Discovering high-performance broadband and broad angle antireflection surfaces by machine learning

Abstract

Eliminating light reflection from the top glass sheet in optoelectronic applications is often desirable across a broad range of wavelengths and large variety of angles. In this paper, we report on a combined simulation and experimental study of single-layer films, nanowire arrays, and nanocone arrays to meet these antireflection (AR) needs. We demonstrate the application of Bayesian learning to the multiobjective optimization of these structures for broadband and broad angle AR and show the superior performance of Bayesian learning to genetic algorithms for optimization. Our simulations indicate that nanocone structures have the best AR performance of these three structures, and we additionally provide physical insight into the AR performance of different structures. Simulations suggest nanocone arrays are able to achieve a solar integrated normal and 65° incidence angle reflection of 0.15% and 1.25%, respectively. A simple and scalable maskless reactive ion etching process is used to create nanocone structures, and etched samples demonstrate a solar integrated normal and 65° reflection of 0.4% and 4.9%, respectively, at the front interface.

Photonics Empowered Passive Radiative Cooling

“…Although conventional structural design methods, including physics‐based approaches and numerical simulations, offer important guidelines, it is not trivial to find the right structures with ideal selective emission spectra. We envisage simple photonic materials and structures designed and optimized by advanced methods such as the inverse design methods, [ 64–75 ] which enable nonintuitive and irregularly shaped structures, outperforming physically or empirically designed structures. Particularly, the artificial neural networks, as one of the most powerful machine learning methods, have shown orders of magnitude faster and much accuracy in optimizing structures in high dimensional space.…”

“…[ 65 ] Another machine learning method, Bayesian learning, was demonstrated to be superior to the conventional genetic method in multiobjective optimization. [ 66 ] This is particularly useful because a highly efficient radiative cooling structure design involves both selective spectra and broad emission angle.…”

Challenges and Prospects of Bio-Inspired and Multifunctional Transparent Substrates and Barrier Layers for Optoelectronics

“…These structures function like a medium where the effective index of refraction can be tuned by adjusting the solid fraction. These structures perform comparably to a single layer film with the ideal index of refraction, 98 but this approach is often used because durable materials with an appropriately low index of refraction may not exist.…”

“…Recently, we used machine learning and optimization together with optical simulations to study the broadband and omnidirectional performance limits of three common antireflection structures including single layer films, nanowire arrays, and nanocone arrays . In this work, we calculated the integrated reflection over the solar spectrum by where R (λ) is the reflection spectrum, b s (λ) is the photon flux density of the AM1.5 global solar spectrum, and λ is the wavelength.…”

Hybrid design of spectral splitters and concentrators of light for solar cells using iterative search and neural networks

“…Higher-level information within a data set, which is presented as weights of the layers, is captured, and thereby, complex network relations between the input data and the output data can be understood. Deep learning proves its adaptability and multi-operational behaviour in various tasks especially in photonics in designing nanophotonic devices [30,31], anti-reflective surfaces [32], and photonics systems [33][34][35][36]. Also, physics-informed neural networks can provide a better understanding of the underlying physics in various tasks [37].…”