Article
Nanoscience & Nanotechnology
Joeri Lenaerts, Hannah Pinson, Vincent Ginis
Summary: Machine learning is proposed for the inverse design of complex nanophotonic components, training neural networks to map input parameters to optical system features. By utilizing Fourier transform in the optimization process, the efficiency of the method is significantly improved. This approach successfully retrieves optimal design parameters for Fabry-Perot resonators and Bragg reflectors, demonstrating potential for more complex nanophotonic components.
Article
Materials Science, Multidisciplinary
Tong Wang, He-Ming Huang, Xiao-Xue Wang, Xin Guo
Summary: An artificial olfactory inference system based on memristive devices has been developed to classify four gases with 10 different concentrations, achieving a high accuracy of 95%. Three strategies are applied to reduce the extracted features from the reservoir computing system in order to reduce device number and power consumption.
Article
Engineering, Electrical & Electronic
Aashu Jha, Chaoran Huang, Thomas Ferreira deLima, Hsuan-Tung Peng, Bhavin Shastri, Paul R. Prucnal
Summary: The demands for bandwidth and energy in neural networks have sparked interest in developing novel neuromorphic hardware, including photonic integrated circuits. However, the channel count of photonic systems is limited by the devices within, affecting the overall throughput and feasibility of high-dimensional input applications. Experimental demonstrations show that photonic crystal nanobeam based synapses can overcome this limitation and increase data throughput, enabling applications such as natural language processing and high-resolution image processing. Additionally, the smaller physical footprint and higher energy efficiency of these synapses offer a path towards realizing highly scalable photonic neural networks.
IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS
(2022)
Article
Engineering, Electrical & Electronic
Keisuke Kojima, Mohammad H. Tahersima, Toshiaki Koike-Akino, Devesh K. Jha, Yingheng Tang, Ye Wang, Kieran Parsons
Summary: This article explores three models for designing nanophonic power splitters using deep neural networks, including a forward regression model, an inverse regression model, and a generative network. These models demonstrate how deep learning can be applied to optimize the design of nanophotonic devices.
JOURNAL OF LIGHTWAVE TECHNOLOGY
(2021)
Article
Nanoscience & Nanotechnology
Gordon H. Y. Li, Ryoto Sekine, Rajveer Nehra, Robert M. Gray, Luis Ledezma, Qiushi Guo, Alireza Marandi
Summary: The introduction of energy-efficient hardware accelerators has become necessary in recent years due to the computational demands of deep learning applications. Optical neural networks have shown promise as a potential solution, but their progress has been limited by a lack of energy-efficient nonlinear optical functions. This study demonstrates an all-optical Rectified Linear Unit (ReLU) using a periodically-poled thin-film lithium niobate nanophotonic waveguide, achieving ultra-low energies and near-instantaneous operation. This provides a practical path towards all-optical and energy-efficient nanophotonic deep learning.
Article
Computer Science, Artificial Intelligence
Alexander J. Dyer, Lewis D. Griffin
Summary: Inferring the connectivity of biological neural networks from neural activation data is challenging, but studying the analogous problem in artificial neural networks can provide insights into the biological case. This study focuses on assigning artificial neurons to locations in the LeNet image classifier. A supervised learning approach based on features derived from the activation correlation matrix is evaluated. The experiments suggest that an image dataset needs to fully activate the network and have minimal confounding correlations for accurate localization, and perfect assignment can be achieved by combining features from multiple image datasets.
Article
Mechanics
Adam Subel, Ashesh Chattopadhyay, Yifei Guan, Pedram Hassanzadeh
Summary: Recent research has shown that deep neural networks trained using properly pre-conditioned data can generate stable and accurate a posteriori LES models, and transfer learning can enable accurate and stable generalization to flows with higher Reynolds numbers.
Review
Chemistry, Multidisciplinary
Lauren A. Warning, Ali Rafiei Miandashti, Lauren A. McCarthy, Qingfeng Zhang, Christy F. Landes, Stephan Link
Summary: Chiral nanophotonic materials are promising for biosensing applications due to their ability to focus light into nanometer dimensions, increasing sensitivity. Recent advances in nanomaterial-enhanced chirality sensing have achieved detection limits as low as attomolar concentrations, showing potential for pharmaceutical, forensic, and medical applications requiring high sensitivity. The review covers the development and application of chiral nanomaterials for detecting biomolecules, supramolecular structures, and environmental stimuli, as well as discussing future prospects for nanophotonic chiral systems.
Article
Engineering, Electrical & Electronic
Luis El Srouji, Yun-Jhu Lee, Mehmet Berkay On, Li Zhang, S. J. Ben Yoo
Summary: Spiking neural networks (SNN) provide a new computational paradigm capable of highly parallelized, real-time processing. Photonic devices and CMOS technology combine to design scalable SNN computing architectures, with optoelectronic circuits and learning algorithms achieving promising results.
IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS
(2023)
Review
Chemistry, Multidisciplinary
Haoran Ren, Stefan A. Maier
Summary: Twisted light, a helical spatial mode carrying orbital angular momentum, has great potential in optical and quantum information applications. However, current experiments are hindered by bulky, expensive, and slow-response optical elements. Nanophotonics offers a solution by providing compact and multifunctional devices for generating and detecting twisted light modes.
ADVANCED MATERIALS
(2023)
Article
Optics
Bowen Ma, Junfeng Zhang, Weiwen Zou
Summary: Neural information representation and processing relies on neural populations instead of single neurons. A comb-based photonic neural population with superior scalability and nonlinear response can effectively utilize the ultra-broad bandwidth of photonics for parallel and nonlinear processing.
PHOTONICS RESEARCH
(2022)
Review
Chemistry, Multidisciplinary
Qin Chen, Xianghong Nan, Mingjie Chen, Dahui Pan, Xianguang Yang, Long Wen
Summary: Recent advances in low-dimensional materials and nanofabrication technologies have led to breakthroughs in nanophotonics field. Spectral routing and filtering schemes offer superior design freedom and efficient spectral utilization, critical for various applications.
ADVANCED MATERIALS
(2021)
Article
Chemistry, Multidisciplinary
Rui Wang, Baicheng Zhang, Guan Wang, Yachen Gao
Summary: Nanophotonics utilize the interaction between light and subwavelength structures to design nanophotonic devices and exhibit unique optical, electromagnetic, and acoustic properties. This paper proposes a method that combines finite-difference time-domain simulations and neural network training to quickly predict reflectance spectra of nanophotonic devices. Experimental results demonstrate the effectiveness of this approach, with most predictions maintaining the main trend observed in simulations.
Article
Engineering, Multidisciplinary
Vahid Keshavarzzadeh, Robert M. Kirby, Akil Narayan
Summary: Inverse problems are common in engineering simulations, and Bayesian inference is a predominant approach to infer unknown parameters. This paper presents a variational inference method that incorporates observation data and the gradient information of the forward map to invert unknown latent parameters. The method utilizes a trained neural network to generate samples for statistical calculations. The effectiveness of the method is demonstrated through examples, and future research directions are discussed.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2022)
Article
Nanoscience & Nanotechnology
George Giamougiannis, Apostolos Tsakyridis, Miltiadis Moralis-Pegios, Christos Pappas, Manos Kirtas, Nikolaos Passalis, David Lazovsky, Anastasios Tefas, Nikos Pleros
Summary: Analog photonic computing is a promising candidate for deep neural network linear operations due to its high bandwidth, low footprint, and low power consumption. However, the limited hardware size and bit precision of electro-optical components pose challenges in surpassing digital processors' performance. In this study, a speed-optimized dynamic precision neural network inference is proposed and experimentally demonstrated using tiled matrix multiplication on a low-radix silicon photonic processor. By dynamically adjusting the compute rates per neural layer, the proposed method achieves a 55% decrease in execution time compared to a fixed-precision scheme without sacrificing accuracy.
