A chip that can classify nearly two billion images per second
Using a deep neural network of optical waveguides, a new chip developed by Penn engineers—smaller than a square centimeter—can detect and classify an image in less than a nanosecond, all without the need for a separate processor or memory unit.