All-Optical Phase Conjugation Using Diffractive Wavefront Processing DOI Creative Commons
Che‐Yung Shen, Jingxi Li,

Tianyi Gan

et al.

arXiv (Cornell University), Journal Year: 2023, Volume and Issue: unknown

Published: Jan. 1, 2023

Optical phase conjugation (OPC) is a nonlinear technique used for counteracting wavefront distortions, with various applications ranging from imaging to beam focusing. Here, we present the design of diffractive processor approximate all-optical operation input fields aberrations. Leveraging deep learning, set passive layers was optimized all-optically process an arbitrary phase-aberrated coherent field aperture, producing output distribution that conjugate wave. We experimentally validated efficacy this by 3D fabricating trained using learning and performing OPC on distortions never seen during its training. Employing terahertz radiation, our physical successfully performed task through shallow spatially-engineered volume axially spans tens wavelengths. In addition transmissive configuration, also created phase-conjugate mirror combining learning-optimized standard mirror. Given compact, scalable nature, can be diverse OPC-related applications, e.g., turbidity suppression aberration correction, adaptable different parts electromagnetic spectrum, especially those where cost-effective engineering solutions do not exist.

Language: Английский

Multiplane quantitative phase imaging using a wavelength-multiplexed diffractive optical processor DOI Creative Commons
Che‐Yung Shen, Jingxi Li, Yuhang Li

et al.

Advanced Photonics, Journal Year: 2024, Volume and Issue: 6(05)

Published: July 25, 2024

Quantitative phase imaging (QPI) is a label-free technique that provides optical path length information for transparent specimens, finding utility in biology, materials science, and engineering. Here, we present QPI of three-dimensional (3D) stack phase-only objects using wavelength-multiplexed diffractive processor. Utilizing multiple spatially engineered layers trained through deep learning, this processor can transform the distributions two-dimensional at various axial positions into intensity patterns, each encoded unique wavelength channel. These patterns are projected onto single field view output plane processor, enabling capture quantitative input located different planes an intensity-only image sensor. Based on numerical simulations, show our could simultaneously achieve all-optical across several distinct by scanning illumination wavelength. A proof-of-concept experiment with 3D-fabricated further validates approach, showcasing successful two terahertz spectrum. Diffractive network-based multiplane designs open up new avenues compact on-chip sensing devices.

Language: Английский

Citations

6
Physics and artificial intelligence: illuminating the future of optics and photonics DOI Creative Commons

Md Sadman Sakib Rahman,

Aydogan Özcan

Advanced Photonics, Journal Year: 2024, Volume and Issue: 6(05)

Published: Oct. 31, 2024

The 2024 Nobel Prize in Physics recognized John Hopfield and Geoffrey Hinton for their pioneering work on artificial neural networks, which profoundly impacted the physical sciences, particularly optics photonics. This perspective summarizes laureates' contributions, highlighting physics-based principles inspiration behind development of modern intelligence (AI) also outlining some emerging major advances achieved photonics enabled by AI.

Language: Английский

Citations

4
Vector vortex beams sorting of 120 modes in visible spectrum DOI Creative Commons
Qi Jia, Yanxia Zhang,

Bojian Shi

et al.

Nanophotonics, Journal Year: 2023, Volume and Issue: 12(20), P. 3955 - 3962

Published: Oct. 1, 2023

Abstract Polarization ( P ), angular index l and radius p ) are three independent degrees of freedom (DoFs) vector vortex beams, which have found extensive applications in various domains. While efficient sorting a single DoF has been achieved successfully, simultaneous all these DoFs compact manner remains challenge. In this study, we propose beam sorter that simultaneously handles the using diffractive deep neural network (D 2 NN), demonstrate robust 120 Laguerre–Gaussian (LG) modes experimentally visible spectrum. Our proposed underscores considerable potential D NN optical field manipulation promises to enhance diverse beams.

Language: Английский

Citations

10
All‐Optical Autoencoder Machine Learning Framework Using Linear Diffractive Processors DOI

Peijie Feng,

Yong Tan,

Mingzhe Chong

et al.

Laser & Photonics Review, Journal Year: 2025, Volume and Issue: unknown

Published: April 13, 2025

Abstract Diffractive deep neural network (D 2 NN), known for its high speed and strong parallelism, is applied across various fields, including pattern recognition, image processing, transmission. However, existing architectures primarily focus on data representation within the original domain, with limited exploration of latent space, thereby restricting information mining capabilities multifunctional integration D NNs. Here, an all‐optical autoencoder (OAE) framework proposed that linearly encodes input wavefield into a prior shape distribution in diffractive space (DLS) decodes encoded back to wavefield. By leveraging bidirectional multiplexing property NN, OAE modelsfunction as encoders one direction decoders opposite direction. The models are three areas: denoising, noise‐resistant reconfigurable classification, generation. Proof‐of‐concept experiments conducted validate numerical simulations. exploits potential representations, enabling single set processors simultaneously achieve reconstruction, representation, This work not only offers fresh insights design optical generative but also paves way developing multifunctional, highly integrated, general intelligent systems.

Language: Английский

Citations

0
Optoelectronic generative adversarial networks DOI Creative Commons
Jumin Qiu, Ganqing Lu, Tingting Liu

et al.

Communications Physics, Journal Year: 2025, Volume and Issue: 8(1)

Published: April 15, 2025

Language: Английский

Citations

0
Integrated input system for partially coherent light control in optical neural networks for remote sensing DOI
Hongmin Li, Xun Liu, Haonan Xu

et al.

Optics & Laser Technology, Journal Year: 2025, Volume and Issue: 189, P. 113055 - 113055

Published: April 28, 2025

Language: Английский

Citations

0
Polarization-selective unidirectional and bidirectional diffractive neural networks for information security and sharing DOI Creative Commons
Ziqing Guo, Zhiyu Tan, Xiaofei Zang

et al.

Nature Communications, Journal Year: 2025, Volume and Issue: 16(1)

Published: May 14, 2025

Information security aims to protect confidentiality and prevent information leakage, which inherently conflicts with the goal of sharing. Balancing these competing requirements is especially challenging in all-optical systems, where efficient data transmission rigorous are essential. Here we propose experimentally demonstrate a metasurface-based approach that integrates phase manipulation, polarization conversion, as well direction- polarization-selective functionalities into diffractive neural networks (DNNs). This enables polarization-controllable switch between unidirectional bidirectional DNNs, thus simultaneously realizing A cascaded terahertz metasurface comprising quarter-wave plates metallic gratings designed function unidirectional-bidirectional classifier imager. By introducing half-wave cascade metasurface, achieve polarization-controlled transition unidirectional-bidirectional-unidirectional modes for classification imaging. Furthermore, high-security exchange framework based on DNNs. The proposed DNNs polarization-switchable unidirectional/bidirectional capabilities offer significant potential privacy protection, encryption, communications, exchange.

Language: Английский

Citations

0
Advances in Mask-Modulated Lensless Imaging DOI Open Access
Yangyundou Wang,

Zhengjie Duan

Electronics, Journal Year: 2024, Volume and Issue: 13(3), P. 617 - 617

Published: Feb. 1, 2024

Lensless imaging allows for designing systems that are free from the constraints of traditional architectures. As a broadly investigated technique, mask-modulated lensless encodes light signals via mask plate integrated with image sensor, which is more compacted, scalability and compressive abilities. Here, we review latest advancements in imaging, reconstruction algorithms, related techniques, future directions applications.

Language: Английский

Citations

3
Integration of Programmable Diffraction with Digital Neural Networks DOI Creative Commons

Md Sadman Sakib Rahman,

Aydogan Özcan

ACS Photonics, Journal Year: 2024, Volume and Issue: 11(8), P. 2906 - 2922

Published: Aug. 12, 2024

Optical imaging and sensing systems based on diffractive elements have seen massive advances over the last several decades. Earlier generations of optical processors were, in general, designed to deliver information an independent system that was separately optimized, primarily driven by human vision or perception. With recent deep learning digital neural networks, there been efforts establish are jointly optimized with networks serving as their back-end. These hybrid (optical + digital) a new "diffractive language" between input electromagnetic waves carry analog process digitized at back-end, providing best both worlds. Such designs can spatially temporally coherent, partially incoherent waves, universal coverage for any varying set point spread functions be given task, executed collaboration networks. In this Perspective, we highlight utility exciting engineered programmed diffraction diverse range applications. We survey some major innovations enabled push–pull relationship analogue wave processing also covering significant benefits could reaped through synergy these two complementary paradigms.

Language: Английский

Citations

3
Review of diffractive deep neural networks DOI
Yichen Sun, Mingli Dong, Mingxin Yu

et al.

Journal of the Optical Society of America B, Journal Year: 2023, Volume and Issue: 40(11), P. 2951 - 2951

Published: Sept. 27, 2023

In 2018, a UCLA research group published an important paper on optical neural network (ONN) in the journal Science . It developed world’s first all-optical diffraction deep (DNN) system, which can perform MNIST dataset classification tasks at near-light-speed. To be specific, adopted terahertz light source as input, established diffractive DNN (D 2 NN) model using Rayleigh-Sommerfeld theory, optimized parameters stochastic gradient descent algorithm, and then used 3D printing technology to make grating built D NN system. This opened new ONN direction. Here, we review analyze development history basic theory of artificial networks (ANNs) ONNs. Second, elaborate holographic elements (HOEs) interconnected by free space describe NN. Then cover nonlinear application scenarios for Finally, future directions challenges are briefly discussed. Hopefully, our work provide support help researchers who study future.

Language: Английский

Citations

7