All-analog photoelectronic chip for high-speed vision tasks

Nature, Год журнала: 2023, Номер 623(7985), С. 48 - 57

Опубликована: Окт. 25, 2023

Abstract Photonic computing enables faster and more energy-efficient processing of vision data 1–5 . However, experimental superiority deployable systems remains a challenge because complicated optical nonlinearities, considerable power consumption analog-to-digital converters (ADCs) for downstream digital vulnerability to noises system errors 1,6–8 Here we propose an all-analog chip combining electronic light (ACCEL). It has systemic energy efficiency 74.8 peta-operations per second watt speed 4.6 (more than 99% implemented by optics), corresponding three one order magnitude higher state-of-the-art processors, respectively. After applying diffractive as encoder feature extraction, the light-induced photocurrents are directly used further calculation in integrated analog without requirement converters, leading low latency 72 ns each frame. With joint optimizations optoelectronic adaptive training, ACCEL achieves competitive classification accuracies 85.5%, 82.0% 92.6%, respectively, Fashion-MNIST, 3-class ImageNet time-lapse video recognition task experimentally, while showing superior robustness low-light conditions (0.14 fJ μm −2 frame). can be across broad range applications such wearable devices, autonomous driving industrial inspections.

Язык: Английский

---

All-optical image denoising using a diffractive visual processor

Light Science & Applications, Год журнала: 2024, Номер 13(1)

Опубликована: Фев. 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.

Язык: Английский

---

Artificial neural networks for photonic applications—from algorithms to implementation: tutorial

Advances in Optics and Photonics, Год журнала: 2023, Номер 15(3), С. 739 - 739

Опубликована: Авг. 3, 2023

This tutorial-review on applications of artificial neural networks in photonics targets a broad audience, ranging from optical research and engineering communities to computer science applied mathematics. We focus here the areas at interface between these disciplines, attempting find right balance technical details specific each domain overall clarity. First, we briefly recall key properties peculiarities some core network types, which believe are most relevant photonics, also linking layer's theoretical design hardware realizations. After that, elucidate question how fine-tune selected model's perform required task with optimized accuracy. Then, review part, discuss recent developments progress for several including multiple aspects communications, imaging, sensing, new materials lasers. In following section, put special emphasis accurately evaluate complexity context transition algorithms implementation. The introduced characteristics used analyze as specific, albeit highly important example, comparing those benchmark signal processing methods. combine description well-known model compression strategies machine learning, novel techniques recently networks. It is stress that although our this methods presented can be handy much wider range scientific applications.

Язык: Английский

---

Resource‐Saving and High‐Robustness Image Sensing Based on Binary Optical Computing

Laser & Photonics Review, Год журнала: 2024, Номер unknown

Опубликована: Окт. 4, 2024

Abstract Computational imaging, as a novel technology utilizing encoded image acquisition, relies on intelligent decoding methods for effective restoration and sensing. Optical computing‐based decoders can efficiently process extract features from pre‐sensor information, reducing the computational burden digital computers. However, mainstream parallel optical neural network (ONN) architectures based wavefront propagation typically possess complex structures high‐precision parameters, which pose challenges in terms of precise fabrication system calibration, well sensitivity to signal‐to‐noise ratios. In this work, binary‐weighted computing engine is proposed with spatial multiplexing aggregation (B‐OSMA), large‐scale passive ONN implementation that achieves high‐efficiency Employing B‐OSMA an decoder, demonstrated categorizing 2% compressive experimented sampling 92.0% 83.8% accuracy MNIST fashion‐MNIST datasets, respectively, approaching performance full‐precision electronic while storage requirements by 97%. Compared conventional ONNs analog weights, exhibits enhanced resilience against systematic errors ambient noise. This work represents significant advancement towards practical applications

Язык: Английский

---

A perspective on the artificial intelligence’s transformative role in advancing diffractive optics

iScience, Год журнала: 2024, Номер 27(7), С. 110270 - 110270

Опубликована: Июнь 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.

Язык: Английский

---

Ln-HOF Nanofiber Organogels with Time-Resolved Luminescence for Programmable and Reliable Encryption

Nano Letters, Год журнала: 2023, Номер 23(24), С. 11916 - 11924

Опубликована: Дек. 6, 2023

Developing tunable luminescent materials for high throughput information storage is highly desired following the explosive growth of global data. Although considerable success has been achieved, achieving programmable encryption remains challenging due to current signal crosstalk problems. Here, we developed long-lived room-temperature phosphorescent organogels enabled by lanthanum-coordinated hydrogen-bonded organic framework nanofibers time-resolved programming. Via modulating coassembled lanthanum concentration and Förster resonance energy transfer efficiency, lifetimes are prolonged facilely manipulated (20–644 ms), realizing encoding space enlargement multichannel data outputs. The aggregated strong interfacial supramolecular bonding endows with excellent mechanical toughness (36.16 MJ m–2) self-healing properties (95.7%), synergistically photostability (97.6% lifetime retention in 10000 fatigue cycles) via suppressing nonradiative decays. This work presents a lifetime-gated strategy lanthanum-coordination regulation that promisingly breaks through limitations responsive materials, opening unprecedented avenues high-level protection.

Язык: Английский

---

Super-resolution diffractive neural network for all-optical direction of arrival estimation beyond diffraction limits

Light Science & Applications, Год журнала: 2024, Номер 13(1)

Опубликована: Июль 10, 2024

Abstract Wireless sensing of the wave propagation direction from radio sources lays foundation for communication, radar, navigation, etc. However, existing signal processing paradigm arrival estimation requires frequency electronic circuit to demodulate and sample multichannel baseband signals followed by a complicated computing process, which places fundamental limit on its speed energy efficiency. Here, we propose super-resolution diffractive neural networks (S-DNN) process electromagnetic (EM) waves directly DOA at light. The multilayer meta-structures S-DNN generate super-oscillatory angular responses in local regions that can perform all-optical with resolutions beyond diffraction limit. spatial-temporal multiplexing passive reconfigurable S-DNNs is utilized achieve high-resolution over wide field view. validated multiple 5 GHz bandwidth latency two four orders magnitude lower than state-of-the-art commercial devices principle. results resolution an order magnitude, experimentally demonstrated times, higher diffraction-limited resolution. We also apply S-DNN’s edge capability, assisted intelligent surfaces, extremely low-latency integrated communication low power consumption. Our work significant step towards utilizing photonic processors facilitate various wireless tasks advantages both paradigms performance computing.

Язык: Английский

---

Terahertz spoof plasmonic neural network for diffractive information recognition and processing

Nature Communications, Год журнала: 2024, Номер 15(1)

Опубликована: Авг. 6, 2024

All-optical diffractive neural networks, as analog artificial intelligence accelerators, leverage parallelism and computation for complex data processing. However, their low space transmission efficiency or large spatial dimensions hinder miniaturization broader application. Here, we propose a terahertz spoof plasmonic network on planar platform direct multi-target recognition. Our approach employs surface plasmon polariton coupler array to construct layer, resulting in compact, efficient, easily integrable architecture. We designed three schemes: basis vector classification, multi-user recognition, MNIST handwritten digit classification. Experimental results reveal that the successfully classifies vectors, recognizes orientation information, directly processes digits using input framework comprising metal grating array, transmitters, receivers. This work broadens application of metamaterials, paving way on-chip integration, intelligent communication, advanced computing systems.

Язык: Английский

---

Compute-First Optical Detection for Noise-Resilient Visual Perception

ACS Photonics, Год журнала: 2025, Номер unknown

Опубликована: Фев. 5, 2025

Язык: Английский

---

All-Optical DCT Encoding and Information Compression Based on Diffraction Neural Network

ACS Photonics, Год журнала: 2025, Номер unknown

Опубликована: Фев. 5, 2025

Язык: Английский

---