Photonics approaches to the implementation of neuromorphic computing

physical limitations on the operation speed of electronic devices has motivated the search for alternative ways to process information. The past few years have seen the development of neuromorphic photonics-a branch of photonics where the physics of optical and optoelectronic devices is combined with mathematical algorithms of artificial neural networks. Such a symbiosis allows certain classes of computation prob,,lems, including some involving artificial intelligence, to be solved with greater speed and higher energy efficiency than can be reached with electronic devices based on the von Neumann architecture. We review optical analog computing, photonic neural networks, and methods of matrix multiplication by optical means, and discuss the advantages and disadvantages of existing approaches.