ABSTRACT Simultaneously sensing and recognizing motion information and time‐varying data are crucial in machine vision applications. Reservoir computing (RC) systems provide a promising framework for processing spatiotemporal signals efficiently. However, a bottleneck of fixed spatiotemporal scales restricts the dynamic application scenarios of RC systems. Here we demonstrate a photosensor based on silicon telluride (SiTe) that has gate‐tunable Schottky barrier under optical stimuli. The photocurrent and relaxation time were customized at various levels by tuning gate voltage without changing light intensity. The programmable response spatialscale and relaxation timescale are attributed to the gate modulation on Schottky barrier height and trap filling level, respectively. Controllable dynamics and time‐dependent response are leveraged to construct a multiscale optoelectronic RC system, delivering efficient and robust perception of the car motion mode and achieving a recognition accuracy of 90%. This work provides an insight of material and device design into RC system for real‐time motion recognition.

Physical reservoir computing (RC) systems have emerged as a prominent research frontier due to their exceptional efficiency in temporal information processing. However, existing implementations, predominantly utilizing resistive devices, face challenges pertaining to power efficiency and dynamic richness. Here, we propose a ferroelectric capacitor-linear capacitor (FC-LC) series device for RC implementation. By leveraging nonlinear polarization switching and back-switching, the FC-LC series device realizes two essential reservoir properties: nonlinearity and fading memory. In addition, the device exhibits an ultralow power consumption, which, along with its direct voltage readout capability, marks a significant advance over resistive reservoir devices. Moreover, the device features bidirectional operation and widely tunable time constants, thereby enhancing reservoir space dimensionality and state richness. Building upon these FC-LC series devices, a ferroelectric capacitive RC system is developed, which demonstrates superior performance in various benchmark tasks. By exploiting the bidirectional operation of the device, the RC system not only delivers enhanced performance in waveform classification but also enables highaccuracy multimodal digit recognition. Through strategically hybridizing the FC-LC series devices with varying time constants, the RC system achieves remarkable performance in Mackey-Glass time-series prediction. Our study paves the way for power-efficient, dynamicrich RC systems capable of handling diverse temporal tasks.

The scaling law of deep learning, which governs the relationship between model size and performance, has led to critical concerns regarding efficiency and sustainability. To address these challenges, this study presents a computational approach using self-organized submillimeter-long tungsten disulfide nanotube cluster as a 3D very-large-scale physical reservoir. The reservoir, with its 0D van der Waals interfaces on the order of 108, or 1.0×1010 mm-3, matches the synaptic quantity and density of the fruit fly's brain. The reservoir demonstrates the capability to perform a wide range of tasks from monomodal challenges to multimodal endeavors such as speech-to-image and medical image generation. The photosensitive mimetic synaptic connections in the very large scale reservoir emulate the optogenetic modulation of neuron circuits in in-vivo biological systems. By integrating the principles of the scaling law, multimodal task capabilities, and mimetic optogenetic mechanisms, this research paves a path toward advanced computing architectures tailored for next-generation energy-efficient artificial intelligence.

Abstract DNA and other biopolymers are being investigated as new computing substrates and alternative to silicon-based digital computers. However, the established top-down design of biomolecular interaction networks remains challenging and does not fully exploit biomolecular self-assembly capabilities. Outside the field of computation, directed evolution has been used as a tool for goal directed optimization of DNA sequences. Here, we propose integrating directed evolution with DNA-based reservoir computing to enable in-material optimization and adaptation. Simulations of colloidal bead networks connected via DNA strands demonstrate a physical reservoir capable of non-linear time-series prediction tasks, including Volterra series and Mackey–Glass chaotic dynamics. Reservoir computing performance, quantified by normalized mean squared error (NMSE), strongly depends on network topology, suggesting task-specific optimal network configurations. Implementing genetic algorithms to evolve DNA-encoded network connectivity effectively identified well-performing reservoir networks. Directed evolution improved reservoir performance across multiple tasks, outperforming random network selection. Remarkably, sequential training on distinct tasks resulted in reservoir populations maintaining performance on prior tasks. Our findings indicate that DNA-bead networks offer sufficient complexity for reservoir computing, and that directed evolution robustly optimizes performance.

Neuromorphic computing systems enable efficient and low-power information processing by emulating the structure and function of the human brain. Among various architectures, reservoir computing (RC) and spiking neural networks (SNNs)...

Reservoir Computing with Pulse-type Hardware Chaotic Neuron Model

Since rich collective dynamical behaviors tend to emerge in networked systems with various interactions between nodes, tracking control for networked coupled systems is far more complicated than that for a single system. In this work, the data-driven tracking control for the single system is extended to the case for networked systems via the control scheme of reservoir computing (RC), and we develop a RC-based control technique for complex dynamical networks, which can control the network of homogeneous nodes, as well as of heterogeneous nodes, into any desired trajectory even for partially observable node states. Numerous simulations on man-made and real networks show that the proposed control scheme is feasible and effective for different network structures and coupling strengths and has good robustness against measurement noise.

Reservoir computing (RC), a brain-inspired neuromorphic algorithm, offers simplicity and efficiency for processing spatiotemporal signals. However, conventional RC systems face limitations in handling diverse temporal scales and spatial complexities due to invariant temporal dynamics. This study introduces a temporally reconfigurable RC system utilizing ultrathin, flexible, all-solid-state electrolyte-gated thin-film transistors (UFLEX TFTs) with high performance: an on/off ratio of ≈107, endurance beyond 2.5 × 104 pulses, and low variability. UFLEX TFTs, based on molybdenum disulfide (MoS2) channels and organic-inorganic hybrid AlOx dielectrics, enable modulation of temporal dynamics via simple electrical signals. The system maintains mechanical flexibility and robust performance after bending tests. By extracting features across varied temporal and spatial scales, it achieves classification accuracies of 90.3% for CIFAR-10 object images and 81.8% for NIH chest X-ray images. This work lays a foundation for flexible neuromorphic hardware systems capable of efficient, high-performance spatiotemporal signal processing.

Model-free forecasting of rogue waves using Reservoir Computing

Understanding and predicting how mechanical systems respond to environmental variability is essential for advancing next-generation robotic systems with physical intelligence. In this study, we investigated the use of echo state networks (ESNs), a representative class of reservoir computing (RC) models, to predict the bifurcation structures of real-world mechanical systems from limited observations. We examined two representative cases: a simulated passive dynamic walking (PDW) robot with hybrid continuous-discrete dynamics and a real-world soft pneumatic artificial muscle (PAM) actuator whose electrical resistance undergoes complex changes under varying loads. To address the challenges posed by the PDW's hybrid dynamics, we proposed a hybrid ESN (HESN) model that integrates a knowledge-based touchdown detection mechanism with an ESN module. The HESN successfully reproduced the route-to-chaos bifurcation structure of the PDW, captured its multi-attractor dynamics, and demonstrated robustness against imperfect domain knowledge. For the PAM, where no reliable physical model is available, a purely data-driven ESN accurately predicted resistance bifurcations across changing environmental conditions. These results highlight the potential of RC models as flexible digital twins for mechanical systems, enabling parameter-aware modeling of bifurcations with limited training data and supporting the design of robust, adaptive robots capable of operating in complex environments.

Reservoir computing (RC) provides a training efficient alternative to recurrent neural networks by fixing recurrent weights and training only a linear readout. In hardware, physical reservoirs harness intrinsic device dynamics...

Atrial fibrillation detection using reservoir computing and anomaly detection method from automated blood pressure measurements

Synchronization of reservoir computers via transmitting invisible signals

Reservoir Computing (RC) is a novel computing framework that simplifies the complexity of recurrent neural networks. By fixing the dynamic connection weights of reservoir nodes and only training the weights of the output layer, it significantly reduces the training cost and diffculty of data prediction. This paper systematically reviews the origin and evolution of the RC method, focuses on analyzing the principles and improvement directions of RC over the past five years, and their variant models. It also provides an overview of their recent applications in time series prediction, memristor technology, and other fields, and forecasts future research trends based on current limitations.

ABSTRACT Generalization across diverse tasks is essential for neuromorphic computing hardware based on memristors. Achieving on‐memristor synaptic modulation from short‐term plasticity (STP) to dynamically tuned long‐term potentiation/depression (LTP/LTD) provides a promising path for adaptive learning, addressing the energy‐function trade‐off in neuromorphic learning systems. Here, we present a solution‐processed Ag/Sb 2 Se 3 /HZO/Si memristor fabricated, exhibiting STP, LTP/LTD, and linear conductance modulation with tunable slope. Integrated into an energy‐efficient reservoir computing (RC) system with adaptive pulse periods, the network achieves 90.33% accuracy on MNIST with enhanced energy efficiency and faster convergence. To enhance learning adaptability, we introduce a memristor‐inspired dynamic learning rate scheduling (DLRS) strategy that leverages conductance tunability for stage‐wise training adaptation. This approach enables rapid convergence and strong generalization, achieving 83.69% test accuracy on Fashion‐MNIST, and a prediction error of 1 × 10 −5 on LSTM‐based time‐series forecasting. The DLRS strategy also improves performance for adaptive neuromorphic learning across diverse driving scenarios in reinforcement learning (RL) tasks. Therefore, this work establishes a critical device‐algorithm interface, showcasing the potential of multi‐modal memristors for reconfigurable, energy‐efficient, and self‐adaptive neuromorphic hardware.

Coupling between quantum or classical degrees of freedom underpins a wide range of physical phenomena, from condensed matter systems to engineered photonic lattices. While positive coupling arising naturally from evanescent interactions has been extensively studied and employed, negative coupling unlocks unique phenomena that are challenging to realize with positive coupling alone. Exciton-polariton micropillars, celebrated for enabling topological lasers, reservoir computing, and quantum simulations, have primarily relied on positive coupling. In this work, we experimentally demonstrate negative coupling between two micropillars using an additional larger micropillar. By combining positively and negatively coupled micropillars, we construct a Su-Schrieffer-Heeger (SSH) topological lattice with a significant topological gap of approximately 15 meV, where band inversion occurs at the center of the Brillouin zone (BZ), rather than at the edges as in conventional SSH lattices. Under non-resonant excitation, we achieve polariton condensation in the in-gap topological edge states at room temperature. Our study introduces a universal method to realize negative coupling in polariton systems, paving the way for novel polaritonic devices based on lattices with arbitrarily controlled coupling signs.

As conventional computing architectures face fundamental physical limitations and the von Neumann bottleneck constrains computational efficiency, neuromorphic systems have emerged as a promising paradigm for next-generation information processing. Memristive neurons, particularly third-order circuits operating near the edge of chaos, exhibit rich neuromorphic dynamics that closely mimic biological neural activities but present significant prediction challenges due to their complex nonlinear behavior. Current approaches typically require complete system state measurements, which is often impractical in real-world neuromorphic hardware implementations where only partial state information is accessible. This paper addresses this critical limitation by proposing an innovative hybrid machine learning framework that integrates a Modified Next-Generation Reservoir Computing (MNGRC) with XGBoost regression. The core novelty lies in its dual-path prediction architecture designed specifically for partial state observability scenarios. The primary path employs NGRC to capture and forecast the system’s temporal dynamics using available state variables and input stimuli, while the secondary path leverages XGBoost as an efficient state estimator to infer unobserved state variables from minimal measurements. This strategic combination enables accurate prediction of diverse neuromorphic patterns with significantly reduced sensor requirements. Experimentally, the framework demonstrates its capability to identify and predict the complex spectrum of neuromorphic behaviors exhibited by the third-order memristive neuron. This includes accurately capturing all 18 distinct neuronal patterns, which are theoretically grounded in Hopf bifurcation analysis near the edge of chaos. Additionally, the framework successfully addresses the inverse problem of input stimulus reconstruction. By achieving accurate prediction of complex dynamics from limited states, our approach represents a key breakthrough, where full state access is often impossible, thereby addressing a critical challenge in edge AI and brain-inspired computing.

Ga2O3 thin-film transistors (TFTs) are resilient to high temperatures and voltages and are suitable for demanding display and sensing applications. Nevertheless, the performance of current Ga2O3 TFTs is constrained by defect-induced impediments to free carrier transport. This work introduces a strategy comprising nitrogen annealing followed by Al2O3 encapsulation via atomic layer deposition, which boosts the mobility and solar-blind UV responsivity of Ga2O3 TFTs by more than 27-fold and 94-fold, respectively. The combined results of high-resolution transmission electron microscopy characterization and computer-aided design simulation ascribe these enhancements to the effective passivation of deep-level defects at the interface, in the bulk, and on the surface of Ga2O3. Furthermore, the competition and synergy between photoconduction and gating in Ga2O3 TFTs yield a gate-voltage-programmable photoresponse, allowing for the control of both the responsivity and response time. Leveraging this, a solar-blind UV in-sensor reservoir computing system based on Ga2O3 TFTs is demonstrated, which achieves over 91.8% accuracy in fingerprint image recognition even under 40% noise. This work integrates an effective defect passivation strategy with a clarified modulation mechanism and further demonstrates its application in neuromorphic computing. The approach presented here shows a broad potential for extension to other wide-bandgap semiconductor systems.

Abstract The growing computational demands of artificial intelligence have accelerated the development of energy-efficient neuromorphic systems capable of processing spatiotemporal information. Reservoir computing (RC) offers a promising approach with low training complexity, particularly when implemented using emerging devices such as memristors. In this work, we present a memristor-based RC system employing vertically stacked Pt/TiO x /Au volatile memristors that inherently exhibit short-term plasticity. These devices enable temporal information encoding via pulse-driven modulation and natural relaxation. Through a modified MNIST classification task, we demonstrate that the system performance deteriorates significantly with delayed readout and small levels of device variation, highlighting the need for robust timing strategies. A virtual memristor model was also developed to evaluate system performance on the Mackey-Glass chaotic time-series forecasting task, achieving up to 93.6% prediction accuracy by tuning the internal time constant. These findings highlight the importance of precise readout control and variation resilience in the design of practical memristor-based RC systems for real-world neuromorphic applications.

A nation's currency plays a vital role in stabilizing the economy and determining exchange rates in global markets. Keeping track of the influential currencies while comparing them with one's own currency becomes essential these days. During this study, we have specifically focused on the exchange rate prediction between the United States dollar (USD) and the Canadian dollar (CAD). This pair is one of the most active currency pairs with significant economic implications for both nations. This paper studies the use of machine learning models for this specific matter, which includes long short-term memory (LSTM), gated recurrent units (GRU), and reservoir computing (RC) echo state network models. They were evaluated not only as individual models but also in various hybrid combinations. A hybrid model which combines LSTM and RC yielded better performance in terms of Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R2). Moreover, the full potential of RC represents a promising direction for future researchers to incorporate into time series analysis. In addition, considering internal and external factors that influence the exchange rate during model development would also give more accurate predictions.