2020 Trends in Silicon Photonics

What's hot. What's not.

Dec 21, 2020

What's Hot

Silicon photonics is one of those rare academic fields where a significant portion of academic research is aimed at near-term commercialization opportunities.* As such, the field as a whole can seem to pivot almost yearly to direct its research energy into the latest hot topic. In 2020 that hot topic was using silicon photonic chips for matrix multiplication. A chip performing this function is sometimes referred to as a "programmable nanophotonic processor," "deep-learning hardware accelerator," "bitcoin-saving energy-efficient photonic co-processor," or similar haute description.

*Perhaps any academic field which has near-term commercialization opportunities will see a significant portion of effort dedicated to such activities?

Historically, and very generally speaking, academic silicon photonics results were in one of four groups: (1) A device design, where device here is a better polarization rotator, or mode combiner, or modulator, or photodetector, or some other fantastical arrangement of silicon, oxygen, nitrogen, germanium, boron, phosphorous, and metal atoms. These are bread-and-butter academic papers because there's almost always at least one interesting thing about any device a grad student might design worth publishing about. (Even devices that didn't work as intended *cough* *cough*). (2) An interesting measurement: first demonstration of [blank] thing or some odd nonlinear optics frequency comb laser. (3) An industry collaboration/bit of marketing that is light on key details but is almost always worldclass, not least because the test and measurement equipment probably cost a million bucks or more. (4) Some kind of integrated system-on-chip. This is the group all the trends fall into.

When I mention that these matrix multiplier integrated chips were a trend this year, I don't mean to say that the majority of published papers or even the majority of integrated systems papers consisted of these chips. A very small sliver of the published work in silicon photonics was dedicated to this kind of work. It became a trend in my mind because we went from approximately dozens or fewer of published papers in 2016 on this topic to literally hundreds and hundreds of publications this last year; and moreover, a significant amount of startup effort.

Because silicon photonics research is so close to real industrial applications, I've been identifying trends by the number of startups founded around these different research areas. In what is an extremely fuzzy reverse chronological order, past trends in silicon photonics have included optical quantum computers, augmented reality, solid-state LIDAR, biosensing, and optical communication. Prior to about 2012, the vast majority--going off of my memory here--of silicon photonics startups focused on optical communication. (Bookham, Luxtera, Lightwire, Kotura, Aurrion, Elenion, Skorpios, Rockley, Acacia, Ranovus, etc....). There are some notable examples of non-optical-communication silicon photonics startups becoming successful: Strobe, started in 2014, acquired by GM Cruise in 2017, and Aeva, also started in 2016, going public via SPAC. I expect several more of these trendy non-comm startups will soon be announced.

Using light to do matrix multiplication isn't exactly new.** Why then, anecdotally, has there been such a sharp rise in the number of academic papers aimed at discussing this trend in the last year? Certainly since the late 2000's and early 2010's, neural networks have become a mainstay in a wide variety of classification problems, with custom integrated circuits being deployed out of a necessity to achieve some sort of competitive advantage. Since 2017, several promising startups have emerged to commercialize the optical integrated circuit for this type of application.

**Sorry that this is behind a paywall. I'm Sure someone Could fIgure out How to get yoU access to this puBlication.

What's Not

As best as I can tell, pretty much everything that was hot is still hot. Augmented reality took a real bruising from Magic Leap's marketing, but silicon photonics is still in the AR game. I also have a (totally unfounded) sense that optical quantum computers had a couple doldrum years, but with the recent optical result showing quantum advantage, even this far-out topic will bring in more investment. Hell, some polymer modulator companies are still going at it. If you have domain expertise in integrated optics and any of the above topics simultaneously with integrated optics, consider founding a company.

At this point, it seems like it will be a real race to see what becomes the high-volume industry that silicon photonics was always meant for. Something with millions and millions of units shipped per year. I previously thought LIDAR would be that application. It still might be, but with the amount of money spent inside the data center, matrix-multiplying co-processors are in hot pursuit.