Alan shared his experiences and insights into the photonics industry and his start-up building journey.
About Welcome to the 34th installment of Foothill Ventures’ Lessons from Founders series. From time to time, we publish an in-depth founder interview, ranging from early-stage entrepreneurs to successful businesses. Our conversations cover their personal journeys, the lessons that shaped them, their visions for the future, and their failures. We also learn more about their companies and about the challenges they try to solve. These insights and lessons are applicable to any entrepreneur — current or future.
Quintessent, established in 2019 in Santa Barbara, develops optical interconnects for artificial intelligence (AI) and machine learning (ML) computing systems. Quintessent’s technology originates from over a decade of research from UCSB Professor John Bowers’ lab, leveraging quantum dot-based lasers and silicon photonic integrated circuits to enable connectivity solutions that are untethered by the scaling limitations of incumbent solutions. Quintessent’s solution is an optical chiplet that enables a new paradigm for highly scalable and power-efficient communication between compute and switch chips across the data center.
Alan, CEO and Co-founder of Quintessent, received his doctoral degree from UC Santa Barbara with a specialization in Electronic and Photonic Materials. Having worked as a photonics consultant for Booz Allen Hamilton advising clients at DARPA and ARPA-E, Alan is an expert in photonics, materials science, and electrical engineering. His co-founder, Professor John Bowers, is a prolific inventor — he has cofounded 5 photonics related start-ups. Quintessent has the audacious goal of solving the data movement bottleneck in computing.
Why we invested in Quintessent: Quintessent is developing a key enabler in the optical compute interconnect market, which has the potential to overtake the existing multi-billion dollar data center networking (Ethernet) market. The company also has early but very strong customer interest from top AI/ML chip companies. Lastly, the founding team are among the best in the world in this field (which we validated with executives at the top AI/ML chip companies). Specifically, Professor John Bowers’s lab is considered the top university lab in silicon photonics and optical communications. We were introduced to Quintessent by Alex Fang, Partner at Entrada Ventures. We consider Alex/ Entrada to be among the savviest investors in photonics (Alex himself earned a PhD in photonics from UCSB, and co-founded Aurrion with Prof Bowers).
What is Quintessent? And What Challenge Does it Aim to Solve?
Quintessent is working on solving the communication bottleneck between compute chips within data centers and accelerated computing clusters. The core problem is that the fundamental unit of computing now has shifted from being a single compute chip, be it a CPU or GPU, to a whole cluster, or even a data center of compute chips, which is to say that solving the most prevalent workloads and problems today, like generative AI, requires the orchestration of thousands, even tens of thousands of compute units in parallel.
The implication is, at the system level, the bottleneck that limits the application performance, runtime, etc., is no longer how fast the individual compute chip can crunch numbers by itself, but it’s how fast they can communicate with each other. And it is actually getting worse. Compute chips get faster and faster with each new generation, but the rate at which they communicate with each other is not keeping up. We can’t actually feed the compute chips fast enough with enough data to keep them fully utilized. So that results in huge inefficiencies in how well you’re utilizing the very expensive computing hardware that data center operators and end users pay for. It also limits the scale of problems you can solve and how fast you can solve, for example, a machine learning training problem.
Quintessent’s Journey: finding focus
When we started off, we were very much a technology company. We had great technology and great innovation from UC Santa Barbara that we wanted to commercialize. There were many different directions that we could have taken the technology because it applied to a broad array of different markets. We looked at LiDAR, our first customer was in the field of optical computing, and of course, we looked at communications. And at the same time, we were also successful in getting government grants and contracts to fund the development of that technology. But each of them was a little different and maybe tailored for a slightly different application. So what ended up happening was we ended up getting pulled in a few different directions. And ultimately that diluted our company’s strategic focus.
One thing that I would definitely do differently from the very get-go is to prioritize finding strategic clarity and direction. We ultimately did end up getting to that point of clarity, which is we want to apply our technology to solve data communication. However, I would have just focused on getting to that point of clarity much, much sooner, so that the entire organization is aligned and pointing in the same direction, rather than getting pulled in a few different directions and not moving as fast in any one of them.
Quintessent’s Big Breakthrough
Fundamentally, we know that the solution (to scale compute IO) requires optical interconnects, which is moving data with light. The reason is that it’s fundamentally more efficient to transfer data (at distances greater than a meter) using photons rather than electrons. And the challenge then becomes, if you just try to brute force scale today’s technologies in order to meet the needs of AI training clusters, and data centers, you pay a very expensive penalty in power, cost, area, and latency. Those are just fundamental limitations of today’s existing technology and a fundamental physical scaling law that we run up against.
What we are working on at Quintessent is we are tackling the problem from the foundational layer, we think there’s a paradigm change needed in order to solve the communication bottleneck between compute chips. Doing so practically requires a new set of foundational technologies to implement (solutions) in a practical manner. And that’s what we are working on at Quintessent. The technological innovations that underpin our solutions are threefold. One is we are leveraging fundamental advances in material science. We use a new material for our semiconductor lasers to make them more efficient, and more reliable. The second is we use novel designs for all of our photonic components that are needed for communication. That enables us to squeeze much more performance, bandwidth, and functionality out of individual chips (8–10 times more performance for the price of one). And finally, we combine all our elements on a single silicon substrate manufactured in silicon-based foundry. So, the net result is that we have a single-chip solution that results in the simplest possible form factor, the lowest possible power consumption, and the highest possible performance for communications. Much of this resulted from early research that was done at UC Santa Barbara and Professor Bowers’ lab.
History of Photonics: recent successes
If you look at the history of photonics, the primary application for photonics has always been to move data. The trend has been that photonics is used to communicate across shorter and shorter distances. And with decreasing distances, the volumes of the photonic interconnect deployments increase. Historically, the market was modest, but I would say it’s always been growing. And it’s now ready to explode, especially with the huge pressure on interconnects that are being created by the proliferation of AI. If you look at the companies in the past that were successful in photonics and in serving those early photonic segments, many of them were notable multibillion-dollar successes. Infinera is one example. Finisar is another one. Lumentum. The commonality among these companies is that they were all vertically integrated. Specifically, they owned their own manufacturing and built up their own manufacturing capacity from the ground up. That’s a very expensive proposition and very capital-intensive. Therefore, it increases the barrier to entry for a successful photonics company. Those are the two main limitations for the historically modest but healthy market.
So, Is Now the Time for Photonics? What’s changed now? Both of those things. In the sense that, firstly, on the market side, there’s now a huge demand for photonic interconnects. Also, there’s a huge opportunity to integrate photonics closer with compute and have them be deployed on a much more pervasive scale. Secondly, the ecosystem is more mature where photonics now resembles the microelectronics industry; you can do a lot of the (photonic) functions with silicon-based materials. Additionally, there’s now a healthy foundry ecosystem that will manufacture and outsource your manufacturing of the photonic substrates and chips, so you don’t have to build it from the ground up. So, fabless companies like Quintessent can benefit from that infrastructure by sharing and amortizing the cost of manufacturing with other customers that use the same foundry. In exchange, we benefit from the shared learnings at a foundry of much tighter process uniformity and much higher yields, that otherwise would be a very expensive investment for a single company. And so with those two shifts, now is a great time to be in the field of Photonics, there’s a great market opportunity, and the ecosystem enables a fabless company like Quintessent to thrive.
Interviewed by Eric Rosenblum. Videographed by Hannah Wu and Heidi Lu.
To get updates on this series, please follow our publication on Medium.
Follow us on LinkedIn: https://www.linkedin.com/company/foothillventures
Follow us on Twitter: @FoothillVenture
Foothill Ventures is a $150M seed-stage technology firm. We back technical founders across software, life sciences, and frontier technologies.
Questions, thoughts, reflections? Let us know in the comments below. We’re always looking for great entrepreneurs and early-stage ideas, and we’re always interested in having a discussion about venture, technology, and anything related. To see more about Foothill Ventures, please visit our website: foothill.ventures.
コメント