Photonic multiplexing techniques for neuromorphic computing

Nanophotonics, Journal Year: 2023, Volume and Issue: 12(5), P. 795 - 817

Published: Jan. 9, 2023

Abstract The simultaneous advances in artificial neural networks and photonic integration technologies have spurred extensive research optical computing (ONNs). potential to simultaneously exploit multiple physical dimensions of time, wavelength space give ONNs the ability achieve operations with high parallelism large-data throughput. Different multiplexing techniques based on these degrees freedom enabled large-scale interconnectivity linear functions. Here, we review recent different approaches multiplexing, present our outlook key needed further advance multiplexing/hybrid-multiplexing ONNs.

Language: Английский

---

High-throughput terahertz imaging: progress and challenges

Light Science & Applications, Journal Year: 2023, Volume and Issue: 12(1)

Published: Sept. 15, 2023

Abstract Many exciting terahertz imaging applications, such as non-destructive evaluation, biomedical diagnosis, and security screening, have been historically limited in practical usage due to the raster-scanning requirement of systems, which impose very low speeds. However, recent advancements systems greatly increased throughput brought promising potential radiation from research laboratories closer real-world applications. Here, we review development technologies both hardware computational perspectives. We introduce compare different types enabling frequency-domain time-domain using various thermal, photon, field image sensor arrays. discuss how algorithms provide opportunities for capturing time-of-flight, spectroscopic, phase, intensity data at high throughputs. Furthermore, new prospects challenges future high-throughput are briefly introduced.

Language: Английский

---

Matrix Diffractive Deep Neural Networks Merging Polarization into Meta‐Devices

Laser & Photonics Review, Journal Year: 2023, Volume and Issue: 18(2)

Published: Oct. 25, 2023

Abstract The all‐optical diffractive deep neural networks (D 2 NNs) framework as a hardware platform is demonstrated to implement various advanced functional meta‐devices with high parallelism and processing speed. However, the design methodology merging trainable polarization modulation neurons into D NNs, which potentially possess higher integration more task‐loading capacity, not yet fully explored. Here, matrix (M‐D are proposed deploy polarization‐sensitive Jones metasurfaces multiplexing perform sophisticated inference tasks well inverse designs for meta‐devices. Three functionalities implemented by M‐D that is, task‐capacity classification, non‐interleaved high‐efficiency eight‐channel regulation, custom‐polarization information cryptographic multiplexing. NNs provide new strategy merge electromagnetic optical field modulators metasurfaces, may drive evolution of toward multi‐task devices.

Language: Английский

---

Snapshot multispectral imaging using a diffractive optical network

Light Science & Applications, Journal Year: 2023, Volume and Issue: 12(1)

Published: April 6, 2023

Abstract Multispectral imaging has been used for numerous applications in e.g., environmental monitoring, aerospace, defense, and biomedicine. Here, we present a diffractive optical network-based multispectral system trained using deep learning to create virtual spectral filter array at the output image field-of-view. This imager performs spatially-coherent over large spectrum, same time, routes pre-determined set of channels onto an pixels plane, converting monochrome focal-plane or sensor into device without any filters recovery algorithms. Furthermore, responsivity this is not sensitive input polarization states. Through numerical simulations, different network designs that achieve snapshot with 4, 9 16 unique bands within visible based on passive spatially-structured surfaces, compact design axially spans ~72 λ m , where mean wavelength band interest. Moreover, experimentally demonstrate 3D-printed creates its plane spatially repeating 2 × = 4 terahertz spectrum. Due their form factor computation-free, power-efficient polarization-insensitive forward operation, imagers can be transformative various sensing parts electromagnetic spectrum high-density wide-area pixel arrays are widely available.

Language: Английский

---

Intelligent optoelectronic processor for orbital angular momentum spectrum measurement

PhotoniX, Journal Year: 2023, Volume and Issue: 4(1)

Published: Feb. 13, 2023

Abstract Orbital angular momentum (OAM) detection underpins almost all aspects of vortex beams’ advances such as communication and quantum analogy. Conventional schemes are frustrated by low speed, complicated system, limited range. Here, we devise an intelligent processor composed photonic electronic neurons for OAM spectrum measurement in a fast, accurate direct manner. Specifically, optical layers extract invisible topological charge information from incoming light shallow layer predicts the exact spectrum. The integration optical-computing promises us compact single-shot system with high speed energy efficiency (optical operations / ~ $${10}^{3}$$ 103 ), neither necessitating reference wave nor repetitive steps. Importantly, our is endowed salient generalization ability robustness against diverse structured adverse effects (mean squared error $$10^{(-5)}$$ xmlns:mml="http://www.w3.org/1998/Math/MathML">10(-5) ). We further raise universal model interpretation paradigm to reveal underlying physical mechanisms hybrid processor, distinct conventional ‘black-box’ networks. Such algorithm can improve up 25-fold. also complete theory optoelectronic network enabling its efficient training. This work not only contributes explorations on physics applications, broadly inspires advanced links between computing effects.

Language: Английский

---

Massively parallel universal linear transformations using a wavelength-multiplexed diffractive optical network

Advanced Photonics, Journal Year: 2023, Volume and Issue: 5(01)

Published: Jan. 9, 2023

We report deep learning-based design of a massively parallel broadband diffractive neural network for all-optically performing large group arbitrarily-selected, complex-valued linear transformations between an input and output field-of-view, each with N_i N_o pixels, respectively. This processor is composed N_w wavelength channels, which uniquely assigned to distinct target transformation. A set arbitrarily-selected can be individually performed through the same at different illumination wavelengths, either simultaneously or sequentially (wavelength scanning). demonstrate that such network, regardless its material dispersion, successfully approximate unique transforms negligible error when number neurons (N) in matches exceeds 2 x N_o. further spectral multiplexing capability (N_w) increased by increasing N; our numerical analyses confirm these conclusions > 180, e.g., ~2000 depending on upper bound approximation error. Massively parallel, wavelength-multiplexed networks will useful designing high-throughput intelligent machine vision systems hyperspectral processors perform statistical inference analyze objects/scenes properties.

Language: Английский

---

Pluggable multitask diffractive neural networks based on cascaded metasurfaces

Opto-Electronic Advances, Journal Year: 2023, Volume and Issue: 7(2), P. 230005 - 230005

Published: July 26, 2023

Optical neural networks have significant advantages in terms of power consumption, parallelism, and high computing speed, which has intrigued extensive attention both academic engineering communities. It been considered as one the powerful tools promoting fields imaging processing object recognition. However, existing optical system architecture cannot be reconstructed to realization multi-functional artificial intelligence systems simultaneously. To push development this issue, we propose pluggable diffractive (P-DNN), a general paradigm resorting cascaded metasurfaces, can applied recognize various tasks by switching internal plug-ins. As proof-of-principle, recognition functions six types handwritten digits fashions are numerical simulated experimental demonstrated at near-infrared regimes. Encouragingly, proposed not only improves flexibility but paves new route for achieving high-speed, low-power versatile systems.

Language: Английский

---

All-optical image classification through unknown random diffusers using a single-pixel diffractive network

Light Science & Applications, Journal Year: 2023, Volume and Issue: 12(1)

Published: March 9, 2023

Classification of an object behind a random and unknown scattering medium sets challenging task for computational imaging machine vision fields. Recent deep learning-based approaches demonstrated the classification objects using diffuser-distorted patterns collected by image sensor. These methods demand relatively large-scale computing neural networks running on digital computers. Here, we present all-optical processor to directly classify through unknown, phase diffusers broadband illumination detected with single pixel. A set transmissive diffractive layers, optimized learning, forms physical network that all-optically maps spatial information input diffuser into power spectrum output light pixel at plane network. We numerically accuracy this framework radiation handwritten digits new diffusers, never used during training phase, achieved blind testing 88.53%. This single-pixel system is based passive layers process can operate any part electromagnetic simply scaling features proportional wavelength range interest. results have various potential applications in, e.g., biomedical imaging, security, robotics, autonomous driving.

Language: Английский

---

Diffractive optical computing in free space

Nature Communications, Journal Year: 2024, Volume and Issue: 15(1)

Published: Feb. 20, 2024

Abstract Structured optical materials create new computing paradigms using photons, with transformative impact on various fields, including machine learning, computer vision, imaging, telecommunications, and sensing. This Perspective sheds light the potential of free-space systems based engineered surfaces for advancing computing. Manipulating in unprecedented ways, emerging structured enable all-optical implementation mathematical functions learning tasks. Diffractive networks, particular, bring deep-learning principles into design operation to functionalities. Metasurfaces consisting deeply subwavelength units are achieving exotic responses that provide independent control over different properties can major advances computational throughput data-transfer bandwidth processors. Unlike integrated photonics-based optoelectronic demand preprocessed inputs, processors have direct access all degrees freedom carry information about an input scene/object without needing digital recovery or preprocessing information. To realize full architectures, diffractive metasurfaces need advance symbiotically co-evolve their designs, 3D fabrication/integration, cascadability, accuracy serve needs next-generation computing, telecommunication technologies.

Language: Английский

---

Universal linear intensity transformations using spatially incoherent diffractive processors

Light Science & Applications, Journal Year: 2023, Volume and Issue: 12(1)

Published: Aug. 15, 2023

Abstract Under spatially coherent light, a diffractive optical network composed of structured surfaces can be designed to perform any arbitrary complex-valued linear transformation between its input and output fields-of-view (FOVs) if the total number ( N ) optimizable phase-only features is ≥~2 i o , where refer useful pixels at FOVs, respectively. Here we report design incoherent processor that approximate in time-averaged intensity FOVs. monochromatic varying point spread function H network, corresponding given, arbitrarily-selected transformation, written as m n ; ′, ′) = | h ′)| 2 same define coordinates Using numerical simulations deep learning, supervised through examples input-output profiles, demonstrate trained all-optically ≥ ~2 . We also networks for processing information multiple illumination wavelengths, operating simultaneously. Finally, numerically performs all-optical classification handwritten digits under illumination, achieving test accuracy >95%. Spatially will broadly designing visual processors work natural light.

Language: Английский

---