SUANPAN: Scalable Photonic Linear Vector Machine DOI Creative Commons
Xue Feng,

Ziyue Yang,

Chen Li

et al.

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: Nov. 25, 2024

Abstract Photonic linear operation is a promising approach to handle the extensive vector multiplications in artificial intelligence (AI) techniques due natural bosonic parallelism and high-speed information transmission of photonics. However, there still no universal scalable photonic computing architecture that can be readily merged with existing electronic digital system. Even though it believed maximizing interaction light beams necessary fully utilize tremendous efforts have been made past decades, achieved dimensionality vector-matrix multiplication very limited difficulty scaling up tightly interconnected or highly coupled optical Here, we propose programmable reconfigurable machine perform only inner product two vectors, formed by series independent basic units, while each unit contains one emitter-detector pair. The elemental values processed vectors are prepared time-space domain encoding. Specifically, encoded output duration continuous light-emitter other as position result obtained sum photocurrents all photodetectors. Since among inside, extreme scalability could simply multiplicating without requiring large-scale analog-to-digital converter digital-to-analog arrays. Our encoding inspired traditional Chinese Suanpan abacus, thus denoted SUANPAN. As proof principle, SUANPAN implemented an 8×8 vertical cavity surface emission laser (VCSEL) array MoTe2 two-dimensional material photodetector array. experimental fidelities for randomly generated products over 98% 1-bit, 2-bit, 4-bit 8-bit quantization 95% 8~80 dimensionalities quantization. Two typical AI tasks Ising non-deterministic polynomial-time (NP)-hard optimization problem neural network visual perception performed demonstrate ability architecture. For problem, 1024-dimensional problems successfully solved, which highest dimensional heuristic algorithm. network, competitive classification accuracy 84~88% MNIST (Modified National Institute Standards Technology) handwritten digit dataset. We believe our proposed capable serving fundamental system potential enhance power future various applications.

Language: Английский

Spectral convolutional neural network chip for in-sensor edge computing of incoherent natural light DOI Creative Commons
Kaiyu Cui, Shijie Rao, Sheng Xu

et al.

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

Published: Jan. 2, 2025

Optical neural networks are considered next-generation physical implementations of artificial networks, but their capabilities limited by on-chip integration scale and requirement for coherent light sources. This study proposes a spectral convolutional network (SCNN) with matter meta-imaging. The optical layer is implemented integrating very large-scale pixel-aligned filters on CMOS image sensor. It facilitates highly parallel vector-inner products incident incoherent natural i.e., the direct information carrier, which empowers in-sensor analog computing at extremely high energy efficiency. To best our knowledge, this first integrated utilizing light. We employ same SCNN chip completely different real-world complex tasks achieve accuracies over 96% pathological diagnosis almost 100% face anti-spoofing video rates. These results indicate feasible scalable edge various portable terminals. have been successfully realized level, yet to operate it requires Here, authors proposed based combining sensor facilitate from

Language: Английский

Citations

2
Universal photonic artificial intelligence acceleration DOI

Segun Ahmed,

Reza Baghdadi,

Mikhail Bernadskiy

et al.

Nature, Journal Year: 2025, Volume and Issue: 640(8058), P. 368 - 374

Published: April 9, 2025

Language: Английский

Citations

2
All-Optical Single-Channel Plasmonic Logic Gates DOI

Zongkun Zhang,

Teng Zhang, Ming-Zhe Chong

et al.

Nano Letters, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 15, 2025

Optical computing, renowned for its light-speed processing and low power consumption, typically relies on the coherent control of two light sources. However, there are challenges in stabilizing maintaining high optical spatiotemporal coherence, especially large-scale computing systems. The coherence requires rigorous feedback circuits numerous phase shifters, introducing system instability complexity. Here we propose an innovative logic gate using a single source, with frequency polarization serving as virtual inputs. Our design leverages frequency-polarization multiplexed metasurfaces to achieve all basic operations by selectively routing surface plasmon polaritons. This single-channel maintains inherent between polarization, thereby considerably eliminating stringent light-source specifications rigid controls resulting higher stability. device showcases unique application potentials on-chip readout encryption information random sequences one-time pad, unlocking fresh prospects protection other simple

Language: Английский

Citations

1
All-optical nonlinear activation functions realized on phase-change photonic integrated circuits with microheaters DOI
Jiyuan Jiang, Bingxin Ding,

Shiyu Li

et al.

Journal of Semiconductors, Journal Year: 2025, Volume and Issue: 46(2), P. 022405 - 022405

Published: Feb. 1, 2025

Abstract Photonic neural networks have garnered significant attention in recent years due to their ultra-high computational speed, broad bandwidth, and parallel processing capabilities. However, compared conventional electronic nonlinear activation function (NAF), progress on efficient easily implementable optical (ONAF) was barely reported. To address this issue, we proposed a programmable, low-loss ONAF device based silicon micro-ring resonator capped with the Antimony selenide (Sb 2 Se 3 ) thin films, indium tin oxide (ITO) used as microheater. Leveraging our self-developed phase-transformation kinetic models, successfully simulated phase-transition behavior of Sb three different ONAFs—ELU, ReLU, radial basis (RBF) were achieved according discernible responses devices under phase-change extents. Classification results from Fashion MNIST dataset demonstrated that these ONAFs can be considered appropriate substitutes for traditional NAF. This indicated bright prospect future photonic networks.

Language: Английский

Citations

1
All-optical Fourier neural network using partially coherent light DOI Creative Commons
Jianwei Qin, Yanbing Liu, Yan Liu

et al.

Chip, Journal Year: 2025, Volume and Issue: unknown, P. 100140 - 100140

Published: March 1, 2025

Language: Английский

Citations

1
Integrated photonic 3D tensor processing engine DOI Creative Commons
Liangjun Lu, Yue Wu,

Ziheng Ni

et al.

Research Square (Research Square), Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 13, 2025

Abstract Optical computing leverages high bandwidth, low latency, and power efficiency, which is considered as one of the most effective solutions for accelerating deep learning tasks. However, mainstream photonic hardware accelerators are primarily optimized two-dimensional (2D) matrix-vector multiplications (MVMs). To implement three-dimensional (3D) convolutional neural networks (CNNs), high-order tensors must be reshaped, duplicated, cached in electrical domain according to size before computation, leading extra memory usage time overheads. Additionally, synchronization across multiple channels depends on external electronic clocks, increases complexity system. In this work, we propose an integrated 3D tensor processing engine (3D-TPE) based interweaving time, wavelength, space. Data caching, realized optical domain, reducing usage, simplifying caching achieved with tunable delay line chip supporting versatile clock frequencies up 200 GHz, accomplished a dual-coupled micro-ring resonators (MRRs) crossbar 3-dB passband width 50 GHz. We verify capabilities 3D-TPE at ranging from 10 GHz 30 perform proof-of-concept experiment LiDAR point cloud image recognition task operating 20 achieving accuracy 97.06%. The proposed anticipated facilitate convolutions, playing important role autonomous driving, healthcare, video analytics, virtual reality, etc.

Language: Английский

Citations

0
Self-driving laboratories, advanced immunotherapies and five more technologies to watch in 2025 DOI Creative Commons

Michael Eisenstein

Nature, Journal Year: 2025, Volume and Issue: 637(8047), P. 1008 - 1011

Published: Jan. 20, 2025

Language: Английский

Citations

0
Thermally Controlled Multi‐Functional Waveguide Photodetector DOI Open Access
Jianing Wang,

Guoyi Tao,

Zimeng Zhang

et al.

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

Published: Feb. 20, 2025

Abstract Photodetectors are one of the fundamental building blocks in integrated photonic systems. They mainly serve to convert optical electrical signals by absorbing photons semiconductors which have a bandgap smaller than photon energy. The constraint on energy relation commonly available semiconductor materials hinders application photonics for some emerging applications. Here novel waveguide detector with on‐chip heater is proposed. Tunable can be achieved via local heating, changes absorption characteristics. Based this mechanism, multi‐functional germanium three different applications including broadband communications, neural networks, and spectral sensing demonstrated. proposed photodetector enables high‐speed detection at extended long wavelengths. In an artificial network, controllable photoresponse allows tailorable nonlinear activation function implemented. It also capable retrieving information single tunable without need any other components. This work not only proposes new structure but identify approach make photodetectors that used integration platforms.

Language: Английский

Citations

0
Biomaterials for neuroengineering: Applications and challenges DOI Creative Commons

Huanghui Wu,

E.J. Feng,

Huazong Yin

et al.

Regenerative Biomaterials, Journal Year: 2025, Volume and Issue: 12

Published: Jan. 1, 2025

Abstract Neurological injuries and diseases are a leading cause of disability worldwide, underscoring the urgent need for effective therapies. Neural regaining enhancement therapies seen as most promising strategies restoring neural function, offering hope individuals affected by these conditions. Despite their promise, path from animal research to clinical application is fraught with challenges. Neuroengineering, particularly through use biomaterials, has emerged key field that paving way innovative solutions It seeks understand treat neurological disorders, unravel nature consciousness, explore mechanisms memory brain’s relationship behavior, tissue engineering, interfaces targeted drug delivery systems. These including both natural synthetic types, designed replicate cellular environment brain, thereby facilitating repair. This review aims provide comprehensive overview biomaterials in neuroengineering, highlighting functional across basic practice. covers recent developments biomaterial-based products, 2D 3D bioprinted scaffolds cell organoid culture, brain-on-a-chip systems, biomimetic electrodes brain–computer interfaces. also explores artificial synapses networks, discussing applications modeling microenvironments repair regeneration, modulation manipulation integration traditional Chinese medicine. serves guide role advancing neuroengineering solutions, providing insights into ongoing efforts bridge gap between innovation application.

Language: Английский

Citations

0
Information Entropy and Its Periodic Features in Hermite–Gaussian Correlated Schell-Model Beams in a Gradient-Index Fiber DOI Creative Commons
Jiayi Yu,

Jifei Huang,

Ruilin Liu

et al.

Photonics, Journal Year: 2025, Volume and Issue: 12(3), P. 198 - 198

Published: Feb. 26, 2025

This paper investigates the evolution of information entropy (IE) in Hermite–Gaussian correlated Schell-model (HGcSM) beams propagating through a gradient-index (GRIN) fiber using Shannon theory. Our results reveal that IE such evolves periodically, with beam order significantly influencing its initial distribution. Compared traditional Gaussian beams, HGcSM exhibit more complex dynamics, characterized by periodically emerging low-entropy regions whose decreases increasing order. Furthermore, fiber’s central refractive index and core radius strongly affect period fluctuation amplitude IE. These findings provide theoretical basis for optimizing partially coherent optical applications.

Language: Английский

Citations

0