Training large-scale optoelectronic neural networks with dual-neuron optical-artificial learning DOI Creative Commons
Xiaoyun Yuan, Yong Wang, Zhihao Xu

et al.

Nature Communications, Journal Year: 2023, Volume and Issue: 14(1)

Published: Nov. 4, 2023

Optoelectronic neural networks (ONN) are a promising avenue in AI computing due to their potential for parallelization, power efficiency, and speed. Diffractive networks, which process information by propagating encoded light through trained optical elements, have garnered interest. However, training large-scale diffractive faces challenges the computational memory costs of diffraction modeling. Here, we present DANTE, dual-neuron optical-artificial learning architecture. Optical neurons model diffraction, while artificial approximate intensive optical-diffraction computations with lightweight functions. DANTE also improves convergence employing iterative global artificial-learning steps local optical-learning steps. In simulation experiments, successfully trains ONNs 150 million on ImageNet, previously unattainable, accelerates speeds significantly CIFAR-10 benchmark compared single-neuron learning. physical develop two-layer ONN system based can effectively extract features improve classification natural images.

Language: Английский

Two-photon polymerization lithography for imaging optics DOI Creative Commons
Hao Wang, Chengfeng Pan, Chi Li

et al.

International Journal of Extreme Manufacturing, Journal Year: 2024, Volume and Issue: 6(4), P. 042002 - 042002

Published: March 20, 2024

Abstract Optical imaging systems have greatly extended human visual capabilities, enabling the observation and understanding of diverse phenomena. Imaging technologies span a broad spectrum wavelengths from x-ray to radio frequencies impact research activities our daily lives. Traditional glass lenses are fabricated through series complex processes, while polymers offer versatility ease production. However, modern applications often require lens assemblies, driving need for miniaturization advanced designs with micro- nanoscale features surpass capabilities traditional fabrication methods. Three-dimensional (3D) printing, or additive manufacturing, presents solution these challenges benefits rapid prototyping, customized geometries, efficient production, particularly suited miniaturized optical devices. Various 3D printing methods demonstrated advantages over counterparts, yet remain in achieving resolutions. Two-photon polymerization lithography (TPL), technique, enables intricate structures beyond diffraction limit via nonlinear process two-photon absorption within liquid resin. It offers unprecedented abilities, e.g. alignment-free fabrication, prototyping almost arbitrary nanostructures. In this review, we emphasize importance criteria performance evaluation devices, discuss material properties relevant TPL, techniques, highlight application TPL imaging. As first panoramic review on topic, it will equip researchers foundational knowledge recent advancements optics, promoting deeper field. By leveraging its high-resolution capability, extensive range, true processing, alongside advances materials, design, envisage disruptive solutions current promising incorporation future applications.

Language: Английский

Citations

30
Optical neural networks: progress and challenges DOI Creative Commons

Tingzhao Fu,

Jianfa Zhang,

Run Cang Sun

et al.

Light Science & Applications, Journal Year: 2024, Volume and Issue: 13(1)

Published: Sept. 20, 2024

Language: Английский

Citations

24
Orbital angular momentum-mediated machine learning for high-accuracy mode-feature encoding DOI Creative Commons
Xinyuan Fang, Xiaonan Hu, Baoli Li

et al.

Light Science & Applications, Journal Year: 2024, Volume and Issue: 13(1)

Published: Feb. 14, 2024

Machine learning with optical neural networks has featured unique advantages of the information processing including high speed, ultrawide bandwidths and low energy consumption because dimensions (time, space, wavelength, polarization) could be utilized to increase degree freedom. However, due lack capability extract features in orbital angular momentum (OAM) domain, theoretically unlimited OAM states have never been exploited represent signal input/output nodes network model. Here, we demonstrate OAM-mediated machine an all-optical convolutional (CNN) based on Laguerre-Gaussian (LG) beam modes diverse diffraction losses. The proposed CNN architecture is composed a trainable mode-dispersion impulse as kernel for feature extraction, deep-learning diffractive layers classifier. resultant selectivity can applied mode-feature encoding, leading accuracy 97.2% MNIST database through detecting weighting coefficients encoded modes, well resistance eavesdropping point-to-point free-space transmission. Moreover, extending target into multiplexed states, realize dimension reduction anomaly detection 85%. Our work provides deep insight mechanism spatial basis, which further improve performances various machine-vision tasks by constructing unsupervised learning-based auto-encoder.

Language: Английский

Citations

20
All-optical image denoising using a diffractive visual processor DOI Creative Commons
Çağatay Işıl, Tianyi Gan,

F. Onuralp Ardic

et al.

Light Science & Applications, Journal Year: 2024, Volume and Issue: 13(1)

Published: Feb. 4, 2024

Abstract Image denoising, one of the essential inverse problems, targets to remove noise/artifacts from input images. In general, digital image denoising algorithms, executed on computers, present latency due several iterations implemented in, e.g., graphics processing units (GPUs). While deep learning-enabled methods can operate non-iteratively, they also introduce and impose a significant computational burden, leading increased power consumption. Here, we an analog diffractive denoiser all-optically non-iteratively clean various forms noise artifacts images – at speed light propagation within thin visual processor that axially spans <250 × λ, where λ is wavelength light. This all-optical comprises passive transmissive layers optimized using learning physically scatter optical modes represent features, causing them miss output Field-of-View (FoV) while retaining object features interest. Our results show these denoisers efficiently salt pepper rendering-related spatial phase or intensity achieving efficiency ~30–40%. We experimentally demonstrated effectiveness this architecture 3D-printed operating terahertz spectrum. Owing their speed, power-efficiency, minimal overhead, be transformative for display projection systems, including, holographic displays.

Language: Английский

Citations

17
Unidirectional imaging using deep learning–designed materials DOI Creative Commons
Jingxi Li, Tianyi Gan, Yifan Zhao

et al.

Science Advances, Journal Year: 2023, Volume and Issue: 9(17)

Published: April 28, 2023

A unidirectional imager would only permit image formation along one direction, from an input field-of-view (FOV) to output FOV B, and in the reverse path, B → A, be blocked. We report first demonstration of imagers, presenting polarization-insensitive broadband imaging based on successive diffractive layers that are linear isotropic. After their deep learning-based training, resulting fabricated form a imager. Although trained using monochromatic illumination, maintains its functionality over large spectral band works under illumination. experimentally validated this terahertz radiation, well matching our numerical results. also created wavelength-selective imager, where two operations, directions, multiplexed through different illumination wavelengths. Diffractive structured materials will have numerous applications in, e.g., security, defense, telecommunications, privacy protection.

Language: Английский

Citations

34
Software-defined nanophotonic devices and systems empowered by machine learning DOI Creative Commons
Yihao Xu, Bo Xiong, Wei Ma

et al.

Progress in Quantum Electronics, Journal Year: 2023, Volume and Issue: 89, P. 100469 - 100469

Published: April 4, 2023

Language: Английский

Citations

23
Human emotion recognition with a microcomb-enabled integrated optical neural network DOI Creative Commons
Junwei Cheng,

Yanzhao Xie,

Yu Liu

et al.

Nanophotonics, Journal Year: 2023, Volume and Issue: 12(20), P. 3883 - 3894

Published: Sept. 29, 2023

Abstract State-of-the-art deep learning models can converse and interact with humans by understanding their emotions, but the exponential increase in model parameters has triggered an unprecedented demand for fast low-power computing. Here, we propose a microcomb-enabled integrated optical neural network (MIONN) to perform intelligent task of human emotion recognition at speed light low power consumption. Large-scale tensor data be independently encoded dozens frequency channels generated on-chip microcomb computed parallel when flowing through microring weight bank. To validate proposed MIONN, fabricated proof-of-concept chips prototype photonic-electronic artificial intelligence (AI) computing engine potential throughput up 51.2 TOPS (tera-operations per second). We developed automatic feedback control procedures ensure stability 8 bits weighting precision MIONN. The MIONN successfully recognized six basic achieved 78.5 % accuracy on blind test set. provides high-speed energy-efficient neuromorphic hardware emotional interaction capabilities.

Language: Английский

Citations

23
A perspective on the artificial intelligence’s transformative role in advancing diffractive optics DOI Creative Commons
Svetlana N. Khonina, Nikolay L. Kazanskiy, Albert Efimov

et al.

iScience, Journal Year: 2024, Volume and Issue: 27(7), P. 110270 - 110270

Published: June 18, 2024

Artificial intelligence (AI) is transforming diffractive optics development through its advanced capabilities in design optimization, pattern generation, fabrication enhancement, performance forecasting, and customization. Utilizing AI algorithms like machine learning, generative models, transformers, researchers can analyze extensive datasets to refine the of optical elements (DOEs) tailored specific applications requirements. AI-driven generation methods enable creation intricate efficient structures that manipulate light with exceptional precision. Furthermore, optimizes manufacturing processes by fine-tuning parameters, resulting higher quality productivity. models also simulate behavior, accelerating iterations facilitating rapid prototyping. This integration into holds tremendous potential revolutionize technology across diverse sectors, spanning from imaging sensing telecommunications beyond.

Language: Английский

Citations

15
Advancements and Applications of Diffractive Optical Elements in Contemporary Optics: A Comprehensive Overview DOI
Svetlana N. Khonina, Nikolay L. Kazanskiy, Р. В. Скиданов

et al.

Advanced Materials Technologies, Journal Year: 2024, Volume and Issue: unknown

Published: Aug. 31, 2024

Abstract Diffractive optical elements (DOEs) represent a revolutionary advancement in modern optics, offering unparalleled versatility and efficiency various applications. Their significance lies their ability to manipulate light waves with intricate patterns, enabling functionalities beyond what traditional refractive optics can achieve. DOEs find widespread use fields such as laser beam shaping, holography, communications, imaging systems. By precisely controlling the phase amplitude of light, generate complex structures, correct aberrations, enhance performance Moreover, compact size, lightweight nature, potential for mass production make them indispensable designing efficient devices diverse industrial scientific From improving systems innovative display technologies, continue drive advancements promising even more exciting possibilities future. In this review, critical importance is illuminated explore profound implications contemporary era.

Language: Английский

Citations

11
Metasurface Enabled Multi‐Target and Multi‐Wavelength Diffraction Neural Networks DOI

Haoxiang Chi,

Xiaofei Zang, Teng Zhang

et al.

Laser & Photonics Review, Journal Year: 2024, Volume and Issue: unknown

Published: Sept. 3, 2024

Abstract Benefiting from low power consumption and high processing speed, there is a growing interest in diffraction neural networks (DNNs), which are typically showcased with 3D printing devices, leading to large volumes, costs, levels of integration. Metasurfaces can desirably manipulate wavefronts electromagnetic waves, providing compact platform for mimicking DNNs novel functions. Although multi‐wavelength multi‐target recognition provides richer more detailed understanding complex environments, existing architectures primarily trained classify single target at specific wavelength. A metasurface approach proposed design multiplexed that multiple targets spatial sequences across various wavelengths channels. To realize multi‐task processing, the dielectric designed based on phase wavelength multiplexing, integrate different tasks such as operating distinct classifying diverse targets. The efficacy this method exemplified through numerical simulation experimental demonstration recognizing two wavelengths, wavelength, dual wavelengths. This enables DNNs, opening new window develop massively parallel versatile artificial intelligence systems.

Language: Английский

Citations

9