Graphene transistor clocked at 427 GHz [pdf] | Hacker News Reader

Graphene transistor clocked at 427 GHz [pdf]

pnas.org

The headline is quite misleading to folks with software backgrounds: "Clocked at 427 GHz" implies that the paper's authors measured something with a signal at 427 GHz. In truth, the authors measured up to 30 GHz and extrapolated a figure-of-merit f_T to be 427 GHz.

For comparison: InP-based HBTs were measured north of 400 GHz back in the 90s, which tells you that Graphene in this paper is behind where InP was 20 years ago. Of course, that doesn't make for sexy headlines.

If you're interest in what the figure-of-merit means: f_T is the theoretical maximum frequency that you can build an amplifier and still have the transistor provide gain. Practicalities limit amplifier design (or digital circuits) to frequencies much lower than f_T.

Nice result. As others have pointed out its not a record or anything, but it is a solid indication that graphene transistors are, in fact, credible. An example of a not really credible transistor are the various organic ones, they have frequency limits in the high kHz to single digit MHz.

But workable graphene transistors gets us one step closer to an all carbon device. Using diamond as a substrate for heat removal, Graphene-on-Diamond (oh snap, those would be GoD devices) could provide for an interesting replacement for more exotic silicon processes. (one could argue that an all carbon process is more exotic still if one chose to).

The benefits over silicon would be faster operation at an equivalent temperature, and the ability to operate at a higher temperature. Pretty much all silicon products melt at a junction temperature around 175 degrees C but carbon based devices should be able to continue to operate well into a few hundred degrees C if I am reading the papers correctly.

Should be an interesting decade.

Great. Now we can finally convert heat radiation (300GHz-430THz) to DC using 1mm long antennas and graphene transistor rectifiers.

I suppose we could, but why would we want to?

We would no longer be slaves to the Carnot cycle. Also, you don't need a temperature gradient anymore. Everything warmer than 0K emits thermal radiation. Think free energy for everyone.

I guarantee you that these rectennas will not work if their temperature is equal to that of the source radiation, and that they will generate waste heat which has to be removed. As for slaves to the Carnot cycle, no one has ever built a Carnot cycle; it is merely an ideal which these rectennas might help us approach more closely.

This is bullshit (specifically, a perpetual motion machine of the 2nd kind).

Suppose you have a heat source T_H in an environment T_C. If you could extract more work from the thermal radiation of T_H than the Carnot limit for this system, you have enough work to run a Carnot heat pump from T_C to T_H that increases the temperature of T_H, and decreases the temperature of T_C -- an violation of the 2nd law. You could use work to increase the heat source's temperature, leading to more work, leading to higher temperatures, in perpetuum. No can do!

Most likely before "free energy for all" you'll have strange IR oriented sensor apps that can analyze the spectrum not just measure total power.

Personally I think it would be fun to see a revival of IR spectroscopy outside of o-chem labs. Its been mostly superseded by NMR in the labs. But imagine an IR spectroscope on a chip. Perhaps for med purposes (what?) or maybe on oil well drilling down hole analysis or something like that.

Or just day to day appliances, like a "smart" household smoke alarm that analyzes the spectrum to identify and ignore tobacco and broiled/grilled meat combustion products but gets really excited about any measurable combustion products of burning furniture or burning paints or burning fabrics. I bet you could cut minutes off fire detection time with one of those, which doesn't sound like much but I bet it would save a lot of lives.

Is this a plausible method of extracting useful amounts of energy, or are we talking about yoctowatts per cubic yottametre?

This look quite promising: http://en.wikipedia.org/wiki/Thermal_radiation#Selected_radi...

We're speaking of radiant heat fluxes in the > 1kW/m^2 range for a sunny day.

I need more than theoretical maximums to tell me this is practical. At what efficiency could it be captured, and at what cost?

If 1 square meter of "thermal panel" were to cost as much as 1 square meter of modern CPU core, and then operate at 1% efficiency, I wouldn't see many practical applications.

Solid state generators would be pretty cool.

Self-powered sensor chips? Micro-transmitters/cameras that don't run down? RFID tags printed onto clothing?

Why wouldn't you?

What about Maxwell demon http://en.wikipedia.org/wiki/Maxwell's_demon ?

Maxwell's Demon applies to heat in the form of brownian motion.

What donquichotte suggests applies to radio waves in the infrared band, basically creating a crystal radio running off light.

We colloquially equate infrared to thermal heat because the former induces physical motion in atoms.

And the suggestion is stark raving awesome.

It's far from just colloquial. Infrared and thermal motion at the same temperature contain the same amount of exergy, meaning that the thermodynamic limits on how much work they can do are the same.

The interesting phrase here is "scalable fabrication". Like the many new battery technologies, there are many ways you can make next-gen transistors, but none of them yet has been manufacturable at large scale.

To me, it's not the transistor that's impressive in modern technology it's the manufacturing process we use to make large scale silicon devices. It's those that are becoming increasingly hard to advance and those that we need to find replacements for.

Funny, I've heard that one of the things making graphene exciting to industry is that it may be easy to adapt existing photolithographic processes to it. I don't know how true that is though.

For anyone curious why this doesn't translate into a 427GHz computer, light can travel about 1.4mm in one cycle at 427GHz, meaning that all synchronized components would have to be located within 1.4mm of each other. The bottleneck for processor speed became the speed of light some years ago.

The speed of light isn't directly why commercially available CPUs stopped getting faster (in terms of clock speed). That's more related to the fact that the increased transistor density and thinner gates (with more leakage) have pushed power dissipation to its limits. If we clock any faster, the CPU will burn up. That's why you can overclock if you use liquid cooling, for example.

That said, I'm sure there are fundamental limits to transistor clock speeds imposed by the speed of light. It's just not why CPUs stopped scaling around 2-4 GHz.

More to the point, between the cycles of the clock in your CPU a signal has to make it's way across a number of gates. This is usually measured in terms of the equivalent of the time it takes a transistor to drive the capacitances of 4 other transistors (FO4s). Your typical CPU these days will have between 10 to 20 FO4s of delay for each actual clock cycle.

I wonder if some the research going on at nasa right now about FTL transportation can be applied at some point?

I doubt it. As far as I understand it, NASA's research centers around the possibility of deforming spacetime to cause an object to fall through space (sorta).

Deforming spacetime wouldn't really put components closer together, or cause light to travel faster, or anything.

You never know with stuff this weird though. Maybe something else will fall out of their research. It's been known to happen when pushing the boundaries of our understanding of physics.

Well if you treat light like an object (if you consider its particle like properties), and deform space-time between the components (or more like deform space-time around the light, so it is of regardless where the components are in euclidean space), wouldn't that theoretically achieve the same thing as making light "move" faster between the components?

Though the way I envision that the components would have be connected by or lie in some kind of vacuum. I mean, these have to be the in the family of tests NASA will have to be conducting on particles before "moving" ships through free space becomes some kind of reality, right?

they could be opening the very gates of hell in that lab

there is no need to keep large synchronous areas though. look up clock domains and asynchronous circuits from the 90s and onwards.

Fascinating, thanks.

You can get a feel for how difficult this is to work with by building a CPU in minecraft.

http://www.technologyreview.com/view/518426/how-to-save-the-...

That means nothing - IBM made a 350 GHz silicon transistor that can go up to 500 GHz at very low temps: http://www.techspot.com/news/21961-ibm-produces-500ghz-silic...

On its own it's useless, though...

427GHz at room temperature isn't nothing.

So... what does the clock rate look like when you've got a network of billions of those things? Propagation delay, rise times, fall times... have they managed to make a simple little CPU out of these graphene transistors yet? Because that would be really cool. The paper's abstract seems to imply that they've found (what they believe to be) a viable way to mass fabricate these, though it makes no mention of an actual network of transistors yet. I can't wait.

How would you even sample a signal like that in the real world?

The best Oscilloscopes you can get right now commercially go to about 65GHz using hybrid chips (http://youtu.be/dx596o8t_TY) and cost $500K...

* http://www.home.agilent.com/en/pd-2108888-pn-DSAX96204Q/infi...

I work in high frequency RF (10s to 100s of GHz). You can use mixers[0] to down convert higher frequency signals and make measurements at lower frequencies.

[0]http://rfic.eecs.berkeley.edu/~niknejad/ee142_fa05lects/pdf/... [pdf]

What are transistors in current gen cpus clocked at?

Intel have their gate delay graphed with 22nm vs. 32nm (and also the switch to 3D gates)... but I'm not sure what the units used here are. It's on page 21 of this PDF: http://www.intel.co.uk/content/dam/www/public/us/en/document...

picoseconds maybe?

Edit: Yea, pico, I think. From another PDF on Intel tech: "At 15nm, the gate delay is 0.39psec, and for 10nm gate length, the gate delay has dropped to 0.11psec"

Deleted comment

...no, that's not how it works.

I realise you're probably being silly on purpose, but.. no.

That is about as far from how it works as you could possibly get.

If we could get a graphene CPU clocked at 10 GHz in the next few years I would already consider it a success.

While this is nice result, it still doesnt solve the biggest problem - scalability. Also this graphene device doesnt turn off.

Great. Year+ old. Thanks for Hacker -> News <-

Reply on news.ycombinator.com