Software-defined nanophotonic devices and systems empowered by machine learning

Abstract

Nanophotonic devices, such as metasurfaces and silicon photonic components, have been progressively demonstrated to be efficient and versatile alternatives to their bulky counterparts, enabling compact and light-weight systems for the application of imaging, sensing, communication and computing. The tremendous advances in machine learning provide new design methods, metrology and functionalities for nanophotonic devices and systems. Specifically, machine learning has fundamentally changed automatic design, measurement and result processing of highly application-specific nanophotonic systems without the need of extensive expert experience. This trend can be well described by the popular concept of “software-defined” infrastructure in information technology, which can decouple specific hardware from end users by virtualizing physical components using software interfaces, making the entire system faster, more flexible and more scalable. In this review, we introduce the concept of software-defined nanophotonics and summarize the interdisciplinary research that bridges nanophotonics and intelligence algorithms, especially machine learning algorithms, in the device design, measurement and system setup. The review is organized in an application-oriented manner, showing how the software-defined scheme is utilized in solving both forward and inverse problems for various nanophotonic devices and systems.