Optimizing broadband metamaterial absorber using deep reinforcement learning

Abstract Optimization of the geometry of broadband metamaterial absorbers is crucial for improving the performance of optoelectronic devices. However, a large number of geometric parameters should be considered to achieve broad absorption, which is time-consuming. Herein, we propose a rapid and simple method for optimizing metamaterial absorbers dedicated to thermal radiation absorption using deep reinforcement learning. Deep reinforcement learning generated an ideal geometry for a broadband metamaterial absorber after 4 h, demonstrating the effectiveness of this technique for the rapid and effective optimization of metamaterial absorbers.