中文 |

Machine Learning Enables Highly Efficient Light Absorption in Tungsten Metasurfaces

Author: WANG Yue |

In a study published in Applied Thermal Engineering, a research team from the Changchun Institute of Optics, Fine Mechanics and Physics (CIOMP) of the Chinese Academy of Sciences (CAS), together with collaborators, developed a machine learning-assisted design strategy for a highly selective tungsten-based metasurface absorber and emitter. By combining artificial intelligence with electromagnetic simulations, the researchers designed a nanostructured surface capable of efficiently absorbing visible and near-infrared light while controlling thermal emission, offering a promising approach to improving solar thermophotovoltaic energy conversion.

Efficient solar energy conversion requires materials that can capture sunlight across a broad wavelength range while minimizing unwanted energy losses. Solar thermophotovoltaic systems convert absorbed sunlight into heat and subsequently emit thermal radiation that can be converted into electricity by photovoltaic cells. However, conventional absorbers often suffer from limited spectral selectivity, sensitivity to incident angles, and reduced performance under demanding operating conditions. Designing nanostructures that simultaneously overcome these limitations remains challenging, particularly when extensive numerical simulations are required to optimize their optical properties.

To address these challenges, the researchers designed a tungsten-based metasurface consisting of a metal–insulator–metal structure with carefully arranged nanoscale resonators. Tungsten was selected for its favorable optical absorption characteristics and suitability for high-temperature applications. The researchers combined finite-difference time-domain simulations with machine learning to investigate how different structural parameters influenced light absorption. They compared six machine learning algorithms and identified XGBoost as the most effective model for predicting absorption performance and guiding structural optimization.

The researchers further investigated the physical mechanisms responsible for the enhanced optical response. By analyzing electromagnetic field distributions and the contributions of different structural components, they found that the combined nanodisk and nanoring resonators supported strong optical interactions across a broad wavelength range. They also evaluated the influence of incident angle and polarization on absorption performance and examined the structure's potential as a selective thermal emitter for solar thermophotovoltaic applications.

The simulation results showed that the optimized metasurface achieved an average absorptivity of 98.3% across the 350–1000 nm wavelength range, with absorption reaching 99.8% in the 450–700 nm region. The structure also maintained over 90% absorption at incident angles up to 60° under both major polarization conditions. As a selective thermal emitter, it achieved a calculated spectral conversion efficiency of 99.4% at 2000 K, indicating strong potential for directing thermal radiation toward wavelengths useful for silicon photovoltaic cells.

This study provides a promising approach to improving solar energy conversion by combining machine learning with advanced metasurface design. The ability to efficiently capture sunlight and control thermal radiation could help improve the performance of solar thermophotovoltaic systems, making more effective use of solar energy. Beyond energy harvesting, this approach could also contribute to the development of advanced photodetectors, optical sensors, and other devices that require precise control of light. The research highlights the potential of artificial intelligence in accelerating the development of high-performance optical materials and opens new possibilities for more efficient and sustainable energy technologies.


Contact

YU Weili

Changchun Institute of Optics, Fine Mechanics and Physics

E-mail:

cas_logo.png



       Copyright @ 吉ICP备06002510号 2007 CIOMP130033