- Neurophos develops photonic AI chips replacing electrons with light for efficiency
- The pilot with Terakraft aims to demonstrate ultra -efficient optical computer
- A single chip promises 100 GPU performance using 1% energy consumption
An American startup is preparing to test an experimental processor that could provide performance on a scale never done on a single chip.
Neurophos, which was out of Duke University with the support of the Metaceptic incubator, associates with the operator of the Norwegian data center Terakraft to manage a pilot of his Optical IA inference platform in 2027.
By combining photonics with metamaterials to reduce optical modulators by a factor of 10,000x, company technology uses optical systolic tables that replace electrons with light latency straps, eliminating latency speeds while operating at clock speeds greater than 100 GHz.
Overcome the energy walls
By combining this with computing architectures in memory, Neurophos thinks that his chips could overcome the energy walls that limit conventional GPUs and TPUs.
Neurophos, which raised $ 7.2 million at the end of 2023, says its technology will allow a single chip to provide the computing power of 100 GPU while using only 1% of energy.
The planned collaboration will see the optical processing units of neurophos deployed at the Green Data Center in Terakraft in Norway. Previously, part of the sauda I hydroelectric plant (disconcerted in 2008), the installation ranks as one of the most effective in the world.
“By hosting the ultra-effective optical chips of Neurophos in our green data center to select business customers, we not only reduce our carbon footprint, but also the fact of raising the infrastructure mark of Econetics,” said Giorgio Sbriglia, Chairman of the Terakraft Board of Directors. “Our mission has always been to be able to fuel the future in a responsible manner, and this collaboration gives life to this vision.”
Patrick Bowen, founder and CEO of Neurophos, added: “Terakraft’s commitment to renewable energies and innovative technologies aligns perfectly with our mission to democratize high performance AI.
If everything goes as planned, the pilot in Norway could mark an early step towards durable ultra-effective AI equipment designed to manage future large-scale workloads.
Neurophos claims that end -to -end simulations validate the performance of its technology, with a roadmap targeting exaflop class computing on a single chip.
Via Enews Europe
