Diwali, the Hindu Festival of Lights...and Photonics, the Silicon Festival of Light

November 7 is the big day in Diwali, the Hindu Festival of Lights, as it symbolizes the victory of light over darkness, and of knowledge over ignorance. All of which made it a perfect day to have Cadence's Silicon Photonics Summit, and celebrate light over electrons, and knowledge of photonics over ignorance of the subject. For a basic introduction to the topic, see my post Silicon Photonics, and for an introduction to Cadence's photonic layout solution (and other curvy stuff), see my post Yoga Is Passé, the Future is CurvyCore.

I'm going to pick two presentations from the day to highlight:

• Vladimir Stojanović of Berkeley Wireless Research Center: More than Moore with Electronic-Photonic Integration
• Rick Stevens of Lockheed Martin: RF Analog Photonic Applications

Vladimir Stojanović

The day's keynote was delivered by Vladimir of the Berkeley Wireless Research Center at UC Berkeley. He went into a lot more technical detail than is appropriate in a blog post like this, but to me, the key message was that they came up with three different approaches to integrating CMOS and photonics, and then built fairly significant test vehicles to understand the advantages and disadvantages.

Back in 2010 or so, he and his team had wondered whether they could get a photonics device in a standard CMOS process with no changes. Even if the device wasn't great, the upside of doing this would be incredible, and they could continuously improve it. They built the first device in 2012 on a 45nm SOI process, and since then in 32nm. This is what they call "zero change".

Next, they considered a More than Moore approach, with any standard CMOS wafer with a photonics wafer (fabricated by CNSE, the Center for Nanoscale Science and Engineering) flipped over and bonded onto the top.

The third approach was "deposited photonics" using bulk CMOS (as opposed to SOI). They manufactured this successfully with Micron at 180nm and with CNSE at 65nm.

They decided to build some realistically complex designs with each approach to see the tradeoffs.

Bulk CMOS peaked at 65nm/40nm and then performance decreased at more advanced nodes due to the higher gate resistance. 32nm/45nm had both the fastest transistors and thick enough silicon bodies to guide the light. Lower dimensions don't have enough thickness for the photonics. They picked the IBM/GLOBALFOUNDRIES 12SOI (45nm) CMOS. This allowed them to build zero-change optics in 45nm, with no process modifications thus getting the closest proximity of electronics and photonics. There is a single substrate removal post-processing step. This gives a monolithic photonics platform with the fastest transistors.

As a demonstrator, they build a chip with zero-change 45nm SOI with photonics integrated into the chip. The chip had a memory band, and a RISC-V processor, and photonics transceivers. The chip could be configured either in "memory mode" or "processor mode" and then a pair of them could run together. It was the world's first processor to communicate with light. The processor/memory interface was completely optical.

Autonomous driving has made lidar important. Most lidar is pulsed: a laser pulse is sent out, and the time difference is measured to when it returns. But Vladimir thinks that it is potentially more attractive to build FMCW (frequency modulated continuous wave). It is less sensitive to shot noise and also less sensitive to background noise. Instead of putting out a chirp, the frequency is ramped in a sawtooth wave, so that the beat frequency can be measured between the outgoing "teeth" and the return.

They built the two chips in IBM 65nm for the CMOS, and CNSE for the photonics. This allowed them to build a complete monolithic lidar system including a beam steering optical phase array and an FMCW receiver. The photonics chip was flipped and bonded to the CMOS chip, as in the above diagram.

Finally, the most recent development is deposited polysilicon photonics platform, deposited onto deep trench oxide. It is the only way to integrate photonics with advanced nodes, workable with any of planar bulk CMOS, FinFET, and ultra-thin-body SOI CMOS.

To wrap up, he made three points:

• Silicon photonics is an enabler of new capabilities, and a good way to think about it is as analogous to a new on-chip inductor or a new on-chip transmission-line.
• It has the potential to revolutionize many applications despite the slowdown in CMOS scaling.
• Deposited polysilicon-photonics is the key to monolithic integration with advanced transistors.

Rick Stevens

Rick turned out to be the second presentation of the day. He works for Lockheed Martin, so is focused on defense applications. From Rick's point of view, the importance is that:

Photonics enables capabilities that otherwise could not be achieved within the same size, weight, and power.

A military aircraft (a "platform" in military speak) has a central core processor and a photonics backbone. If you don't centralize the main processing, then you end up with a lot of duplication of processing resources. The picture above shows the possibilities. Using fiber like this has the advantage of less loss per distance, wider bandwidth, EMI immunity, and non-conductive (lightning strikes, etc). Today, coax is used. It has limited distance due to losses, limited bandwidth due to losses, and needs equalizers to adapt to frequency-dependent losses.

Another practical problem with coax is that it is easily damaged if it is bent on too tight of a radius. Replacing it (which also can't bend it much) is a 10-day exercise that involves taking all the "stealth" surfaces off the platform, replacing the cable, replacing everything, then putting it in an anechoic chamber to make sure it is still stealthy. Rick loves the photo to the right that shows the difference more impressively than any number of words. The black coax is the old way of doing things, with cables that cannot be bent too much. The yellow fiber can be coiled so tightly that it can be wrapped around the old solution. The fiber is also higher bandwidth. The quarter gives you the scale.

This is what gets people so excited about putting this on the platform.

Yet another advantage is that fiber is fiber. There is no need to re-cable the platform to upgrade all the electronics. New electronics can run over the "old" fibers without having to strip down the platform. In the military, platforms can last a long time, with the electronics going through several generations. The most extreme example I know is the B-52 bomber, operating since the 1950s, with the last delivery in the early '60s, and is expected to last until the 2050s. Since its maiden flight was in 1952 (I think that is a coincidence, and not where the name came from) that would make it 100 years old.

The really big gains come with an integrated solution. On the left is a discrete solution (still with fiber, not coax though). On the right is the integrated solution. Rick had brought the device in the picture and passed it around for us all to handle. Okay, sorry, you had to be there. Some of these devices are located on the leading edge of the aircraft and there is not a lot of room out there. For stealth reasons, it also needs to be thermally isolated and so that also requires pumping coolant out there. The integrated version contains a thermo-electric cooler (or TEC) that he'd still like to get rid off for obvious power/reliability reasons.

Obviously, these solutions need to work in harsh environments and extreme conditions (war is messy) and that needs to be tested as they move toward deployment. The proof of concept is this year. But:

It won't be on a platform any sooner than two years, since we have to convince the platforms to take the risk. But a couple of years is pretty quick for a military application.

The diagrams above show the roadmap. At the top, today with coax. Second down, fiber cable with the large discrete electronics. Third down, the integrated solution. All these use analog. The last, which "we are a step or two away from being able to do" is to do all the IF conversion out at the receiver (at the end of the wing, say), and transmit digitally over the fiber.

Summary

I think these two presentations make a nice matched pair. Vladimir showed a lot of technical detail about how to integrate CMOS and photonics on the same chip. Rick gave a dramatic demonstration of what the upside from doing this could be, at least in the context of stealth aircraft. You can use your imagination to take all this and extend to, say, automotive applications.

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.