Deep Learning-Enabled On-Demand Inverse Design of Daytime Radiative Coolers with Dual-Polarization Color Control

Abstract

Daytime radiative cooling is an environmentally friendly approach that reaches below ambient temperature by simultaneously reflecting solar irradiance and dissipating heat to the cold universe through the atmospheric window (AW). Conventional daytime radiative coolers typically exhibit white appearances due to their broadband solar reflection, limiting their aesthetic applicability. Here, we propose colored daytime radiative coolers (CRCs) comprising an optimized top thermal emitter and an inverse designed metal–insulator–metal (MIM) metasurface reflector. The multilayer top emitter is optimized using a genetic algorithm, achieving a high mean emissivity of 98.9% within the AW. The bottom elliptic MIM metasurface is inverse designed using a deep neural network to penalize solar absorption while enabling on-demand structural color generation with polarization control. By appropriately specifying the color targets, the framework enables the inverse design of both conventional CRCs operating under unpolarized illumination and polarization-dependent CRCs that exhibit distinct visible appearances under orthogonal x- and y-polarized illumination, while remaining fully functional under unpolarized light within a single device geometry, enabling color modulation without physical reconfiguration. The proposed CRCs achieve cooling powers of up to 25 W m–2 and demonstrate positive cooling performance across diverse global climates, offering a versatile solution for visually appealing and energy-efficient thermal management.