Optimized Design of Multi-layer Nano-photonic Structures for Selective Absorption Applications by Artificial Neural Netw
- PDF / 3,300,677 Bytes
- 7 Pages / 595.276 x 790.866 pts Page_size
- 23 Downloads / 177 Views
Optimized Design of Multi‑layer Nano‑photonic Structures for Selective Absorption Applications by Artificial Neural Networks Meijie Chen1 · Dan Pang1 · Xingyu Chen1 · Hongjie Yan1 · Ping Zhou1 Received: 18 July 2020 / Accepted: 3 November 2020 © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Data-driven method based on the machine learning provides an efficient tool to optimize nano-structure for the target spectrum applications. Artificial neural network (NN) can be trained with limited samples to approximate the complex physical simulation with the high accuracy. In this work, we use the artificial neural network to approximate the absorbance spectra of multi-layer structure. Results show that the forward design of NN can predicate the spectrum accurately based on the mean square error as the cost function. For inverse design and optimized design processes, the predicted results show great based on the physically realized spectrum. However, the ideal full absorption spectrum or selective absorption spectrum is usually needed for different photonic applications, which is not suitable based on the reverse network. An optimized network based on the forward and inverse network was designed to predicate the ideal selective or full absorption spectra, which provides a way to optimize complex photonic structures for different applications. Keywords Neural network · Optimized design · Nano-structure · Selective spectrum
Introduction Nano-structures with the selective or full absorption performance are widely used in solar thermal conversion [1], photovoltaic, and other photonic devices [2, 3], which increasingly relies on the complex nano-structure design to achieve the better performance at target wavelengths. With the increasing structural complexity, the design process is difficult due to large considered parameters. A typical design process is based on the forward calculation based on the Maxwell equation and it usually starts with the random structure with the calculation of electromagnetic response and then changes the structure size to update the design by comparing with the target response. There are many different ways to solve this process including genetic algorithm [4], adjoint method [5], and optimization of the specific geometric parameters [6]. For the complex nano-structure, it would take lots of computation power and time based on the genetic algorithm and optimization of specific geometric parameters by searching the parameters step by step based on the * Meijie Chen [email protected] 1
School of Energy Science and Engineering, Central South University, Changsha 410083, China
calculation of Maxwell equation. On the other hand, the adjoint method is difficult to set up with the poor effectiveness since it requires a deep understanding in the photonic. Data-driven method based on the machine learning provides an efficient tool to optimize the nano-structure for the target spectrum. Artificial neural network (NN) can be trained with the limited samples to approxim
Data Loading...
