Light-Emitting Particles Yield Faster Computing 65
schliz writes to tell us that researchers at the University of California San Diego are developing new transistors based on particles called 'excitons' in an attempt to speed up the interaction between computing and communications signals. "Excitons are formed by linking a negatively-charged electron with a positively-charged 'hole'. An exciton decays when the electron and hole combine, emitting a flash of light in the process. By joining exciton-based transistors to form several types of switches, the UCSD physicists were able to achieve switching times on the order of 200 picoseconds."
200 ps switch times != fast. (Score:4, Interesting)
Hmmm.... 200 ps switch times.
A modern processor operating at 2GHz has one clock cycle every 500ps. A signal leaving a flop and travelling to another flop typically goes through about 20 gate delays, yielding a switch time of 500/20=25ps.
Tell me again how this is faster?
Re:Holes vs. Positrons (Score:2, Interesting)
That may have been the initial model. but today, positron is anti-electron and in semiconductor industry a hole is lack of electron. I haven't read the article (this is /. after all) but IMO the flash of light here is not due to annihilation of matter but due to the electron entering a lower energy state.
Re:Here are some better articles (Score:2, Interesting)
Also, the article sourced in the submission is apparently using a copyrighted image, that of Pink Floyd's Dark Side of the Moon 20th anniversary cover. IANAL, but it doesn't look like it's covered under fair use, either.
Re:Here are some better articles (Score:3, Interesting)
You are correct. (Score:5, Interesting)
In an organic photocell an incoming photo will excite an electron. The positive and negative charges (electron and hole) will be "linked" together (i.e. they will move around together). In this state they are not useful. However, if you can separate them and draw them in different directions, then you'll get a current. They can only be drawn apart if you create a situation in which it is energetically favorable for them to separate, usually by attracting them to high and low work function contacts. Therefore, in a photocell of this type, you sandwich two materials together - one in which it is easy for holes to move, but difficult for electrons, and one in which it is easy for electrons to move, but difficult for holes (called the hole transport and electron transport layers). Then, you put a bias across the layers by using two dissimilar contact materials, one high work function and one low work function. Note that one contact needs to be transparent (ITO is most common) so photons can get to the middle layers.
Anyway, when an exciton is created it goes on a random walk through the material in which it is created, and will eventually collapse. The 'exciton diffusion length' is the distance over which your average exciton will move before collapsing. You want any created excitons to be within a diffusion length of the interface between the hole and electron transport layers. When the exciton hits an interface, it separates and the charges move towards their respective contacts. Put a load across the contacts, and you've got a working circuit (assuming excitons are being created).
This is a mildly simplified explanation, but it works.
Anyway, you can go the other way - imagine injecting an electron in one side and a hole into the other. You could choose your materials such that they would meet up at the interface and collapse together, emitting a photon. This is an OLED, and is conceptually similar to the photocells I just described.
So now imagine that you make it so that either the hole or electron transport layer is semiconducting. You could set up your device such that a dielectric layer and then a 'gate' contact are touching the transport layers along an axis perpendicular to the nominal current flow through your device (like in a thin-film transistor). Then, the layers would only transport charge (like in a transistor) and hence emit light (like in an OLED) when a voltage is applied to the gate contact. Then you have a thin-film device across which you put a bias that only emits light (and draws current) when it is switched 'on' by the gate contact.
In other words, you've combined a TFT with an OLED. Very, very slick.
Re:excitons are not supposed to be faster switcher (Score:1, Interesting)
Individual electrons travel at much slower speeds in a wire, on the order of molecules in a gas between collisions.
Re:FYI: An exiton is just an imaginary particle. (Score:3, Interesting)
You're right, this is basically a fancy LED. The difference seems to be that they're exercising control over when the holes and electrons recombine, in order to switch it off and on much faster.
In a regular LED, you have to flood the diode with minority carriers, and wait for them to recombine spontaneously, which presumably has a nice long time constant, so you can't actually turn it off and on very easily...
In this exciton thing, they're letting the minority carriers combine into excitons first, which (somehow) lets them control the recombination more precisely, and allows them to switch it off and on directly with the current.
In high-speed communications, they use lasers, which are basically LED's where the recombination is accelerated by stimulated emission, instead of spontaneous emission. You can look at the exciton thing as an attempt to do the same thing more cheaply, and on a smaller scale.
Re:Here are some better articles (Score:3, Interesting)
I wonder if it would be possible to etch hollow channels on a circuit board and have photons bounce through them rather than just have electrons running through wires.
Re:excitons are not supposed to be faster switcher (Score:2, Interesting)
You are looking for this link:
http://en.wikipedia.org/wiki/Velocity_of_propagation
Electromagnetic waves travel at the speed of light in a vacuum. In copper wire it can be as low as 40% of the speed of light.
If I recall, converting from electricity to light and back was really slow, so this will help switching speeds, and thus internet bandwidth.
Re:200 ps switch times != fast. (Score:3, Interesting)
Compact 5GHz on-die optical links would be a godsend for global routing. Global routes typically take several cycles for all but toplevel metals. Even at the top of the stack it's getting harder to get everything where it needs to go in a single cycle due to higher RCs as dies scale down. If this technology is viable I'm sure they'll improve on that delay time as well.