Inverse design for integrated photonics using deep neural network

Focusing on nanophotonic power splitters, we show that a generative neural network can design a series of devices that achieve nearly arbitrary target performance, with an excellent capability to generalize training data produced by the adjoint method.