Session: 17-15-01: Society-Wide Micro/Nano Poster Forum

Paper Number: 99183

99183 - Design Transparent Radiative Cooler Using Machine Learning and Quantum Computing 

Passive radiative cooler, which allows the radiation emission through an atmospheric window (8 < λ < 13 μm), has attracted much attention as a solution to the climate change challenges, owing to their potential energy- and refrigerant-free cooling capability. However, most passive radiative coolers have high reflection to the whole solar spectrum (0.3 < λ < 2.5 μm), making them opaque. Hence, these reflective coolers cannot be used for window applications. Considering that building or automobile windows lose a significant amount of cooling energy, transparent radiative coolers that have high transmission to visible light but high emission to the atmospheric window have been developed for windows. These coolers generally utilize photonic structures to possess high transmission in visible regime or low transmission in non-visible regime. However, it is challenging to design photonic structures having both optical properties (i.e., high visible light transmission and low non-visible light transmission).

In this work, an active learning design framework including a machine-learning model and quantum annealing is used to efficiently design a transparent radiative cooler with an optimal photonic structure and an infrared emissive layer. A photonic structure is divided into n-layers, where each layer can be one of four materials (silicon oxide (SiO₂), silicon nitride (Si₃N₄), aluminum oxide (Al₂O₃), and titanium dioxide (TiO₂)), making the layer configurations expressible by binary vectors (i.e., [00], [01], [10], and [11] for SiO₂, Si₃N₄, Al₂O₃, and TiO₂, respectively). We use transfer matrix method to calculate the optical properties of given binary vectors, which is then used to train a factorization machine (FM) model. The FM learns the Hamiltonian matrix for the quantum annealing (QA), and then the QA predicts the optimal binary vector. The framework can efficiently find optimal structure although there are an astronomic number of design options according to the number of layers. For example, a 12-layered photonic structures have 4¹² (= 16,777,216) possible configurations. The optimally designed cooler has high visible light transmission and low ultraviolet/near-infrared light transmission to maximize cooling performance and high emission in the atmospheric window range. We experimentally measure the optical properties of the designed cooler and demonstrate its passive cooling ability under steady and dynamic environments. In addition, we calculate the energy saving for cooling when using the designed cooler instead of normal windows, and the results show that the cooler has a potential energy saving of ~ 86.3 MJ/m² annually in hot regions. We believe that the designed transparent radiative cooler can be applied for cooling windows and the active learning design framework can be efficiently used for functional material design in general.

Presenting Author: Seongmin Kim University of Notre Dame

Presenting Author Biography: Seongmin Kim received his Ph.D. in Mechanical Engineering at Pohang University of Science<br/>and Technology in 2021. He is currently a postdoctoral researcher in the Aerospace and<br/>Mechanical Engineering Department at the University of Notre Dame. His research interests<br/>include the fabrication of micro/nanostructured functional materials, development of a design<br/>platform for nano materials using machine learning and quantum computing, and their<br/>applications.

Authors:

Seongmin Kim University of Notre Dame
Seunghyun Moon University of Notre Dame
Eungkyu Lee Kyung Hee University
Tengfei Luo University of Notre Dame