“Computing with Light: The Future of Photonics and the New OPSIS Lab”
A Conversation with Michael Hochberg, Executive Director, OPSIS Lab, University of Washington; hosted by Robert Anderson, Director Technology Transfer, Illinois Institute of Technology
MH: How do you replace computers that use electrons with computers that use photons? The answer is, it’s a bad idea. Photons don’t talk to each other. Making a photonic computer fights the physics.
BA: What’s the problem your’e solving and how far is up?
MH: We’re trying to make it cheap to manipulate photons. What’s remarkable about Moore’s law is that the marginal cost of adding another function to an electronic chip is negligible. In optics, the marginal cost of complexity remains very very high. We are repurposing the infrastructure developed to make electronic chips and by accident have built circuits, that can build all components of an optical system on the same chip as a transistor.
The first big impact: making it cheaper to move data. Optics starting to become dominant in high end computing, etc. All the cost is in digging the ditch. If you can move data seamlessly, you’ll see much higher bandwidth. The first place this will have an impact on consumers is in cloud.
Good news: it will become radically cheaper to move large vols of data long distances at very high speeds. With electrons, when you make everything closer and closer together, it becomes harder and harder to cool. That’s not true with photons. In the next few years, there will be disruptive change in the field of data transmission. The bad nes is that systems will have to be reengineered.
Militarily, DOD is a big funder of this. The first products are already coming out fo the market. The next gen will be analog– advanced signal processing that you can’t do in electronics. Eg radar systems, ISR.
BA: Where are you on progress as a field?
MH: Photonics as a field is about 50 years behind electronics. They have one function. There’s some integration, but generally built around idea that each component will be built in its optimal data system. We don’t have best-in-class devices for any of major device categories, but we can scale complexity. Most complex optical circuits, the doubling time is about 12 months– faster than electrons. Talking about being able to put thousands of devices down on a single chip. Tapping into infrastructure, rather than building it new.
MA: Costs of entering field?
MH: “Business model at the moment is kind of a catastrophe”. Spending billions of dollars trying to build things. Tweaks are Very expensive and you have to convince someone that it’s worth breaking process. Opsys is trying to change economic model through development of shuttle ecosystem. In late 70s it became possible for someone to design circuits and only pay for part of fab that their particular project was going to use. If you’re doing small circuits, you can develop little pieces and just pay for the area you’re using. That requires you have the ecosystem in place for everyone to share design rules, which is what we’re developing.
MA: How are you protecting your IP?
MH: It’s a challenge that I don’t have to be smart enough to solve. Semiconductor industry has drawn a wall between process IP and designers,w ho use PDK to design systems. Lines are blurred when your design needs tweaks, which is where Intel has an advantage (no walls). There are exceptions to the basic scheme, but the point is to create a system where all levels can contribute their specialties and have ownership, yet still maintain ownership. Physics is different, but the business scheme is the same.
We as a community have demonstrated every major photonic building block. Resonators, filters, all of the pieces you’d use to build an optical system, you can do. The one challenge is the laser,– can bond little pieces, create double layered chips. All of these pieces have been demonstrated together in the same platforms with transistors, but we are missing the ecosystem. Don’t have stable processes accessible to the industry.
Community is pretty friendly, there’s a consensus that photonics is on the way– Si as a plausible platform.
Been working on a lab on a chip — you can fold a very long optical pathlength into avery small area, using a circular design. 100 microns of path life, with light going around it 1000s or millions of times.
3rd generation of application: With a very small amount of fluid, you’re going to be able to do 100s of tests on the same chip.