MIT Reports 400 GHz Graphene Transistor Possible With 'Negative Resistance' 123
An anonymous reader writes "The idea is to take a standard graphene field-effect transistor and find the circumstances in which it demonstrates negative resistance (or negative differential resistance, as they call it). They then use the dip in voltage, like a kind of switch, to perform logic. They show how several graphene field-effect transistors can be combined and manipulated in a way that produces conventional logic gates. Graphene-based circuit can match patterns and it has several important advantages over silicon-based versions. Liu and co can build elementary XOR gates out of only three graphene field-effect transistors compared to the eight or more required using silicon. That translates into a significantly smaller area on a chip. What's more, graphene transistors can operate at speeds of over 400 GHz."
2000's called... (Score:1)
... they want their GHz war back...
Re: 2000's called... (Score:5, Insightful)
And? We're at similiar Ghz to back then but not because we want to be.
Re: (Score:3)
Re: (Score:1)
There's more to performance than just clock speed. 2 Ghz on two processors is faster than 2 Ghz on one, provided your software uses two processors. In fact, I'd say that a 2 Ghz quad core would be better than a 3Ghz single core (again assuming your software is actually using four cores....).
Re: (Score:1)
Go a little bit further, I prefer a GPU with 1K~2K+ cores running at 800MHz than a single Core at 400GHz for some purpose (simulations, image processing,...). It's not a game on who has the highest clock rate anymore but rather on who is able to fetch data at the fastest speed, who is able prevent algorithm from doing too much branching, instructions replays etc. and generally who is able to maximize throughput...
Re: (Score:2)
It's not a game on who has the highest clock rate anymore
Indeed. Also, when we look at normal CPUs, it's amazing how much more processing power we can get today from a 2GHz chip (e.g. an Intel Core) compared to a 2GHz chip a decade ago (e.g. a Pentium 4 chip). Same clock frequency.
Re: (Score:2)
Speak for yourself. My processor will very happily go up to 4.1ghz without overclocking (Not bad for a mid-range processor).
AMD is doing their best to keep clock speeds going strong which would explain why they have the current world record.
Re: (Score:2)
But they don't beat their own Athlon II/Phenom II gen which ran at only 3GHz.
Re: (Score:2)
Speak for yourself. My processor will very happily go up to 4.1ghz without overclocking (Not bad for a mid-range processor).
I was only talking about performance per clock.
Of course, as you say, today chips can work at higher clock rates too, which also improves performance.
Re: (Score:2)
I was comparing only the performance of a single core.
Of course, as you say, adding more cores will improve overall performance too.
I'm starting to wonder if slashdotters read the comments they are replying to anymore. :)
Re: (Score:1)
Re: (Score:2, Interesting)
Good, less stupid marketing trying to convince us that their product is faster then the other by using some set of obscure benchmarks to compare each other.
Intel Carbon 50Ghz vs Intel Carbon 100Ghz I now what is faster. ... 2 cores, 4 cores, 8 cores.... I can't barely figure out which chips from the same product line are better then the others.
Vs.
Intel Core i4, i5, i7 Sandy Bridge, Ivory Bridge
Back in the good old days
286, 386, 486 and the number of Megahertz If your 386 had a faster megahertz then you
Re: (Score:3)
I'm all for a nice and straightforward means to compare chips, but you're wrong that clock speed was a good way to do that - at least after about 1998 or so. There's just way too many other ways in which a chip can be faster or slower. Cache size, cache speed, cache prediction, instruction size, data path latency, pipelining, hyperthreading, multiple cores, etc, etc, etc.
The Pentium IV really put a stick in the idea of comparing clock speed because they actually made it do less work in each cycle so they co
1990's called... (Score:2)
Jokes called... (Score:2)
In Soviet Russia (Score:1)
The GHz war didn't end (Score:2)
The GHz war didn't end it just got to the point where pursuing higher clock speeds caused performance due to other architectural constraints. So the focus became on getting more efficient on every clock cycle which we did and then we hit a wall there too.
Now the focus is on paralleling tasks but guess what, now we're hitting a wall as to how to make effective use of those cores, and some of us wish the GHz war would be back so we can get some faster clock speeds again.
Re:2000's called... (Score:5, Funny)
Re: (Score:2)
Re: (Score:2)
And again.. (Score:2)
Cynicism aside, the research is exciting, but it's not likely to bear fruit any time soon.
Re: (Score:2)
Well, if you want something to feed your cynicism, it's pretty reasonably supported that graphene causes cancer, if it gets in your body.
Re: (Score:1)
I read a research paper that said if you took graphene in suppository form it could not only cure cancer but freshen your breath too.
Re:And again.. (Score:5, Interesting)
You're cynicism is valid in this case. This is just rehashing research from 20 years ago on negative differential resistance (NDR) two-terminal devices. CMOS won out because it scales much better. Graphene is a horrible material for traditional logic; it has no bandgap. Graphene switches have an on-off current ratio of ~3 (tiny and useless), whereas similar sized silicon-based MOSFETs have on-off current ratios of ~1000.
There is some interesting work on making a new kind of logic with graphene-based BiSFETs, but it's still not possible to actually fabricate them. In contrast, neuristors, which are another interesting form of nanoscale logic, have been fabricated by HP labs from Mott-insulator based memristors. If I had to put my money on a replacement for CMOS, it would be these neuristors. However, there are still huge engineering challenges that lay ahead. Nonetheless, the Mott-insulator based memristors are already being commercially developed for high density, solid-state memory, with the hopes of eventually replacing flash memory.
Re:And again.. (Score:5, Informative)
You're cynicism is valid in this case. This is just rehashing research from 20 years ago on negative differential resistance (NDR) two-terminal devices.
Logic based on two terminal NDR devices has been around for more than 50 years (tunnel diodes, neon tubes, ...). Its big problem is input-output isolation: cascading elements is tricky. But these guys are using four terminal devices in a three terminal NDR mode, so they don't have that problem.
Graphene switches have an on-off current ratio of ~3 (tiny and useless),
Well, that depends. The ECL gates that Cray used for their early supercomputers had nearly constant current. Some specialized applications still use ECL. If you're changing state frequently, low static current may not actually save power. So, if this new technology ever becomes practical, you'll see it in fast clocking cores where essentially every gate and flip flop is busy all of the time. The surrounding support circuits will still be silicon.
Re: (Score:2)
So, if this new technology ever becomes practical, you'll see it in fast clocking cores where essentially every gate and flip flop is busy all of the time. The surrounding support circuits will still be silicon.
Heh, so the internal logic is running at 400GHZ, and the rest of the chip is running at 10GHZ? Is that even practical?
Re: (Score:1)
Your CPU already operates at different speeds for different instructions, uses the cache at a slower rate, uses your RAM at an even slower rate and your disk an an even slower rate.
Re: (Score:2)
Heh, so the internal logic is running at 400GHZ, and the rest of the chip is running at 10GHZ? Is that even practical?
That's the way most systems have worked since the 486@66MHz days.
Re: (Score:2)
So, if this new technology ever becomes practical, you'll see it in fast clocking cores where essentially every gate and flip flop is busy all of the time. The surrounding support circuits will still be silicon.
Heh, so the internal logic is running at 400GHZ, and the rest of the chip is running at 10GHZ? Is that even practical?
Yes. Clock speed mismatch in different parts of a system is common with current technology. Cores commonly clock faster than memories, and much faster than many peripherals.
Re: (Score:2)
Re: (Score:1)
Heh, so the internal logic is running at 400GHZ, and the rest of the chip is running at 10GHZ? Is that even practical?
Some other folks have covered the fact that it's not unusual to interface wildly differing clock rates. The other thing to keep in mind is that whenever device researchers quote ludicrously high speeds, they're talking about the speed at which an individual transistor can switch states (because that's what their research is focused on). Real logic circuits are always at least an order of magnitude slower. They're built up from many transistors, many of which are organized serially, so the switching delay
Re: (Score:2)
If you exclude the integration issues, then you can go further back to 1959 with the GE Tunnel Diode Manual which discusses two-terminal negative differential resistance logic.
Since the speed was limited by the lead inductance of the discrete parts, I wonder how well an integrated version would perform. I suspect integration density would be limited by power dissipation as usual but integrated tunnel diode logic would sure be fast.
Re: (Score:2)
Re: (Score:3)
I'm really hoping it's isolinear chips.
Re: (Score:3)
Re: (Score:2)
Re: (Score:3)
Re: (Score:3)
FYI, here's [wordpress.com] the full post where the graph came from.
Re: (Score:2)
However we got more Cores. So there is more parallel processing going on.
Bad if you are still using DOS, but we learned to open up a bunch of apps without worrying anymore.
Re: (Score:2)
Re: (Score:2)
The focus shifted from more MHZ to doing more with them. Mostly parallelisation, more specialised hardware abilities and more efficient pipelining.
Still sucks for those algorithms that can't be made parallel, though.
Re: (Score:2)
Re: (Score:3)
A LONG time. Graphene is not even close to prime time yet. It leaks current like a colander leaks water and has such low gain (current or voltage take your pick) to make it nearly useless as a switch. Graphene *might* find use in broadband RF amps at lower power, but it's going to waste huge amounts of energy and you won't get much gain in the process. I'm not sure what value it will be, even in that application.
They are going to have to come up with some modifications to the graphene crystal structure
Re: (Score:2)
they solved that in 2012. Do try to keep up.
Re: (Score:2)
Not really. They have some improvements to the current leakage issue, but we are not quite there yet.
From my understanding (and that is admittedly somewhat limited, not being active in research on this) the state of research is that they can build working logic with graphene, but that they still require huge amounts of current and dissipate huge amounts of heat compared to the current silicon designs. The issue is the uncontrolled current leakage and a woefully low gain of a graphene transistor. I suspec
Re: (Score:2)
Your graphene question has been addressed by bobbied, and the answer to your other question is: right now. Modern GPUs can produce constant 60+FPS on crysis 1 at full details quite easily. In fact, they can do that for crysis 3 (think nvidia titans in sli). It'll just cost you quite a lot. But it's already doable.
The next stage is going to be 4k rendering, and for that, we're not quite at 60FPS+ yet. Though there we are not only hitting the output of modern GPUs, but even the GPUmonitor interface starts to
how much watt do you need for 400Ghz? (Score:1)
k.T.
Re: (Score:2)
Doesn't matter, it's negative resistance so you'll get back more than you put in.
Insecure Passwords will Be Toast ... (Score:1)
Once these transistors make up a functional computer &/or computer network.
Plenty of jobs for security specialists are ensured.
Re: (Score:2)
Re: (Score:2)
I suspect they wont immediately. More likely they will cut costs and instead of having 4 or 8 cores on your CPU, you'll have ONE, just clocked higher. a single core is a LOT less complex to program for, and it will run fast enough that it doesn't matter for some time. ditto for GPUs. I suspect you'll see the designs simplified, consuming less power, etc long before we see similar complexity designs to today on graphene.
A major reason is that the software just isn't there yet.
Re: (Score:2)
Why couldn't they make an x86-64 compatible CPU using this (or in fact, any other) CPU technology?
Perhaps the first few CPU's will be custom a ISA, but after they learn how to build full CPU's with this technology, there is nothing stopping them from making ARM or x86-64 compatible CPU's.
Re: (Score:2)
Even if it is a custom ISA, at 400GHz I'm pretty sure it can just emulate any existing CPU without even breaking a sweat.
Re: (Score:2)
And they'll a a linux distro in a week.
And Apple will change their architecture 2 years later.
Re: (Score:1)
When your DRAM still has nanosecond-level access times, it's still pretty relevant.
Re: (Score:2)
Re: (Score:1)
Obligitory XKCD (Score:4, Funny)
http://xkcd.com/678/ [xkcd.com]
Not negative resistance (Score:5, Informative)
"in which it demonstrates negative resistance (or negative differential resistance, as they call it)"
Negative resistance and negative differential resistance are not the same thing. Negative resistance would mean the current flows against the voltage. Negative differential resistance just means that the current goes down when you increase voltage.
The first one is not possible (unless you've got an external energy source driving the current) because it would imply a perpetuum mobile. The second is unusual, but doesn't violate any fundamental laws of the universe.
Re:Not negative resistance (Score:4, Funny)
How can you get anyone to read an article on science if you don't convince them you are violating the fundamental laws of the universe?
Re: (Score:2)
How can you get anyone to read an article on science if you don't convince them you are violating the fundamental laws of the universe?
Heck, that's half the reason to read Hitchhiker's Guide to the Galaxy.
Re:Not negative resistance (Score:5, Interesting)
Negative differential resistance just means that the current goes down when you increase voltage.
Interestingly, the entire electric grid is developing NDR, and that is a big problem for power companies. In the old days, if there was too much demand for electricity, or if transformers were overheating, the power company could reduce the voltage (a "brown-out") and the current would fall. But with more and more switching power supplies in electronics and fluorescent lights, that doesn't work as well anymore. The switching power supply in your computer and CFLs will compensate for reduced voltage by increasing the duration of the "on" phase of the switch, thus drawing additional current, the opposite of normal resistance.
Re:Not negative resistance (Score:4, Interesting)
I'm waiting to see what abuses show-off hackers can carry out with that. Not hacking the grid itsself, but the devices. Think a virus that infects the top five lines of electric car in 2030 and tells them all to flatten their batteries one night, then all simutainously kick in fast-charge mode at precisely the peak of the normal morning cuppa-tea surge. From low load to a couple hundred gigawatt above normal in five seconds. Grid wouldn't be able to react in time, substations would shut down automatically to prevent damage, could take hours to bring everything back up manually.
Re: (Score:2)
The UK national grid does not do this (lower the voltage) and I don't think ever has. As load increases, frequency begins to drop and there are various things that happen if the frequency can't be maintained within tolerances. Large industrial users with things that can go without power for a while without a problem have contracts with the National Grid to have frequency cut-offs. Think of a furnace that takes a week to get to temperature - being without power for a half hour doesn't really matter, so the f
The arc (Score:5, Interesting)
A little-known example of negative differential resistance is the common electric arc. In an arc, as the current increases the arc gets "fatter" (wider), and so the voltage across the arc decreases. Increasing current with decreasing voltage is negative differential resistance. This enables oscillations, which were first encountered as audio noise in electric arc lighting in the mid-1800s. These led to William Duddell's "Singing Arc" [wikipedia.org], in which Duddell added a tuned circuit to the negative resistance, creating a stable audio tone. The next step was obvious; he wired a keyboard to the arc and made the first electronic music.
Danish physicist Valdemar Poulsen took Duddell's audio oscillator and, by placing the arc in a transverse magnetic field, and in a hydrogen atmosphere (and somehow not getting blown up in the process), moved the frequency of oscillation up into the low radio range, around 500 kHz or so. This was the arc radio transmitter. It differed from the more common spark transmitter in that the arc's output oscillation was continuous, while that of the spark transmitter was a damped (decaying) oscillation.
The arc transmitter caught the attention of Cyril Elwell, of Palo Alto, California, who arranged to obtain the rights to the arc from Poulsen, and started commercial production of it with his company, the Federal Telegraph Company. The arc transmitter became a big success in World War One, when transmitters as large as 1 MW (one million watts) output were installed by 1918.
Much as the Fairchild Semiconductor Company spawned several successful companies in Silicon Valley in the 1960s, Federal did so, too, 50 years earlier; refugees from Federal formed well-known companies like Magnavox and Litton Industries.
Re: (Score:2)
Danish physicist Valdemar Poulsen took Duddell's audio oscillator and, by placing the arc in a transverse magnetic field, and in a hydrogen atmosphere (and somehow not getting blown up in the process), moved the frequency of oscillation up into the low radio range, around 500 kHz or so. This was the arc radio transmitter. It differed from the more common spark transmitter in that the arc's output oscillation was continuous, while that of the spark transmitter was a damped (decaying) oscillation.
I learned something on Slashdot, my day is done.
(I'm a software geek, so my electronics only had to go as far as a wheatstone bridge [wikipedia.org]. Which is kind of embarrassing when you consider that my grandfather was an electrical engineer, I'll bet he could have whipped up an arc transmitter for fun.)
How was the estimation of 400 GHz made? (Score:2)
And what prevents silicon transistors from operating at frequencies over 400 GHz in theory? I'd much very like to know the answer before gasping in excitement. Something is telling me this estmiation has very little to do with the current technological level we have now...
Re:How was the estimation of 400 GHz made? (Score:5, Informative)
>And what prevents silicon transistors from operating at frequencies over 400 GHz in theory?
http://en.wikipedia.org/wiki/Electron_mobility [wikipedia.org]
Simply put, electrons (and holes, if you're looking at the other way) can only move so quickly through a given material.
Re: (Score:2)
The article you have linked to does not provide the definitive answer as to what the relation between the estimate in maximum frequency and electron mobility is. So it is not clear to me that silicon transistors cannot achieve 400 GHz. It is intuitive to conclude that the faster an electron can move through a material, the faster it can oscillate, and the higher you can crank up the frequency, so max estimation in frequency for Si is most likely lower than the one in graphene. But, again, I cannot find an e
Re: (Score:2)
The article has the damn formula in it.
What you mean to say is you aren't smart enough to use it.
The actual number will depend on the doping.
for example:
Doping cm-3 = 10 to the -14
Electron Mobility (cm2 V-sec) ~ 1500
Hole Mobility (cm2 V-sec) ~ 450
doping has a big impact on the numbers.
Even if you could get Silicon to 400Ghz, the amount of power and heat would be a lot hire the graphene
If you need it simpler then that, then you're too ignorant to have an opinion on this topic.
Yeah, I'm rude but I am tired of
Re: (Score:2)
The article has the damn formula in it.
What formula? I asked for the estimation of the maximum frequency for a silicon-based transistor. A formula, along with the way it could be derived, would do. I looked for the word "frequency" in the article, yet I did not find anything directly describing said equation.
Doping cm-3 = 10 to the -14
Electron Mobility (cm2 V-sec) ~ 1500
Hole Mobility (cm2 V-sec) ~ 450
All right, you copied some numbers from the article. How are they related to the estimated maximum frequency of a transistor?
Even if you could get Silicon to 400Ghz, the amount of power and heat would be a lot hire the graphene
I asked if it was theoretically possible. The article implied it was not, and that was the reason for my question. A
teleportation (Score:2)
that's what's great about bandgaps in silicon. The electron doesn't travel, it teleports.
Re: (Score:2)
>And what prevents silicon transistors from operating at frequencies over 400 GHz in theory?
http://en.wikipedia.org/wiki/Electron_mobility [wikipedia.org]
Simply put, electrons (and holes, if you're looking at the other way) can only move so quickly through a given material.
It's a bit more complicated than that if you are looking for an "ultimate" limit.
Electron mobility is important when considering the charge drift-velocity limited performance of a transistor (e.g., in a typical sub-micron device of today), but for nano-scale transistors, ballistic charge transport is predicted to be an important factor in charge transfer. Basically instead of operating in an ohmic region where charge motion is modeled by a drift equations and limited by channel scattering effects, in balli
Re: (Score:2)
Re: (Score:2)
Silicon transistors can run at 500GHz* - but that's hitting the limits of electron mobility, and it needed liquid helium cooling.
Sometimes, liquid nitrogen just doesn't cut it.
* http://gtresearchnews.gatech.edu/newsrelease/half-terahertz.htm [gatech.edu]
Re: (Score:2)
Thank you, I did not notice your comment before posting the same link. I wonder how much better graphene will do at such frequencies and temperatures and how much of a breakthrough there is behind this summary blurb.
Re: (Score:2)
Answering my own question: nothing prevents silicon transistors from working over 400 GHz. IBM & GeorgiaTech have already done that.
http://gtresearchnews.gatech.edu/newsrelease/half-terahertz.htm [gatech.edu]
http://www-03.ibm.com/press/us/en/pressrelease/19843.wss [ibm.com]
As it has been mentioned by user mc6809e in another comment, certain transistors have long since reached 1 THz, but I'm unqualified in the area and can't find the appropriate article or key words.
Keeping my excitement for some other occasion.
Summary incorrect - Don't need 8 transistors in Si (Score:3, Informative)
Two NMOS transistors and a resistor can perform an XOR in Si. I remember interviewing at Intel in 1980, and every damn interview question was about XOR gates. First was an XOR gate in TTL, then an XOR gate in CMOS, and finally an XOR gate in NMOS. Apparently I passed all three questions, 'cause they offered me a job.
Re:Summary incorrect - Don't need 8 transistors in (Score:5, Funny)
NAND? Don't leave us hanging.
Re: (Score:2)
That wasn't the only part of the summary that was wrong; about the only part that was correct was the part that stated that they were able to perform an XOR in graphene with 3 FETs due to negative differential resistance.
Not MIT's work (Score:4, Informative)
MIT may have reported it, but the research comes out of UC Riverside. Give credit where credit's due; interesting research isn't only done at MIT, Stanford, or Cambridge.
Tunnel Diodes (Score:3)
Re: (Score:2)
Also I think they had to be made by hand, from Germanium, which made them expensive and unstable.
#import <std_ex_wife_joke>
Silicon is already there (Score:5, Informative)
Silicon transistors with sub picosecond switching times were fabricated in 2002. That's in the THz range.
What holds back processors today is mostly the RC delay of metal wires.
Re: (Score:3)
Thank you for clearing this up. I've asked above why silicon (or silicon with some additions) cannot reach the theoretical threshold of 400 GHz the summary seems to make such a big deal of and didn't get a clear answer. Turns out it can and that theoretical frequency, as somehow expected, is not the limiting factor for modern processors.
Re: (Score:2)
So people showing you the link with the math doesn't cut it, but some random post that agrees with you does?
I bet you believe your mother when she said you where a smart boy.
Re: (Score:2, Informative)
Stop being an arrogant asshole. The wikipedia article someone linked to him doesn't actually have any math relating charge carrier mobility to device switching speed. There's no way any reasonable person could expect him to accept it as an explanation in an informal context -- it's like getting angry at someone who asks why airplanes don't fall out of the sky, and isn't immediately satisfied by a link to a page full of basic fluid physics which doesn't explain (or even mention) lift.
As for your so-called