The light-based chip can only perform selected tasks at present such as image recognition, but can operate much faster than current products on the market.
Right now it seems they depend on direct optical input, so it’d be harder to adapt it to other deep-learning techniques. It might be useful down the line for stand-alone computer vision parts for micro-controllers. If they manage to develop this photonic computation tech to a commercial and more general-purpose scale, it could seriously revolutionise the GPU industry too, specially on the energy efficiency front. Really impressive stuff.
The paper in question for the nerds: https://www.nature.com/articles/s41586-023-06558-8
Right now it seems they depend on direct optical input, so it’d be harder to adapt it to other deep-learning techniques. It might be useful down the line for stand-alone computer vision parts for micro-controllers. If they manage to develop this photonic computation tech to a commercial and more general-purpose scale, it could seriously revolutionise the GPU industry too, specially on the energy efficiency front. Really impressive stuff.