Determining optical constants of 2D materials with neural networks from multi-angle reflectometry data

Abstract

Synthetically generated multi-angle reflectometry data is used to train a neural network based learning system to estimate the refractive index of atomically thin layered materials in the visible part of the electromagnetic spectrum. Unlike previously developed regression based optical characterization methods, the prediction is achieved via classification by using the probabilities of each input element belonging to a label as weighting coefficients in a simple analytical formula. Various types of activation functions and gradient descent optimizers are tested to determine the optimum combination yielding the best performance. For the verification of the proposed method's accuracy, four different materials are studied. In all cases, the maximum error is calculated to be less than 0.3%. Considering the highly dispersive nature of the studied materials, this result is a substantial improvement in terms of accuracy and efficiency compared to traditional approaches.

Machine Learning Exercises on 1-D Electromagnetic Inversion

“…Second, we can build neuralnetworks with complex activation functions to improve the accuracy [29], but since they are computationally expensive, they might slow down the training process significantly, especially, if the network is implemented with a programming language which is not specifically designed for scientific computing, such as Python. Third, simple post-processing methods, such as taking the average of the predicted values in the last few epochs [4], and advanced methods like the neuralnetwork based image processing algorithm implemented in [14] can increase the accuracy by balancing over and underestimates and increasing the contrast among multiple objects, respectively. Another method for a slightly higher accuracy is using a non-uniform grid for the training.…”

“…Machine learning has become a popular subject in the computational electromagnetics (CEM) society as well. Researchers have proposed using machine learning to solve advanced CEM problems in device design [1]- [3], material characterization [4], geophysical prospecting [5], [6], and electromagnetic inversion [3], [5], [7]- [16], which attempts to estimate the distribution of physical properties in a domain of interest from antenna measurements collected outside of that domain. Since the inversion problems are nonlinear, nonunique, and ill-posed [17], [18], electromagnetic inversion has been one of the most challenging subjects studied by the CEM society over the past decades.…”

Attention mechanisms for broadband feature prediction for electromagnetic and photonic applications

“…We have witnessed a surge of interest in the use of artificial intelligence, specifically neural networks, to enhance the understanding and application of light-matter interactions. [1][2][3][4][5][6][7][8] Neural networks have shown promise in various photonics applications, including inverse photonic design, material and device characterization, optical sensing, image processing and classification, and optical communication.…”

Machine learning prediction of magnetic properties of Fe-based metallic glasses considering glass forming ability

“…Recently, machine learning (ML) has been applied to material science and became a promising tool for the development of new materials with desired properties. For example, based on extensive published research data, ML models have been trained to predict the GFA of metallic glasses [27][28][29][30], mechanical properties of metallic glasses [31,32], glass transition temperature [33], hardness of high-entropy alloys [34], optical constants of 2D materials [35], and morphology of nanoscale metal-organic frameworks [36]. In the past few decades, a large number of FeMGs have been discovered, so that sufficient data have been produced, and they could meet the data needs of ML.…”