Neuromorphic Computing and Engineering

Inspired by the functioning of the human brain, organic synaptic transistors represent a promising avenue for developing neuromorphic technologies. However, achieving sustainability while maintaining performance and functionality remains a critical challenge. Here, we report on an innovative strategy where synthetic melanin (Mel)—a natural pigment known for its improved ionic–electronic coupling, high volumetric capacitance, and environmentally friendly characteristics—is blended with benchmark poly(3,4-ethylenedioxythiophene) polystyrene sulfonate (PEDOT:PSS) to fabricate synaptic transistors. Mel/PEDOT:PSS blends having different amounts of Mel have been systematically evaluated as semiconducting layer in organic electrochemical transistors. Our findings reveal that Mel incorporation at different concentrations enables tunable synaptic responses, such as enhanced memory retention and access to multiple memory states. These effects arise from the unique properties of Mel which modulate the charge density of PEDOT:PSS in a controlled manner. This approach demonstrates the potential for developing highly stable, multi-level memory materials for organic neuromorphic devices while addressing sustainability goals. We believe our strategy can open new avenues via the integration of natural and bio-inspired materials into organic semiconductors towards the development of sustainable neuromorphic technologies.

Epilepsy poses a significant global health challenge, driving the need for reliable diagnostic tools like scalp electroencephalogram (EEG), subscalp EEG, and intracranial EEG (iEEG) for accurate seizure detection, localization, and modulation for treating seizures. However, these techniques often rely on feature extraction techniques such as Short Time Fourier Transform (STFT) for efficiency in seizure detection. Drawing inspiration from brain architecture, we investigate biologically plausible algorithms, specifically emphasizing time-domain inputs with low computational overhead. Our novel approach features two hidden layer dendrites with Leaky Integrate-and-Fire (dLIF) spiking neurons, containing fewer than 300K parameters and occupying a mere 1.5 MB of memory. Our proposed network is tested and successfully generalized on four datasets from the USA and Europe, recorded with different front-end electronics. USA datasets are scalp EEG in adults and children, and European datasets are iEEG in adults. All datasets are from patients living with epilepsy. Our model exhibits robust performance across different datasets through rigorous training and validation. We achieved AUROC scores of 81.0% and 91.0% in two datasets. Additionally, we obtained AUPRC and F1 Score metrics of 91.9% and 88.9% for one dataset, respectively. We also conducted out-of-sample generalization by training on adult patient data, and testing on children’s data, achieving an AUROC of 75.1% for epilepsy detection. This highlights its effectiveness across continental datasets with diverse brain modalities, regardless of montage or age specificity. It underscores the importance of embracing system heterogeneity to enhance efficiency, thus eliminating the need for computationally expensive feature engineering techniques like Fast Fourier Transform (FFT) and STFT.

Percolating Networks of Nanoparticles (PNNs) are promising systems for neuromorphic computing due to their brain-like network structure and dynamics. In particular, electrical spiking in PNNs meets criteria for criticality, which is thought to be the operating point for biological brains and associated with optimal computation. Previous work showed through simulations that spiking PNNs can be used as the core stochastic component in a probabilistic computing scheme. Here, we demonstrate a route to experimental implementation of an integer factorization algorithm. We outline an important modification to the algorithm previously used and demonstrate factorization of up to six-digit integers. Finally, we explore the effect of criticality in the context of the integer factorization task by comparing critical and non-critical systems. We show significant differences in the probability distribution of states generated by the critical and non-critical systems, though all systems correctly factorize the integer of interest.

While it is important to make implantable brain-machine interfaces wireless to increase patient comfort and safety, the trend of increased channel count in recent neural probes poses a challenge due to the concomitant increase in the data rate. Extracting information from raw data at the source by using edge computing is a promising solution to this problem, with integrated intention decoders providing the best compression ratio. Recent benchmarking efforts have shown recurrent neural networks to be the best solution. Spiking Neural Networks (SNN) emerge as a promising solution for resource efficient neural decoding while Long Short Term Memory (LSTM) networks achieve the best accuracy. In this work, we show that combining traditional signal processing techniques, namely signal filtering, with SNNs improve their decoding performance significantly for regression tasks, closing the gap with LSTMs, at little added cost. Results with different filters are shown with Bessel filters providing best performance. Two block-bidirectional Bessel filters have been used–one for low latency and another for high accuracy. Adding the high accuracy variant of the Bessel filters to the output of ANN, SNN and variants provided statistically significant benefits with maximum gains of ≈5% and 8% in R² for two SNN topologies (SNN_Streaming and SNN_3D). Our work presents state of the art results for this dataset and paves the way for decoder-integrated-implants of the future.

Spiking neural networks (SNNs) express higher biological plausibility and excel at learning spatiotemporal features while consuming less energy than conventional artificial neural networks, particularly on neuromorphic hardware. The leaky integrate-and-fire (LIF) neuron stands out as one of the most widely used spiking neurons in deep learning. However, its sequential information processing leads to slow training on lengthy sequences, presenting a critical challenge for real-world applications that rely on extensive datasets. This paper introduces the parallelizable LIF (ParaLIF) neuron, which accelerates SNNs by parallelizing their simulation over time, for both feedforward and recurrent architectures. Compared to LIF in neuromorphic speech, image and gesture classification tasks, ParaLIF demonstrates speeds up to 200 times faster and, on average, achieves greater accuracy with similar sparsity. When integrated into state-of-the-art architectures, ParaLIF’s accuracy matches or exceeds the highest performance reported in the literature on various neuromorphic datasets. These findings highlight ParaLIF as a promising approach for the development of rapid, accurate and energy-efficient SNNs, particularly well-suited for handling massive datasets containing long sequences.

In contrast to the traditional principle of periodic sensing, neuromorphic engineering pursues a paradigm shift towards bio-inspired event-based sensing, where events are primarily triggered by a change in the perceived stimulus. We show in a rigorous mathematical way that information encoding by means of Threshold-Based Representation based on either Leaky Integrate-and-Fire or Send-on-Delta is linked to an analog-to-spike conversion that guarantees maximum sparsity while satisfying an approximation condition based on the Alexiewicz norm.

Neuromorphic hardware systems emulate the parallel neural networks of the human brain, and synaptic weight storage elements are crucial for enabling energy-efficient information processing. They must represent multiple data states and be able to be updated analogously. In order to realize highly controllable synaptic devices, replacing the high-k gate dielectric in conventional transistor structures with either solid-electrolytes that facilitate bulk ionic motion or ferroelectric oxide allows for steady adjustment of channel currents in response to gate-voltage signals. This approach, in turn, accelerates backpropagation algorithms used for training neural networks. Furthermore, because the channel current in electrochemical random-access memory (ECRAM) is influenced by the number of mobile ions (e.g. Li⁺, O²⁻, H⁺ or Cu⁺) passing through the electrolytes, these synaptic device candidates have demonstrated an excellent linear and symmetrical channel current response when updated using an identical pulse scheme. In the latter case, which is known as the ferroelectric field-effect transistor (FeFET), the number of electrons accumulated near the channel rapidly varies with the degree of the alignment of internal dipoles in thin doped ferroelectric HfO2. This leads to a multilevel state. Based on the working principles of these two promising candidates, enabling gate-controlled ion-transport primarily in electrolytes for ECRAM and understanding the relationship between polarization and the ferroelectric layer in FeFETs are crucial to improve their properties. Therefore, this paper aims to present our recent advances, highlighting the engineering approaches and experimental findings related to ECRAM and FeFET for three-terminal synaptic devices.

The field of neuromorphic engineering addresses the high energy demands of neural networks through brain-inspired hardware for efficient neural network computing. For on-chip learning with spiking neural networks, neuromorphic hardware requires a local learning algorithm able to solve complex tasks. Approaches based on burst-dependent plasticity have been proposed to address this requirement, but their ability to learn complex tasks has remained unproven. Specifically, previous burst-dependent learning was demonstrated on a spiking version of the ‘exclusive or’ problem (XOR) using a network of thousands of neurons. Here, we extend burst-dependent learning, termed ‘Burstprop’, to address more complex tasks with hundreds of neurons. We evaluate Burstprop on a rate-encoded spiking version of the MNIST dataset, achieving low test classification errors, comparable to those obtained using backpropagation through time on the same architecture. Going further, we develop another burst-dependent algorithm based on the communication of two types of error-encoding events for the communication of positive and negative errors. We find that this new algorithm performs better on the image classification benchmark. We also tested our algorithms under various types of feedback connectivity, establishing that the capabilities of fixed random feedback connectivity is preserved in spiking neural networks. Lastly, we tested the robustness of the algorithm to weight discretization. Together, these results suggest that spiking Burstprop can scale to more complex learning tasks and is therefore likely to be considered for self-supervised algorithms while maintaining efficiency, potentially providing a viable method for learning with neuromorphic hardware.

Spiking Neural Networks (SNNs) have emerged as a compelling, energy-efficient alternative to traditional Artificial Neural Networks (ANNs) for static image tasks such as image classification and segmentation. However, in the more complex video classification domain, SNN-based methods fall considerably short of ANN-based benchmarks due to the challenges in processing dense frame sequences. To bridge this gap, we propose ReSpike, a hybrid framework that synergizes the strengths of ANNs and SNNs to tackle action recognition tasks with high accuracy and low energy cost. By partitioning film clips into RGB image Key Frames, which primarily capture spatial information, and event-like Residual Frames, which emphasize temporal dynamics cues, ReSpike leverages ANN for processing spatial features and SNN for modeling temporal features. In addition, we propose a multi-scale cross-attention mechanism for effective feature fusion. Compared to state-of-the-art SNN baselines, our ReSpike hybrid architecture demonstrates significant performance improvements (e.g. >30% absolute accuracy improvement on HMDB-51, UCF-101, and Kinetics-400). Furthermore, ReSpike achieves comparable performance with prior ANN approaches while bringing better accuracy-energy tradeoff. Code is shared at https://github.com/Intelligent-Computing-Lab-Yale/ReSpike.

Spiking neural networks (SNNs) offer rich temporal dynamics and unique capabilities, but their training presents challenges. While backpropagation through time with surrogate gradients is the defacto standard for training SNNs, it scales poorly with long time sequences. Alternative learning rules and algorithms could help further develop models and systems across the spectrum of performance, bio-plausibility, and complexity. However, these alternatives are not consistently implemented with the same, if any, SNN framework, often complicating their comparison and use. To address this, we introduce Slax, a JAX-based library designed to accelerate SNN algorithm design and evaluation. Slax is compatible with the broader JAX and Flax ecosystem and provides optimized implementations of diverse training algorithms, enabling direct performance comparisons. Its toolkit includes methods to visualize and debug algorithms through loss landscapes, gradient similarities, and other metrics of model behavior during training. By streamlining the implementation and evaluation of novel SNN learning algorithms, Slax aims to facilitate research and development in this promising field.

Programming recurrent spiking neural networks (RSNNs) to robustly perform multi-timescale computation remains a difficult challenge. To address this, we describe a single-shot weight learning scheme to embed robust multi-timescale dynamics into attractor-based RSNNs, by exploiting the properties of high-dimensional distributed representations. We embed finite state machines into the RSNN dynamics by superimposing a symmetric autoassociative weight matrix and asymmetric transition terms, which are each formed by the vector binding of an input and heteroassociative outer-products between states. Our approach is validated through simulations with highly nonideal weights; an experimental closed-loop memristive hardware setup; and on Loihi 2, where it scales seamlessly to large state machines. This work introduces a scalable approach to embed robust symbolic computation through recurrent dynamics into neuromorphic hardware, without requiring parameter fine-tuning or significant platform-specific optimisation. Moreover, it demonstrates that distributed symbolic representations serve as a highly capable representation-invariant language for cognitive algorithms in neuromorphic hardware.

Current bionic hands lack the ability of fine force manipulation for grasping fragile objects due to missing human neuromuscular compliance in control. This incompatibility between prosthetic devices and the sensorimotor system has resulted in a high abandonment rate of hand prostheses. To tackle this challenge, we employed a neuromorphic modeling approach, biorealistic control, to regain human-like grasping ability. The biorealistic control restored muscle force regulation and stiffness adaptation using neuromorphic modeling of the neuromuscular reflex units, which was capable of real-time computing of model outputs. We evaluated the dexterity of the biorealistic control with a set of delicate grasp tasks that simulated varying challenging scenarios of grasping fragile objects in daily activities of life, including the box and block task, the glass box task, and the potato chip task. The performance of the biorealistic control was compared with that of proportional control. Results indicated that the biorealistic control with the compliance of the neuromuscular reflex units significantly outperformed the proportional control with more efficient grip forces, higher success rates, fewer break and drop rates. Post-task survey questionnaires revealed that the biorealistic control reduced subjective burdens of task difficulty and improved subjective confidence in control performance significantly. The outcome of the evaluation confirmed that the biorealistic control could achieve superior abilities in fine, accurate, and efficient grasp control for prosthetic users.

Spintronic devices -based on magnetic solitons, such as the domain wall motion and the skyrmions, have shown a significant potential for applications in energy-efficient data storage and beyond CMOS computing architectures. Based on the magnetic multilayer hetero-structures, we propose a magnetic skyrmion-magnetic tunnel junction device structure, mimicking leaky integrate and fire LIF neuron characteristics. The device is controlled by spin-orbit torque-SOT driven skyrmion motion in the ferromagnetic thin film. The modified leaky integrate and fire LIF neuron-like characteristics are shown using the combination of SOT and the skyrmion position dependence of the demagnetization energy. The device characteristics are modeled as the modified LIF neuron. The LIF neuron is one of the fundamental spiking neuron models; we integrate the model in the three-layer spiking neural network (SNN) and convolutional CSNN framework to test these spiking neuron models to classify the MNIST and FMNIST datasets. In both architectures, the network achieves classification accuracy above 97.10%. Additionally, the LIF neuron latency is in ns; thus, when integrated with the CMOS, the proposed device structures and associated systems exhibit an excellent future for energy-efficient neuromorphic computing.

This work introduces a neuromorphic compression based neural sensing architecture with address-event representation inspired readout protocol for massively parallel, next-gen wireless implantable brain machine interface (iBMI). The architectural trade-offs and implications of the proposed method are quantitatively analyzed in terms of compression ratio (CR) and spike information preservation. For the latter, we used metrics such as root-mean-square error and correlation coefficient (CC) between the original and recovered signals to assess the effect of neuromorphic compression on the spike shape. Furthermore, we use accuracy, sensitivity, and false detection rate to understand the effect of compression on downstream iBMI tasks, specifically, spike detection. We demonstrate that a data CR of 15–265 per channel can be achieved by transmitting address-event pulses for two different biological datasets. The CR further increases to 200– 50K per channel, 50 × more than in prior works, by the selective transmission of event pulses corresponding to neural spikes. A CC of ≈0.9 and spike detection accuracy of over 90% were obtained for the worst-case analysis involving 10K-channel simulated recording and typical analysis using 100 or 384-channel real neural recordings. We also analyzed the collision handling capability for up to 10K channels and observed no significant error, indicating the scalability of the proposed pipeline. We also present initial results to show the ability of intention decoders to work directly on the events generated by the neuromorphic front-end.

spiking neural networks (SNNs) and neuromorphic models are believed to be more efficient in general and have more biological realism than the activation functions typically used in deep neural networks, transformer models and generative AI. SNNs have local learning rules, are able to learn on small data sets, and can adapt through neuromodulation. Although research has shown their advantages, there are still few compelling practical applications, especially at the edge where sensors and actuators need to be processed in a timely fashion. One reason for this might be that SNNs are much more challenging to understand, build, and operate due to their intrinsic properties. For instance, the mathematical foundation involves differential equations rather than basic activation functions. To address these challenges, we have developed CARLsim++, an integrated toolbox that facilitates the creation of neuromorphic applications. It extends the highly efficient CARLsim open-source SNN simulator. CARLsim++ encapsulates the mathematical intrinsics and low-level C++ programming by providing a graphical user interface for users to easily create their SNNs and a means to configure sensors and actuators for robotics and other edge devices. These can be accurately simulated before deploying on physical devices. CARLsim++ can lead to rapid development of neuromorphic applications for simulation or edge processing. We introduce CARLsim++ with a closed loop robotic demonstration using neuromorphic computing.

Event-based machine learning promises more energy-efficient AI on future neuromorphic hardware. Here, we investigate how the recently discovered Eventprop algorithm for gradient descent on exact gradients in spiking neural networks (SNNs) can be scaled up to challenging keyword recognition benchmarks. We implemented Eventprop in the GPU-enhanced neural networks framework (GeNN) and used it for training recurrent SNNs on the Spiking Heidelberg Digits (SHD) and Spiking Speech Commands (SSC) datasets. We found that learning depended strongly on the loss function and extended Eventprop to a wider class of loss functions to enable effective training. We then tested a large number of data augmentations and regularisations as well as exploring different network structures; and heterogeneous and trainable timescales. We found that when combined with two specific augmentations, the right regularisation and a delay line input, Eventprop networks with one recurrent layer achieved state-of-the-art performance on SHD and good accuracy on SSC. In comparison to a leading surrogate-gradient-based SNN training method, our GeNN Eventprop implementation is 3× faster and uses 4× less memory. This work is a significant step towards a low-power neuromorphic alternative to current machine learning paradigms.

One of the ways neuromorphic applications emulate the processing performed by the brain is by using spikes as inputs instead of time-varying analog stimuli. Therefore, these time-varying stimuli have to be encoded into spikes, which can induce important information loss. To alleviate this loss, some studies use population coding strategies to encode more information using a population of neurons rather than just one neuron. However, configuring the encoding parameters of such a population is an open research question. This work proposes an approach based on maximizing the mutual information between the signal and the spikes in the population of neurons. The proposed algorithm is inspired by the information-theoretic framework of Partial Information Decomposition. Two applications are presented: blood pressure pulse wave classification, and neural action potential waveform classification. In both tasks, the data is encoded into spikes and the encoding parameters of the neuron populations are tuned to maximize the encoded information using the proposed algorithm. The spikes are then classified and the performance is measured using classification accuracy as a metric. Two key results are reported. First, adding neurons to the population leads to an increase in both mutual information and classification accuracy beyond what could be accounted for by each neuron separately, showing the usefulness of population coding strategies. Second, the classification accuracy obtained with the tuned parameters is near-optimal and it closely follows the mutual information as more neurons are added to the population. Furthermore, the proposed approach significantly outperforms random parameter selection, showing the usefulness of the proposed approach. These results are reproduced in both applications.

Current bionic hands lack the ability of fine force manipulation for grasping fragile objects due to missing human neuromuscular compliance in control. This incompatibility between prosthetic devices and the sensorimotor system has resulted in a high abandonment rate of hand prostheses. To tackle this challenge, we employed a neuromorphic modeling approach, biorealistic control, to regain human-like grasping ability. The biorealistic control restored muscle force regulation and stiffness adaptation using neuromorphic modeling of the neuromuscular reflex units, which was capable of real-time computing of model outputs. We evaluated the dexterity of the biorealistic control with a set of delicate grasp tasks that simulated varying challenging scenarios of grasping fragile objects in daily activities of life, including the Box and Block Task, the Glass Box Task, and the Potato Chip Task. The performance of the biorealistic control was compared with that of proportional control. Results indicated that the biorealistic control with the compliance of the neuromuscular reflex units significantly outperformed the proportional control with more efficient grip forces, higher success rates, fewer break and drop rates. Post-task survey questionnaires revealed that the biorealistic control reduced subjective burdens of task difficulty and improved subjective confidence in control performance significantly. The outcome of the evaluation confirmed that the biorealistic control could achieve superior abilities in fine, accurate, and efficient grasp control for prosthetic users.

Spiking Neural Networks (SNNs) offer rich temporal dynamics and unique capabilities, but their training presents challenges. While backpropagation through time (BPTT) with surrogate gradients is the defacto standard for training SNNs, it scales poorly with long time sequences. Alternative learning rules and algorithms could help further develop models and systems across the spectrum of performance, bio-plausibility, and complexity. However, implementing and evaluating these alternatives at scale is cumbersome and error-prone, requiring repeated reimplementations. To address this, we introduce Slax, a JAX-based library designed to accelerate SNN algorithm design and evaluation. Slax is compatible with the broader JAX and Flax ecosystem and provides optimized implementations of diverse training algorithms, enabling direct performance comparisons. Its toolkit includes methods to visualize and debug algorithms through loss landscapes, gradient similarities, and other metrics of model behavior during training. By streamlining the implementation and evaluation of novel SNN learning algorithms, Slax aims to facilitate research and development in this promising field.

The effort addresses the research activity around the usage of non-volatile memories (NVM) for storage of ‘weights’ in neural networks and the resulting computation through these memory crossbars. In particular, we focus on the CMOS implementations of, and comparisons between, memristor/resistive random access memory (RRAM) devices, and floating-gate (FG) devices. A historical perspective for illustrating FG and memristor/RRAM devices enables comparison of nonvolatile storage (addressing issues related to resolution, lifetime, endurance etc), feedforward computation (different variants of vector matrix multiplication, tradeoffs between power dissipation and signal to noise ratio etc), programming (addressing issues of selectivity, peripheral circuits, charge trapping etc), and learning algorithms (continuous time LMS or batch update), in these systems. We believe this historical perspective is necessary and timely given the increasing interest in using computation in memory with NVM for a wide variety of memory bound applications.

Neuromorphic computing has recently gained significant attention as a promising approach for developing energy-efficient, massively parallel computing systems inspired by the spiking behavior of the human brain and natively mapping spiking neural networks (SNNs). Effective training algorithms for SNNs are imperative for increased adoption of neuromorphic platforms; however, SNN training continues to lag behind advances in other classes of ANN. In this paper, we reduce this gap by proposing an innovative encoder–decoder technique that leverages sparse coding and the locally competitive algorithm (LCA) to provide an algorithm specifically designed for neuromorphic platforms. Using our proposed Dataset-Scalable Exemplar LCA-Decoder we reduce the computational demands and memory requirements associated with training SNNs using error backpropagation methods on increasingly larger training sets. We offer a solution that can be scalably applied to datasets of any size. Our results show the highest reported top-1 test accuracy using SNNs on the ImageNet and CIFAR100 datasets, surpassing previous benchmarks. Specifically, we achieved a record top-1 accuracy of 80.75% on ImageNet (ILSVRC2012 validation set) and 79.32% on CIFAR100 using SNNs.

Recently, the use of bio-inspired learning techniques such as Hebbian learning and its closely-related spike-timing-dependent plasticity (STDP) variant have drawn significant attention for the design of compute-efficient AI systems that can continuously learn on-line at the edge. A key differentiating factor regarding this emerging class of neuromorphic continual learning system lies in the fact that learning must be carried using a data stream received in its natural order, as opposed to conventional gradient-based offline training, where a static training dataset is assumed available a priori and randomly shuffled to make the training set independent and identically distributed (i.i.d). In contrast, the emerging class of neuromorphic CL systems covered in this survey must learn to integrate new information on the fly in a non-i.i.d manner, which makes these systems subject to catastrophic forgetting. In order to build the next generation of neuromorphic AI systems that can continuously learn at the edge, a growing number of research groups are studying the use of sparse and predictive Coding (PC)-based Hebbian neural network architectures and the related spiking neural networks (SNNs) equipped with STDP learning. However, since this research field is still emerging, there is a need for providing a holistic view of the different approaches proposed in the literature so far. To this end, this survey covers a number of recent works in the field of neuromorphic CL based on state-of-the-art sparse and PC technology; provides background theory to help interested researchers quickly learn the key concepts; and discusses important future research questions in light of the different works covered in this paper. It is hoped that this survey will contribute towards future research in the field of neuromorphic CL.

Speech enhancement improves communication in noisy environments, affecting areas such as automatic speech recognition (ASR), hearing aids, and telecommunications. With these domains typically being power-constrained and event-based, and often requiring low latency, neuromorphic algorithms–particularly spiking neural networks (SNNs)–hold significant potential. However, current effective SNN solutions require a long temporal window to calculate Short Time Fourier Transforms (STFTs) and thus impose substantial latency, typically around 32 ms, which is too long for applications such as hearing aids. Inspired by the Dual-Path Recurrent Neural Network (DPRNN) in deep neural networks (DNNs), we develop a two-phase time-domain streaming SNN fframework for speech enhancement, named Dual-Path Spiking Neural Network (DPSNN). DPSNNs achieve low latency by replacing the STFT and inverse STFT (iSTFT) in traditional frequency-domain models with a learned convolutional encoder and decoder. In the DPSNN, the first phase uses Spiking Convolutional Neural Networks (SCNNs) to capture temporal contextual information, while the second phase uses Spiking Recurrent Neural Networks (SRNNs) to focus on frequency-related features. In addition, threshold-based activation suppression, along with L1 regularization loss, is applied to specific non-spiking layers in DPSNNs to further improve their energy efficiency. Evaluating on the Voice Cloning Toolkit (VCTK) Corpus and Intel N-DNS Challenge dataset, our approach demonstrates excellent performance in speech objective metrics, along with the very low latency (approximately 5 ms) required for applications like hearing aids.

Artificial nociceptors, mimicking human-like stimuli perception, are of significance for intelligent robotics to work in hazardous and dynamic scenarios. One of the most essential characteristics of the human nociceptor is its self-adjustable attribute, which indicates that the threshold of determination of a potentially hazardous stimulus relies on environmental knowledge. This critical attribute has been currently omitted, but it is highly desired for artificial nociceptors. Inspired by these shortcomings, this article presents, for the first time, a self-directed channel memristor-based self-reconfigurable nociceptor, capable of perceiving hazardous pressure stimuli under different temperatures and demonstrates key features of tactile nociceptors, including ‘threshold,’ ‘no-adaptation,’ and ‘sensitization.’ The maximum amplification of hazardous external stimuli is 1000%, and its response characteristics dynamically adapt to current temperature conditions by automatically altering the generated modulation schemes for the memristor. The maximum difference ratio of the response of memristors at different temperatures is 500%, and this adaptability closely mimics the functions of biological tactile nociceptors, resulting in accurate danger perception in various conditions. Beyond temperature adaptation, this memristor-based nociceptor has the potential to integrate different sensory modalities by applying various sensors, thereby achieving human-like perception capabilities in real-world environments.