Glass processors aim to boost edge AI

Glass processors aim to boost edge AI

Technology News |
By Nick Flaherty

Cognifiber has developed a process for building machine learning systems with glass processors.

The Israeli startup has developed a technique for in-fibre computing, using analog machine learning networks for processing in a fibre optic cable. It has extended this to processors built on a glass substrate with the deposition of dopants using masks to generate a desired pattern of waveguides.

“For now, we design and test the implementation of the linear weighting matrix using a combination of standard glass and Erbium-doped glass. Such devices are expected to reach up to 3.2 TOPs per each weighted input using less than 0.5 Watt per input channel,” said Dr. Eyal Cohen, Co-founder & CEO of Cognifiber.

Even with in-fibre processing, which Cohen says can deliver a 100-fold boost in computing capabilities, there is still a reliance on semiconductors to conduct various operations of control and training. The development of the glass photonic chips, beyond downsizing, may provide a replacement for today’s silicon ones, while reducing manufacturing costs, power consumption, and the removal of bandwidth bottlenecks

This could provide edge AI processing in smaller devices with lower power, he says, reducing a data centre rack (such as those shown above) to a 4U server that is 18cm high and deployable in any office.

“The downsizing potential using glass-based photonic chips in conjunction with our proprietary fibres promises to bring superb-performance servers to the edge, removing many existing bottlenecks while dramatically reducing power consumption,” said Cohen. “Anything that generates vast amounts of data every second, such as connected vehicles, automated trains, or fleet management of large shipment drones can respond in real-time to events without reliance on data centres.”

Related edge AI articles

Other articles on eeNews Europe



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


Linked Articles