U.S. markets close in 2 hours 6 minutes
  • S&P 500

    +11.45 (+0.25%)
  • Dow 30

    +135.23 (+0.38%)
  • Nasdaq

    +22.65 (+0.16%)
  • Russell 2000

    +18.99 (+1.06%)
  • Crude Oil

    +1.13 (+1.48%)
  • Gold

    +8.70 (+0.42%)
  • Silver

    +0.20 (+0.78%)

    -0.0012 (-0.11%)
  • 10-Yr Bond

    -0.0630 (-1.45%)

    +0.0018 (+0.14%)

    -0.2320 (-0.16%)
  • Bitcoin USD

    -70.92 (-0.19%)
  • CMC Crypto 200

    -3.48 (-0.44%)
  • FTSE 100

    -31.78 (-0.43%)
  • Nikkei 225

    -87.17 (-0.26%)

Stealthy startup Neurophos raises $7M to create metamaterials-based chips for AI

Neurophos co-founder Andrew Traverso works on an optics experiment in the lab. (Neurophos Photo)
Neurophos co-founder Andrew Traverso works on an optics experiment in the lab. (Neurophos Photo)

A semi-stealthy startup called Neurophos says it’s raised $7 million in seed funding to support the development of a chip that makes use of metamaterials for heavy-duty AI applications. And although the company’s HQ is in Austin, Texas, it has plenty of connections to Seattle-area tech leaders.

Founded in 2020, Neurophos was one of the first companies to receive pre-seed support from MetaVC Partners, a metamaterials-centric venture fund backed by Microsoft co-founder Bill Gates and former Microsoft executive Nathan Myhrvold. Neurophos’ co-founder and CEO, Patrick Bowen, previously contributed his expertise to Seattle-area metamaterials ventures such as Kymeta and Lumotive.

Tom Driscoll — the founder and chief technology officer of yet another Gates-backed metamaterials venture, Kirkland, Wash.-based Echodyne — is listed on Neurophos’ website as its CTO and co-founder. Kymeta’s former CEO, Nathan Kundtz, is listed as a board member.

The aforementioned ventures all rely on the exotic properties of metamaterials — electronic arrays that are structured to bend light in a variety of wavelengths, in a variety of ways, without the need for moving parts. Bowen told GeekWire that such properties could reduce the size and the energy requirements for photonic chips that could be tailor-made for artificial intelligence platforms like ChatGPT.

The key has to do with the scale of the chip’s optical modulators, which are analogous to the transistors on a conventional computer chip.

“We have revisited what the ‘optical transistor’ looks like,” Bowen explained. “The optical modulators that are available today are massive. We are able to use metamaterials to make optical modulators that are 8,000 times smaller, which means you can fit 8,000 times more of them on a chip.”

Bowen said pre-seed funding has sustained Neurophos through the past three years of startup incubation. “We raised $1.8 million with MetaVC, and then we brought in another $1 million in convertible debt for the incubation,” he said. Now the venture is moving on to the next stage.

Although Bowen was forthcoming about the pre-seed investments, he said he couldn’t immediately identify the lead investor for the fresh round of seed funding round, due to confidentiality concerns. (When we do learn the identity of the investors, we’ll update this report with the details.)

The seed round will help Neurophos develop a market-ready chip.

“It’s been three years since we’ve been in stealth,” Bowen said. “So, we built a benchtop proof-of-concept [processor], but in the meantime, the architecture has evolved so much that that benchtop proof-of-concept is no longer relevant to what we’re doing. What is relevant is our metasurface design. That has been designed to manufacture, and the purpose of the seed round is for us to ‘tape-out’ chips.”

For years, experts on the chip industry have said that high-bandwidth photonic chips like the ones that Neurophos is developing will play an increasing role in delivering the processing speeds and energy efficiency required for artificial intelligence and machine learning applications.

“The recent arrival of large generative AI models such as ChatGPT and Lambda has put a spotlight on the importance of networking bandwidth when using high-performance AI/ML clusters used to train such models,” Joris Van Campenhout, the director of Imec’s optical I/O program, said in a recent Electronic Design interview.

So far, Neurophos has gotten by with a small team of executives and a wide array of engineering contractors. “The reason why we chose Austin [for the company’s headquarters] was really for access to hardware engineers,” Bowen said.

Bowen said the company is now aiming to expand its full-time workforce to “around 10 within the next year.” And that’s just the beginning of the anticipated ramp-up to commercial operations. “We’re hoping to hit market by 2027,” he said.

Bowen is well aware that the technological fields known as photonic computing and analog computing have been marked by lots of unmet expectations. He said the rise and fall of those expectations seem to have followed a classic curve known as the Gartner Hype Cycle.

“Right now, analog computer is in a little bit of a Trough of Disillusionment,” he said. “There are a lot of these companies that have either failed, or almost failed, or have pivoted. But I like to think of Neurophos as being on the Slope of Enlightenment on the other side, where we realized that the only way to really solve this problem is to go all the way back to the fundamental physics of the optical transistor/modulator, and rebuild from the ground up.”

More from GeekWire: