The industry’s most innovative people 2024 Mark Filer
Organisation: Stealth start-up Role: Area Tech Lead, Photonics Based in: Los Altos, California, US Education: MSc in EE, Georgia Institute of Technology
Mark Filer is currently working on photonic interconnects for artificial intelligence (AI) and machine learning (ML) systems.
What was the most surprising thing you found in the course of your latest research? The increasingly massive bandwidth requirements for AI/ML systems compared with traditional general purpose cloud infrastructure. GPU to GPU communications require massive optical "pipes" across thousands (or tens of thousands) of nodes, often requiring optics to the server. If the demand for AI applications continues to grow, we should see a discernible inflection point in the photonics industry compared with historical growth trajectories.
What’s your biggest research priority in the coming year? There are a lot of great tools being put in the toolbox by research, academia, and industry to construct networks supporting AI/ML applications. Examples include co- packaging, linear (direct) drive photonics, optical circuit switching and all-photonic fabrics. Understanding exactly where and how those tools may be best applied to deliver the most performant and efficient AI solutions will be critical.
In your opinion, what is the most significant photonics technology to emerge in the past 12 months and why? While the concept of linear or direct drive optical engines is not new, their application in traditional data centre transceivers has received a lot of attention in the last 12 months. The potential power savings (~50% at the module level, ~25% at the system level) and cost savings (from eliminating optical digital signal processors (DSPs)) can't be ignored. However, there are obvious challenges with operationalising a pluggable transceiver and switching ecosystem where the transceivers and host ASICs (which are effectively providing the DSP functionality with their high-performing SerDes) may all come from unique suppliers. Further is the 800G/112G-SerDes ecosystem "one and done" phenomenon, where future SerDes rates (224G+) and pluggable optical PMDs may no longer be performant enough to leverage linear drive pluggables.
What are the biggest challenges or threats to the industry in the next 12 months? How can these be overcome? Supply chain issues still aren't fully resolved
“GPU to GPU communications require massive optical 'pipes' across thousands (or tens of thousands) of nodes, often requiring optics to the server”
from the Covid-era bottlenecks, with some parts still experiencing excessive lead times. Additional fab capacity slated to come online in the next 12-18 months, along with expanding manufacturing capabilities (Asia, Central, and North America) to support the onslaught of AI/ML-driven bandwidth growth, should help mitigate the vestiges of supply constraint on the photonics side.
Which photonics sector do you see as having the greatest opportunity for growth in the next 12 months? Co-packaged optical transceivers, particularly for what's been termed "optical IO" applications, seems poised for growth in the next 12-24 months. Up to now, there have been solid POCs developed, but that killer app for these interfaces has yet to emerge. As we see the continued demand for AI systems driving ultra-high optical connectivity requirements, applications for such high density, power efficient photonic implements should solidify. This should be further bolstered by commercialisation of composable, chiplet-based UCIe systems.
What is your proudest moment in photonics so far? Helping to "jump start" a category of
photonic interconnect called "DCI". I was working as principal optical network architect at Microsoft at the time, and we were banking the Azure cloud network on a regional architecture that required massively bandwidth-dense point-to-point DWDM links interconnecting data centres together in a given metropolitan geographic area. Achieving the required inter-DC capacities simply could not be done with traditional transport hardware (cost, space, and power constraints prevented it), and so we partnered with Inphi to develop a router-pluggable 100G DWDM module in QSFP28 form-factor which was capable of closing 120km links. This was widely deployed at Microsoft and enabled their disaggregated regional architecture, with hundreds of thousands of ports still employed and operational today. It also paved the way for development of the now ubiquitous OIF 400ZR ecosystem, which was essentially the same concept scaled up from 100 to 400G. Follow-on developments of 800ZR and the recently announced 1600ZR projects are well under way.
What advice would you give to someone who wants to get to where you are/do the job you do? I was really fortunate to have done stints at several well-known hyperscalers, including AWS, Microsoft, and Google. In those roles, I was exposed to what scale means – architecting, designing, deploying, operating, and troubleshooting these extensive networks underpinning the giant public clouds. It definitely gave me a different appreciation for operational simplicity and standards than what I had coming from system vendors earlier in my career. Also along the way, I had the opportunity to meet super smart people, both within my companies, but also in the supplier space and academia throughout the photonics industry. I've learned so much from so many, and benefited by having a deep network of folks I can go to when I don't know the answers (which happens often!). So I'd highly encourage anyone looking to "get where I am" to put themselves out there – be involved in industry conferences, events, and consortia, and be intentional about building relationships. Those relationships are critical – ultimately the people you learn from and impact are what matters in the end.
Where can people see you in person? OFC, ECOC, OCP, and (hopefully, eventually) OIF.
2024 Photonics100 39
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56
