Profile: Luminous Powers AI Optics at the Edge

A startup named Luminous Computing is focused on a technique for powering artificial intelligence (AI)-equipped edge devices that could prove to be a game changer.

First, the science: Back in October 2022, a team of 16 experts from NTT Research Inc., the Massachusetts Institute of Technology (MIT), and Nokia (NYSE: NOK) authored a paper titled “Delocalized Photonic Deep Learning on the Internet’s Edge” that appeared in Science. Why the technology it describes for edge applications wasn’t widely publicized until last week is a question that, put to NTT Research, went unanswered as of this writing. Anyway, the article outlined how newly designed smart optical transceivers can be deployed in sensors, smartphones, content delivery networks, and even aircraft to tackle the hefty processing and associated power consumption of deep neural networks (DNNs).

What is a DNN, you ask? It’s a kind of AI tool that mimics human reasoning to make decisions. The so-called matrix algebra involved in creating and deploying a DNN hikes the power required so high that IoT gear such as sensors and smartphones can’t process it. Instead, the edge devices must rely on separate servers to crunch the data. Even if these servers are located close to the device, latency and security are affected, the paper’s authors said.

The technology described in Science was dubbed Netcast thanks to its use of wave division multiplexing (WDM). It has been successful in experiments, the paper’s authors said, and will support edge device data rates at today’s electronic speeds while consuming “orders of magnitude” less power.

A “Luminous” Startup Emerges

While it may be a while before Netcast is commercialized, a bevy of the optical scientists who authored the NTT Research/MIT paper seems to have shifted from both companies to Luminous Computing, whose website states its mission to “build the world’s most powerful AI supercomputer” by “[bringing] optics to the heart of computer architecture.” Luminous will do this in part, it states, by applying optical technology (a la Netcast) to devices that can’t run sophisticated AI today (e.g., edge equipment for IoT).

Luminous was founded in 2018. It is headquartered in Mountain View, Calif., and has raised over $105 million in Series A and other funding from Gigafund, Bill Gates, 8090 Partners, Neo, Third Kind Venture Capital, Alumni Ventures Group, Strawberry Creek Ventures, Horsley Bridge, Modern Venture Partners, among others.

Luminous has 122 employees listed on LinkedIn, including several of the authors of the NTT Research paper who also worked at Elenion Technologies LLC, an optical component startup that was bought by Nokia in 2020. (Notably, it’s not clear whether Nokia Ventures invested in Luminous, though we’ve asked. No answer yet. Stay tuned.) Following are those authors, listed in alpha order:

Tom Baehr-Jones, VP of Engineering, was a co-founder of Luxtera and a former director of R&D at Elenion.

Michael Hochberg, Luminous president, was the CTO of Elenion and subsequently CTO of optical subsystems at Nokia.

Ari Novack, Luminous VP of system hardware design, was director of advanced module design at Elenion and then leader, advanced module design, at Nokia.

Matthew Streshinsky, is VP of Packaging, Photonics, and Mixed-Signal at Luminous. He was also involved in the founding of Elenion.

With a lineup like this, Luminous could become a major player in the edge computing space. Still, it's produced nothing commercial just yet. And there is competition from the likes of NVIDIA (Nasdaq: NVDA) and Intel (Nasdaq: INTC), both of which are keenly exploring – and funding – the use of optics in chip-level accelerators. Powering AI with reduced power is a key goal for other chip firms.

For the record: Another startup, Lightmatter, also has at least two authors of the Netcast paper on its staff: Darius Bunandar, chief scientist at the startup, worked on his MIT thesis with oversight from Dirk Englund, another paper author, who is now a technical advisor to Lightmatter and has founded his own startup, QuEra Computing. Lightmatter, funded in part by Hewlett Packard Enterprise (NYSE: HPE), has focused on optically networked chiplets and has a suite of products listed on its site.

Startup Profile: Luminous Computing

HQ location: Mountain View, California

Employees: 122 on LinkedIn

CEO and Co-Founder: Marcus Gomez (ex-Google, Mayo Clinic, Tinder)

Target market: edge compute AI processing

Prominent investors: Gigafund, Bill Gates, 8090 Partners, Neo, Third Kind Venture Capital, Alumni Ventures Group, Strawberry Creek Ventures, Horsley Bridge, Modern Venture Partners, among others.

Funding raised to date: $105 million (via Series A) and reportedly $21 more in unspecified round(s).