- Neurophos develops photonic AI chips that replace electrons with light for efficiency
- Pilot with terakraft aims to demonstrate sustainable ultra -effective optical computing
- Single chip promises 100 GPUS —YYDAVABLE using 1% energy consumption
An American startup is preparing to test an experimental processor that could deliver performance on a scale that was never obtained before a single chip.
Neurophos, who was spun out by Duke University with the backing of Incubatormetacept, collaborates with the Norwegian data center operator Terakraft to run a pilot of his optical AI -Inference Platform in 2027.
By combining photonics with metam materials to shrink optical modulators with a factor of 10,000x, the company’s technology uses optical systolic arrays, replacing electrons with light and removing latency bottlenecks while running at watches over 100 GHz.
Overcome energy walls
By combining this with computer-in-memory architectures, Neurophos believes its chips could overcome the energy eggs that limit conventional GPUs and TPUs.
Neurophos, who raised $ 7.2 million at the end of 2023, claims its technology will allow a single chip to deliver the calculation of 100 GPUs, while only spent 1% of the energy.
The planned collaboration will see Neurophos’ optical treatment units implemented at Terakraft’s Green Data Center in Norway. Formerly part of Sauda in the Hydropower system (closed down in 2008), the plant ranks as one of the most efficient in the world.
“By hosting Neurophos’ ultra-efficient optical chips in our green data center for selected company clients, we not only reduce our carbon footprint, but also raise the bar for energy-efficient AI infrastructure,” said Giorgio Sbriglia, chairman of the board of Terakraft. “Our mission has always been to run the future responsible, and this collaboration brings this vision to life.”
Patrick Bowen, Neurophos founder and CEO added, “Terakraft’s obligation to renewable energy and innovative technologies perfectly adapts to our mission to democratize AI with high-performance AI. By implementing our 100x more efficient inference chips in Terakraft’s Green Data Center, we prove that AI’s exponential growth can achieve sustainable,”
If everything goes as planned, the pilot in Norway could mark an early step towards sustainable ultra-efficient AI-Hardware designed to handle future workloads on scale.
Neurophos claims that end-to-end-simulations validate its Technology’s performance, with a roadmap targeted at Exaflop-Class Computing on a single chip.
Via EENEWS Europe
