Deep Learning and Natural Language Processing for Accelerated Inverse Design of Optical Metamaterials
Optical metamaterial device design has enjoyed a long track of success over the past 50 years leading to the manipulation of light over a wide range of wavelengths spanning the ultraviolet to the far infrared. The manipulation of light over such wavelengths has already led to many technological advancements such as the design of selective radiative absorbers for solar energy, daytime passive cooling using deep space, and optical invisibility cloaks for defense applications. Further disruptive advancements in energy, defense, computing, and biomedical fields could be enabled or enhanced by future optical metamaterial devices. These technologies could lead to increased energy efficiency and hence reduced national primary energy consumption, cheap long duration energy storage, and next generation solid-state heat engines. But historically the methods to invent and develop all of these devices have been time- consuming and based mostly on intuition and iteration. Finding an optimal design can take years. In this project we developed a machine learning-based algorithm capable of automatically generating device designs to produce desired optical properties, reducing the design cycle life in certain situations to be almost instantaneous.