KentuckyFC writes "A team at IBM has built the first high quality graphene transistors and clocked them running at 26 GHz . That doesn't quite knock silicon off its perch. The fastest silicon transistors are an order of magnitude faster than that but the record is held by indium phosphide transistors which have topped 1000 GHz. But it's not bad for a new kid on the block. It took silicon 40 years to get this far. By contrast, the first graphene transistor was built only last year. IBM says 'the work represents a significant step towards the realization of graphene-based electronics.' (Abstract)."
ok so I wasn't thinking, forgot about my tags, and now I look like the fool... should look like this..
What you're after is something like...
1) Graphene Transistors Run Linux
2) Year of Linux on the Desktop
3) Imagine a Beowulf cluster of Graphene Transistors running Linux
4) In Soviet Russia, Year of Linux on the Desktop runs Beowulf cluster of Graphene Transistors
5) But does it run Vista / Crysis ?
6)...
7) Profit!
Both of ya are new here, otherwise one of you would have added a car analogy AND something about hot graphene transistors poured down your pants by Natalie Portman.
Or have we finally let that whole meme-complex die? I'm getting old, I can't remember anymore.
It's a lot harder to get a switching transistor (for digital circuitry) to operate at high speeds than for a transistor to show gain as an RF amplifier.
26 GHz is incredible for switching circuitry, but it's nothing if you're talking RF signals nowadays. I'm guessing that this was an RF amp given the comments of other transistors being faster in the article summary.
There is a comment about "clocked at" which implies digital switching, but that could easily be a clueless journalist that has no idea of the difference between transistors in clocked digital circuitry and transistors as RF amplifiers.
I searched for "clock" in the paper on arxiv and got no results! The abstract there is more informative:
ABSTRACT Top-gated graphene transistors operating at high frequencies (GHz) have been fabricated and their characteristics analyzed. The measured intrinsic current gain shows an ideal 1/f frequency dependence, indicating an FET-like behavior for graphene transistors. The cutoff frequency f_T is found to be proportional to the dc transconductance g_m of the device, consistent with the relation f_T=g_m/(2piC_G). The peak f_T increases with a reduced gate length, and f_T as high as 26 GHz is measured for a graphene transistor with a gate length of 150 nm. The work represents a significant step towards the realization of graphene-based electronics for high-frequency applications.
This is precisely the kind of innovation we need to get to the kind of AI we want. Giving us the ability to do the right-hemisphere's job is what these kinds of transistor speeds will give us.
Speed has nothing to do with artificial Intelligence. Unless by AI you mean systems that look up 'responses' to mimic what an AI might do.
Learn the difference. AI will come from mimicking the brain. In fact, I will go so far to say that we will have AI before we know everything about the brain. Limited testing has shown that a neural network designed to model the brain behave like the brain. Very limited tests at this point.
The brain, we must never forget, consists of two independent hemispheres that work together--via the corpus callosum--but whose functions and methodology are different.
The left hemisphere, which is bigger and faster and evolved earlier, is mostly discrete storage locations, optimized towards the storage of individual bits of information. This same left hemisphere is optimized toward the processing of linear-sequential pattern-streams of information, such as those found in language and the maintenance of words.
Current computers work in a manner more similar to the way our older left brain hemisphere works.
The right hemisphere, which evolved later than the left, is smaller and consists more of axon/dendrite interconnections, making he right hemisphere better optimized towards the processing of visual-simultaneous pattern streams.
Virtually no current computer system even attempts to model the visual-simultaneous pattern stream processing that is done by the right hemisphere. That consists of taking in patterns of data points and comparing those to known shape and other sub shapes and the associations that are introduced and the recursive processing of gleaning information from images.
The human ear has about 30,000 neurons leading data to the brain. The human eye has about 1,000,000 neurons leading data to the brain. You can see it's an order-of-magnitude harder problem and so yes, it needs more speed!
OK, if you are an undergrad deciding on your choice for thesis and postgrad studies, graphene is great. There is a lot of companies, including Nokia, that pour tons and tons of money into graphene research. It's the easiest grant money to get these days.
That said, there's a reason you don't see much GaAS integrated circuits, even though GaAs has been around for decades, and has much higher carrier mobility (and therefore top speeds) than silicon: it's hard to devise a good IC technology for GaAs. For graphene the problems are way, way bigger than that even. I have seen some attempts of my colleagues (I research in nanosci) at fabricating graphene transistors, and while they can make discrete components with a certain limited rate of success, integration is not even on the horizon. Maybe other people around the world use technologies that are more promising, but it will take a great effort to knock silicon off the top spot for the time being. In fact, I predict a brighter immediate future for Si/Ge and some III/V group compounds as the successors of pure Si, as the next big thing in IC tech.
That said, there's a reason you don't see much GaAS integrated circuits, even though GaAs has been around for decades, and has much higher carrier mobility (and therefore top speeds) than silicon: it's hard to devise a good IC technology for GaAs.
We used to say the same thing about SiGe, but that's starting to make its way into CMOS technologies. Standard 100% bulk Si is hitting the wall of what's possible. Even if geometries are 10-20% bigger, but provide better switching speeds, on currents or lower off
Although there is a practical limit to how far a single electron can go, electrical signals don't consist of a single electron going from one end of the wire to the other. Instead, it's like a game of miniature billiards, with electrons lined up in the wire. You pop one in one end, and another falls out the other end almost instantaneously.
There's a practical (and theoretical) limit to how fast this force propagates, too. Fortunately, that's quite high. I don't think electric propagation time is or will be the practical limit on transistor speed.
The limit on signal transmission speed is relativistic, and about one foot per nanosecond. So the maximum characteristic distance of a chip clocked at 1GHz is about a foot. 10GHz is about an inch. A pentium is about square, and about half an inch on a side. Asynchronous electronics can operate with higher frequency signals, though timing and lead length are still considerations in such devices at really high frequencies.
Actually, the propagation speed can be pretty close to c. The speed of an electron is pretty darn slow (on the order of inches per hour, IIRC), but the propagation speed of an electromagnetic wave (which is what actually does stuff. it's like a hose full of marbles. you push a marble in, and another one pops out the other end.) is about 0.96c in good copper.
I should clarify - Speed of an electron in a conductor in very slow. a CRT's electron gun is a low energy particle accelerator, so it tosses the electrons around significantly faster.
Yes, but unless you can make them break the speed of light there is going to be a very hard limit on how far you can send the signal within an oscillation. At 1 Ghz the signal can travel around 30 centimeters before next cycle, at 5 Ghz you are down to 6 cm (compared to speed of light, since they are going slightly slower mileage will vary), when things go fast enough "almost instantaneously" is quite a long time.
Because each of these interactions takes time, the signal actually propagates significantly more slowly than the speed of light. The standard rule of thumb is c/2.
You're right that it's not the speed of an electron that matters.
However, according to relativity, information itself cannot propagate faster than the speed of light. Using your "billiards" analogy, even though the cue ball doesn't have to make it across the table, the 8 ball can't "know" (or in any way react to the fact) that the cue ball started moving any sooner than an object, moving at the speed of light, could cross the table.
The speed of light is fast, but on the timescales we're discussing it does not translate to "almost instantaneous".
However, the position of the sun does get transmitted to the earth faster than the speed of light. Its called aberration, and the instantaneous position of hte sun is 20 arc seconds ahead of the visible (8.3 minute light lagged) position that you see in the sky. Astronomers are unable to point their telescopes in the correct direction if they assume gravity effects travel at the speed of light. they get the correct position if they assume it is instantaneous (at least for stuff in our star system).
I've heard this argued both ways about gravity, and I don't disbelieve what you're saying; but it's a bit off-topic since transistors don't operate on gravity.
Correct... although electrons actually don't propagate that rapidly though a wire, GP was fundamentally correct in that an electron doesn't have to travel the entire length of the wire to transmit the signal. The added electrons at one end of the wire force electrons out the other, and the electrical force is transmitted through the wire at the speed of light. Push an electron into the wire at one end, and you should expect an electron to come out the other end after a delay of wire length/c.
1000 nm light has a frequency of 3e15 Hz, or 3000 THz. The real thing with optics is to be able to do the processing on light signals instead of electron signals, even in this case the transistors would run at tens to hundreds of GHz. The switching frequency they are talking about here is basically how small they have gotten the internal resistances and capacitances so that the time constant is very very very short. Running one transistor at that kind of speed is one thing, running one hundred million is so
You're never going to have clock frequencies in the light range, for the simple fact that light waves are shorter than the diameter of an atom and thus bigger than any transistor.
Luckily switching light doesn't require transistors that fast. For example, an LCD display switches light directly, without first converting it to electrons. That uses electricity to switch light, but the idea has already been extended to switching light with light in the lab.
You mean photon? We know a hell of a lot more about photons than "it's not an electron".
Don't confuse you not knowing what a photon is with physicists not knowing what a photon is. Don't confuse not knowing what something "is" with the inability to make working devices with them.
Except that although nobody exactly knows what a foton is, it is known not to be an electron. And these transistors happen to be designed for the latter...
It's a flat mattress that sits very low to the ground and are very popular in Japan. Everyone knows that!
Say it with me...SiliCONE.
SiliCON boobs would be ridiculously uncomfortable.
Of course, if you used the hydrogenated amorphous variant, you might be able to work out a way to turn them into flat panels as well.
Since it's Slashdot, I'll leave the next joke for someone else.
At some point, we have to conclude that we are good. Silicon is likely the best material for chips, and will continue to stay that way. other materials have been tried (Germanium was the first) but silicon took precedence because it was cheap and efficient, and I don't see any reason to change that
Silicon sucks.
Pretty much the only redeeming feature it has is that its cheap. when you compare the material properties of Si to GaAs, IIRC, GaAs is better in every way. Unfortunately its also about 100 times as expensive. At least it was back in the mid 90s when I last studied that.
Pretty much the only redeeming feature it has is that its cheap.
I have two words for you: native oxide. Yes, silicon is cheap, but don't underestimate the value of being able to easily grow silicon dioxide on top of it. I would say that that is the main reason that silicon has completely dominated the industry. Of course, with Intel moving to high-k dielectrics, it may be the case that that could change soon.
At some point, we have to conclude that we are good. Gasoline is likely the best energy source for cars, and will continue to stay that way. Other sources have been tried (electricity was the first) but gasoline took precedence because it was cheap and efficient, and I don't see any reason to change that.
At some point, we have to conclude that we are good. Tiger hide is likely the best material for clothing, and will continue to stay that way. Other sources have been tried (leaves were the first) but tiger hide took precedence because it was warmer and less scratchy, and I don't see any reason to change that.
in other words (Score:5, Funny)
pencil < pen < sliderule < calculator < computer < supercomputer < pencil
Re: (Score:2, Insightful)
+1 Most beautiful use of inequalities
Yes but... (Score:3, Funny)
Running 26 GHz is nice, but... Does it run Linux ?
Re:Yes but... (Score:5, Funny)
1. In Soviet Russia,
2. Slashdot is pants
3. Imagine a beowulf cluster of those
4.
5. Profit!
P.S. You must be new here.
Parent
Re: (Score:2)
In South Korea, only old people aggregate memes.
Re: (Score:3, Interesting)
What you're after is something like...
1) Graphene Transistors Run Linux
2) Year of Linux on the Desktop
3) Imagine a Beowulf cluster of Graphene Transistors running Linux
4) In Soviet Russia, Year of Linux on the Desktop runs Beowulf cluster of Graphene Transistors
5) But does it run Vista / Crysis ?
6)
7) Profit!
You, Sir, appear to be the new one here.
I shall now let myself out whilst I
Re: (Score:2)
Both of ya are new here, otherwise one of you would have added a car analogy AND something about hot graphene transistors poured down your pants by Natalie Portman.
Or have we finally let that whole meme-complex die? I'm getting old, I can't remember anymore.
(Does slashdot have a +1, Nostalgia mod?)
Re: (Score:3, Funny)
(Does slashdot have a +1, Nostalgia mod?)
No, but I think it needs one ... as well as -1, Off My Lawn.
Ignore speed of light: (Score:3, Funny)
Allow/Deny?
pretty sweet (Score:5, Informative)
IBM and Columbia are working together on this. Their grant calls for them to push this up to 50 THz.
Oh, and what was done last year was a single electron transistor... normal transistors were available just about as soon as graphene was, in 2004.
Digital switching or signal amplification? (Score:5, Insightful)
It's a lot harder to get a switching transistor (for digital circuitry) to operate at high speeds than for a transistor to show gain as an RF amplifier.
26 GHz is incredible for switching circuitry, but it's nothing if you're talking RF signals nowadays. I'm guessing that this was an RF amp given the comments of other transistors being faster in the article summary.
There is a comment about "clocked at" which implies digital switching, but that could easily be a clueless journalist that has no idea of the difference between transistors in clocked digital circuitry and transistors as RF amplifiers.
Re: (Score:3, Informative)
I searched for "clock" in the paper on arxiv and got no results! The abstract there is more informative:
ABSTRACT
Top-gated graphene transistors operating at high frequencies (GHz) have been fabricated and their characteristics analyzed. The measured intrinsic current gain shows an ideal 1/f frequency dependence, indicating an FET-like behavior for graphene transistors. The cutoff frequency f_T is found to be proportional to the dc transconductance g_m of the device, consistent with the relation f_T=g_m/(2piC_G). The peak f_T increases with a reduced gate length, and f_T as high as 26 GHz is measured for a graphene transistor with a gate length of 150 nm. The work represents a significant step towards the realization of graphene-based electronics for high-frequency applications.
Not again... (Score:2)
Re: (Score:2)
Artificial Intelligence, Here We Come! (Score:2)
Re: (Score:2)
Speed has nothing to do with artificial Intelligence. Unless by AI you mean systems that look up 'responses' to mimic what an AI might do.
Learn the difference.
AI will come from mimicking the brain. In fact, I will go so far to say that we will have AI before we know everything about the brain.
Limited testing has shown that a neural network designed to model the brain behave like the brain. Very limited tests at this point.
Re:Artificial Intelligence, Here We Come! (Score:4, Informative)
The brain, we must never forget, consists of two independent hemispheres that work together--via the corpus callosum--but whose functions and methodology are different.
The left hemisphere, which is bigger and faster and evolved earlier, is mostly discrete storage locations, optimized towards the storage of individual bits of information. This same left hemisphere is optimized toward the processing of linear-sequential pattern-streams of information, such as those found in language and the maintenance of words.
Current computers work in a manner more similar to the way our older left brain hemisphere works.
The right hemisphere, which evolved later than the left, is smaller and consists more of axon/dendrite interconnections, making he right hemisphere better optimized towards the processing of visual-simultaneous pattern streams.
Virtually no current computer system even attempts to model the visual-simultaneous pattern stream processing that is done by the right hemisphere. That consists of taking in patterns of data points and comparing those to known shape and other sub shapes and the associations that are introduced and the recursive processing of gleaning information from images.
The human ear has about 30,000 neurons leading data to the brain. The human eye has about 1,000,000 neurons leading data to the brain. You can see it's an order-of-magnitude harder problem and so yes, it needs more speed!
Parent
Graphene from CO2? (Score:2)
Graphene is great for young scientists.. but.... (Score:5, Informative)
OK, if you are an undergrad deciding on your choice for thesis and postgrad studies, graphene is great. There is a lot of companies, including Nokia, that pour tons and tons of money into graphene research. It's the easiest grant money to get these days.
That said, there's a reason you don't see much GaAS integrated circuits, even though GaAs has been around for decades, and has much higher carrier mobility (and therefore top speeds) than silicon: it's hard to devise a good IC technology for GaAs. For graphene the problems are way, way bigger than that even. I have seen some attempts of my colleagues (I research in nanosci) at fabricating graphene transistors, and while they can make discrete components with a certain limited rate of success, integration is not even on the horizon. Maybe other people around the world use technologies that are more promising, but it will take a great effort to knock silicon off the top spot for the time being. In fact, I predict a brighter immediate future for Si/Ge and some III/V group compounds as the successors of pure Si, as the next big thing in IC tech.
Re: (Score:3, Interesting)
We used to say the same thing about SiGe, but that's starting to make its way into CMOS technologies. Standard 100% bulk Si is hitting the wall of what's possible. Even if geometries are 10-20% bigger, but provide better switching speeds, on currents or lower off
Even faster CPUs (Score:2, Funny)
I tell ya, Graphene-based CPUS will even be able to run Vista at a decent clip.
-Z
Re: (Score:3, Informative)
Re:Practical limit (Score:4, Informative)
There's a practical (and theoretical) limit to how fast this force propagates, too. Fortunately, that's quite high. I don't think electric propagation time is or will be the practical limit on transistor speed.
Parent
Re: (Score:3, Informative)
The limit on signal transmission speed is relativistic, and about one foot per nanosecond. So the maximum characteristic distance of a chip clocked at 1GHz is about a foot. 10GHz is about an inch. A pentium is about square, and about half an inch on a side. Asynchronous electronics can operate with higher frequency signals, though timing and lead length are still considerations in such devices at really high frequencies.
Re:Practical limit (Score:5, Informative)
Actually, the propagation speed can be pretty close to c. The speed of an electron is pretty darn slow (on the order of inches per hour, IIRC), but the propagation speed of an electromagnetic wave (which is what actually does stuff. it's like a hose full of marbles. you push a marble in, and another one pops out the other end.) is about 0.96c in good copper.
Parent
Re: (Score:3, Informative)
I should clarify - Speed of an electron in a conductor in very slow. a CRT's electron gun is a low energy particle accelerator, so it tosses the electrons around significantly faster.
Re: (Score:2, Insightful)
The GP is suggesting that at some point that almost becomes very real and very important.
Re:Practical limit (Score:4, Informative)
Yes, but unless you can make them break the speed of light there is going to be a very hard limit on how far you can send the signal within an oscillation. At 1 Ghz the signal can travel around 30 centimeters before next cycle, at 5 Ghz you are down to 6 cm (compared to speed of light, since they are going slightly slower mileage will vary), when things go fast enough "almost instantaneously" is quite a long time.
Parent
Re: (Score:2, Informative)
Because each of these interactions takes time, the signal actually propagates significantly more slowly than the speed of light. The standard rule of thumb is c/2.
Re:Practical limit (Score:5, Insightful)
You're right that it's not the speed of an electron that matters.
However, according to relativity, information itself cannot propagate faster than the speed of light. Using your "billiards" analogy, even though the cue ball doesn't have to make it across the table, the 8 ball can't "know" (or in any way react to the fact) that the cue ball started moving any sooner than an object, moving at the speed of light, could cross the table.
The speed of light is fast, but on the timescales we're discussing it does not translate to "almost instantaneous".
Parent
Re: (Score:3, Interesting)
However, the position of the sun does get transmitted to the earth faster than the speed of light. Its called aberration, and the instantaneous position of hte sun is 20 arc seconds ahead of the visible (8.3 minute light lagged) position that you see in the sky. Astronomers are unable to point their telescopes in the correct direction if they assume gravity effects travel at the speed of light. they get the correct position if they assume it is instantaneous (at least for stuff in our star system).
Re: (Score:3, Insightful)
I've heard this argued both ways about gravity, and I don't disbelieve what you're saying; but it's a bit off-topic since transistors don't operate on gravity.
Re: (Score:3, Interesting)
it's the same speed as the speed of light...
Correct... although electrons actually don't propagate that rapidly though a wire, GP was fundamentally correct in that an electron doesn't have to travel the entire length of the wire to transmit the signal. The added electrons at one end of the wire force electrons out the other, and the electrical force is transmitted through the wire at the speed of light. Push an electron into the wire at one end, and you should expect an electron to come out the other end after a delay of wire length/c.
However, GP inc
Re:Are we getting into light spectrum territory no (Score:3, Informative)
Re: (Score:2)
Not even close. (Score:2, Interesting)
Luckily switching light doesn't require transistors that fast. For example, an LCD display switches light directly, without first converting it to electrons. That uses electricity to switch light, but the idea has already been extended to switching light with light in the lab.
Re:Are we getting into light spectrum territory no (Score:5, Insightful)
You mean photon? We know a hell of a lot more about photons than "it's not an electron".
Don't confuse you not knowing what a photon is with physicists not knowing what a photon is. Don't confuse not knowing what something "is" with the inability to make working devices with them.
Parent
Re: (Score:3, Insightful)
Don't confuse not knowing what something "is" with the inability to make working devices with them.
But... nobody knows exactly what the meaning of "is" is.
Re:Are we getting into light spectrum territory no (Score:4, Funny)
Except that although nobody exactly knows what a foton is, it is known not to be an electron. And these transistors happen to be designed for the latter...
It's a flat mattress that sits very low to the ground and are very popular in Japan. Everyone knows that!
Parent
Re: (Score:2)
Flat panel (Score:2)
Re:Why? (Score:5, Interesting)
At some point, we have to conclude that we are good. Silicon is likely the best material for chips, and will continue to stay that way. other materials have been tried (Germanium was the first) but silicon took precedence because it was cheap and efficient, and I don't see any reason to change that
Silicon sucks.
Pretty much the only redeeming feature it has is that its cheap. when you compare the material properties of Si to GaAs, IIRC, GaAs is better in every way. Unfortunately its also about 100 times as expensive. At least it was back in the mid 90s when I last studied that.
Parent
Re: (Score:2)
Just because there are better materials then Silicon, doesn't mean silicon sucks.
Re: (Score:3, Informative)
I have two words for you: native oxide. Yes, silicon is cheap, but don't underestimate the value of being able to easily grow silicon dioxide on top of it. I would say that that is the main reason that silicon has completely dominated the industry. Of course, with Intel moving to high-k dielectrics, it may be the case that that could change soon.
Re: (Score:2, Insightful)
Re:Why? (Score:5, Insightful)
Parent
Re: (Score:3, Insightful)
Sarcasm is hard to do online.
Re: (Score:2)
Re: (Score:2, Informative)
Re: (Score:3, Funny)