Physical reservoir computing on a soft bio-inspired swimmer.

Bio-inspired Autonomous Underwater Vehicles with soft bodies provide significant performance benefits over conventional propeller-driven vehicles; however, it is difficult to control these vehicles due to their soft underactuated bodies. This study investigates the application of Physical Reservoir Computing (PRC) in the swimmer's flexible body to perform state estimation. This PRC informed state estimation has potential to be used in vehicle control. PRC is a type of recurrent neural network that leverages the nonlinear dynamics of a physical system to predict a nonlinear spatiotemporal input-output relationship. By embodying the neural network into the physical structure, PRC can process the response to an environment input with high computational efficiency. This study uses a soft bio-inspired propulsor embodied as a physical reservoir. We evaluate its ability to predict different state estimation tasks including hydrodynamic forces and benchmark computational tasks in response to the forcing applied to the artificial muscles during actuation. The propulsor's nonlinear fluid-structural dynamics act as the physical reservoir and the kinematic feedback serves as the reservoir readouts. We show that the bio-inspired underwater propulsor can predict the hydrodynamic thrust and benchmark tasks with high accuracy under specific input frequencies. By analyzing the frequency spectrum of the input, readouts, and target signals, we demonstrate that the system's dynamic response determines the frequency contents relevant to the task being predicted. The propulsor's ability to process information stems from its nonlinearity, as it is responsible to transform the input signal into a broader spectrum of frequency content at the readouts. This broad band of frequency content is necessary to recreate the target signal within the PRC algorithm, thereby improving the prediction performance. The spectral analysis provides a unique perspective to analyze the nonlinear dynamics of a physical reservoir and serves as a valuable tool for examining other types of vibratory systems for PRC. This work serves as a first step towards embodying computation into soft bio-inspired swimmers.

Dynamical predictive coding with reservoir computing performs noise-robust multi-sensory speech recognition.

Multi-sensory integration is a perceptual process through which the brain synthesizes a unified perception by integrating inputs from multiple sensory modalities. A key issue is understanding how the brain performs multi-sensory integrations using a common neural basis in the cortex. A cortical model based on reservoir computing has been proposed to elucidate the role of recurrent connectivity among cortical neurons in this process. Reservoir computing is well-suited for time series processing, such as speech recognition. This inquiry focuses on extending a reservoir computing-based cortical model to encompass multi-sensory integration within the cortex. This research introduces a dynamical model of multi-sensory speech recognition, leveraging predictive coding combined with reservoir computing. Predictive coding offers a framework for the hierarchical structure of the cortex. The model integrates reliability weighting, derived from the computational theory of multi-sensory integration, to adapt to multi-sensory time series processing. The model addresses a multi-sensory speech recognition task, necessitating the management of complex time series. We observed that the reservoir effectively recognizes speech by extracting time-contextual information and weighting sensory inputs according to sensory noise. These findings indicate that the dynamic properties of recurrent networks are applicable to multi-sensory time series processing, positioning reservoir computing as a suitable model for multi-sensory integration.

A Near-Infrared Retinomorphic Device with High Dimensionality Reservoir Expression.

Physical reservoir-based reservoir computing (RC) systems for intelligent perception have recently gained attention because they require fewer computing resources. However, the system remains limited in infrared (IR) machine vision, including materials and physical reservoir expression power. Inspired by biological visual perception systems, the study proposes a near-infrared (NIR) retinomorphic device that simultaneously perceives and encodes narrow IR spectral information (at ≈980 nm). The proposed device, featuring core-shell upconversion nanoparticle/poly (3-hexylthiophene) (P3HT) nanocomposite channels, enables the absorption and conversion of NIR into high-energy photons to excite more photo carriers in P3HT. The photon-electron-coupled dynamics under the synergy of photovoltaic and photogating effects influence the nonlinearity and high dimensionality of the RC system under narrow-band NIR irradiation. The device also exhibits multilevel data storage capability (≥8 levels), excellent stability (≥2000 s), and durability (≥100 cycles). The system accurately identifies NIR static and dynamic handwritten digit images, achieving recognition accuracies of 91.13% and 90.07%, respectively. Thus, the device tackles intricate computations like solving second-order nonlinear dynamic equations with minimal errors (normalized mean squared error of 1.06 × 10⁻3 during prediction).

Dual-Mode Semiconductor Device Enabling Optoelectronic Detection and Neuromorphic Processing with Extended Spectral Responsivity.

High-performance semiconductor devices capable of multiple functions are pivotal in meeting the challenges of miniaturization and integration in advanced technologies. Despite the inherent difficulties of incorporating dual functionality within a single device, a high-performance, dual-mode device is reported. This device integrates an ultra-thin Al2O3 passivation layer with a PbS/Si hybrid heterojunction, which can simultaneously enable optoelectronic detection and neuromorphic operation. In mode 1, the device efficiently separates photo-generated electron-hole pairs, exhibiting an ultra-wide spectral response from ultraviolet (265 nm) to near-infrared (1650 nm) wavelengths. It also reproduces high-quality images of 256 × 256 pixels, achieving a Q-value as low as 0.00437 µW cm- 2 at a light intensity of 8.58 µW cm- 2. Meanwhile, when in mode 2, the as-assembled device with typical persistent photoconductivity (PPC) behavior can act as a neuromorphic device, which can achieve 96.5% accuracy in classifying standard digits underscoring its efficacy in temporal information processing. It is believed that the present dual-function devices potentially advance the multifunctionality and miniaturization of chips for intelligence applications.

Micromagnetic analysis of magnetic vortex dynamics for reservoir computing.

Reservoir computing (RC) has generated significant interest for its ability to reduce computational costs compared to traditional neural networks. The performance of the RC element is quantified by its memory capacity (MC) and prediction capability. In this study, we utilize micromagnetic simulations to investigate a magnetic vortex based on a permalloy ferromagnetic layer and its dynamics in RC. The nonlinear dynamics of the vortex core (VC), driven by continuous oscillating magnetic fields and binary digit data as spin-polarized current pulses, are analyzed. The highest MC observed is 4.1, corresponding to the nonlinear VC dynamics. Additionally, the prediction capability is evaluated using the Nonlinear Auto-Regressive Moving Average 2 task, demonstrating a normalized mean squared error of 0.0241 highlighting the time-series data prediction performance of the vortex as a reservoir.

An Antiferromagnetic Neuromorphic Memory Based on Perpendicularly Magnetized CoO.

Antiferromagnets (AFMs) are ideal materials to boost neuromorphic computing toward the ultrahigh speed and ultracompact integration regime. However, developing a suitable AFM neuromorphic memory remains an aspirational but challenging goal. In this work, we construct such a memory based on the CoO/Pt heterostructure, in which the collinear insulating AFM CoO shows a strong perpendicular anisotropy facilitating its electrical readout and writing. Utilizing the unique nonlinear response and bipolar fading memory properties of the device, we demonstrate a multidimensional reservoir computing beyond the traditional binary paradigm. These results are expected to pave the way toward next-generation fast and massive neuromorphic computing.

Photonic deep residual time-delay reservoir computing.

Time-delay reservoir computing (TDRC) represents a simplified variant of recurrent neural networks, employing a nonlinear node with a feedback mechanism to construct virtual nodes. The capabilities of TDRC can be enhanced by transitioning to a deep architecture. In this work, we propose a novel photonic deep residual TDRC (DR-TDRC) with augmented capabilities. The additional time delay added to the residual structure enables DR-TDRC superior to traditional deep structures across various benchmark tasks, especially in memory capability and almost an order of magnitude improvement in nonlinear channel equalization. Additionally, a specifically designed clipping algorithm is utilized to counteract the damage of redundant layers in deep structures, enabling the extension of the deep TDRC to dozens rather than just a few layers, with higher performance. We experimentally demonstrate the proof-of-concept with a 4-layer DR-TDRC containing 960 interrelated neurons (240 neurons per layer), based on four injection-locked distributed feedback lasers. We confirm the potential for scalable deep RC with elevated performance. Our results provide a feasible approach for expanding deep photonic computing to satisfy the boosting demand for artificial intelligence.

Physical Reservoir Computing Using van der Waals Ferroelectrics for Acoustic Keyword Spotting.

Acoustic keyword spotting (KWS) plays a pivotal role in the voice-activated systems of artificial intelligence (AI), allowing for hands-free interactions between humans and smart devices through information retrieval of the voice commands. The cloud computing technology integrated with the artificial neural networks has been employed to execute the KWS tasks, which however suffers from propagation delay and the risk of privacy breach. Here, we report a single-node reservoir computing (RC) system based on the CuInP2S6 (CIPS)/graphene heterostructure planar device for implementing the KWS task with low computation cost. Through deliberately tuning the Schottky barrier height at the ferroelectric CIPS interfaces for the thermionic injection and transport of the electrons, the typical nonlinear current response and fading memory characteristics are achieved in the device. Additionally, the device exhibits diverse synaptic plasticity with an excellent separation capability of the temporal information. We construct a RC system through employing the ferroelectric device as the physical node to spot the acoustic keywords, i.e., the natural numbers from 1 to 9 based on simulation, in which the system demonstrates outstanding performance with high accuracy rate (>94.6%) and recall rate (>92.0%). Our work promises physical RC in single-node configuration as a prospective computing platform to process the acoustic keywords, promoting its applications in the artificial auditory system at the edge.

A potential mechanism for Gibsonian resonance: behavioral entrainment emerges from local homeostasis in an unsupervised reservoir network.

While the cognitivist school of thought holds that the mind is analogous to a computer, performing logical operations over internal representations, the tradition of ecological psychology contends that organisms can directly "resonate" to information for action and perception without the need for a representational intermediary. The concept of resonance has played an important role in ecological psychology, but it remains a metaphor. Supplying a mechanistic account of resonance requires a non-representational account of central nervous system (CNS) dynamics. Towards this, we present a series of simple models in which a reservoir network with homeostatic nodes is used to control a simple agent embedded in an environment. This network spontaneously produces behaviors that are adaptive in each context, including (1) visually tracking a moving object, (2) substantially above-chance performance in the arcade game Pong, (2) and avoiding walls while controlling a mobile agent. Upon analyzing the dynamics of the networks, we find that behavioral stability can be maintained without the formation of stable or recurring patterns of network activity that could be identified as neural representations. These results may represent a useful step towards a mechanistic grounding of resonance and a view of the CNS that is compatible with ecological psychology.

Volatile and Nonvolatile Programmable Iontronic Memristor with Lithium Imbued TiOx for Neuromorphic Computing Applications.

We demonstrate a lithium (Li) imbued TiOx iontronic device that exhibits synapse-like short-term plasticity behavior without requiring a forming process beforehand or a compliance current during switching. A solid-state electrolyte lithium phosphorus oxynitride (LiPON) behaves as the ion source, and the embedding and releasing of Li ions inside the cathodic like TiOx renders volatile conductance responses from the device and offers a natural platform for hardware simulating neuron functionalities. Besides, these devices possess high uniformity and great endurance as no conductive filaments are present. Different short-term pulse-based phenomena, including paired pulse facilitation, post-tetanic potentiation, and spike rate-dependent plasticity, were observed with self-relaxation characteristics. Based on the voltage excitation period, the time scale of the volatile memory can be tuned. Temperature measurement reveals the ion displacement-induced conductance channels become frozen below 220 K. In addition, the volatile analog devices can be configured into nonvolatile memory units with multibit storage capabilities after an electroforming process. Therefore, on the same platform, we can configure volatile units as nonlinear dynamic reservoirs for performing neuromorphic training and the nonvolatile units as the weight storage layer. We proceed to use voice recognition as an example with the tunable time constant relationship and obtain 94.4% accuracy with a minimal training data set. Thus, this iontronic platform can effectively process and update temporal information for reservoir and neuromorphic computing paradigms.

High-Dimensional Physical Reservoir with Back-End-of-Line-Compatible Tin Monoxide Thin-Film Transistor.

This work demonstrates a physical reservoir using a back-end-of-line compatible thin-film transistor (TFT) with tin monoxide (SnO) as the channel material for neuromorphic computing. The electron trapping and time-dependent detrapping at the channel interface induce the SnO·TFT to exhibit fading memory and nonlinearity characteristics, the critical assets for physical reservoir computing. The three-terminal configuration of the TFT allows the generation of higher-dimensional reservoir states by simultaneously adjusting the bias conditions of the gate and drain terminals, surpassing the performances of typical two-terminal-based reservoirs such as memristors. The high-dimensional SnO TFT reservoir performs exceptionally in two benchmark tests, achieving a 94.1% accuracy in Modified National Institute of Standards and Technology handwritten number recognition and a normalized root-mean-square error of 0.089 in Mackey-Glass time-series prediction. Furthermore, it is suitable for vertical integration because its fabrication temperature is <250 °C, providing the benefit of achieving a high integration density.

Complex dynamics in nonlinear small time-delayed optoelectronic oscillator and application in fast reservoir computing and pulse generation.

We investigated a time-delayed optoelectronic oscillator (OEO) that displays a wide range of complex dynamic behavior under small time delay. The phase-space trajectory distributions in different dynamic regimes were compared which brings a new perspective on the underlying mechanism of the transition process. It was found that bifurcation is always possible no matter how small the time delay is even if the universal adiabatic approximation model is invalid. Hereby we proposed a versatile simple oscillator which has a potential capacity as memory carrier and high-dimensional state spatial mapping ability that brings 1000 times computing-efficiency improvements of reservoir computing over the large time delay one. Furthermore, we demonstrated a new approach for a tunable optoelectronic pulse generator (repetition rate at 0.2 MHz and 0.25 GHz) which depends critically on time-delayed input electrical pulse. The proposed oscillator is also a promising system for the applications of fast chaos-based communication.

Infinite-dimensional reservoir computing.

Reservoir computing approximation and generalization bounds are proved for a new concept class of input/output systems that extends the so-called generalized Barron functionals to a dynamic context. This new class is characterized by the readouts with a certain integral representation built on infinite-dimensional state-space systems. It is shown that this class is very rich and possesses useful features and universal approximation properties. The reservoir architectures used for the approximation and estimation of elements in the new class are randomly generated echo state networks with either linear or ReLU activation functions. Their readouts are built using randomly generated neural networks in which only the output layer is trained (extreme learning machines or random feature neural networks). The results in the paper yield a recurrent neural network-based learning algorithm with provable convergence guarantees that do not suffer from the curse of dimensionality when learning input/output systems in the class of generalized Barron functionals and measuring the error in a mean-squared sense.

Sparse reservoir computing with vertically coupled vortex spin-torque oscillators for time series prediction.

Spin torque nano-oscillators possessing fast nonlinear dynamics and short-term memory functions are potentially able to achieve energy-efficient neuromorphic computing. In this study, we introduce an activation-state controllable spin neuron unit composed of vertically coupled vortex spin torque oscillators and aV-Isource circuit is proposed and used to build an energy-efficient sparse reservoir computing (RC) system to solve nonlinear dynamic system prediction task. Based on micromagnetic and electronic circuit simulation, the Mackey-Glass chaotic time series and the real motor vibration signal series can be predicted by the RC system with merely 20 and 100 spin neuron units, respectively. Further study shows that the proposed sparse reservoir system could reduce energy consumption without significantly compromising performance, and a minimal response from inactivated neurons is crucial for maintaining the system's performance. The accuracy and signal processing speed show the potential of the proposed sparse RC system for high-performance and low-energy neuromorphic computing.

Reservoir computing with a random memristor crossbar array.

Physical implementations of reservoir computing (RC) based on the emerging memristors have become promising candidates of unconventional computing paradigms. Traditionally, sequential approaches by time-multiplexing volatile memristors have been prevalent because of their low hardware overhead. However, they suffer from the problem of speed degradation and fall short of capturing the spatial relationship between the time-domain inputs. Here, we explore a new avenue for RC using memristor crossbar arrays with device-to-device variations, which serve as physical random weight matrices of the reservoir layers, enabling faster computation thanks to the parallelism of matrix-vector multiplication as an intensive operation in RC. To achieve this new RC architecture, ultralow-current, self-selective memristors are fabricated and integrated without the need of transistors, showing greater potential of high scalability and three-dimensional integrability compared to the previous realizations. The information processing ability of our RC system is demonstrated in asks of recognizing digit images and waveforms. This work indicates that the 'nonidealities' of the emerging memristor devices and circuits are a useful source of inspiration for new computing paradigms.

Multi-scale dynamics by adjusting the leaking rate to enhance the performance of deep echo state networks.

Introduction: The deep echo state network (Deep-ESN) architecture, which comprises a multi-layered reservoir layer, exhibits superior performance compared to conventional echo state networks (ESNs) owing to the divergent layer-specific time-scale responses in the Deep-ESN. Although researchers have attempted to use experimental trial-and-error grid searches and Bayesian optimization methods to adjust the hyperparameters, suitable guidelines for setting hyperparameters to adjust the time scale of the dynamics in each layer from the perspective of dynamical characteristics have not been established. In this context, we hypothesized that evaluating the dependence of the multi-time-scale dynamical response on the leaking rate as a typical hyperparameter of the time scale in each neuron would help to achieve a guideline for optimizing the hyperparameters of the Deep-ESN. Method: First, we set several leaking rates for each layer of the Deep-ESN and performed multi-scale entropy (MSCE) analysis to analyze the impact of the leaking rate on the dynamics in each layer. Second, we performed layer-by-layer cross-correlation analysis between adjacent layers to elucidate the structural mechanisms to enhance the performance. Results: As a result, an optimum task-specific leaking rate value for producing layer-specific multi-time-scale responses and a queue structure with layer-to-layer signal transmission delays for retaining past applied input enhance the Deep-ESN prediction performance. Discussion: These findings can help to establish ideal design guidelines for setting the hyperparameters of Deep-ESNs.

Memristors with Tunable Volatility for Reconfigurable Neuromorphic Computing.

Neuromorphic computing promises an energy-efficient alternative to traditional digital processors in handling data-heavy tasks, primarily driven by the development of both volatile (neuronal) and nonvolatile (synaptic) resistive switches or memristors. However, despite their energy efficiency, memristor-based technologies presently lack functional tunability, thus limiting their competitiveness with arbitrarily programmable (general purpose) digital computers. This work introduces a two-terminal bilayer memristor, which can be tuned among neuronal, synaptic, and hybrid behaviors. The varying behaviors are accessed via facile control over the filament formed within the memristor, enabled by the interplay between the two active ionic species (oxygen vacancies and metal cations). This solution is unlike single-species ion migration employed in most other memristors, which makes their behavior difficult to control. By reconfiguring a single crossbar array of hybrid memristors, two different applications that usually require distinct types of devices are demonstrated - reprogrammable heterogeneous reservoir computing and arbitrary non-Euclidean graph networks. Thus, this work outlines a potential path toward functionally reconfigurable postdigital computers.

Reconfigurable Resistive Switching Memory for Telegraph Code Sensing and Recognizing Reservoir Computing Systems.

Reservoir computing (RC) system is based upon the reservoir layer, which non-linearly transforms input signals into high-dimensional states, facilitating simple training in the readout layer-a linear neural network. These layers require different types of devices-the former demonstrated as diffusive memristors and the latter prepared as drift memristors. The integration of these components can increase the structural complexity of RC system. Here, a reconfigurable resistive switching memory (RSM) capable of implementing both diffusive and drift dynamics is demonstrated. This reconfigurability is achieved by preparing a medium with a 3D ion transport channel (ITC), enabling precise control of the metal filament that determines memristor operation. The 3D ITC-RSM operates in a volatile threshold switching (TS) mode under a weak electric field and exhibits short-term dynamics that are confirmed to be applicable as reservoir elements in RC systems. Meanwhile, the 3D ITC-RSM operates in a non-volatile bipolar switching (BS) mode under a strong electric field, and the conductance modulation metrics forming the basis of synaptic weight update are validated, which can be utilized as readout elements in the readout layer. Finally, an RC system is designed for the application of reconfigurable 3D ITC-RSM, and performs real-time recognition on Morse code datasets.

Energy-Efficient Edge and Cloud Image Classification with Multi-Reservoir Echo State Network and Data Processing Units.

In an era dominated by Internet of Things (IoT) devices, software-as-a-service (SaaS) platforms, and rapid advances in cloud and edge computing, the demand for efficient and lightweight models suitable for resource-constrained devices such as data processing units (DPUs) has surged. Traditional deep learning models, such as convolutional neural networks (CNNs), pose significant computational and memory challenges, limiting their use in resource-constrained environments. Echo State Networks (ESNs), based on reservoir computing principles, offer a promising alternative with reduced computational complexity and shorter training times. This study explores the applicability of ESN-based architectures in image classification and weather forecasting tasks, using benchmarks such as the MNIST, FashionMnist, and CloudCast datasets. Through comprehensive evaluations, the Multi-Reservoir ESN (MRESN) architecture emerges as a standout performer, demonstrating its potential for deployment on DPUs or home stations. In exploiting the dynamic adaptability of MRESN to changing input signals, such as weather forecasts, continuous on-device training becomes feasible, eliminating the need for static pre-trained models. Our results highlight the importance of lightweight models such as MRESN in cloud and edge computing applications where efficiency and sustainability are paramount. This study contributes to the advancement of efficient computing practices by providing novel insights into the performance and versatility of MRESN architectures. By facilitating the adoption of lightweight models in resource-constrained environments, our research provides a viable alternative for improved efficiency and scalability in modern computing paradigms.