An on-chip photonic deep neural network for image classification

(Source: Tima Miroshnichenko)   Researchers at the University of Pennsylvania have developed an optical chip that can process almost 1.7 billion images per second. It uses a neural network that processes visual information directly without memory read/writes. Despite advances in photonic computation, the lack of scalable on-chip optical non-linearity and the loss of photonic devices limit the scalability of optical deep networks. The Penn researchers integrated an end-to-end photonic deep neural network (PDNN) that they report can perform sub-nanosecond image classification—and it does it through direct processing of the optical waves impinging on the on-chip pixel array as they propagate