Resonant-Tunnelling Diode Reservoir Computing System for Image 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 high-accuracy 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, dynamic-rich RC systems capable of handling diverse temporal tasks.

Reservoir computing has attracted much attention for its easy training process as well as its ability to deal with temporal data. A reservoir computing system consists of a reservoir part represented as a sparsely connected recurrent neural network and a readout part represented as a simple regression model. In machine learning tasks, the reservoir part is fixed and only the readout part is trained. Although reservoir computing has been mainly applied to time series prediction and recognition, it can be applied to image recognition as well by considering an image data as a sequence of pixel values. However, to achieve a high performance in image recognition with raw image data, a large-scale reservoir including a large number of neurons is required. This is a bottleneck in terms of computer memory and computational cost. To overcome this bottleneck, we propose a new method which combines reservoir computing with untrained convolutional neural networks. We use an untrained convolutional neural network to transform raw image data into a set of smaller feature maps in a preprocessing step of the reservoir computing. We demonstrate that our method achieves a high classification accuracy in an image recognition task with a much smaller number of trainable parameters compared with a previous study.

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.

Artificial neural networks built with optoelectronic synaptic devices have been proven to process visual information effectively. However, it takes great latency time and energy consumption, especially facing dynamic visual information, due to the separated optical sensor, memory, and process unit. Reservoir computing (RC) based on optoelectronic synaptic devices provides an in-sensor RC for processing temporal information efficiently. It achieves efficient computation by sensing and processing optical signals directly with optoelectronic synaptic devices. Optoelectronic synaptic devices shine in visual information processing, whose application in visual sensing and processing will provide a viable hardware solution for in-sensor computing. Therefore, the application of optoelectronic synaptic devices in reservoir computing has prompted increasing attention. Herein, for promoting the application of physical reservoir computing (PRC) with optoelectrical synapses in machine vision, synaptic plasticity will be introduced first in this work and then illustrate the basic functions of optoelectronic synapses as well as their application in reservoir computing further, and provide a perspective on PRC with optoelectronic synapses in the final.

Physical reservoir computing, which is a promising method for the implementation of highly efficient artificial intelligence devices, requires a physical system with nonlinearity, fading memory, and the ability to map in high dimensions. Although it is expected that spin wave interference can perform as highly efficient reservoir computing in some micromagnetic simulations, there has been no experimental verification to date. Herein, reservoir computing is demonstrated that utilizes multidetected nonlinear spin wave interference in an yttrium‐iron‐garnet single crystal. The subject computing system achieves excellent performance when used for hand‐written digit recognition, second‐order nonlinear dynamical tasks, and nonlinear autoregressive moving average (NARMA). It is of particular note that normalized mean square errors for NARMA2 and second‐order nonlinear dynamical tasks are 1.81 × 10−2 and 8.37 × 10−5, respectively, which are the lowest figures for any experimental physical reservoir so far reported. Said high performance is achieved with higher nonlinearity and the large memory capacity of interfered spin wave multidetection.

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.

Physical reservoir computing is a computational paradigm that enables spatiotemporal pattern recognition to be performed directly in matter. The use of physical matter leads the way toward energy‐efficient devices capable of solving machine learning problems without having to build a system of millions of interconnected neurons. Proposed herein is a high‐performance “skyrmion mixture reservoir” that implements the reservoir computing model with multidimensional inputs. This implementation solves spoken digit classification tasks with an overall model accuracy of 97.4% and a < 1% word error ratethe best performance ever reported for in materio reservoir computers. Due to the quality of the results and the low‐power properties of magnetic texture reservoirs, it is evident that skyrmion fabrics are a compelling candidate for reservoir computing.

With the growing use of artificial neural networks (ANNs) in temporal data processing tasks, the cost of training for complex ANNs is an escalating concern. Physical reservoir computing (RC), a variation of recurrent neural networks (RNNs), obviate the need for most data intensive matrix vector multiplication in the recurrent layer by evolving the RC’s internal states with the inherent nonlinear dynamics and short-term memory. In this study, we show that magnetic skyrmion confined in a fixed geometry forming the soft layer of a magnetic tunnel junction (MTJ) can work as a RC and perform autonomous long-term prediction of temporal data. Our proposed skyrmion reservoir allows for manipulation of spin dynamics with ultra energy efficient voltage controlled magnetic anisotropy modulation (VCMA) method. Furthermore, the boundary effect on the skyrmion from the geometric edges provides necessary consistency property of the reservoir. We employ our proposed reservoir for the modeling and prediction of the chaotic time series such as Mackey-Glass and dynamic time-series data, such as household building energy loads. For autonomous run, the predicted output is fed to the input of the reservoir. By comparing our spintronic physical RC approach with energy load forecasting algorithms, such as LSTMs and RNNs, we conclude that the proposed framework presents good performance in achieving high predictions accuracy, while also requiring low memory and energy both of which are at a premium in hardware resource and power constrained edge applications. Further, the proposed approach is shown to require very small training datasets and at the same time being at least 16× energy efficient compared to the sequence to sequence LSTM for accurate household load predictions. Higher endurance, fast processing and well-established technology (i.e., MTJ) to integrate with CMOS technology makes such spintronic reservoir attractive over other emergent memory device-based reservoirs.

An electrocardiogram (ECG) is a medical test that records the electrical activity of the heart over a period of time. This non-invasive and painless test makes ECG an essential tool in cardiology for diagnosing and monitoring heart health. Phonocardiogram (PCG) are typically used as an adjunct to the expertise of healthcare professionals rather than as a standalone diagnostic tool. By applying ECG signals to predict the corresponding PCG signals, no additional test for PCG is needed. These predictions can assist doctors in making informed decisions about patient care and treatment plans by merely collecting ECG signals. Physical reservoir computing (RC), as a bio-inspired algorithm, has received growing research interest. Under the term neuromorphic, physical RC processes the raw signals in the analogue domain with in-memory computing, thus reducing massive power consumption. At the same time, RC is a special type of Recurrent Neural Network (RNN) that is suitable for time-dependent signal processing. The ECG/PCG dataset we used here was experimentally collected by 10 people. We applied a recently proposed physical RC architecture called Rotating Neuron Reservoir and achieved an average NRMSE of 0.3690 for prediction, aiming to propose and construct a neuromorphic processing core for the ECG-to-PCG prediction task.

Physical reservoir computing (RC) is a recently introduced framework for information processing using the complex dynamics of physical systems. In this paper, a physical reservoir based on a molecular network of polyoxometalate (POM) decorated single-walled carbon nanotubes (SWCNT) with PBMA composite is fabricated. By the lab-built hardware platform, we experimentally demonstrate its excellent performance in a time series prediction benchmark with large short-term memory capacity (MC), indicating SWCNT/POM network as a promising substrate for reservoir computing because abundant inner charge and discharge in junctions leading to special electron transport characteristics that make rich dynamic and high-dimensional mapping properties appear in the POM-decorated SWCNT composite structure.

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.

We present a complex network of interconnected optical structures for information processing. This network is an implementation of reservoir computing, a novel method in the field of machine learning. Reservoir computing can be used for example in classification problems such as speech and image recognition, or for the generation of arbitrary patterns, tasks which are usually very hard to generalize. A nanophotonic reservoir can be constructed to perform optical signal processing. Previously, simulations demonstrated that a reservoir consisting of Semiconductor Optical Amplifiers (SOA) can outperform traditional software-based reservoirs for a speech task. Here we propose a network of coupled photonic crystal cavities. Because of the resonating behaviour, a lot of power is stored in the cavity, which gives rise to interesting nonlinear effects. Simulations are done using a novel software tool developed at Ghent University, called Caphe. We train this network of coupled resonators to generate a periodic pattern using a technique called FORCE. It is shown that photonic reservoirs can outperform classical software-based reservoirs on a pattern generation task.

Reservoir computing leverages the dynamic properties of a fixed, randomly connected neural network to facilitate simplified training and enhanced computational efficiency. Many forms of physical reservoir computers have been proposed. In this paper, we use a three-dimensional (3D)-printed reservoir computer as the design environment, develop analytic models to describe its performance, and validate the models through simulations. This approach offers practical insights for designing physical reservoirs with targeted computational capabilities and enables the assessment of the influence of reservoir parameters such as scale or material choice, on performance metrics, including speed and power consumption. Additionally, the proposed approach may be employed to optimally design physical reservoir computers to solve specific problems. This work contributes to the understanding of physical RC systems by providing a detailed analysis of the physical basis that connects computational performance with multidomain physical interactions at the device level. The methods and results from this work not only propel the development of future 3D-printed physical RC systems but also serves as a framework for evaluating and designing diverse physical RC models based on other approaches. Published by the American Physical Society 2024

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

A theoretical model for numerical simulation of the nonlinear spin-wave transient processes in magnonic active-ring oscillators (MAROs) with variable gain has been developed. The model employs the formalism of the Landau–Ginzburg equation to describe the nonlinear propagation of spectrally narrow magnonic wave packets in a magnetic film. We show that the model allows one to simulate ring's operation above the self-generation threshold as a magnonic physical reservoir computer for which the control of ring gain is employed as a method of data input into the physical reservoir. Performance of the reservoir computer was evaluated by carrying out numerical simulations using the developed model. To this end, we simulated the completion of the short-term memory and the parity-check tasks by the model. We found that the simulation results are in good agreement with experimental data. This evidences that the constructed model can be used for investigating physics underlying the performance of the MARO as a physical reservoir computer and for reservoir optimization with the final goal of maximizing reservoir performance.