Slashdot Mirror
Germanium Diodes Mean Progress Toward Silicon-Chip Lasers
David Orenstein writes "Teams at Stanford and MIT have each reported getting
strong light signals from germanium-based diodes on silicon at room temperature. Engineers have long sought to do this because, with further refinement into lasers, such diodes would allow for optical interconnects on chips. Optical interconnects could operate much faster and with less power than electrical (metal) ones that are becoming bottlenecks on current chips."
Wow, plant-based electronics! This will surely usher in a new age of biological computers that will be able to . . .
What? It's not a geranium diode?
Uh, how 'bout that new version of Firefox? Pretty snazzy, eh.
The news is that they've found a way to grow 'em on silicon, which lends itself well to chip production.
I know for sure that I used Germanium diodes before and I'm pretty sure Germanium-based LED's have been developed before. Dunno what the news is.
They seem to have improved on Germanium LEDs by doping them differently to the point where the can look into using photons to transmit information around a silicon chip in place of electrons. I imagine they will look into building light pipes out of silicon, ie, little optical fibres.
OT: somebody should teach ascribe how to use the title tag.
http://michaelsmith.id.au
Germanium semiconductors are old news(in fact, I have this vague impression that they might have gotten germanium working in fairly common use earlier than silicon); but, according to TFA, germanium-based light emitters built into silicon structures under more or less reasonable production and operation conditions, is what is new. This isn't about discrete components; but about structures built into larger silicon ICs.
Now, lets have that lead to jobs for the west, rather than simply giving the tech to China. All fo this American paid for RD, should require that the work stay in the west.
Do you really want to deny the West the advances in manufacturing that the Chinese have contributed?
It's a global economy now. Get used to it.
John
Now, lets have that lead to jobs for the west, rather than simply giving the tech to China. All fo this American paid for RD, should require that the work stay in the west.
Do you really want to deny the West the advances in manufacturing that the Chinese have contributed?
It's a global economy now. Get used to it.
Because copying Henry Ford and making up for that with cheap near-slave labor is a grand contribution. By that standard, the plantation owners of the Old South were models of efficiency and innovation. Hell, they had a technique so "advanced", they didn't have to pay their workers at all! I'd say that one-ups the China model, though not by very much.
And if you want to dumb this down by talking about lithography and the use of it by the Chinese to make integrated circuits, just ask yourself whether it was the Chinese who invented it, or the Chinese who merely have a gigantic economic advantage for its use (which they copied) since they are competing against Western societies that won't accept slave wages.
The promise of making a laser from indirect bandgap semiconductors, then gathering investors, then losing the investors' money goes back to the Sixties at least.
Some scientists showed off SiC blue LEDs in the '60s that shown brilliantly like laser light, but were not the read deal. The real blue room-temperature laser had to wait for Nakamura and a direct bandgap material.
Doping, adding nitrogen, and adding defects to the lattice to produce more light is nothing new. Look at your stop lights. It's working there, but don't count on these indirect materials suddenly turning into lasers. No need to hold your breath.
A quick scientific note. Photons have a lot of energy, but not much momentum. You get hot on a sunny day, but not blown over by the sun. Electrons fall almost directly down in the bandgap diagram to produce light. This makes direct-gap semiconductors useful for lasers. The trick one can use is to provide momentum-shifting impurities to the lattice of an indirect bandgap crystal. The electron creates a photon by dropping directly down, but some other mechanism shifts the electron momentum to create an overall diagonal transition. It's not efficient, but it works.
2-3GHz is around the frequency that FR4 (fiber glass) material for building PCB starting to become lossy. You know some of the cheap plastic gets hot in the microwave oven, that's because the material become lossy and change the energy into heat. Same principle here.
We played with 3GHz and was already pushing it back then.
So transmitting that type of signals outside a chip for a long distance ~ 30-40cm to a backplane onto another card for say a router core or a blade server is going to take a bit more work if you want frequencies well into the tens of GHz.
The real benefit is you wont have to worry about cross talk or other electromagnetic interference. The short haul of the board level optical interconnects means we can have very high speed chip to chip interconnects without worrying too much about trace routing or length. And LED's are quite efficient when it comes to turning to electrical power into light. Metal wires at high frequencies develop a high resistance which has to be overcome by using more energy.
Having said that you are entirely right in your main observation. The main problem for germanium has always been fabrication; no germanium ICs. This is because there is no germanium equivalent of planar technology. It has been known for a long time that if this could be overcome there would be a role for germanium. It's just that, as with so many apparently breakthrough technologies, making it happen turns out to be very hard.
From scarped cliff or quarried stone she cries "A thousand types are gone, I care for nothing, no not one."
That's cool, but with modern chip designs using electron tunneling for some of the effects, it can't be used chip-wide. On the other hand, light can cross through light, so you would be able to avoid tediously long tracks currently required.
There may be some additional interest in the aerospace industry for this. Optical circuits on the chips aren't going to be so affected by radiation, and by having more real-estate available for redundant components and optimal placement, they can improve the resistance to radiation considerably.
Not sure how much heat this'll cut down on, as the transistors are the big heat-producers. On the other hand, better placement means more even heat production which means they should be able to push the designs a little bit further.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
I can see this technology being able to be used to help with inter-chip communication, perhaps to help with running more tasks in parallel, or locking/unlocking memory segments shared by the CPUs.
The only thing I see that would be a limit is having to mux/demux a lot of signals before they get put on the fiber optic cable. However fiber optic cables have a lot of bandwidth, so this may not be a big issue.
It would be nice if silicon chip lasers could replace most signal circuits on a PC board. Mainly because it would allow positioning of components to allow for better cooling and heat dissipation. Ultimately, if several fiber optic connections can replace the hundreds (going on thousands) of pins needed on a CPU to the motherboard, it would be a great advance in reliability.
Fiber optics on chips isn't new though. I remember talk about the PowerPC 603 having the ability to have this for better SMP communication.
The Stanford team's abstract is at
http://www.opticsinfobase.org/oe/abstract.cfm?uri=oe-17-12-10019
and the full paper is downloadable there
The MIT abstract is at
http://www.opticsinfobase.org/ol/abstract.cfm?URI=ol-34-11-1738
but you have to pay to read the paper
well - they do know how to use the title tag - they actually used the title "Untitled Document"
The real benefit is you wont have to worry about cross talk or other electromagnetic interference. The short haul of the board level optical interconnects means we can have very high speed chip to chip interconnects without worrying too much about trace routing or length. And LED's are quite efficient when it comes to turning to electrical power into light. Metal wires at high frequencies develop a high resistance which has to be overcome by using more energy.
I'm not convinced. You can still get electromagnetic interference with light - look at TV remotes. Of course, if you use fibre optic cable it's not a problem, but that's akin to using coaxial cable to route electrical signals. While it would be possible to embed coaxial structures in PCBs to eliminate the possibility of cross-talk and noise, in practice this would be prohibitively expensive and the same result can be achieved with stripline and careful routing. The question is, what does an optical PCB look like? You'll still need copper for power distribution so the optical PCB will need to tolerate soldering temperatures. Do you have a layer of interwoven fibre optic cables? How do these interface with the components such that there is tolerance in the size and position of the terminals? Do you use mirrors and optical waveguides embedded in the substrate? If so how do you make this cost effective to manufacture? If you use fibre optics, you have a minimum bend radius, so you open up a whole new set of routing problems. While there are obviously clear benefits in theory, when it comes to actually implementing this as a cost effective PCB interconnect you'll have a whole set of new problems to deal with, and it's unlikely to be anywhere close in cost to gluing layers of copper and plastic together.
If I seem short sighted, it is because I stand on the shoulders of midgets
that was my WTF moment for today. I'll get some coffee now. sorry for the interruption.
Its actually very common to use coaxial like structures on PCB boards by placing sensitive signal traces between ground layers in a grounded copper pour. This works very well for shielding analog signals, but it doesn't work for high frequency digital signals. The grounded copper surrounding the trace creates a significant amount of capacitance that needs to be overcome every time there is a change in state. Once you get into the hundreds or thousands of megahertz, you start to consume more and more power to keep the rise times to acceptable levels. Eventually, you get to a point of diminishing returns where you're consuming huge amounts of power to drive signals and spending extremely large amounts of design time trying to minimize inductance and capacitance amongst traces - this is basically where we are now.
Also notice how we've all but abandoned parallel buses in PCs. This is because it is almost impossible with copper to ensure the multiple signals of a parallel bus will arrive at their destination at the same time - creating what is called a race condition. At higher frequencies, you'll start seeing data bits from the same message arriving before and after the clock signal which results in an unreliable interconnect. We have reached this point years ago, and that's why everything is using serial today - SATA, PCIe, USB, etc - serial buses eliminate the possibility for a race condition at the cost of requiring much higher clock speeds for the same throughput. Optical interconnects on the other hand could be used to much, much higher frequencies before we start seeing race conditions thus allowing for a very high speed parallel bus.
I don't think adding fiber to a PCB would really increase the cost of a PCB all that much, at least not when it becomes a common practice. I don't think the fabrication process would be all that much different from copper in the end; I don't see it being much more than routing a transparent plastic instead of copper. Granted, you'd be more limited in your routing paths, but that's nothing an extra layer wouldn't fix. So yeah, I guess it would increase costs a bit, at least at first, but it would probably reach the commodity level pricing we are at now pretty quickly. (The real cost increase will probably be the connectors and not the PCB itself.)
The point is we have just about maxed out copper and there is no other long term option but to move to optical interfaces.