Ultrabroadband and band-selective thermal meta-emitters by machine learning

Area of Science:

• Nanophotonics
• Materials Science
• Machine Learning

Background:

• Thermal nanophotonics is crucial for energy and information technologies.
• Precise spectral engineering of thermal emitters is limited by trial-and-error methods.
• Machine learning shows promise in nanophotonic material design.

Purpose of the Study:

• To develop a general machine learning methodology for designing high-performance nanophotonic emitters.
• To achieve ultrabroadband control and precise band selectivity in thermal emitters.
• To overcome limitations of traditional design approaches, including predefined geometries and local optimization traps.

Main Methods:

• An unconventional machine learning paradigm for multiparameter optimization with sparse data.
• Inverse design of metastructure and material combinations for spectral tailoring.
• A three-plane modeling method for designing three-dimensional (3D) meta-emitters, moving beyond 2D structures.

Main Results:

• Designed a multitude of ultrabroadband and band-selective thermal meta-emitters.
• Demonstrated dual design capabilities: automated inverse design and 3D emitter design.
• Presented seven proof-of-concept meta-emitters with superior optical and radiative cooling performance.

Conclusions:

• The proposed framework enables global optimization through expanded geometric freedom and dimensionality.
• A generalizable fabrication framework for 3D nanophotonic materials was provided.
• The approach surpasses current state-of-the-art designs and offers a comprehensive materials database.