There was a time when we thought a 50 MHz 486 was something to get excited about. In comparison, the computer this post was written on clocks in at about 3.8 GHz, which these days, isn’t an especially fast machine. But researchers at the University of Rochester and the Friedrich-Alexander-Universität Erlangen-Nürnberg want to blow the doors off even the fastest modern CPUs. By using precise lasers and graphene, they are developing logic that can operate at nearly 1 petahertz (that’s 1,000,000 GHz).
These logic gates use a pair of very short-burst lasers to excite electrical current in graphene and gold junctions. Illuminating the junctions very briefly creates charge carriers formed by electrons excited by the laser. These carriers continue to move after the laser pulse is gone. However, there are also virtual charge carriers that appear during the pulse and then disappear after. Together, these carriers induce a current in the graphene. More importantly, altering the laser allows you to control the direction and relative composition of the carriers. That is, they can create a current of one type or the other or a combination of both.
This is the key to creating logic gates. By controlling the real and virtual currents they can be made to add together or cancel each other out. You can imagine that two inputs that cancel each other out would be a sort of NAND gate. Signals that add could be an OR or AND gate depending on the output threshold.
[Ignacio Franco], the lead researcher, started working on this problem in 2007 when he started thinking about generating electrical currents with lasers. It would be 2013 before experiments bore out his plan and now it appears that the technique can be used to make super fast logic gates.
We often pretend our logic circuits don’t have any propagation delays even though they do. If you could measure it in femtoseconds, maybe that’s finally practical. Then again, sometimes delays are useful. You have to wonder how much the scope will cost that can work on this stuff.
https://paeantosmac.wordpress.com/2015/08/13/technology-optical-computers/
I wonder what this will allow… very simple, very fast PD controllers that don’t need the integral term because they react just that fast?
No, the delay in these loops is mainly caused by the intrinsic delays of the sensors and actuators. The computation time is negligible.
The integral term would still be needed to reduce steady state error which is its main purpose anyway, it doesn’t matter how fast your computer is, it depends on the actual hardware and how it responds to input signals. The integral term isn’t going anywhere.
thank you!
“These logic gates use a pair of very short-burst lasers to excite electrical current in graphene and gold junctions.”
Sounds like a quasi-optical electronic device. As if modern electronics aren’t complicated enough.
finally, all those sharks with laser beams on their heads can use math to better their position in life.
Deliberately waited in the hope that someone else would post about sharks. Left satisfied.
Bloatware writers say…. Challenge Accepted!
While transmission distances will be “short”, flexures and vibrations will effect operation — so maybe sensors will be one of their main applications (e.g. super-MEMS)?
Considering other such “break-throughs” I’ve “seen” in my life (google: ovonics in the 70s), I probably won’t be around to see this come to fruition, at least as envisioned.
Agree, I probably won’t be around either… But I do remember when I got a 100Mhz 486DX CPU and thought this would be the last CPU I’ll need!!! Boy was I wrong.
OK, pretty neat to be so fast. And the structures needed are not too ridiculously small: 1 petahertz is 0.3 mm wavelength.
But the energy required! if each logic operation requires two or three photons, that’s about 5 eV per logic operation (a lower bound — likely to be much higher). Do this at petahertz rates and you’re looking at a half milliwatt *per gate*. Even a very modest CPU or array of a million gates will eat hundreds of watts.
Now, though, your structures are likely to be *huge* (wavelength-size), so the areal power density is actually quite low.
Clearly not for general-purpose computing, but maybe a niche in sampling or some such application.
How would they deal with heat?
Deploy it on Pluto.
Or Ura…
nah, forget it.
LOL!
Great idea! If it’s on Pluto, no need for bulky, expensive cooling equipment. May notice a slight increase in latency, though, if we want to use the results here.
I think you made an error in your calculation. The wavelength of 300μm (0.3mm) corresponds to 1THz, not 1PHz. 1PHz is 300nm, or blue/ultraviolet light wavelength. This is nearish the size of individual transistors.
Size of a transistor??? Did you mean a decade ago?
I think you meant size of 3600 5nm function (60 x 60)
Great, I was working on something like this abou 5-10 years ago. Still have the notes. They are finally catching up with me.