Artificial Neuron Research Articles

Self‐Compliant, Variation‐Suppressed Memristor Implemented with Carbon Nanotube/hBN/Silver Nanowire Cross‐Point Structure

AbstractImplementing memristors for neuromorphic computing demands ultralow power, suppressed variations, and compact structure. Previously reported artificial neurons are mostly demonstrated with micrometer size in which the stochastic formation/rupture of conductive filaments leads to significant temporal and spatial variations. Additionally, external current compliance is commonly applied to ensure volatile switching behavior, inevitably increasing design complexity and power consumption due to auxiliary circuitry. Here, an ultra‐scaled volatile memristor is demonstrated using carbon nanotube (CNT)/hBN/silver nanowire (Ag NW) cross‐point structure with a conducting area of only 120 nm2. Owing to the nanoscale geometry for ion migration, the memristor exhibits suppressed cycle‐to‐cycle and device‐to‐device variations. Self‐compliant memristive behavior is achieved, simplifying the overall system design. Furthermore, the power consumption of the cross‐point memristor‐based neuron is drastically reduced. The results provide guidelines for tailoring the critical electrical behavior of geometry‐scaled memristor, generating practical understanding of ultra‐scaled memristor and its potential application in neuromorphic computing.

Fast gradient-free activation maximization for neurons in spiking neural networks

Elements of neural networks, both biological and artificial, can be described by their selectivity for specific cognitive features. Understanding these features is important for understanding the inner workings of neural networks. For a living system, such as a neuron, whose response to a stimulus is unknown and not differentiable, the only way to reveal these features is through a feedback loop that exposes it to a large set of different stimuli. The properties of these stimuli should be varied iteratively in order to maximize the neuronal response. To utilize this feedback loop for a biological neural network, it is important to run it quickly and efficiently in order to reach the stimuli that maximizes certain neurons’ activation with the least number of iterations possible. Here we present a framework with an efficient design for such a loop. We successfully tested it on an artificial spiking neural network (SNN), which is a model that simulates the asynchronous spiking activity of neurons in living brains. Our optimization method for activation maximization is based on the low-rank Tensor Train decomposition of the discrete activation function. The optimization space is the latent parameter space of images generated by SN-GAN or VQ-VAE generative models. To our knowledge, this is the first time that effective AM has been applied to SNNs. We track changes in the optimal stimuli for artificial neurons during training and show that highly selective neurons can form already in the early epochs of training and in the early layers of a convolutional spiking network. This formation of refined optimal stimuli is associated with an increase in classification accuracy. Some neurons, especially in the deeper layers, may gradually change the concepts they are selective for during learning, potentially explaining their importance for model performance. The source code of our framework, MANGO (for Maximization of neuronal Activation via Non-Gradient Optimization) is available on GitHub (https://github.com/iabs-neuro/mango)

Controllable frequency tunability in constriction-based spin Hall nano-oscillators for neuromorphic computing

Abstract The capability to precisely tune auto-oscillation frequency in spin Hall nano-oscillators (SHNOs) holds great promise across diverse applications, from data communication to neuromorphic computing. This study systematically investigates the controllable frequency tunability of nano-constriction-based SHNOs across distinct operational regimes. Through the independent tuning of critical magnetodynamical parameters-effective magnetization ( μ 0 M eff ) and magnetic damping (α)-we unveil varying impacts on frequency tunability with changes in out-of-plane (OOP) field strengths and constriction widths. At large OOP fields, μ 0 M eff predominantly governs frequency tunability, yielding a current tunability of 1 GHz/mA-four times greater than observed at the lowest μ 0 M eff . At low OOP fields, although we observe a remarkably high frequency-tunability of 4 GHz mA−1, μ 0 M eff alters the onset of transition from a linear-like mode to a spin-wave bullet mode. In contrast, α primarily affects the threshold current with less impact on frequency tunability. Moreover, we demonstrate that M eff-driven frequency tuning enables control over the mutual synchronization of SHNOs, mimicking synaptic-like interactions between artificial neurons. Our findings greatly extend the versatility of SHNOs for oscillator-based neuromorphic computing and provide key insights into the intricate auto-oscillation dynamics under varying OOP fields.

Low-power artificial neuron networks with enhanced synaptic functionality using dual transistor and dual memristor.

Artificial neurons with bio-inspired firing patterns have the potential to significantly improve the performance of neural network computing. The most significant component of an artificial neuron circuit is a large amount of energy consumption. Recent literature has proposed memristors as a promising option for synaptic implementation. In contrast, implementing memristive circuitry through neuron hardware presents significant challenges and is a relevant research topic. This paper describes an efficient circuit-level mixed CMOS memristor artificial neuron network with a memristor synapse model. From this perspective, the paper describes the design of artificial neurons in standard CMOS technology with low power utilization. The neuron circuit response is a modified version of the Morris-Lecar theoretical model. The suggested circuit employs memristor-based artificial neurons with Dual Transistor and Dual Memristor (DTDM) synapse circuit. The proposed neuron network produces a high spiking frequency and low power consumption. According to our research, a memristor-based Morris Lecar (ML) neuron with a DTDM synapse circuit consumes 12.55 pW of power, the spiking frequency is 22.72 kHz, and 2.13 fJ of energy per spike. The simulations were carried out using the Spectre tool with 45 nm CMOS technology.

Volatile MoS2 Memristors with Lateral Silver Ion Migration for Artificial Neuron Applications

Layered 2D semiconductors have shown enhanced ion migration capabilities along their van der Waals (vdW) gaps and on their surfaces. This effect can be employed for resistive switching (RS) in devices for emerging memories, selectors, and neuromorphic computing. To date, all lateral molybdenum disulfide (MoS2)‐based volatile RS devices with silver (Ag) ion migration have been demonstrated using exfoliated, single‐crystal MoS2 flakes requiring a forming step to enable RS. Herein, present volatile RS with multilayer MoS2 grown by metal‐organic chemical vapor deposition (MOCVD) with repeatable forming‐free operation is presented. The devices show highly reproducible volatile RS with low operating voltages of ≈2 V and fast‐switching times down to 130 ns considering their micrometer‐scale dimensions. The switching mechanism is investigated based on Ag ion surface migration through transmission electron microscopy, electronic transport modeling, and density functional theory. Finally, a physics‐based compact model is developed and the implementation of the volatile memristors as artificial neurons in neuromorphic systems is exploredd.

Real-Time Unsupervised Learning and Image Recognition via Memristive Neural Integrated Chip Based on Negative Differential Resistance of Electrochemical Metallization Cell Neuron Device.

Spiking neurons are essential for building energy-efficient biomimetic spatiotemporal systems because they communicate with other neurons using sparse and binary signals. However, the achievable high density of artificial neurons having a capacitor for emulating the integrate function of biological neurons has a limit. Furthermore, a low-voltage operation (<1.0V) is essential for connecting with modern complementary metal-oxide-semiconductor-field-effect-transistor-based (C-MOSFET-based) integrated circuits. Here, a capacitorless memristive-neural integrated chip (MnIC) based on the negative differential resistance of the electrochemical metallization cell designed using a 28-nm C-MOSFET process in a foundry is reported. The fabricated MnIC exhibits extremely low-voltage operation (<0.7V) via the rupture dynamics of Ag filaments formed in the GeS2 chalcogenide layer, with a nonlinear increase in the action potential in a manner similar to a human sensory system. Moreover, to construct a fully-structured spiking neural network (SNN), an oxygenated amorphous carbon-based (α-COx-based) synaptic device having 32 multi-level conductance states is designed. The designed MnIC and α-COx-based synaptic device demonstrate real-time unsupervised learning via a spike-timing-dependent plasticity learning rule with an SNN. Using the trained SNN, the real-time hand-written digit image of a cell phone obtained from a live webcam is successfully classified, which suggests practical applications for brain-like neuromorphic chips.

Integrated artificial neurons from metal halide perovskites.

Hardware neural networks could perform certain computational tasks orders of magnitude more energy-efficiently than conventional computers. Artificial neurons are a key component of these networks and are currently implemented with electronic circuits based on capacitors and transistors. However, artificial neurons based on memristive devices are a promising alternative, owing to their potentially smaller size and inherent stochasticity. But despite their promise, demonstrations of memristive artificial neurons have so far been limited. Here we demonstrate a fully on-chip artificial neuron based on microscale electrodes and halide perovskite semiconductors as the active layer. By connecting a halide perovskite memristive device in series with a capacitor, the device demonstrates stochastic leaky integrate-and-fire behavior, with an energy consumption of 20 to 60 pJ per spike, lower than that of a biological neuron. We simulate populations of our neuron and show that the stochastic firing allows the detection of sub-threshold inputs. The neuron can easily be integrated with previously-demonstrated halide perovskite artificial synapses in energy-efficient neural networks.

Flexible Neuromorphic Electronics for Wearable Near-Sensor and In-Sensor Computing Systems.

Flexible neuromorphic architectures that emulate biological cognitive systems hold great promise for smart wearable electronics. To realize neuro-inspired sensing and computing electronics, artificial sensory neurons that detect and process external stimuli must be integrated with central nervous systems capable of parallel computation. In near-sensor computing, synaptic devices, and sensors are used to emulate sensory neurons and receptors, respectively. In contrast, in in-sensor computing, a single multifunctional device serves as both the receptor and neuron. Bio-inspired cognitive systems efficiently detect and process stimuli through data structuring techniques, significantly reducing data volume and enabling the extension of neuromorphic applications to smart wearable systems. To construct wearable near- and in-sensor computing, it is crucial to develop artificial sensory neurons and central nervous synapses that replicate the biological functionalities. Additionally, the integrated systems must exhibit high mechanical flexibility and integration density. This review addresses research on flexible bio-inspired cognitive systems, classified into near- and in-sensor computing. It covers fundamental aspects, including biological cognitive processes, the required components, and the structures for each component, as well as applications for wearable smart systems. Finally, it offers perspectives on future research directions for flexible neuromorphic electronics in smart wearable systems connected to the next-generation Internet of Things.

On the hidden layer-to-layer topology of the representations of reality realised within neural networks

PurposeConsider an information processing algorithm that is designed to process an input data object onto an output data object via a number of successive internal {\it layers} and mappings between them. The possible activation state within each layer can be represented as a cube within Euclidean space of a high dimension (e.g. equal to the number of artificial neurons at that level). Multiple instances of such input objects produce a point cloud within each layer’s cube: this is the “representation of the reality” at that layer, as sampled by the set of input objects.Design/methodology/approachMost neural networks reduce the dimension of each layer’s cube from layer to successive layer. This gives the false impression of refining the inner representations of reality, distilling it down to fewer dimensions from which to discriminate or to infer outcomes (whatever is the aim). However, the representation of reality realised within each layer’s cube is a manifold, a curved subset embedded within it and of much lower dimension. Investigations show that such manifolds may not always be reducing in their local dimension. Instead, the manifold may become folded over and over, filling up further dimensions and creating non-realistic (unforeseeable) proximities.FindingsWe discuss some of the likely consequences of these relatively unforeseen characteristics and, in particular, the possible vulnerability of such algorithms to non-realistic perturbations. We consider a possible response to this issue.Practical implicationsNew forms of calibration are necessary, using geometric/topological loss functions, as opposed to simple (variation-limiting) regularisation terms.Originality/valueWe apply persistent homology methods to understand how the images of the point cloud (representing the sampled reality) change as they pass from layer to layer.

Polariton lattices as binarized neuromorphic networks

We introduce a novel neuromorphic network architecture based on a lattice of exciton-polariton condensates, intricately interconnected and energized through nonresonant optical pumping. The network employs a binary framework, where each neuron, facilitated by the spatial coherence of pairwise coupled condensates, performs binary operations. This coherence, emerging from the ballistic propagation of polaritons, ensures efficient, network-wide communication. The binary neuron switching mechanism, driven by the nonlinear repulsion through the excitonic component of polaritons, offers computational efficiency and scalability advantages over continuous weight neural networks. Our network enables parallel processing, enhancing computational speed compared to sequential or pulse-coded binary systems. The system’s performance was evaluated using diverse datasets, including the MNIST dataset for image recognition and the Speech Commands dataset for voice recognition tasks. In both scenarios, the proposed system demonstrates the potential to outperform existing polaritonic neuromorphic systems. For image recognition, this is evidenced by an impressive predicted classification accuracy of up to 97.5%. In voice recognition, the system achieved a classification accuracy of about 68% for the ten-class subset, surpassing the performance of conventional benchmark, the Hidden Markov Model with Gaussian Mixture Model.

Shape Anisotropy-Dependent Leaking in Magnetic Neurons for Bio-Mimetic Neuromorphic Computing.

Spiking neural networks seek to emulate biological computation through interconnected artificial neuron and synapse devices. Spintronic neurons can leverage magnetization physics to mimic biological neuron functions, such as integration tied to magnetic domain wall (DW) propagation in a patterned nanotrack and firing tied to the resistance change of a magnetic tunnel junction (MTJ), captured in the domain wall-magnetic tunnel junction (DW-MTJ) device. Leaking, relaxation of a neuron when it is not under stimulation, is also predicted to be implemented based on DW drift as a DW relaxes to a low energy position, but it has not been well explored or demonstrated in device prototypes. Here, we study DW-MTJ artificial neurons capable of leaky integrate-and-fire (LIF) behavior and demonstrate geometry-dependent leaking dynamics that results in repeatable, tunable LIF operation. Studying the behavior of five different device designs, we show tuning the geometry, stimulating fields and currents, and location of electrical contacts results in a wide range of neuron behavior. Additionally, implementation of an asymmetric notch allows for nonlinear pinning which increased expressivity without sacrificing leaking. The measured behavior is implemented in a simulated spiking neural network that outperforms a 1D model of continuous DW motion and approaches the performance of an ideal LIF activation function. The results show that the analog LIF capability of DW-MTJ neurons combines many desirable neuron functions into a single device, which can result in varied forms of multifunctional neuromorphic computing.

In-Plane Polarization-Triggered WS2-Ferroelectric Heterostructured Synaptic Devices.

To date, various kinds of memristors have been proposed as artificial neurons and synapses for neuromorphic computing to overcome the so-called von Neumann bottleneck in conventional computing architectures. However, related working principles are mostly ascribed to randomly distributed conductive filaments or traps, which usually lead to high stochasticity and poor uniformity. In this work, a heterostructure with a two-dimensional WS2 monolayer and a ferroelectric PZT film were demonstrated for memristors and artificial synapses, triggered by in-plane ferroelectric polarization. It is noted that the properties of the WS2/PZT heterostructures, including photoluminescence (PL) and conductivity, can be effectively tuned by in-plane polarization. In contrast to conventional memristors, the resistance switch of our memristors relies on the dynamic regulation of Schottky barriers at the WS2/metal contacts by ferroelectric polarization. PL characterizations verified the existence of lateral fields inside the WS2 originating from the polarization of the PZT. In particular, such memristors can emulate neuromorphic functions, including threshold-driven spiking, excitatory postsynaptic current, paired-pulse promotion (PPF), and so on. The results indicate that the WS2/PZT heterostructures with in-plane polarization are promising for the hardware implementation of artificial neural networks.

Dynamic Control of Weight-Update Linearity in Magneto-Ionic Synapses.

Multifunctional hardware technologies for neuromorphic computing are essential for replicating the complexity of biological neural systems, thereby improving the performance of artificial synapses and neurons. Integrating ionic and spintronic technologies offers new degrees of freedom to modulate synaptic potentiation and depression, introducing novel magnetic functionalities alongside the established ionic analogue behavior. We demonstrate that magneto-ionic devices can perform as synaptic elements with dynamically tunable depression linearity controlled by an external magnetic field, a functionality reminiscent of neuromodulation in biological systems. By applying magnetic fields we significantly reduce the nonlinearity of synaptic depression, transitioning from an exponential dependence to a linear response at higher fields. Neural network simulations reveal that this magnetically induced linearity enhancement improves learning accuracy across a wide range of learning rates, which is retained after the magnetic field is removed. These findings highlight the versatility and promise of magneto-ionic devices for developing tunable synaptic elements for neuromorphic hardware.

Multielement Filament Memristor Enabling Multifunctional Neuromorphic Device

AbstractMemristors exhibit changes in internal resistance in response to external voltage, introducing new functionalities to electronic devices. This enables diverse applications in non‐volatile memory, neuromorphic devices, sensors, and computing systems, highlighting their growing importance in electronics. These applications leverage various mechanisms underlying memristors. Therefore, understanding these mechanisms and discovering new memristive mechanisms are essential for overcoming implementation challenges and developing emerging applications. Here, a new type of memristor is introduced comprising an Ag/Ag:Cu‐islands/HfO2/Pt structure, characterized by a hybrid mechanism and its potential for multifunctional applications. The memristor combines the metallic filament (Ag/Cu alloy) of the electrochemical metallization (ECM) mechanism with the oxygen vacancy filament of the valence change memory (VCM) mechanism, achieving both the high on/off ratio of ECM and the analog characteristics of VCM with enhanced reliability. Both resistive and threshold switching characteristics are shown by controlling the compliance current, making the device applicable to artificial synapses and neurons. Notably, this device exhibits heat‐responsive nociceptor characteristics, positioning it as a promising candidate for next‐generation neuromorphic devices.

A Transient Photoelectric Spiking Neuron Based on a Highly Robust MgO Composite Threshold Switching Memristor for Selective UV Perception

AbstractThe biological photoreceptors in the retina convert light information into spikes, inspiring the emergence of artificial photoelectric spiking neurons. However, due to the lack of biocompatible and biodegradable characteristics, artificial photoelectric spiking neurons based on threshold switching (TS) devices are not available for bio‐integrated optical medical diagnostics and neuromorphic computing. Here, an artificial photoelectric spiking neuron integrated with a physically transient memristor and photodetector for UV perception is proposed. The transient memristor with a MgO:Mg resistive layer implemented by the co‐sputtering process of MgO and Mg targets shows highly robust TS performance, while the ZnO‐based transient photodetector can selectively detect UV light at power densities below 10 mW cm−2. More interestingly, the frequency of the firing spikes generated by artificial photoelectric spiking neuron increases with the enhancement of UV light intensity. In addition, the recognition accuracy of UV information extracted from the surrounding environment reaches ≈99.8% by spiking neural network consisting of photoelectric spiking neuron when the object that blended into the background are not easily detected. This work demonstrates that the functions of the biological photoreceptors may be truly mimicked by artificial photoelectric spiking neuron with transiency, expanding its application in optical disease diagnosis and implantable visual neuromorphic computing.

Synergistic effects of fly ash and graphene oxide composites at high temperatures and prediction using ANN and RSM approach

Fly ash (FA) is a consequence of burning coal and is widely used in construction because of its pozzolanic qualities, which increase the strength and longevity of materials. Graphene oxide (GO) is a functionalized version of graphene with low electrical conductivity, high mechanical strength, and a large surface area. By examining the behavior of fly ash and GO composites at high temperatures, new materials with improved mechanical and functional qualities that are appropriate for a range of industrial uses can be created. By improving the material quality of cement and reducing material use while boosting durability, adding graphene oxide to cement offers an opportunity to drastically lower carbon emissions. However, the entire effect is dependent on the GO emissions, manufacturing procedures, and viability from an economic standpoint; to fully reap the benefits of this novel strategy for the environment, more research and development are necessary. this paper primarily examined the effects of high temperatures on the mechanical, microstructural, and thermal properties of concrete when fly ash was replaced with 20% by weight of the cement, graphene oxide were added by 0.08% by the weight of the cement, and their combinations i.e. (20% FA + 0.08% GO) by the weight of the cement were tested at various temperatures 200, 400, 600, and 800 °C for 4 h. The ideal temperature decided by the mechanical characteristics is 200 °C, and to understand the microstructure of graphene oxide (GO) material is essential for understanding its performance, stability, and the principles underlying its behavior, particularly at elevated temperatures. Machine learning tools such as Response Surface Methodology (RSM) and Artificial Neuron Network (ANN) have been used to forecast the mechanical properties of concrete.

An organic electrochemical neuron for a neuromorphic perception system

Human perception systems are highly refined, relying on an adaptive, plastic, and event-driven network of sensory neurons. Drawing inspiration from Nature, neuromorphic perception systems hold tremendous potential for efficient multisensory signal processing in the physical world; however, the development of an efficient artificial neuron with a widely calibratable spiking range and reduced footprint remains challenging. Here, we report an efficient organic electrochemical neuron (OECN) with reduced footprint (<37 mm2) based on high-performance vertical OECT (vOECT) complementary circuitry enabled by an advanced n-type polymer for balanced p-/n-type vOECT performance. The OECN exhibits outstanding neuronal characteristics, capable of producing spikes with a widely calibratable state-of-the art firing frequency range of 0.130 to 147.1 Hz. Leveraging this capability, we develop a neuromorphic perception system that integrates mechanical sensors with the OECN and integrates them with an artificial synapse for tactile perception. The system successfully encodes tactile stimulations into frequency-dependent spikes, which are further converted into postsynaptic responses. This bioinspired design demonstrates significant potential to advance cyborg and neuromorphic systems, providing them with perceptual capabilities.

An Artificial L‐ReLU Neuron with Asymmetric Diffusive Memristor for High‐Accuracy Neuromorphic Systems

AbstractArtificial neurons with leaky rectified linear unit (L‐ReLU) function can effectively process negative information, enhancing the neuromorphic systems capbility to handle negative values. Memristive devices show great potential in building compact and bio‐plausible artificial neurons, however, a neuron device that supports L‐ReLU functions is still lacking. In this work, a compact L‐ReLU neuron is proposed based on a bipolar asymmetrical diffusive memristor. Utilizing intercalation and leveraging the migration diffusion ratio of Ag ions, devices meeting the characteristics of the L‐ReLU function are fabricated (Ag/TaOx/SiOx/Pt). Thus, the constructed neuron can encode the positive and negative input information into positive and negative spikes, respectively, with L‐ReLU‐like response function. In addition, to address the problem that the frequency only has positive values, a vector frequency (VF) is defined to describe the frequency of the positive and negative spikes. Moreover, a convolutional neural network and You‐Only‐Look‐Once version 3 (YOLOv3) network are constructed with the L‐ReLU neurons for gesture language recognition and target detection tasks, respectively. The network based on L‐ReLU neurons can avoid negative information squandering, showing a higher inference accuracy than the network with ReLU neurons. The results show that the neurons can offer a promising platform for building high‐accuracy neuromorphic systems.

Achieving neuronal dynamics with spike encoding and spatial-temporal summation in vanadium-based threshold switching memristor for asynchronous signal integration.

Artificial neuronal devices that emulate the dynamics of biological neurons are pivotal for advancing brain emulation and developing bio-inspired electronic systems. This paper presents the design and demonstration of an artificial neuron circuit based on a Pt/V/AlOx/Pt threshold switching memristor (TSM) integrated with an external resistor. By applying voltage pulses, we successfully exhibit the leaky integrate-and-fire (LIF) behavior, as well as both spatial and spatiotemporal summation capabilities, achieving the asynchronous signal integration. Notably, the Pt/V/AlOx/Pt TSM demonstrates ultrafast switching speeds (on/off times ∼165 ns/310 ns) and remarkable stability (endurance >102 cycles with cycle-to-cycle variations <2.5%). These attributes render the circuit highly suitable as a spike generator in neuromorphic computing applications. The Pt/V/AlOx/Pt TSM-based spike encoder can output current spikes at frequencies ranging from approximately 200 kHz to 800 kHz. The modulation of output spike frequency is achievable by adjusting the external resistor and capacitor within the spike encoder circuit, providing considerable operational flexibility. Additionally, the Pt/V/AlOx/Pt TSM boasts a lower threshold voltage (Vth ∼ 0.84 V) compared to previously reported VOx-based TSMs, leading to significantly reduced energy consumption for spike generation (∼2.75 nJ per spike).

Retina-inspired artificial optoelectronic neurons with broad spectral response for visual image pre-processing

Retina-inspired artificial optoelectronic neurons with broad spectral response for visual image pre-processing