Inverse design of photonic topological state via machine learning

Abstract

The photonics topological state plays an important role in recent optical physics and has led to devices with robust properties. However, the design of optical structures with the target topological states is a challenge for current research. Here, we propose an approach to achieve this goal by exploiting machine learning technologies. In our work, we focus on Zak phases, which are the topological properties of one-dimensional photonics crystals. After learning the principle between the geometrical parameters and the Zak phases, the neural network can obtain the appropriate structures of photonics crystals by applying the objective Zak phase properties. Our work would give more insights into the application of machine learning on the inverse design of the complex material properties and could be extended to other fields, i.e., advanced phononics devices.