Deep Learning the Electromagnetic Properties of Metamaterials—A Comprehensive Review

Abstract

Deep neural networks (DNNs) are empirically derived systems that have transformed traditional research methods, and are driving scientific discovery. Artificial electromagnetic materials (AEMs)—including electromagnetic metamaterials, photonic crystals, and plasmonics—are research fields where DNN results valorize the data driven approach; especially in cases where conventional methods have failed. In view of the great potential of deep learning for the future of artificial electromagnetic materials research, the status of the field with a focus on recent advances, key limitations, and future directions is reviewed. Strategies, guidance, evaluation, and limits of using deep networks for both forward and inverse AEM problems are presented.

Prediction and Inverse Design of Structural Colors of Nanoparticle Systems via Deep Neural Network

“…Machine learning is a statistics technology that trains a machine by telling it what to do, and has proven to be particularly good at solving the problems of classification and regression [22,23]. In the field of nanophotonics, machine learning has made great progress in many applications such as pattern recognition, optical imaging, and structure design [24][25][26][27]. Most recently, machine learning techniques have been utilized to inversely design the structure and material to achieve the desired optical and color properties [28][29][30][31][32][33].…”

Simulation Intelligence: Towards a New Generation of Scientific Methods

“…A recent study in the scope of surrogate-based optimization is the neural-adjoint (NA) method [495] which directly searches the global space via gradient descent starting from random initialization, such that a large number of interactions are required to converge and its solutions are easily trapped in the local minima [496]. Although the neural-adjoint method boosts the performance by down selecting the top solutions from multiple starts, the computational cost is significantly higher, especially for high-dimensional problems [497].…”

SUTD-PRCM Dataset and Neural Architecture Search Approach for Complex Metasurface Design

“…These applications are made possible by the rapid advancement in micro-and even nano-fabrication technologies and computational modeling over the past decades. In the design of complex metasurfaces, machine learning (ML) methods like deep learning (DL) techniques has demonstrated unprecedented performance in providing rapid yet accurate prediction [26]. Particularly, DL technique has been mainly applied for forward modeling and inverse design generation [27][28][29][30].…”