Neuromorphic Research Articles

Effects of AlOx Sub‐Oxide Layer on Conductance Training of Passive Memristor for Neuromorphic Computing

AbstractMemristors are recognized as crucial devices for the hardware implementation of neuromorphic computing. The conductance training process of memristors has a direct impact on the performance of neuromorphic computing. However, memristor breakdown and conductance decay still hinder the precise training process of neural networks based on passive memristor. Here, AlOx/LiNbO3 (LN) memristors are designed by inserting a AlOx sub‐oxide layer between the single‐crystalline LN thin film with oxygen vacancies (OVs) and Pt layer. Under the same training conditions, lower conductance and self‐compliance current effects are observed in AlOx/LN memristor. Slight spontaneous decay of conductance is achieved after the removal of the external stimulation. To explore the effects of AlOx sub‐oxide layer on the prevention of device breakdown and suppression of conductance decay, the memristive mechanism of devices with and without AlOx layer is revealed via time‐of‐flight secondary ion mass spectrometer (ToF‐SIMS). It is reasonable to believe that the AlOx inserting layer in memristors can serve as a self‐compliance current layer to inhibit device breakdown and provide the OVs reservoir to suppress conductance decay. These results offer new possibilities and theoretical grounds for achieving more reliable and precise conductance training of passive memristors.

Scalable Synaptic Transistor Memory from Solution-Processed Carbon Nanotubes for High-Speed Neuromorphic Data Processing.

Neural networks as a core information processing technology in machine learning and artificial intelligence demand substantial computational resources to deal with the extensive multiply-accumulate operations. Neuromorphic computing is an emergent solution to address this problem, allowing the computation performed in memory arrays in parallel with high efficiencies conforming to the neural networks. Here, scalable synaptic transistor memories are developed from solution-sorted carbon nanotubes. The transistors exhibit a large switching ratio of over 105, a significant memory window of ≈12V arising from charge trapping, and low response delays down to tens of nanoseconds. These device characteristics endow highly stabilized reconfigurable conductance states, successful emulation of synaptic functions, and a high data processing speed. Importantly, the devices exhibit uniform characteristic metrics, e.g., with a 1.8% variation in the memory window, suggesting an industrial-scale manufacturing capability of the fabrication. Using the memories, a hardware convolution kernel is designed and parallel image processing is demonstrated at a speed of 1M bit per second per input channel. Given the efficacy of the convolution kernel, a promising prospect of the memories in implementing neuromorphic computing is envisaged. To explore the potential, large-scale convolution kernels are simulated and high-speed video processing is realized for autonomous driving.

Mott Memristors for Neuromorphics

AbstractNeuromorphic computing has emerged as a key solution for overcoming the challenge of von Neumann bottleneck, offering a pathway to more efficient and biologically inspired computing systems. A crucial advancement in this field is the utilization of Mott insulators, where the metal‐insulator transition (MIT) elicits substantial alterations in material properties, infusing renewed vigor into the progression of neuromorphic systems. This review begins by explaining the MIT mechanisms and the preparation processes of Mott insulators, followed by an introduction of Mott memristors and memristor arrays, showing different types of multidimensional integration styles. The applications of Mott memristor in neuromorphic computing are then discussed, which include artificial synapse designs and various artificial neuron architectures for sensory recognition and logic calculation. Finally, facing challenges and potential future directions are outlined for utilizing Mott memristors in the advancement of neuromorphic computing. This review aims to provide a thorough understanding of the latest advancements in Mott memristors and their applications, offering a comprehensive reference for further research in related areas, and contributing to bridging the gap between traditional silicon‐based electronics and future brain‐inspired architectures.

HARNESSING GROUP THEORY FOR SIMPLIFIED ELECTRIC CIRCUIT ANALYSIS

Electric circuit analysis is fundamental to advancing modern technology, yet traditional methods often become complex and inefficient when applied to large-scale or intricate networks. Group theory, a branch of abstract algebra renowned for its capacity to exploit symmetry, has shown promise as a solution to this challenge. By applying group theoretical principles to circuit analysis, particularly through symmetry operations and transformation groups, engineers can simplify complex systems, reducing computational demands and revealing deeper insights into circuit behavior. Through systematic symmetry-based simplifications, group theory can decrease the number of equations required by up to 40% in certain configurations, enhancing both analytical efficiency and practical implementation. However, challenges include the mathematical complexity inherent to group theory, the limitations in analyzing highly irregular or nonlinear circuits, and the need for compatible computational tools. Recommendations include the integration of group theoretical approaches into standard circuit analysis software, focused training for electrical engineers, and further research to address the application of these methods in emerging technologies such as quantum and neuromorphic computing. Therefore, this review aims to elucidate the practical potential of group theory in transforming electric circuit analysis, presenting a streamlined approach that balances theoretical rigor with engineering feasibility.

Self-Powered Infrared-Detectable BP/Ta2NiS5 Heterojunction and Its Application in Energy-Efficient Optoelectronic Synapses.

The development of energy-efficient and high-performance optoelectronic devices is crucial for the advancement of modern optoelectronic and microelectronic systems. Although the self-powered devices and optoelectronic synapses based on 2D heterojunction show great application prospects, the high energy consumption and infrared band detection of self-powered optoelectronic synapses are still an urgent problem to be solved. In this report, a BP/Ta2NiS5 heterojunction is constructed to achieve infrared detection by leveraging differences in Fermi energy levels. This heterojunction exhibits a high specific detectivity of 6.57×1010, 2.6×1010, and 1.12×1010 Jones and responsivity of 20, 10.6, and 5.9mAW-1 for 1064, 1550, and 2200nm infrared light at 0 bias voltage, respectively. In addition, under the 2200nm light, by applying an ultra-low bias voltage of 800µV, the heterojunction exhibits ultra-low power and energy consumption of 28.8pW and 0.64pJ, successfully simulates a variety of synaptic behaviors under infrared light, and demonstrates its image perception and image memory capabilities. These findings position the BP/Ta2NiS5 heterojunction as an ideal candidate for a multifunctional optoelectronic device crucial for advanced photodetection, neuromorphic computing, and artificial intelligence.

Room-Temperature Superconductivity Heats Up

Workable technology could all but eliminate energy losses while significantly advancing neuromorphic AI computing.

Solving Max-Cut Problem Using Spiking Boltzmann Machine Based on Neuromorphic Hardware with Phase Change Memory.

Efficiently solving combinatorial optimization problems (COPs) such as Max-Cut is challenging because the resources required increase exponentially with the problem size. This study proposes a hardware-friendly method for solving the Max-Cut problem by implementing a spiking neural network (SNN)-based Boltzmann machine (BM) in neuromorphic hardware systems. To implement the hardware-oriented version of the spiking Boltzmann machine (sBM), the stochastic dynamics of leaky integrate-and-fire (LIF) neurons with random walk noise are analyzed, and an innovative algorithm based on overlapping time windows is proposed. The simulation results demonstrate the effective convergence and high accuracy of the proposed method for large-scale Max-Cut problems. The proposed method is validated through successful hardware implementation on a 6-transistor/2-resistor (6T2R) neuromorphic chip with phase change memory (PCM) synapses. In addition, as an expansion of the algorithm, several annealing techniques and bias split methods are proposed to improve convergence, along with circuit design ideas for efficient evaluation of sampling convergence using cell arrays and spiking systems. Overall, the results of the proposed methods demonstrate the potential of energy-efficient and hardware-implementable approaches using SNNs to solve COPs. To the best of the author's knowledge, this is the first study to solve the Max-Cut problem using an SNN neuromorphic hardware chip.

Spin-torque nano-oscillators and their applications

Spin-torque nano-oscillators (STNOs) have emerged as an intriguing category of spintronic devices based on spin transfer torque to excite magnetic moment dynamics. The ultra-wide frequency tuning range, nanoscale size, and rich nonlinear dynamics have positioned STNOs at the forefront of advanced technologies, holding substantial promise in wireless communication, and neuromorphic computing. This review surveys recent advances in STNOs, including architectures, experimental methodologies, magnetodynamics, and device properties. Significantly, we focus on the exciting applications of STNOs, in fields ranging from signal processing to energy-efficient computing. Finally, we summarize the recent advancements and prospects for STNOs. This review aims to serve as a valuable resource for readers from diverse backgrounds, offering a concise yet comprehensive introduction to STNOs. It is designed to benefit newcomers seeking an entry point into the field and established members of the STNOs community, providing them with insightful perspectives on future developments.

Flexible TiO2-WO3−x hybrid memristor with enhanced linearity and synaptic plasticity for precise weight tuning in neuromorphic computing

Tungsten oxide (WO3)-based memristors show promising applications in neuromorphic computing. However, single-layer WO3 memristors suffer from issues such as weak memory performance and nonlinear conductance variations. In this work, a functional layer based on the hybrids of WO3−x and TiO2 is proposed for constructing flexible memristors featuring outstanding synaptic characteristics. Applying diverse electrical stimulations to the memristor enables a range of synaptic functions, elucidating its conduction mechanism through the conductive filament model. The incorporation of TiO2 not only enhances the memristor’s memory characteristics but makes its conductance more linear, symmetrical and uniform during the long-term changes. Furthermore, in view of the enhanced device performance by TiO2 doping, the potential of this device for simple behavioral simulation and processing of complex computing problems is explored. The “learning-forgetting-relearning” characteristics and device integrability are visually demonstrated. Applying the device to a convolutional neural network, the recognition accuracy of MNIST handwritten digits reaches 98.7%.

Effect of Temperature on TiO2/P‐Si‐Based Memristor Devices as Optical/Electrical Synapse

Light‐based modulation on resistive switching has opened a new route for development of the devices for advanced applications like integrated switching with memory functionality, neuromorphic computing, and humanoid robotics. This article presents the wafer‐scalable process to fabricate titanium dioxide (TiO2)/porous silicon (P‐Si)‐based heterostructures for memristor applications. The resistive switching performance of the memristor is analyzed using electrical/optical stimulation. The fabricated devices offer enhanced performance metrics during optical stimulation when compared to electrical stimulation. The memristor device offers the paired pulse facilitation (PPF) index of ≈519% (during optical stimulation), which is 3.4 times of the electrical stimulation. Further, the stability and reliability measurements are performed at different temperatures especially (room temperature to 348 K) under 450 nm illumination. The testing results prove that the developed TiO2/P‐Si‐based memristors are suitable for the next‐generation computation systems development.

Ferroelectric Polarization Enhanced Optoelectronic Synaptic Response of a CuInP2S6 Transistor Structure.

Neuromorphic computing can simulate brain function and is a pivotal element in next-generation computing, providing a potential solution to the limitations brought by the von Neumann bottleneck. Optoelectronic synaptic devices are highly promising tools for simulating biomimetic nervous systems. In this study, we developed an optoelectronic neuromorphic device with a transistor structure constructed using ferroelectric CuInP2S6. Essential synaptic behaviors in this device are observed in response to light and electrical stimuli. The optoferroelectric coupling is revealed, and the highly tunable gate modulation of the charge carrier is realized in a single device. On this basis, the light adaptation of the biological eyes and smarter Pavlovian dogs was implemented successfully and enhanced by ferroelectric polarization. The gate voltage application promotes the migration of additional Cu+ ions in the in-plane direction, thus enhancing the synaptic performance of electrical stimulation. Meanwhile, the processing ability of convolutional kernel noise images in ferroelectric devices has been achieved. Our results offer the important observation and application of ferroelectric polarization-enhanced synaptic properties of a transistor structure and have great potential in promoting the development of two-dimensional van der Waals materials and devices.

A Sliding-Kernel Computation-In-Memory Architecture for Convolutional Neural Network.

Presently described is a sliding-kernel computation-in-memory (SKCIM) architecture conceptually involving two overlapping layers of functional arrays, one containing memory elements and artificial synapses for neuromorphic computation, the other is used for storing and sliding convolutional kernel matrices. A low-temperature metal-oxide thin-film transistor (TFT) technology capable of monolithically integrating single-gate TFTs, dual-gate TFTs, and memory capacitors is deployed for the construction of a physical SKCIM system. Exhibiting an 88% reduction in memory access operations compared to state-of-the-art systems, a 32 × 32 SKCIM system is applied to execute common convolution tasks. A more involved demonstration is the application of a 5-layer, SKCIM-based convolutional neural network to the classification of the modified national institute of standards and technology (MNIST) dataset of handwritten numerals, achieving an accuracy rate of over 95%.

Ultra-Low Power Consumption Artificial Photoelectric Synapses Based on Lewis Acid Doped WSe2 for Neuromorphic Computing.

Capitalizing on the extensive spectral capacity and minimal crosstalk properties inherent in optical signals, photoelectric synapses are poised to assume a pivotal stance in the realm of neuromorphic computation. Herein, a photoelectric synapse based on Lewis acid-doped semiconducting tungsten diselenide (WSe2) is introduced, exhibiting tunable short-term and long-term plasticity. The device consumes a mere 0.1 fJ per synaptic operation, which is lower than the energy required by a single synaptic event observed in the human brain. Furthermore, these devices demonstrate high-pass filtering capabilities, highlighting their potential in image-sharpening applications. In particular, by synergistically modulating the photoconductivity and electrical gate bias, versatile logic capabilities are demonstrated within a single device, enabling it to flexibly perform both Boolean AND and OR gate operations. This work demonstrates a viable approach for Lewis acid-treated TMDs to realize multifunctional photoelectric synapses for neuromorphic computing.

120 GOPS Photonic tensor core in thin-film lithium niobate for inference and in situ training

Photonics offers a transformative approach to artificial intelligence (AI) and neuromorphic computing by enabling low-latency, high-speed, and energy-efficient computations. However, conventional photonic tensor cores face significant challenges in constructing large-scale photonic neuromorphic networks. Here, we propose a fully integrated photonic tensor core, consisting of only two thin-film lithium niobate (TFLN) modulators, a III-V laser, and a charge-integration photoreceiver. Despite its simple architecture, it is capable of implementing an entire layer of a neural network with a computational speed of 120 GOPS, while also allowing flexible adjustment of the number of inputs (fan-in) and outputs (fan-out). Our tensor core supports rapid in-situ training with a weight update speed of 60 GHz. Furthermore, it successfully classifies (supervised learning) and clusters (unsupervised learning) 112 × 112-pixel images through in-situ training. To enable in-situ training for clustering AI tasks, we offer a solution for performing multiplications between two negative numbers.

Modeling of the metal–insulator transition temperature in alio-valently doped VO2 through symbolic regression

The correlated semiconductor vanadium dioxide (VO2) exhibits an insulator–metal transition (IMT) near room temperature, which is of interest in various device applications. Precise IMT temperature control is crucial to determine the use cases across technologies such as thermochromic windows, actuators for robots or neuronal oscillators. Doping the cation or anion sites can modulate the IMT by several tens of degrees and control hysteresis. However, modeling the effects of control parameters (e.g., doping concentration, type of dopants) is challenging due to complex experimental procedures and limited data, hindering the use of traditional data-driven machine learning approaches. Symbolic regression (SR) can bridge this gap by identifying nonlinear expressions connecting key input parameters to target properties, even with small data sets. In this work, we develop SR models to capture the IMT trends in VO2 influenced by different dopant parameters. Using experimental data from the literature, our study reveals a dual nature of the IMT temperature with varying tungsten (W) doping concentrations. The symbolic model captures data trends and accounts for experimental variability, providing a complementary approach to first-principles calculations. Our feature-driven analysis across a broader class of dopants informs selectivity and provides qualitative insights into tuning phase transition properties valuable for neuromorphic computing and thermochromic windows.

Wrinkled Rhenium Disulfide for Anisotropic Nonvolatile Memory and Multiple Artificial Neuromorphic Synapses.

Two-dimensional materials are emerging as potential solutions for high-density nonvolatile memory and efficient neuromorphic computing. However, integrating multidimensional memory and an ideal linear weight updating synapse in a simple device configuration to achieve versatile biomimetic neuromorphic systems remains challenging. Here, we introduce a wrinkled rhenium disulfide (ReS2) transistor, where the wrinkled structure facilitates the carrier trapping/detrapping at the dielectric interface, thus enabling the fusion of nonvolatile memory and both electronic and optoelectronic synaptic functionalities. As a nonvolatile memory, anisotropic wrinkled ReS2 can yield three distinct sets of data across three crystal orientations under identical programming operations. Each set demonstrates exceptional retention and endurance properties. As a neuromorphic synapse, it realizes the linear and symmetric updates of conductance states up to 9 bits and 8 bits, the ultra-low-energy consumption of 75 fJ and 2.5 pJ under the electrical and optical stimuli, respectively. The artificial neural network (ANN) based on electronic synapses gives a superior recognition accuracy of 92.9% for the original handwritten digits. The anisotropic synaptic responses and multiwavelength sensitivities of optoelectronic synapses enable them to execute advanced memory and recognition functions for complex images that encompass a variety of pattern features or color information. This underscores its substantial potential for integration into efficient biomimetic visual systems.

Spike-based rotation detection using a self-powered triboelectric sensor with mimicry of a vestibular system

The biological vestibular system in semicircular canals detects head rotation with high energy efficiency through its unique spike-based processing. We propose a self-powered vestibular sensory neuron inspired by the biological vestibular system and utilizing a triboelectric nanogenerator (TENG) as a rotation sensor and a neuron transistor (neuristor) as a spiking neuron. This artificial vestibular neuron simultaneously detects the angular velocity and encodes it into spikes. When the angular velocity that is applied to the TENG increases, the spiking frequency increases, and this is a desired property of the vestibular neuron. Therefore, the vestibular neuron can serve as an input neuron in a spiking neural network (SNN). We also demonstrate a fully hardware-based neuromorphic system that is attractive for its improved energy efficiency to recognize rotation axes. Compared to a vestibular system based on conventional von-Neumann computing, that based on the proposed neuromorphic computing can effectively reduce power consumption by adopting spike transmission and employing the self-powered sensor. The proposed system is promising for applications related to rotation that demands high energy-efficiency.

Leveraging Tunability of Localized-Interfacial Memristors for Efficient Handling of Complex Neural Networks.

A scalable (<130 nm) resistive switching memristor that features both filamentary and interfacial switching aimed at neuromorphic computing is developed in this study. The typically perceived noise or volatility was effectively harnessed as a controlled mechanism for interfacial switching. The multilayer structure for the proposed memristor enhances switching stability by curbing ionic overmigration and mitigating leakage paths. Furthermore, the memristors showcased their reliability by demonstrating more than 15 M cycles in the filamentary mode and 1 M pulses in the interfacial mode. Additionally, retention tests at 85 °C for 104 s confirmed the stability across different states, affirming its reliability as a nonvolatile CMOS-compatible element. While many studies validate performance solely on the MNIST data set, this work also evaluates more complex data sets, demonstrating the robustness of the demonstrated memristor in supervised learning. Specifically, supervised learning simulations on MNIST and fashion MNIST data sets indicated a high learning rate with <4% deviations from numerical training, while offline inference trained on CIFAR-10 and CIFAR-100 data sets revealed <2.5% and <7% deviations caused by programing error accumulation, even with increased memristor counts for these highly complex data sets. Unsupervised learning via spike-timing-dependent plasticity further highlights the potential of the developed memristor in bridging artificial and biological paradigms, offering a significant advance toward efficient and biologically inspired computing architectures.

From Solid to Fluid: Novel Approaches in Neuromorphic Engineering.

Neuromorphic engineering is rapidly developing as an approach to mimicking processes in brains using artificial memristors, devices that change conductivity in response to the electrical field (resistive switching effect). Memristor-based neuromorphic systems can overcome the existing problems of slow and energy-inefficient computing that conventional processors face. In the Introduction, the basic principles of memristor operation and its applications are given. The history of switching in sandwich structures and granular metals is reviewed in the Historical Overview. Particular attention is paid to the fundamental articles from the pre-memristor era (the 1960s-70s), which demonstrated the first evidence of resistive switching and predicted the filamentary mechanism of switching. Multi-dimensionality in neuromorphic systems: Despite the powerful computational abilities of traditional memristor arrays, they cannot repeat many organizational characteristics of biological neural networks, i.e., their multi-dimensionality. This part reviews the unconventional nanowire- and nanoparticle-based neuromorphic systems that demonstrate incredible potential for use in reservoir computing due to the unique spiking change in conductance similar to firing in neurons. Liquid-based neuromorphic devices: The transition of neuromorphic systems from solid to liquid state broadens the possibilities for mimicking biological processes. In this section, ionic current memristors are reviewed and, the working principles of which bring us closer to the mechanisms of information transmittance in real synapses. Nanofluids: A novel direction in neuromorphic engineering linked to the application of nanofluids for the formation of reconfigurable nanoparticle networks with memristive properties is given in this section. The Conclusion t summarizes the bullet points of the Review and provides an outlook on the future of liquid-state neuromorphic systems.

Linearly programmable two-dimensional halide perovskite memristor arrays for neuromorphic computing.

The exotic properties of three-dimensional halide perovskites, such as mixed ionic-electronic conductivity and feasible ion migration, have enabled them to challenge traditional memristive materials. However, the poor moisture stability and difficulty in controlling ion transport due to their polycrystalline nature have hindered their use as a neuromorphic hardware. Recently, two-dimensional (2D) halide perovskites have emerged as promising artificial synapses owing to their phase versatility, microstructural anisotropy in electrical and optoelectronic properties, and excellent moisture resistance. However, their asymmetrical and nonlinear conductance changes still limit the efficiency of training and accuracy of inference. Here we achieve highly linear and symmetrical conductance changes in Dion-Jacobson 2D perovskites. We further build a 7 × 7 crossbar array based on analogue perovskite synapses, achieving a high device yield, low variation with synaptic weight storing capability, multi-level analogue states with long retention, and moisture stability over 7 months. We explore the potential of such devices in large-scale image inference via simulations and show an accuracy within 0.08% of the theoretical limit. The excellent device performance is attributed to the elimination of gaps between inorganic layers, allowing the halide vacancies to migrate homogeneously regardless of grain boundaries. This was confirmed by first-principles calculations and experimental analysis.