UK funds startup Lumai for optical AI

UK funds startup Lumai for optical AI

Business news |
By Peter Clarke

UK startup company Lumai Ltd. (Oxford, England) has been awarded a £1.1 million grant by the UK government to commercialize its work on all-optical neural networks.

A spin-off from Oxford University, Lumai was founded in November 2021 as OxONN Ltd. –  standing for Oxford optical neural networks. The name was formally changed to Lumai in February 2023.

The company said the grant would be support Lumai in building and launching optical neural networks for high-performance computing and machine vision. The grant, received in conjunction with the University of Oxford, follows on from previous funding of £1.75 million from IP Group plc and Runa Capital, announced in January 2022.

Lumai was spun out of the experimental optics research group of Professor Alex Lvovsky with the vision of developing end-to-end all-optical neural networks. The company was co-founded by Xianxin Guo, who serves as head of research, senior researcher Tom Barrett and PhD student James Spall. Experienced business executive Tim Weil has been recruited as CEO.

Laser beams through displays

Lumai’s technology makes use of free-space optics to perform basic operations by shining laser beams through addressable displays to perform the function of multiplication. Addition or subtraction can be performed by converging beams of light through a lens. This brings the advantages of low energy consumption and almost zero latency. The 2D nature of displays provides for layer-by-layer parallel computation and even greater parallel computation can be supported by wavelength multiplexing.

The company claims that its laser-based systems will be able to scale more easily than conventional electronic systems. Such optical neural networks can be 1000x faster – and much more sustainable – than existing transistor-based digital electronics. This is particular valuable for some of the most complex generative AI software that has billions of parameters that require training.

On its website it states that backpropagation is performed by shining laser beams into the neural network to complete optical training. There are also references to predominantly-optical self-learning systems that require no a priori digital modelling.

At the time Guo said in a statement: “Having already demonstrated the world’s largest optical matrix-vector multipliers to serve as the backbone of our ONNs, we are now making breakthroughs on developing viable schemes for training ONNs in an all-optical setting, a world-first.”

Related links and articles:

News articles:

Former Dialog CEO helps photonic AI startup raise $11.5 million

Canadian startup backed for photonic tensor processor

Startup raises $56 million for optoelectronic AI platform

Photonic FPGA developer ships SmartLight Processor

Photonics startup aims to create the world’s most powerful AI supercomputer

If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :    eeNews on Google News


Linked Articles