Ultralow-power reservoir computing based on bidirectionally operable ferroelectric capacitors with tunable time constants

In‐sensor computing hardware with signal sensing and dynamic signal processing capabilities—inspired by the human sensory system—have attracted interest as the rapid proliferation of data sensing and computation in the era of big data. Here, an in‐sensor reservoir computing (RC) system with integrated functions of signal sensing, preprocessing, and RC in a single device is developed, by constructing a full quantum dot optoelectronic memristor (FQDOM) based on a ZnO QDs/CdSe QDs/ZnO QDs heterojunction. The device shows nonlinear and short‐term memory behavior in response to both electrical stimuli and optical signals covering the spectra from ultraviolet to red light. The FQDOM can achieve color perception. The in‐sensor RC system can sense varied types of input signal, exhibiting multisensory fusion capability by fashion‐MNIST classification. More importantly, these proposed FQDOMs can preprocess signal and then send for RC in sensor, which effectively reduces the bit error rate of RC systems. The processing capability of the in‐sensor RC system for temporal signals is further demonstrated by a dynamic gesture perception recognition task, showing an accuracy of up to 92.59%. This research provides an effective way to design advanced in‐sensor computational systems.

Analysis and fully memristor-based reservoir computing for temporal data classification

Reservoir computing based encryption-then-compression scheme of image achieving lossless compression

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.

A neuromorphic reservoir computing (RC) system using a semiconductor nanolaser (SNL) with double phase conjugate feedbacks (PCF) is proposed for the first time and demonstrated numerically. The prediction performance of such RC system is investigated via Santa Fe chaotic time series prediction task. The Purcell cavity-enhanced spontaneous emission factor F and the spontaneous emission coupling factor β are included in the rate equations, and the influences of F and β on the prediction performance of such RC system are analyzed extensively. For the purpose of comparison, the prediction performance of SNL-based RC system with single PCF is also considered. The simulation results indicate that, compared with the SNL-based RC system with single PCF, enhanced prediction performance can be obtained for the SNL-based RC system with double PCF. Moreover, the influences of bias current, the modulation depth of input signal, feedback strength, as well as feedback delay, are also taken into account. The proposed SNL-based RC system subject to double PCF in this paper has the potential to develop the RC-based neuromorphic photonic integrated circuit.

Real-time processing of space-and-time-variant signals is imperative for perception and real-world problem-solving. In the brain, spatio-temporal stimuli are converted into spike trains by sensory neurons and projected to the neurons in subcortical and cortical layers for further processing. Reservoir Computing (RC) is a neural computation paradigm that is inspired by cortical Neural Networks (NN). It is promising for real-time, on-line computation of spatio-temporal signals. An RC system incorporates a Recurrent Neural Network (RNN) called reservoir, the state of which is changed by a trajectory of perturbations caused by a spatio-temporal input sequence. A trained, non- recurrent, linear readout-layer interprets the dynamics of the reservoir over time. Echo-State Network (ESN) [1] and Liquid-State Machine (LSM) [2] are two popular and canonical types of RC system. The former uses non-spiking analog sigmoidal neurons – and, more recently, Leaky Integrator (LI) neurons – and a normalized random connectivity matrix in the reservoir. Whereas, the reservoir in the latter is composed of Leaky Integrate-and-Fire (LIF) neurons, distributed in a 3-D space, which are connected with dynamic synapses through a probability function. The major difference between analog neurons and spiking neurons is in their neuron model dynamics and their inter-neuron communication mechanism. However, RC systems share a mysterious common property: they exhibit the best performance when reservoir dynamics undergo a criticality [1–6] – governed by the reservoirs’ connectivity parameters, |λmax| ≈ 1 in ESN, λ ≈ 2 and w in LSM – which is referred to as the edge of chaos in [3–5]. In this study, we are interested in exploring the possible reasons for this commonality, despite the differences imposed by different neuron types in the reservoir dynamics. We address this concern from the perspective of the information representation in both spiking and non-spiking reservoirs. We measure the Mutual Information (MI) between the state of the reservoir and a spatio-temporal spike-trains input, as well as that, between the reservoir and a linearly inseparable function of the input, temporal parity. In addition, we derive Mean Cumulative Mutual Information (MCMI) quantity from MI to measure the amount of stable memory in the reservoir and its correlation with the temporal parity task performance. We complement our investigation by conducting isolated spoken-digit recognition and spoken-digit sequence-recognition tasks. We hypothesize that a performance analysis of these two tasks will agree with our MI and MCMI results with regard to the impact of stable memory in task performance. It turns

Reservoir computing as one of the artificial neural networks can process input signals in various ways, thereby showing strength in modeling data that changes over time. The reservoir is utilized in various fields because it is particularly energy efficient in learning and can exhibit powerful performance with relatively few trainings cost. This study emphasizes the significant advantages of the hafnium zirconium oxide (HZO) film in reservoir applications by controlling the depolarization field. The decay time of HZO-based ferroelectric memory devices is investigated, focusing on the impact of both ferroelectric layer thickness and interlayer (IL) thickness on physical reservoir computing system. Devices with HZO film thicknesses of 10, 15, and 20 nm were fabricated and characterized. Among these, the 15 nm HZO film demonstrated optimal thickness, exhibiting excellent ferroelectric properties, including enhanced orthorhombic phase (o-phase) formation and reliable short-term memory characteristics. When the optimized device for decay time is integrated into a reservoir computing system, it achieved a remarkable average accuracy of 93.42% in image recognition tasks, emphasizing its capability for high-precision computations.

In this work, we report a flexible doped hafnium oxide memcapacitor for building a noise-robust reservoir computing (RC) system. The nonlinearity and fading memory properties required for RC could be obtained from the capacitively coupled polarization switching and charge trapping in devices. The flexible oxide-based RC system shows a maximum accuracy of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\sim$</tex-math> </inline-formula> 91.4% in the music genres (GTZAN Genre Collection dataset) classification task. The high noise robustness of the system has been verified by adding different signal-to-noise ratios (SNRs) of environment noise (NOISEX-92 dataset) to the raw music. The accuracy remains over 87% even with an SNR of 0 dB with respect to six types of noises. Our results are expected to open up opportunities for flexible electronics with high computing capability.

Reservoir computing (RC) has attracted attention as an efficient recurrent neural network architecture due to its simplified training, requiring only its last perceptron readout layer to be trained. When implemented with memristors, RC systems benefit from their dynamic properties, which make them ideal for reservoir construction. However, achieving high performance in memristor-based RC remains challenging, as it critically depends on the input preprocessing method and reservoir size. Despite growing interest, a comprehensive evaluation that quantifies the impact of these factors is still lacking. This paper systematically compares various preprocessing methods for memristive RC systems, assessing their effects on accuracy and energy consumption. We also propose a parity-based preprocessing method that improves accuracy by 2-6% while requiring only a modest increase in device count compared to other methods. Our findings highlight the importance of informed preprocessing strategies to improve the efficiency and scalability of memristive RC systems.

Reservoir computing (RC) is a computational framework that utilizes a nonlinear, memory-based kernel to map time series inputs into a high-dimensional space for information processing. Most hardware-based RC systems treat their physical reservoirs as black boxes, with the roles of their complex nonlinear dynamics empirically understood and adjusted by tuning physical conditions. However, theoretically understanding the reservoir's nonlinearity in relation to its computational capabilities is crucial for ensuring consistent performance and adaptive system design. In this paper, we propose a magnetization-saturation-based physical reservoir with interpretable nonlinearity, where spin dynamics follows a hyperbolic tangent function and its nonlinear scaling is analytically governed by the damping constant of the magnetic material. Our spin reservoir consists of a magnetic thin film segmented into multiple subregions with different damping constants. This design enables not only nonlinearity diversity through a variety of nonlinear scaling but also short-term memory through hysteresis interactions of neighboring spins. We numerically validate our spin RC by demonstrating its high performance in nonlinear waveform transformation and temporal exclusive-OR (XOR) tasks, which aligns with our theoretical understanding of the system. We expect this work to contribute to the advancement of reliable and configurable spintronic RC systems. Published by the American Physical Society 2025

Physical reservoir computing utilizes a physical system as a computational resource. This nontraditional computing technique can be computationally powerful, without the need of costly training. Here, a Hopf oscillator is implemented as a reservoir computer by using a node-based architecture; however, this implementation does not use delayed feedback lines. This reservoir computer is still powerful, but it is considerably simpler and cheaper to implement as a physical Hopf oscillator. A non-periodic stochastic masking procedure is applied for this reservoir computer following the time multiplexing method. Due to the presence of noise, the Euler–Maruyama method is used to simulate the resulting stochastic differential equations that represent this reservoir computer. An analog electrical circuit is built to implement this Hopf oscillator reservoir computer experimentally. The information processing capability was tested numerically and experimentally by performing logical tasks, emulation tasks, and time series prediction tasks. This reservoir computer has several attractive features, including a simple design that is easy to implement, noise robustness, and a high computational ability for many different benchmark tasks. Since limit cycle oscillators model many physical systems, this architecture could be relatively easily applied in many contexts.

We numerically investigate the performance of Chinese herbal medicine (CHM) recognition using a time delay reservoir computing (TDRC) system consist of a vertical-cavity surface-emitting laser (VCSEL) subjected to optical feedback. A database of 960 images from 12 CHM categories is used to evaluate the recognition performance, most of which contain many CHM samples and have different backgrounds. The CHM image feature vector is masked and optically injected into the VCSEL, and the VCSEL output in two polarization modes are sampled as the virtual node states for readout. After testing the CHM recognition performance of TDRC under injecting the same information into two polarization modes of the VCSEL, we further test the performance of parallel processing by injecting half of the image information into each of the two polarization modes. By concatenating the virtual node states collected from two modes for readout, the recognition speed is doubled. Meanwhile, the influence of key parameters on recognition performance are investigated and the relationship between system idle state and recognition performance is also examined. The results show that, under parallel processing and orthogonal optical feedback, the TDRC achieves the minimum recognition error of 1.7% at a reservoir processing rate of 2×10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">9</sup> images per second.

Detection and recognition of latent fingerprints play crucial roles in identification and security. However, the separation of sensor, memory, and processor in conventional ex-situ fingerprint recognition system seriously deteriorates the latency of decision-making and inevitably increases the overall computing power. In this work, a photoelectronic reservoir computing (RC) system, consisting of DUV photo-synapses and nonvolatile memristor array, is developed to detect and recognize the latent fingerprint with in-sensor and parallel in-memory computing. Through the Ga-rich design, we achieve amorphous GaOx (a-GaOx) photo-synapses with an enhanced persistent photoconductivity (PPC) effect. The PPC effect, which induces nonlinearly tunable conductivity, renders the a-GaOx photo-synapses an ideal deep ultraviolet (DUV) photoelectronic reservoir, thus mapping the complex input vector into a dimensionality-reduced output vector. Connecting the reservoirs and a memristor array, we further construct an in-sensor RC system for latent fingerprint identification. The system maintains over 90% recognition accuracy for latent fingerprint within 15% stochastic noise level via the proposed dual-feature strategy. This work provides a subversive prototype system of DUV in-sensor RC for highly efficient recognition of latent fingerprints.

Reservoir computing (RC) systems are computational tools for information processing that can be fully implemented in optics. Here, we experimentally and numerically show that an optically pumped laser subject to optical delayed feedback can yield similar results to those obtained for electrically pumped lasers. Unlike with previous implementations, the input data are injected at a time interval that is much larger than the time-delay feedback. These data are directly coupled to the feedback light beam. Our results illustrate possible new avenues for RC implementations for prediction tasks.

Improving the processing rate of VCSEL-based reservoir computing through adopting variable polarization information injection