Si/inp Hybrid Laser + Waveguide

[Ishtvan]

Silicon photonics advances, check it. There are some caveats, one of which Mario mentioned himself in a presentation I saw: It remains to be seen whether it's actually more cost-effective than other Si laser methods, because when you do the wafer bonding you essentially bond a whole wafer of InP onto the Si, and have to etch away all of the InP in the places you're not using it. So you're wasting a lot of InP material unless you have a very high density of components on the chip that use it. Still, pretty cool that they got it working.

 

http://www.engineering.ucsb.edu/bowers/index.php

 

http://www.intel.com/research/platform/sp/hybridlaser.htm

[Ishtvan]

What the hell, no response?? Let me spell it out more clearly: POSSIBLY CHEAP INTEGRATED LASERS AND MODULATORS FOR OPTICAL INTERCONNECTS AND STUFF!! Now we just need those SiGe detector breakthroughs.

[ascottk]

I have no idea what that means . . .

[Gildoran]

As far as I can tell, it means bigger, faster busses without the problems of accidently generating radio interferance (meaning being able to transfer more data per second to a GFX card, for example), and perhaps better processors. (which is a good thing, since I've been under the impression that we're nearing the maximum potential of what electronics the size of a fingernail can do) Or maybe, it might make the equipment at either end of a fiber-optic cable much cheaper?

 

Ishtvan, are any of these guesses anywhere near correct?

[SneaksieDave]

Haha.. having just watched, I was about to ask. We could probably all do a bunch of research, but as long as we have our very own physicist, let's ask him!

 

Ishtvan:

1. In simplest layman's terms (okay, it doesn't have to be uber simple, but I assume the best description would be one that assumes no knowledge at all), what's the fundamental difference between today's tech and this (e.g., using silicon in a completely different way... using light (lasers) instead of electrons... etc.)? Why is it so much faster? They're talking on the order of 1000 times faster... How realistic is that for the consumer, and when?

2. What does this mean for AMD? Are they going to be completely barred out of this tech by patents (which pretty much guarantees they're done for)?

3. Why does John's voice change so dramatically from his opening statement to where he starts explaining? Maybe they're helium lasers!!1

[Ishtvan]

I'm not an expert on the technological implications, but basically, Silicon is cheap, and we have infrastructure for producing very large scale integrated (VLSI) Silicon structures, so this could allow us to produce integrated optics structures cheaply, and integrate optical functions with electronic functions (which we all know Si is good at) on the same Si substrate. One example of this is the optical interconnect bus for connecting chips in a processor, or connecting multiple processor cores in a supercomputer.

 

I don't know much about the systems-level stuff, but I was told by a researcher on this project that we'll hit the electronic bus limit in ~5 years as CPU development continues. I'm not sure what exactly this limit is, could be due to ohmic heating, or electron repulsion, or interference between two closely spaced wires. Photons have the advantage of not heating as much, they don't repel eachother, and they don't generate or respond to electromagnetic interference. Also, if you get multiple wavelengths muxed together, you can send much more data over one bus (like WDM in telecomm). Anyway, the main application Intel is interested in is chip to chip optical interconects.

 

The other application would be potentially making integrated optics devices more cheap, as it would allow things like optical transceivers to be manufactured with an Si substrate and Si manufacturing infrastructure. The thing they're demonstrating in that video is kind've like this - several lasers integrated on one Si chip. They could probably design them to emit at different wavelengths and mux them if they wanted. (Although I think Intel just sold the branch of their company that makes transceivers - whoops)

 

Anyway, to answer Dave's question specifically:

1. Silicon is not a good material for making lasers. In the past, commercial integrated optics that needed an active (laser) component have been made with materials like III-V semiconductors such as GaAs, InP. These materials are more expensive, and the infrastructure for processing them is not as developed, and also they are not typically used as a platform for electronic devices (GaAs is used for some high speed electronics).

 

Silicon is fine for guiding light, and Intel has also demonstrated an Si modulator for modulating light (there's still room for improvement in that one). So before this, we didn't have a good way of generating the light on the same Si substrate (although Intel did also made a purely Si laser using a different technique, but it is not as efficient). Now, with the wafer bonding technology, we can stick the InP on and use it to get optical gain. They've basically designed a laser cavity in Si with the optical gain contributed by InP stuck on top of it. Now we can generate light on the Si substrate, and since we can also modulate it, it opens the door for some integrated optics devices made entirely on Si substrates, and for adding optics capabilities to the current CMOS structures like CPUs.

 

As Mario says, it's not about trying to beat out the performance of other materials with Si, it's about trying to get 90% of the performance for 5% of the cost.

 

2. Not sure what it means for AMD. They never publically announced an optical interconnect research program. Either they've been working on it in secret or they've been waiting for Intel to do it first and then they'll be the second mover and modify it enough so that they can avoid patenet infringement (the laser industry is full of different ways of doing the same thing solely for patent reasons).

 

Keep in mind though that this is University research, so it's still pretty far away from actual technology. It'll probably be at least a few years before this is actually implemented in actual CPU products that compete with AMD products.

 

3. I dunno why John's voice changed there, but there was an editing cut there too. They could've changed microphones, or filled the room with helium. Fun fact: John likes tree-skiing.

[Ishtvan]

Ishtvan, are any of these guesses anywhere near correct?

 

Yeah, both guesses are actually correct, as far as I know. Although Si devices still have a way to go in performance before it can equal III-V semiconductors in terms of next-gen optical network backbone devices.