New photonic chip recognizes 2 billion images per second

Scientists from the University of Pennsylvania have succeeded in making a chip, measuring just 9.3mm across, which is the first "deep neural network" implemented within an integrated scalar photonic device capable of recognizing an image. in less than 0.57 nanoseconds. Such a speed would make it possible to classify as much as 1.75 billion images per second.

This impressive result was achieved by combining photonics technology, which is increasingly popular in ultra-high frequency applications (thanks to the fact that components based on the do not suffer from the resistance and heat problems that plague traditional microelectronics), with neural networks. Scientists were the first to simulate neurons in an optical space, all to the advantage of speed and power consumption (compared to more common silicon-based alternatives such as CPU and GPU).



Photo Credit: Nature Journal