration.
Even id some smart people come up with a Fab clean enough, a way to etch small enough, and a process of moving them without ANY cracks (cracks so small it ha
Re: (Score:2, Informative)
The other day we discussed what sounds like similar research [slashdot.org] by a group of scientists at Tohoku University; that team did not produce transistors, however.
Surely that is some sort of joke. From the summary of the Tokyo University article:
A new paper entitled Epitaxial Graphene on Silicon toward Graphene-Silicon Fusion Electronics published by a group of physicists at Tohoku University in Japan has demonstrated that they can grow graphene on a silicon substrate and pair that technique with conventional lithography to create a graphene-on-silicon field effect transistor.
Not to mention that article is a myriad of highly moderated comments admonishing the staleness of graphene on silicon transistors.
From reading what you quoted, it's not certain that Tohoku produced anything, at least not a graphene transistor. They did however demonstrate that they can grow graphene on a silicon substrate, and that they can pair that technique with conventional lithography to create a graphene-on-silicon field effect transistor. It's just not clear that they did create a graphene transistor, or at least anything comparable to what IBM apparently is producing.
Re:Didn't Produce Transistors? Oh Come On! (Score:5, Informative)
Note that the Tohoku group grew graphene on silicon while IBM produced graphene transistors on silicon carbide. These are complementary efforts, not competing ones.
Re: (Score:3, Informative)
Re: (Score:2)
Commercially Viable (Score:5, Insightful)
Just remember. (Score:5, Informative)
Look where they are now.
Re: (Score:3, Insightful)
You are forgetting the exponentinal acceleration of progress.
So the duration between 1925 and when they were first used, is not linearly comparable to the duration between now and when those graphene ones will be first used.
Re: (Score:2)
Re: (Score:2)
Do you really mean this one:
http://www.google.com/patents?printsec=abstract&zoom=4&id=Ts5KAAAAEBAJ&output=text&pg=PA2 [google.com]
filed in 1919 and granted in 1928?
(That was simply the first one that had the word transistor in it in my search.)
If not, could you give a link to which one you mean? Thanks.
yes, can you make a billion for $10? (Score:2)
My prediction (Score:5, Funny)
Year 2173:
"Hidrogen-Unobtanium polycomposites seems like a viable replacement until quantum computing gets to desktop."
Re: (Score:3, Funny)
Year 2173:
"Hidrogen-Unobtanium polycomposites seems like a viable replacement until quantum computing gets to desktop."
I came here from the year 2242 to tell you that you're wrong.
Re: (Score:3, Funny)
can you tell me when 6 digit /. UIDs will become popular?
Re: (Score:2)
Re: (Score:2)
In 2032 right after the event we know as "The electrosoccertux-Hamster-Beanbag Incident" that changed the rules of underwater violin racing forever.
But I've already said too much.
Re: (Score:3, Funny)
... and all we need to do get some is to get some stupid natives out of their tree house.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
I think you got some Upsidasium in your Unobtanium if it's hovering.
Re: (Score:2)
How long until you can buy it? (Score:4, Funny)
"This is not pie-in-the-sky stuff, this is real," he says. "This development is really going to turn into a communications device not too long from now."
So, I won't be playing Crysis on this transistor next month, but I might be using it to make a phone call "not too long from now".
Re:How long until you can buy it? (Score:4, Interesting)
My VLSI professor was in the forefront of the industry. He had some very good contract with some good R&D firms. One day, he told us that copper might one day replace aluminum as wires in chips. The lower resistance would make a big difference, but nobody had overcome the increased reactance yet. The next day, IBM announced that they had figured it all out. A year later, copper interconnect was being used in chips, and 6 months later, in iBooks. The same professor in a subsequent class was discussing SOI with similar promises of improvements, and similar "nobody has it figured out yet". A few weeks later, IBM came through again with an announcement. 2 years later, there it was in products.
With game changers like SOI and copper, IBM has gone to market in much less than 5 years.
As a former circuit designer, and still a CPU engineer, I can say without hesitation that I don't care about graphene. The transistors aren't the big factor anymore. Sure, smaller transistors are good to increase transistors per die, and reduce the distance between them, but wire RC delay is the big deal. Even if the Ioff goes down and Ion goes up, the speed of the chip isn't going to change much.
Things aren't going to get much better than copper -- it's very good already. Even if they upgraded to slightly lower resistance silver (and talk about a reactive metal!), the delay wouldn't change much. Lower K dielectric would help too. There are some minor improvements that can be done, but we're probably talking 5% here and there, and they probably don't add up to 20%.
Architecture changes are going to be important, from instruction optimization to multiple cores. The automated synthesis tools available also have an amazing amount of potential improvement -- placement and routing is a field with a lot of graph theory headroom. There is a world of difference still between "good enough" synthesis and what can be done by a well trained technician.
Re: (Score:2)
Another chip designer here ...
Anything that lets us make transistors faster without paying a huge cost in leakage (power consumption when not switching) is a win. It sucks to go from one process node to the next and get almost no performance benefit unless you're willing to pay a leakage penalty. SOI rocks, but not everybody has it. Therefore, I'm a bit more excited about graphene than you are.
Re: (Score:2)
As a former circuit designer, and still a CPU engineer, I can say without hesitation that I don't care about graphene. The transistors aren't the big factor anymore. Sure, smaller transistors are good to increase transistors per die, and reduce the distance between them, but wire RC delay is the big deal. Even if the Ioff goes down and Ion goes up, the speed of the chip isn't going to change much.
Sure it does. Current circuit speed is still (despite predictions) dominated by capacitance. This includes both load capacitance on the transistors themselves (which, mind you, is still not trivial compared to interconnect) and load capacitance on the metal itself.
To decrease rise and fall time you can either decrease capacitance (shorter wires) or increase the drive current, which faster transistors do.
And while transistor frequency scaling isn't overwhelmingly dominant as they were back in 0.35um, they st
Re: (Score:2)
Re: (Score:2)
3D chips (Score:5, Interesting)
To be honest I'm more interested in seeing proper 3D chips become reality. If you find some affordable way to produce chips with, say 10 000 layers, then processing power per volume unit would increase rapidly.
I think the major obstacle is going to be what to do about heat. The center of such a chip-stack would probably get quite hot so you probably want to run some form of liquid cooling through the chip itself. Alternatively materials like silicon carbide or diamond might be able to cope better with the high power density.
Re: (Score:2)
The center of such a chip-stack would probably get quite hot so you probably want to run some form of liquid cooling through the chip itself.
Once you're creating enough layers, there's nothing preventing the designers to create a 3D structure that's similar to that of a heatsink. Basically, it'll be designed with more surface area so that it can be cooled effectively. Probably a batch of fin-like elements that are connected together. And you wouldn't have to run liquid through it -- just fill the spaces between the semiconductor material with a better heat-transferring material (like copper, or eventually artificial diamond), and have that conne
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Quite right. [usatoday.com]
Re: (Score:3, Interesting)
3D chip manufacturing would be interesting. As well as having a possible stepping stone towards universal fabrication, you would also have a great increase in the potential number of connections between processing elements. Connectivity is one of the main divides between silicon and neural tissue, so this may have implications for artificial intelligence. Two singularities for the price of one!
Re: (Score:2)
3D chips will not take us any closer to true AI.
We still don't truly understand the nature of intelligence, and we won't be able to manufacture it unless we can define it formally in some mathematical/logical notation.
I've done some work with neural networks, and we can simulate neurons with any number of connections (inputs and outputs), but having a bunch of neurons work together discerning things is not intelligence.
Re: (Score:2)
I think that a kind of fractal volume for the CPU (which will maximize the surface between the cooling fluid and the heating parts of the CPU) could be pretty cool (no pun intended!), but quite hard to manufacture.
Re: (Score:2)
Heat is caused by power dissipation. Not only has IBM been researching how to integrate cooling channels inside of chips, they've also been researching ways to create circuits that recycle as many of the electrons as possible, to avoid dissipating heat. Hopefully they'll work something out that will enable 3D chips. The connectivity possibilities are intriguing.
IBM still does do basic research, and they seem to believe in liberal licensing, judging by the 2 terabyte hard drives that can be had. Hopefull
Re: (Score:2)
Re: (Score:2)
The problem is that processing power doesn't scale linearly with number of transistors you can fit in an area. That's the primary concern over the frequency scaling of silicon. You can cram more transistors in some space but if they can't run faster, your options are: 1. more cache 2. dual core 3. more specialized functions.
None of these will universally speed up computing like frequency scaling will.
Re: (Score:2)
Re: (Score:2)
Why would running liquid through the chip not be able to control the temperature?
In nano-scale channels, liquid doesn't run.
Re: (Score:2)
Sounds cheap (Score:4, Funny)
It was bad enough when computers were made out of mere sand, now they will be made out of coal?
Can't they make computers out of sapphires or something so I can feel sophisticated when I buy it?
silicon on sapphire (Score:4, Insightful)
Re: (Score:2)
Re: (Score:2)
Diamonds are made from Coal too. Just say it is made from diamonds and you are all set.
Re: (Score:3, Funny)
Re: (Score:2)
Just get DeBeers to sell people coal as if it were something valuable.
They did it with the now nearly worthless diamonds. So coal should not be hard for them.
Re: (Score:2)
Bad / Incorrect Article (Score:3, Insightful)
"The prototype devices, made from atom-thick sheets of carbon, operate at 100 gigahertz"
Define operate? This sounds like the cut-off frequency, which is 100s of GHz for Si CMOS. How is 200GHz 100GHz? And no, this does not mean it can switch this fast. If it can switch this fast, it would likely operate into the THz, and we would be interested in using it for THz applications. Maybe operate is maximum stable oscillation frequency? Ft? Fmax? It's sure as hell not a switching frequency, despite what the article tells us.
"Growing transistors on a wafer not only leads to better performance, it's also more commercially feasible"
Growing transistors on a wafer? As compared to what? A waffle?
Done reading... moving on...
Imagine the speed (Score:2)
Re: (Score:2)
Gotcha.
Stupid question (Score:2)
graphene provides a promising potential replacement because electrons move through the material much faster than they do through silicon
Could someone elaborate on that statement? I assume that they mean that an electron will move through the material with "less interference", like light traveling through space will be "faster" (to reach its destination) than if it were traveling through matter. Is that what they mean?
Re: (Score:2)
In addition to fewer scattering events, I believe the energy required to affect the electron bonds on graphene is less than on silicon, so you reach the energy level faster, so you move the electron along faster.
Re: (Score:2)
Graphene in its conductive state has a much lower resistance/area than silicon semiconductors. There's also far less scattering.
This means more electrons can move through a piece of graphene than a piece of silicon of the same size per second.
Interconnects (Score:4, Interesting)
Graphene will probably be at least as important as a replacement for metallic interconnects as for transistors. Much of the area of a chip is covered by interconnects they are responsible for much of the heat and delay.
Military Application? (Score:3, Insightful)
Re:Military Application? (Score:4, Funny)
Re: (Score:2, Informative)
They mean the gate dielectric (which is used for the majority of transistor designs, silicon or otherwise) not that the transistors need insulation from the environment - graphene is more sensitive to the dielectric material (ie the enivronment around the transitior) than silicon. Extreme external (ie military) environment is irrelevant as the entire chip is packaged up anyway.
Re: (Score:2)
Look up "hermetic seal" in Wikipedia.
Re: (Score:2)
Oh yes, I worked on some pretty dainty equipment in the Army,the abbreviated Guided Missile test set for the Hawk Missile [wikipedia.org], just starting the truck meant 8 hours of work getting the equipment back into alignment.
Other Applications (Score:2)
I wonder what a fuzz box made of these would sound like...
Re: (Score:2)
Hear hear!
For those interested: basic guitar fuzzboxes (like the venerable Fuzz Face [ning.com]) are simple, design arround one or two transistors. The sound is heavily dependent on the type transistor used - old Germanium devices have a nicer, more musical sound than modern Silicon ones; this depends on how they clip the signal when amplifying, basically. I'd love to hear one using these new devices...
9x faster, not 10x faster (Score:4, Informative)
These transistors are only about 9x faster than silicon, not 10x faster as the Slashdot headline claims.
Re:9x faster, not 10x faster (Score:5, Funny)
These transistors are only about 9x faster than silicon, not 10x faster as the Slashdot headline claims.
Oh, well, in that case don't even bother.
hold yer horses (Score:5, Informative)
overrated (Score:2)
if the channel can pinch *almost* open/shut at 100Ghz, then the transistor can switch a lot faster than silicon, too.
Re: (Score:2)
Not really. You can have switching without reaching saturation. It just happens that we like to have digital circuits in saturation because it helps noise immunity.
Remember that there's always an upper clamp: your supply voltage.
Graphene is ok for now (Score:2)
graphene seems like a viable replacement until quantum computing gets to desktop
With everyone quitting smoking, we've run out of dead people's lungs to scrape carbon out of, so we've reached the limits of carbon-based CPUs and had to switch to graphene.
But the extra pencils from companies going paperless will only last so long. When we run out, we will have to switch to making quantum CPUs. Hopefully by then, making quantums will be a lot cheaper.
Beware! (Score:2)
All right! Now we have a chip that we can get rid of using an eraser!
Integration scale (Score:2)
The summary doesn't mention it, but is the integration scale potentially competitive? I'd assume so, since it's supposed to be commercially viable, but of course I didn't RTFA.
The Future (Score:2)
The future of computing is gallium arsenide^h^h^h^h^h^h^h^h^h^h^h^h^h^h^h^hphotonics^h^h^h^h^h^h^h^h^hmolecular switches^h^h^h^h^h^h^h^h^h^h^h^h^h^h^h^h^h^hquantum whatnot^h^h^h^h^h^h^h^h^h^h^h^h^h^h^hummmmmmm^h^h^h^h^h^h^h^hgraphene?^h!
Silicon is still faster (Score:3, Interesting)
Graphene is still very much a lab technology which isn't anywhere near ready for commercial production of devices. It may turn out to replace Silicon one day, but guess what, people keep doing amazing shit with silicon because it's still the cheapest material system for fabrication.
Apologies to those without IEEE access, but here is a paper discussing a recent 150GHz Silicon CMOS amplifier: A 1.1V 150GHz amplifier with 8dB gain and +6dBm saturated output power in standard digital 65nm CMOS using dummy-prefilled microstrip lines [ieee.org]. That's pretty awesome in my book. It's pushing the amplifier very close to fmax of the actual transistors, but it works and it's in a commercial silicon process.
There are always applications where we can do better systems with more expensive materials like GaAs, GaN, InP, Graphene, etc... but silicon is cheap and easily mass-produced, so lots of engineers work on pushing it to incredible performance.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Moore's law is about quantity of transistors, not speed of computing, the two just tend to be highly correlated.
However if you go with the alternative speed formulation of "popular Moore's Law", this will be right on the curve if the first large-scale products, with speeds about like the current lab rates, come out in about three years. And it will track the curve for at least another three or four years if the expected speed improvements work out and take that long to deploy, and the limits are where expe
Re: (Score:2)
I could dream about better language translation software.