Magnetization Vector Rotation Reservoir Computing Operated by Redox Mechanism DOI
Wataru Namiki, Daiki Nishioka, Takashi Tsuchiya

et al.

Nano Letters, Journal Year: 2024, Volume and Issue: 24(15), P. 4383 - 4392

Published: March 21, 2024

Physical reservoir computing is a promising way to develop efficient artificial intelligence using physical devices exhibiting nonlinear dynamics. Although magnetic materials have advantages in miniaturization, the need for field and large electric current results high power consumption complex device structure. To resolve these issues, we propose redox-based utilizing planar Hall effect anisotropic magnetoresistance, which are phenomena described by different functions of magnetization vector that do not be applied. The expressive this based on compact all-solid-state redox transistor higher than previous reservoir. normalized mean square error second-order equation task was 1.69 × 10–3, lower memristor array (3.13 10–3) even though number nodes fewer half array.

Language: Английский

Photonic multiplexing techniques for neuromorphic computing DOI Creative Commons
Yunping Bai, Xingyuan Xu, Mengxi Tan

et al.

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

Citations

129
The physics of optical computing DOI
Peter L. McMahon

Nature Reviews Physics, Journal Year: 2023, Volume and Issue: 5(12), P. 717 - 734

Published: Oct. 9, 2023

Language: Английский

Citations

118
Hands-on reservoir computing: a tutorial for practical implementation DOI Creative Commons
Matteo Cucchi, Steven Abreu, Giuseppe Ciccone

et al.

Neuromorphic Computing and Engineering, Journal Year: 2022, Volume and Issue: 2(3), P. 032002 - 032002

Published: July 1, 2022

Abstract This manuscript serves a specific purpose: to give readers from fields such as material science, chemistry, or electronics an overview of implementing reservoir computing (RC) experiment with her/his system. Introductory literature on the topic is rare and vast majority reviews puts forth basics RC taking for granted concepts that may be nontrivial someone unfamiliar machine learning field (see example reference Lukoševičius (2012 Neural Networks: Tricks Trade (Berlin: Springer) pp 659–686). unfortunate considering large pool systems show nonlinear behavior short-term memory harnessed design novel computational paradigms. offers framework circumvents typical problems arise when traditional, fully fledged feedforward neural networks hardware, minimal device-to-device variability control over each unit/neuron connection. Instead, one can use random, untrained where only output layer optimized, example, linear regression. In following, we will highlight potential hardware-based networks, advantages more traditional approaches, obstacles overcome their implementation. Preparing high-dimensional system well-performing task not easy it seems at first sight. We hope this tutorial lower barrier scientists attempting exploit tasks typically carried out in artificial intelligence. A simulation tool accompany paper available online 7 https://github.com/stevenabreu7/handson_reservoir .

Language: Английский

Citations

104
Emerging opportunities and challenges for the future of reservoir computing DOI Creative Commons
Min Yan, Can Huang, Peter Bienstman

et al.

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

Published: March 6, 2024

Abstract Reservoir computing originates in the early 2000s, core idea being to utilize dynamical systems as reservoirs (nonlinear generalizations of standard bases) adaptively learn spatiotemporal features and hidden patterns complex time series. Shown have potential achieving higher-precision prediction chaotic systems, those pioneering works led a great amount interest follow-ups community nonlinear dynamics systems. To unlock full capabilities reservoir towards fast, lightweight, significantly more interpretable learning framework for temporal substantially research is needed. This Perspective intends elucidate parallel progress mathematical theory, algorithm design experimental realizations computing, identify emerging opportunities well existing challenges large-scale industrial adoption together with few ideas viewpoints on how some might be resolved joint efforts by academic researchers across multiple disciplines.

Language: Английский

Citations

64
Large-scale photonic chiplet Taichi empowers 160-TOPS/W artificial general intelligence DOI

Zhihao Xu,

Tiankuang Zhou, Muzhou Ma

et al.

Science, Journal Year: 2024, Volume and Issue: 384(6692), P. 202 - 209

Published: April 11, 2024

The pursuit of artificial general intelligence (AGI) continuously demands higher computing performance. Despite the superior processing speed and efficiency integrated photonic circuits, their capacity scalability are restricted by unavoidable errors, such that only simple tasks shallow models realized. To support modern AGIs, we designed Taichi-large-scale chiplets based on an diffractive-interference hybrid design a distributed architecture has millions-of-neurons capability with 160-tera-operations per second watt (TOPS/W) energy efficiency. Taichi experimentally achieved on-chip 1000-category-level classification (testing at 91.89% accuracy in 1623-category Omniglot dataset) high-fidelity intelligence-generated content up to two orders magnitude improvement paves way for large-scale advanced tasks, further exploiting flexibility potential photonics AGI.

Language: Английский

Citations

55
Physical reservoir computing with emerging electronics DOI
Xiangpeng Liang, Jianshi Tang, Ya‐Nan Zhong

et al.

Nature Electronics, Journal Year: 2024, Volume and Issue: 7(3), P. 193 - 206

Published: March 12, 2024

Language: Английский

Citations

53
A perspective on physical reservoir computing with nanomagnetic devices DOI Creative Commons
D. A. Allwood, Matthew O. A. Ellis, David Griffin

et al.

Applied Physics Letters, Journal Year: 2023, Volume and Issue: 122(4)

Published: Jan. 23, 2023

Neural networks have revolutionized the area of artificial intelligence and introduced transformative applications to almost every scientific field industry. However, this success comes at a great price; energy requirements for training advanced models are unsustainable. One promising way address pressing issue is by developing low-energy neuromorphic hardware that directly supports algorithm's requirements. The intrinsic non-volatility, non-linearity, memory spintronic devices make them appealing candidates devices. Here, we focus on reservoir computing paradigm, recurrent network with simple algorithm suitable computation since they can provide properties non-linearity memory. We review technologies methods conclude critical open issues before such become widely used.

Language: Английский

Citations

44
Reducing reservoir computer hyperparameter dependence by external timescale tailoring DOI Creative Commons
Lina Jaurigue, Kathy Lüdge

Neuromorphic Computing and Engineering, Journal Year: 2024, Volume and Issue: 4(1), P. 014001 - 014001

Published: Jan. 10, 2024

Abstract Task specific hyperparameter tuning in reservoir computing is an open issue, and of particular relevance for hardware implemented reservoirs. We investigate the influence directly including externally controllable task timescales on performance sensitivity approaches. show that need optimisation can be reduced if are tailored to task. Our results mainly relevant temporal tasks requiring memory past inputs, example chaotic timeseries prediction. consider various methods approach demonstrate universality our message by looking at both time-multiplexed spatially-multiplexed computing.

Language: Английский

Citations

20
Physical deep learning with biologically inspired training method: gradient-free approach for physical hardware DOI Creative Commons
M. Nakajima, Katsuma Inoue, Kenji Tanaka

et al.

Nature Communications, Journal Year: 2022, Volume and Issue: 13(1)

Published: Dec. 26, 2022

Abstract Ever-growing demand for artificial intelligence has motivated research on unconventional computation based physical devices. While such devices mimic brain-inspired analog information processing, the learning procedures still rely methods optimized digital processing as backpropagation, which is not suitable implementation. Here, we present deep by extending a biologically inspired training algorithm called direct feedback alignment. Unlike original algorithm, proposed method random projection with alternative nonlinear activation. Thus, can train neural network without knowledge about system and its gradient. In addition, emulate this scalable hardware. We demonstrate proof-of-concept using an optoelectronic recurrent reservoir computer. confirmed potential accelerated competitive performance benchmarks. Our results provide practical solutions acceleration of neuromorphic computation.

Language: Английский

Citations

68
Role of delay-times in delay-based photonic reservoir computing [Invited] DOI Creative Commons
Tim Hülser, Felix Köster, Lina Jaurigue

et al.

Optical Materials Express, Journal Year: 2022, Volume and Issue: 12(3), P. 1214 - 1214

Published: Feb. 1, 2022

Delay-based reservoir computing has gained a lot of attention due to the relative simplicity with which this concept can be implemented in hardware. However, unnecessary constraints are commonly placed on relationship between delay-time and input clock-cycle, have detrimental effect performance. We review existing literature subject introduce delay-based manner that demonstrates no predefined clock-cycle is required for work. Choosing delay-times independent one gains an important degree freedom. Consequently, we discuss ways improve performance formed by delay-coupled oscillators show impact tuning such systems.

Language: Английский

Citations

56