The University of Washington today will debut a new program that aims to help researchers and startups get their hands on silicon photonic chips -- aiming to usher in a new generation of faster, low-power computer processors that carry data using light, replacing or supplementing electrons.

Intel is contributing $250,000 to the program. Attendees at a launch event this afternoon on the UW campus will include Intel CTO Justin Rattner, and CalTech semiconductor pioneer Carver Mead.

The program, based at the UW's new Institute for Photonic Integration, will offer a foundry service called Optoelectronic Systems Integration in Silicon, or OpSIS. Michael Hochberg, a UW assistant professor of electrical engineering and director of the new institute, explained the ins and outs of the technology, and the initiative, in an interview.

Q: For people who may not be familiar with them, what are silicon photonic chips?

Hochberg: Basically, the idea is that you can take all of the infrastructure that’s used to make silicon CMOS (complementary metal–oxide–semiconductor) chips for electronics — all of this really complex infrastructure that goes into building these multibillion dollar plants for making things like CPUs -- and you can repurpose it for making circuits that do all of the same things that they already do from an electronic standpoint, but that can also move photons around. That comes out of a combination of some fortuitous coincidences in terms of the physical properties of the materials involved, and out of the ability to be clever about the design side, as well.

Q: So in a regular chip, electrons are moving around, but in this type of chip, light is moving around? Is that it?

Hochberg: Yeah, exactly. Well, in this type of chip, electrons and light are moving around. Sometimes interacting with each other. Sometimes you’ll have data that’s carried just optically, sometimes you’ll have data that’s carried just electrically, and then you’ll have devices that translate data between the two domains.

Q: What are the practical implications of that for computer systems?

Hochberg: The biggest and most obvious near-term one is that right now, a lot of the way you design computers is limited by the fact that you have a lot of loss as you transmit high-bandwidth electrical signals over any substantial distance. That’s why your on-chip bandwidth, say your core-to-core bandwidth, is very very high. But your chip-to-memory bandwidth is substantially lower, and your bandwidth going from, say, your CPU out to your USB to your iPhone is even lower. So the further you have to push your data in the electrical domain, the more energy it costs, which translates into heat, to push the same data a longer distance. In the optical domain that really is not true in quite the same way, because it turns out that once you eat the overhead of translating data from electronics to optics, the losses are measured in tiny fractions of your power per kilometer, if you’re in a fiber or small fractions of your power per centimeter if you’re using an on-chip waveguide. What that means is that once you translate the data into an optical signal, and put it into a fiber, you can move components further away from each other, without having to eat a big energy cost in order to do it.

Q: So if this technology takes off, how might the average person using a system with one of these chips notice the difference compared with a system that’s based on a standard integrated circuit?

Hochberg: For instance, one of the things we think is going to be possible is making terabit scale links from, say, a computer to a computer or a computer to your iPod. What your talking about is 100x increase in the speed with which you can move data from place to place, just on the scale of a USB cable. So if now you’re waiting minutes or at least several seconds to move, say, a single episode of a TV show — you’re talking about moving an entire season or filling your entire iPod over the course of only a few seconds or even less.

Q: Why the UW for this?

Hochberg: This is an initiative that I’ve been pushing forward to try to create this shared infrastructure. We have a very good relationship with Intel. We have a pretty good history of doing leading-edge work in photonics through several centers that have existed on campus, most notably the (Science & Technology Center), which is an NSF center. We’ve got the Intel lab right off campus, which makes it very easy to collaborate. We’re funded by the Air Force, and I have a very good relationship with BAE (Systems, which will build the chips). Really the Air Force and BAE and Intel are the three organizations that have come together to make this happen.

Q: What will happen in the lab, who will be involved, and what will the outcomes be?

Hochberg: The idea is that if you’re someone, say, at a startup company that wants to build a product using silicon photonics, you would be able to come to us and we would give you a design manual that says how to use our processes. Then you would design your circuits within the constraints of those design rules. We would aggregate that together with a bunch of other users who are doing the same thing, send it out for fabrication, and then when the chips come back, we would do some testing to make sure that they are what they should be. And then we would send you the chips that have your devices on them.

Q: Would there be sharing of those particular chips?

Hochberg: Well, the idea is that you share the process, but the individual chips get broken apart, and they get sent to different people. Each chip only has the design that that person put into it.

Q: Gotcha. It’s like group buying at Costco. You break up the pack of tomato sauce.

Hochberg: Exactly. You and your neighbors buy the 24-pack of tomato sauce, and you only get the two that you need. But the cool thing is that, in this case, you’re paying the discount price but getting to make up a custom sauce, with your favorite ingredients, for your two bottles. So in some ways it’s the best of both worlds.

Q: For startups, is there a cost? How does it work economically for them?

Hochberg: There is, there’s a cost in making these chips. The cost of actually running the wafers and setting up the process is probably 90 percent of the total investment to make a set of chips. And so the cost of running something complex with us, as opposed to doing it yourself, is going to be between one and two orders of magnitude lower than what it would cost you to do it yourself, because you’re sharing the sunk costs with lots of other users.

Q: Have you already had interest from startups, or are you actively looking for folks?

Hochberg: We actually have more people who want to run stuff with us than we have bandwidth to run it right now. The first run we already have a number of people involved, and we’ve got a bunch of people expressing an interest in the second one. We’re still trying to figure out what the exact numbers are going to work out to. We’re doing this in real time — we’re trying to develop the process as we bring people in. So the first few runs are what are called risk lots, which means that we’re not actually guaranteeing that they’re going to work. We’re guaranteeing that we’re going to make our best effort to make them work. But because we haven’t run this process before, there’s no hard-and-fast guarantee that we can make that it will actually work as designed. We’re only bringing people in who understand those risks.

Q: Do you expect startups to result from this, as well?

Hochberg: Yes, I think we do. We have one startup that we’re doing that will probably be a user of these processes, called Portage Bay Photonics. I certainly expect that there will be opportunities for other startup activity around this.

Q: Related to that, is this the kind of thing that people should be looking at on a regional economic level and saying, hey, this is potentially a future for the region’s tech industry, if this stuff takes off and if Seattle is a center for it?

Hochberg: Well, I think we have a pretty good opportunity if we make the right kind of efforts to claim this as an industry. We have the advantage that we’re relatively close to Intel’s TMG (Technology and Manufacturing Group) organization down in Hillsboro. They’re Intel’s premier fabrication development organization. We’ve got a fair number of photonics startups and companies in the area. I think we could try to claim this as a strategic initiative, but obviously the state funding situation for doing strategic initiatives like that is not as strong as it might be at the moment. I’m funded out of the Washington State STAR researchers program, which has commercialization as an explicit goal, and that’s been pretty helpful in getting this started. But to really claim this as a regional cluster will require a pretty substantial investment of resources.

Image via University of Washington