Integrated photonic neuromorphic computing: opportunities and challenges

Nature Reviews Electrical Engineering, Год журнала: 2024, Номер 1(6), С. 358 - 373

Опубликована: Июнь 6, 2024

Язык: Английский

---

Optical phase encoding in a pulsed approach to reservoir computing

Optics Letters, Год журнала: 2024, Номер 49(8), С. 2097 - 2097

Опубликована: Март 5, 2024

The exploitation of the full structure multimode light fields enables compelling capabilities in many including classical and quantum information science. We exploit data-encoding on optical phase pulses a femtosecond laser source for photonic implementation reservoir computing protocol. Rather than intensity detection, data-reading is done via homodyne detection that accesses combinations an amplitude field. Numerical experimental results nonlinear autoregressive moving average (NARMA) tasks dynamic predictions are shown. discuss perspectives quantum-enhanced protocols.

Язык: Английский

---

Spiking Reservoir Computing Based on Stochastic Diffusive Memristors

Advanced Electronic Materials, Год журнала: 2024, Номер unknown

Опубликована: Авг. 21, 2024

Abstract Reservoir computing (RC), a type of recurrent neural network, is particularly well‐suited for hardware implementation in edge computing. It shown that RC based on dynamic memristors potentially offers much lower power consumption and reduced computation times than digital electronics. However, challenges such as stochasticity read noise these devices can impair its performance. Furthermore, the external analog‐to‐digital (ADC) readout circuits may require substantial area energy. In this work, it experimentally demonstrated population stochastic diffusive Ag:SiO x effectively construct spiking reservoir system. This system demonstrates remarkable resilience to delivers exceptional performance across range computational tasks, achieving 98% accuracy waveform classification normalized root mean square error (NRMSE) 0.154 time‐series prediction. Further simulations reveal certain degree device actually enhances Without using ADC converters, hybrid memristor‐CMOS designed significantly compared fully systems.

Язык: Английский

---

Photonic Spiking Reservoir Computing System Based on a DFB-SA Laser for Pattern Recognition

ACS Photonics, Год журнала: 2025, Номер unknown

Опубликована: Янв. 17, 2025

In the field of information processing, traditional computing methods encounter limitations in handling increasingly complex tasks and meeting growing performance requirements. Reservoir computing, as a new paradigm, has demonstrated excellent time series prediction tasks. However, photonic reservoir still needs improvement certain aspects, such high computational complexity relatively high-power consumption for processing. our work, spiking system based on single distributed feedback laser with saturable absorber (DFB-SA laser) is numerically experimentally. Owing to advantages DFB-SA, easy manufacture, faster response time, better control over gain current reverse voltage. Compared continuous signals, pulse emission requires less energy, helping reduce power consumption. The reported effectively implements DFB-SA shows its successful application nonlinear classification achieves favorable lower input dimensionality, opening avenues future processing systems machines integrated absorption regions providing approach lightweight architectures.

Язык: Английский

---

Photonic spiking neural network built with a single VCSEL for high-speed time series prediction

Communications Physics, Год журнала: 2025, Номер 8(1)

Опубликована: Март 20, 2025

Язык: Английский

---

Versatile and Robust Reservoir Computing with PWM‐Driven Heterogenous R–C Circuits

Advanced Science, Год журнала: 2025, Номер unknown

Опубликована: Май 14, 2025

Abstract Physical reservoir computing (PRC) holds great promise for low‐latency, energy‐efficient information processing, yet current implementations often suffer from limited flexibility, adaptability, and environmental stability. Here, a PRC system based on pulse‐width modulation (PWM)‐encoded resistor‐capacitor ( R – C ) circuits is introduced, achieving exceptional versatility robustness. By leveraging customizable nonlinearities dynamic timescales, this achieves state‐of‐the‐art performance across diverse tasks, including chaotic time‐series forecasting (NRMSE = 0.015 Mackey‐Glass) complex multiscale tasks (94% accuracy multiclass heartbeat classification). Notably, the design reduces relative errors by 98.4% different device batches under temperature variations compared to memristor‐based reservoirs. These features position approach as scalable, adaptive, solution edge in environments, paving way robust practical analog systems.

Язык: Английский

---

Multimode Fabry-Perot laser as a reservoir computing and extreme learning machine photonic accelerator

Neuromorphic Computing and Engineering, Год журнала: 2023, Номер 3(4), С. 044003 - 044003

Опубликована: Окт. 11, 2023

Abstract In this work, we introduce Fabry–Perot lasers as neuromoprhic nodes in the context of time-delayed reservoir computing and extreme learning machine (ELM) for processing temporal signals high-speed classification images. By exploiting multi-wavelength emission capabilities lasers, additional can be introduced, thus raising computational power without sacrificing speed. An experimental validation concept using a ELM is presented targeting time depedent task such channel equalization 50 km 28 Gbaud ‘PAM-4’ transmission, offering hard-decision forward error correction compatible performance. Additionally, neuromorphic has been further strengthened by modifying data entry technique parallelelly assigning different samples input signal to modes so significantly reduce speed penalty. Numerical simulations revealed that alternative insertion offer reduction delay physical footprint 75% compared conventional approach same symbols all Fairy–Perot modes. Moreover, similar scheme ‘MNIST’ image were able numerically achieve 255.1 Mimages s −1 accuracy up 95.95%.

Язык: Английский

---

The influence of timescales and data injection schemes for reservoir computing using spin-VCSELs

Communications Physics, Год журнала: 2024, Номер 7(1)

Опубликована: Ноя. 14, 2024

Abstract Reservoir computing with photonic systems promises fast and energy efficient computations. Vertical emitting semiconductor lasers two spin-polarized charge-carrier populations (spin-VCSEL), are good candidates for high-speed reservoir computing. With our work, we highlight the role of internal dynamic coupling on prediction performance. We present numerical evidence critical impact different data injection schemes timescales. A central finding is that dynamics all dynamical degrees freedom can only be utilized if an appropriate perturbation via input chosen as scheme. If encoded optical phase difference, carrier not addressed but instead a faster rate possible. find strong correlations performance system response time underlying delay-induced bifurcation structure, which allows to transfer results other physical systems.

Язык: Английский

---

Magnetization reversal by multiple optical pulses for photonic spiking neuron with the leaky integrate and fire model

Japanese Journal of Applied Physics, Год журнала: 2024, Номер 63(7), С. 072002 - 072002

Опубликована: Июнь 3, 2024

Abstract Photonic accelerators are anticipated to be the next generation of hardware processors, replacing traditional electronic accelerators. In current photonic based on artificial neural networks, integrated circuits incorporated with leverage their strengths: used perform linear calculations, while nonlinear calculations. However, this architecture requires optoelectric conversion at each layer and is unable superiority light. We propose a novel spiking neuron magneto-optical synapse an all-optical network. This study experimentally demonstrates that magnetization reversal CoFeB, which occurs during thermal accumulation owing multiple optical pulses, similar behavior leaky fire model neurons.

Язык: Английский

---

BP-based supervised learning algorithm for multilayer photonic spiking neural network and hardware implementation

Optics Express, Год журнала: 2023, Номер 31(10), С. 16549 - 16549

Опубликована: Апрель 20, 2023

We introduce a supervised learning algorithm for photonic spiking neural network (SNN) based on back propagation. For the algorithm, information is encoded into spike trains with different strength, and SNN trained according to patterns composed of numbers output neurons. Furthermore, classification task performed numerically experimentally in SNN. The neuron vertical-cavity surface-emitting laser which functionally similar leaky-integrate fire neuron. results prove demonstration implementation hardware. To seek ultra-low power consumption delay, it great significance design implement hardware-friendly networks realize hardware-algorithm collaborative computing.

Язык: Английский

---