The money was awarded in conjunction with The University of Oxford, and follows funding from IP Group and Runa Capital.
Its intellectual property covers optical matrix multiplication, electro-optic neural network training and all-optical neural network training.
“Optics is a promising platform for high-speed and energy-efficient computing, because some mathematical computation heavily used in real life can be naturally realised using optical interference and diffraction almost for free. One simple example is that a Fourier transform can be achieved with a single lens,” a spokeswoman for the company told Electronics Weekly. “At Lumai, we build optical computing hardware dedicated to fast matrix multiplication. This is achieved by modulating the intensities of a series of laser beams using the mathematical values of the matrix elements, and summing the element-wise products with a lens.”
The inherently parallel architecture outputs an answer within 1ns as the light travels through the system. “Compared to other optical multiplication schemes, our technology supports much more parallel operations and is capable of a much higher computing speed,” claimed the spokeswoman.
Special-purpose optical processors such as neural networks can be built using the matrix multipliers.
“We have created a stack of intellectual property enabling us to build an all-optical neural network that’s minimally-reliant on electronics, with optical linear layers, optical non-linear activation layers and an optical backpropagation algorithm,” according to the company.
The funding will support building and launching optical neural networks for computing and machine vision.
Lumai’s website can be found here
