A Nanoscale Bistable Resistor for an Oscillatory Neural Network.

Coupled oscillators construct an oscillatory neural network (ONN) by mimicking the interactions among neurons in the human brain. This work demonstrates a fully CMOS-based oscillator consisting of a bistable resistor (biristor), which shares a structure identical with that of a metal-oxide-semiconductor field-effect transistor, except for the use of a gate electrode. The biristor-based oscillator (birillator) generates oscillating voltage signals in the form of spikes due to a single transistor latch phenomenon. When two birillators are connected with a coupling capacitor, they become synchronized with a phase difference of 180°. These coupled oscillation characteristics are experimentally investigated for an ONN. As practical applications of the ONN with coupled birillators, edge detection and vertex coloring are conducted by encoding information into phase differences between them. The proposed fully CMOS-based birillators are advantageous for low power consumption, high CMOS compatibility, and a compact footprint area.

Bio-Inspired Organic Synaptor with In Situ Ion-Doped Ultrathin Polyelectrolyte Containing Acetylcholine-Like Cation.

An ion-based synaptic transistor (synaptor) is designed to emulate a biological synapse using controlled ion movements. However, developing a solid-state electrolyte that can facilitate ion movement while achieving large-scale integration remains challenging. Here, a bio-inspired organic synaptor (BioSyn) with an in situ ion-doped polyelectrolyte (i-IDOPE) is demonstrated. At the molecular scale, a polyelectrolyte containing the tert-amine cation, inspired by the neurotransmitter acetylcholine is synthesized using initiated chemical vapor deposition (iCVD) with in situ doping, a one-step vapor-phase deposition used to fabricate solid-state electrolytes. This method results in an ultrathin, but highly uniform and conformal solid-state electrolyte layer compatible with large-scale integration, a form that is not previously attainable. At a synapse scale, synapse functionality is replicated, including short-term and long-term synaptic plasticity (STSP and LTSP), along with a transformation from STSP to LTSP regulated by pre-synaptic voltage spikes. On a system scale, a reflex in a peripheral nervous system is mimicked by mounting the BioSyns on various substrates such as rigid glass, flexible polyethylene naphthalate, and stretchable poly(styrene-ethylene-butylene-styrene) for a decentralized processing unit. Finally, a classification accuracy of 90.6% is achieved through semi-empirical simulations of MNIST pattern recognition, incorporating the measured LTSP characteristics from the BioSyns.

2Memristor-1Capacitor Integrated Temporal Kernel for High-Dimensional Data Mapping.

Compact but precise feature-extracting ability is core to processing complex computational tasks in neuromorphic hardware. Physical reservoir computing (RC) offers a robust framework to map temporal data into a high-dimensional space using the time dynamics of a material system, such as a volatile memristor. However, conventional physical RC systems have limited dynamics for the given material properties, restricting the methods to increase their dimensionality. This study proposes an integrated temporal kernel composed of a 2-memristor and 1-capacitor (2M1C) using a W/HfO2/TiN memristor and TiN/ZrO2/Al2O3/ZrO2/TiN capacitor to achieve higher dimensionality and tunable dynamics. The kernel elements are carefully designed and fabricated into an integrated array, of which performances are evaluated under diverse conditions. By optimizing the time dynamics of the 2M1C kernel, each memristor simultaneously extracts complementary information from input signals. The MNIST benchmark digit classification task achieves a high accuracy of 94.3% with a (196×10) single-layer network. Analog input mapping ability is tested with a Mackey-Glass time series prediction, and the system records a normalized root mean square error of 0.04 with a 20×1 readout network, the smallest readout network ever used for Mackey-Glass prediction in RC. These performances demonstrate its high potential for efficient temporal data analysis.

A Bioinspired Artificial Gustatory Neuron for a Neuromorphic Based Electronic Tongue.

A novel biomimicked neuromorphic sensor for an energy efficient and highly scalable electronic tongue (E-tongue) is demonstrated with a metal-oxide-semiconductor field-effect transistor (MOSFET). By mimicking a biological gustatory neuron, the proposed E-tongue can simultaneously detect ion concentrations of chemicals on an extended gate and encode spike signals on the MOSFET, which acts as an input neuron in a spiking neural network (SNN). Such in-sensor neuromorphic functioning can reduce the energy and area consumption of the conventional E-tongue hardware. pH-sensitive and sodium-sensitive artificial gustatory neurons are implemented by using two different sensing materials: Al2O3 for pH sensing and sodium ionophore X for sodium ion sensing. In addition, a sensitivity control function inspired by the biological sensory neuron is demonstrated. After the unit device characterization of the artificial gustatory neuron, a fully hardware-based E-tongue that can classify two distinct liquids is demonstrated to show a practical application of the artificial gustatory neurons.

A Vertical Silicon Nanowire Based Single Transistor Neuron with Excitatory, Inhibitory, and Myelination Functions for Highly Scalable Neuromorphic Hardware.

A single transistor neuron (1T-neuron) is demonstrated by using a vertically protruded nanowire from an 8 in. silicon (Si) wafer. The 1T-neuron adopts a gate-all-around structure to completely surround the Si nanowire (Si-NW) to make a floating body and allow aggressive downscaling. The Si-NW is composed of an n+ drain at the top, n+ source at the bottom, and p-type floating body at the middle, which are self-aligned vertically. Thus, it occupies a small footprint area. The gate controls an excitatory/inhibitory function. In addition, myelination of a biological neuron that changes membrane capacitance is mimicked by an inherently asymmetric source/drain structure. Two spiking frequencies at the same input current are controlled by whether the neuron is myelinated or unmyelinated. Using the vertical 1T-neuron, pattern recognition is demonstrated with both measurements and semiempirical circuit simulations. Furthermore, handwritten numbers in the MNIST database are recognized with accuracy of 93% by software-based simulations. Applicability of the vertical 1T-neuron to various neural networks is verified, including a single-layer perceptron, multilayer perceptron, and spiking neural network.

Bioinspired Photoresponsive Single Transistor Neuron for a Neuromorphic Visual System.

Realizing a neuromorphic-based artificial visual system with low-cost hardware requires a neuromorphic device that can react to light stimuli. This study introduces a photoresponsive neuron device composed of a single transistor, developed by engineering an artificial neuron that responds to light, just like retinal neurons. Neuron firing is activated primarily by electrical stimuli such as current via a well-known single transistor latch phenomenon. Its firing characteristics, represented by spiking frequency and amplitude, are additionally modulated by optical stimuli such as photons. When light is illuminated onto the neuron transistor, electron-hole pairs are generated, and they allow the neuron transistor to fire at lower firing threshold voltage. Different photoresponsive properties can be modulated by the intensity and wavelength of the light, analogous to the behavior of retinal neurons. The artificial visual system can be miniaturized because a photoresponsive neuronal function is realized without bulky components such as image sensors and extra circuits.

A high-dimensional in-sensor reservoir computing system with optoelectronic memristors for high-performance neuromorphic machine vision.

In-sensor reservoir computing (RC) is a promising technology to reduce power consumption and training costs of machine vision systems by processing optical signals temporally. This study demonstrates a high-dimensional in-sensor RC system with optoelectronic memristors to enhance the performance of the in-sensor RC system. Because optoelectronic memristors can respond to both optical and electrical stimuli, optical and electrical masks are proposed to improve the dimensionality and performance of the in-sensor RC system. An optical mask is employed to regulate the wavelength of light, while an electrical mask is used to control the initial conductance of zinc oxide optoelectronic memristors. The distinct characteristics of these two masks contribute to the representation of various distinguishable reservoir states, making it possible to implement diverse reservoir configurations with minimal correlation and to increase the dimensionality of the in-sensor RC system. Using the high-dimensional in-sensor RC system, handwritten digits are successfully classified with an accuracy of 94.1%. Furthermore, human action pattern recognition is achieved with a high accuracy of 99.4%. These high accuracies are achieved with the use of a single-layer readout network, which can significantly reduce the network size and training costs.

Spatiotemporal Data Processing with Memristor Crossbar-Array-Based Graph Reservoir.

Memristor-based physical reservoir computing (RC) is a robust framework for processing complex spatiotemporal data parallelly. However, conventional memristor-based reservoirs cannot capture the spatial relationship between the time-varying inputs due to the specific mapping scheme assigning one input signal to one memristor conductance. Here, a physical "graph reservoir" is introduced using a metal cell at the diagonal-crossbar array (mCBA) with dynamic self-rectifying memristors. Input and inverted input signals are applied to the word and bit lines of the mCBA, respectively, storing the correlation information between input signals in the memristors. In this way, the mCBA graph reservoirs can map the spatiotemporal correlation of the input data in a high-dimensional feature space. The high-dimensional mapping characteristics of the graph reservoir achieve notable results, including a normalized root-mean-square error of 0.09 in Mackey-Glass time series prediction, a 97.21% accuracy in MNIST recognition, and an 80.0% diagnostic accuracy in human connectome classification.

Cross-Wired Memristive Crossbar Array for Effective Graph Data Analysis.

Graphs adequately represent the enormous interconnections among numerous entities in big data, incurring high computational costs in analyzing them with conventional hardware. Physical graph representation (PGR) is an approach that replicates the graph within a physical system, allowing for efficient analysis. This study introduces a cross-wired crossbar array (cwCBA), uniquely connecting diagonal and non-diagonal components in a CBA by a cross-wiring process. The cross-wired diagonal cells enable cwCBA to achieve precise PGR and dynamic node state control. For this purpose, a cwCBA is fabricated using Pt/Ta2O5/HfO2/TiN (PTHT) memristor with high on/off and self-rectifying characteristics. The structural and device benefits of PTHT cwCBA for enhanced PGR precision are highlighted, and the practical efficacy is demonstrated for two applications. First, it executes a dynamic path-finding algorithm, identifying the shortest paths in a dynamic graph. PTHT cwCBA shows a more accurate inferred distance and ≈1/3800 lower processing complexity than the conventional method. Second, it analyzes the protein-protein interaction (PPI) networks containing self-interacting proteins, which possess intricate characteristics compared to typical graphs. The PPI prediction results exhibit an average of 30.5% and 21.3% improvement in area under the curve and F1-score, respectively, compared to existing algorithms.

Memristive Crossbar Array-Based Probabilistic Graph Modeling.

Modern graph datasets with structural complexity and uncertainties due to incomplete information or data variability require advanced modeling techniques beyond conventional graph models. This study introduces a memristive crossbar array (CBA)-based probabilistic graph model (C-PGM) utilizing Cu0.3Te0.7/HfO2/Pt memristors, which exhibit probabilistic switching, self-rectifying, and memory characteristics. C-PGM addresses the complexities and uncertainties inherent in structural graph data across various domains, leveraging the probabilistic nature of memristors. C-PGM relies on the device-to-device variation across multiple memristive CBAs, overcoming the limitations of previous approaches that rely on sequential operations, which are slower and have a reliability concern due to repeated switching. This new approach enables the fast processing and massive implementation of probabilistic units at the expense of chip area. In this study, the hardware-based C-PGM feasibly expresses small-scale probabilistic graphs and shows minimal error in aggregate probability calculations. The probability calculation capabilities of C-PGM are applied to steady-state estimation and the PageRank algorithm, which is implemented on a simulated large-scale C-PGM. The C-PGM-based steady-state estimation and PageRank algorithm demonstrate comparable accuracy to conventional methods while significantly reducing computational costs.

Hyperplane tree-based data mining with a multi-functional memristive crossbar array.

This study explores the stochastic and binary switching behaviors of a Ta/HfO2/RuO2 memristor to implement a combined data mining approach for outlier detection and data clustering algorithms in a multi-functional memristive crossbar array. The memristor switches stochastically with high state dispersion in the stochastic mode and deterministically between two states with low dispersion in the binary mode, while they can be controlled by varying operating voltages. The stochastic mode facilitates the parallel generation of random hyperplanes in a tree structure, used to compress spatial information of the dataset in the Euclidian space into binary format, still retaining sufficient spatial features. The ensemble effect from multiple trees improved the classification performance. The binary mode facilitates parallel Hamming distance calculation of the binary codes containing spatial information, which measures similarity. These two modes enable efficient implementation of the newly proposed minority-based outlier detection method and modified K-means method on the same hardware. Array measurements and hardware simulations investigate various hyperparameters' impact and validate the proposed methods with practical datasets. The proposed methods show linear O(n) time complexity and high energy efficiency, consuming <1% of the energy compared to digital computing with conventional algorithms while demonstrating software-comparable performance in both tasks.

Spike-Based Neuromorphic Hardware for Dynamic Tactile Perception with a Self-Powered Mechanoreceptor Array.

A self-powered mechanoreceptor array is demonstrated using four mechanoreceptor cells for recognition of dynamic touch gestures. Each cell consists of a triboelectric nanogenerator (TENG) for touch sensing and a bi-stable resistor (biristor) for spike encoding. It produces informative spike signals by sensing a force of an external touch and encoding the force into the number of spikes. An array of the mechanoreceptor cells is utilized to monitor various touch gestures and it successfully generated spike signals corresponding to all the gestures. To validate the practicality of the mechanoreceptor array, a spiking neural network (SNN), highly attractive for power consumption compared to the conventional von Neumann architecture, is used for the identification of touch gestures. The measured spiking signals are reflected as inputs for the SNN simulations. Consequently, touch gestures are classified with a high accuracy rate of 92.5%. The proposed mechanoreceptor array emerges as a promising candidate for a building block of tactile in-sensor computing in the era of the Internet of Things (IoT), due to the low cost and high manufacturability of the TENG. This eliminates the need for a power supply, coupled with the intrinsic high throughput of the Si-based biristor employing complementary metal-oxide-semiconductor (CMOS) technology.

A stochastic photo-responsive memristive neuron for an in-sensor visual system based on a restricted Boltzmann machine.

In-sensor computing has gained attention as a solution to overcome the von Neumann computing bottlenecks inherent in conventional sensory systems. This attention is due to the ability of sensor elements to directly extract meaningful information from external signals, thereby simplifying complex data. The advantage of in-sensor computing can be maximized with the sampling principle of a restricted Boltzmann machine (RBM) to extract significant features. In this study, a stochastic photo-responsive neuron is developed using a TiN/In-Ga-Zn-O/TiN optoelectronic memristor and an Ag/HfO2/Pt threshold-switching memristor, which can be configured as an input neuron in an in-sensor RBM. It demonstrates a sigmoidal switching probability depending on light intensity. The stochastic properties allow for the simultaneous exploration of various neuron states within the network, making identifying optimal features in complex images easier. Based on semi-empirical simulations, high recognition accuracies of 90.9% and 95.5% are achieved using handwritten digit and face image datasets, respectively. In addition, the in-sensor RBM effectively reconstructs abnormal face images, indicating that integrating in-sensor computing with probabilistic neural networks can lead to reliable and efficient image recognition under unpredictable real-world conditions.

High-Dimensional Physical Reservoir with Back-End-of-Line-Compatible Tin Monoxide Thin-Film Transistor.

This work demonstrates a physical reservoir using a back-end-of-line compatible thin-film transistor (TFT) with tin monoxide (SnO) as the channel material for neuromorphic computing. The electron trapping and time-dependent detrapping at the channel interface induce the SnO·TFT to exhibit fading memory and nonlinearity characteristics, the critical assets for physical reservoir computing. The three-terminal configuration of the TFT allows the generation of higher-dimensional reservoir states by simultaneously adjusting the bias conditions of the gate and drain terminals, surpassing the performances of typical two-terminal-based reservoirs such as memristors. The high-dimensional SnO TFT reservoir performs exceptionally in two benchmark tests, achieving a 94.1% accuracy in Modified National Institute of Standards and Technology handwritten number recognition and a normalized root-mean-square error of 0.089 in Mackey-Glass time-series prediction. Furthermore, it is suitable for vertical integration because its fabrication temperature is <250 °C, providing the benefit of achieving a high integration density.

Heterogeneous density-based clustering with a dual-functional memristive array.

In the big data era, the requirement for data clustering methods that can handle massive and heterogeneous datasets with varying distributions increases. This study proposes a clustering algorithm for data sets with heterogeneous density using a dual-mode memristor crossbar array for data clustering. The array consists of a Ta/HfO2/RuO2 memristor operating in analog or digital modes, controlled by the reset voltage. The digital mode shows low dispersion and a high resistance ratio, and the analog mode enables precise conductance tuning. The local outlier factor is introduced to handle a heterogeneous density, and the required Euclidean and K-distances within the given dataset are calculated in the analog mode in parallel. In the digital mode, clustering is performed based on the connectivity among data points after excluding the detected outliers. The proposed algorithm boasts linear time complexity for the entire process. Extensive evaluations of synthetic datasets demonstrate significant improvement over representative density-based algorithms, and the datasets with heterogeneous density are clustered feasibly. Finally, the proposed algorithm is used to cluster the single-molecule localization microscopy data, demonstrating the feasibility of the suggested method for real-world problems.

Implementation of Bayesian networks and Bayesian inference using a Cu0.1Te0.9/HfO2/Pt threshold switching memristor.

Bayesian networks and Bayesian inference, which forecast uncertain causal relationships within a stochastic framework, are used in various artificial intelligence applications. However, implementing hardware circuits for the Bayesian inference has shortcomings regarding device performance and circuit complexity. This work proposed a Bayesian network and inference circuit using a Cu0.1Te0.9/HfO2/Pt volatile memristor, a probabilistic bit neuron that can control the probability of being 'true' or 'false.' Nodal probabilities within the network are feasibly sampled with low errors, even with the device's cycle-to-cycle variations. Furthermore, Bayesian inference of all conditional probabilities within the network is implemented with low power (<186 nW) and energy consumption (441.4 fJ), and a normalized mean squared error of ∼7.5 × 10-4 through division feedback logic with a variational learning rate to suppress the inherent variation of the memristor. The suggested memristor-based Bayesian network shows the potential to replace the conventional complementary metal oxide semiconductor-based Bayesian estimation method with power efficiency using a stochastic computing method.

In-Materia Annealing and Combinatorial Optimization Based on Vertical Memristive Array.

Due to its area and energy efficiency, a memristive crossbar array (CBA) has been extensively studied for various combinatorial optimization applications, from network problems to circuit design. However, conventional approaches include heavily burdening software fine-tuning for the annealing process. Instead, this study introduces the "in-materia annealing" method, where the inter-layer interference of vertically stacked memristive CBA is utilized as an annealing method. When mapping combinatorial optimization problems into the configuration layer of the CBA, exponentially decaying annealing profiles are generated in nearby noise layers. Moreover, in-materia annealing profiles can be controlled by changing compliance current, read voltage, and read pulse width. Therefore, the annealing profiles can be arbitrarily controlled and generated individually for each cell, providing rich noise sources to solve the problem efficiently. Consequently, the experimental and simulation of Max-Cut and weighted Max-Cut problems achieve notable results with the minimum software burden.

Vertically Stackable Ovonic Threshold Switch Oscillator Using Atomic Layer Deposited Ge0.6Se0.4 Film for High-Density Artificial Neural Networks.

Nanodevice oscillators (nano-oscillators) have received considerable attention to implement in neuromorphic computing as hardware because they can significantly improve the device integration density and energy efficiency compared to complementary metal oxide semiconductor circuit-based oscillators. This work demonstrates vertically stackable nano-oscillators using an ovonic threshold switch (OTS) for high-density neuromorphic hardware. A vertically stackable Ge0.6Se0.4 OTS-oscillator (VOTS-OSC) is fabricated with a vertical crossbar array structure by growing Ge0.6Se0.4 film conformally on a contact hole structure using atomic layer deposition. The VOTS-OSC can be vertically integrated onto peripheral circuits without causing thermal damage because the fabrication temperature is <400 °C. The fabricated device exhibits oscillation characteristics, which can serve as leaky integrate-and-fire neurons in spiking neural networks (SNNs) and coupled oscillators in oscillatory neural networks (ONNs). For practical applications, pattern recognition and vertex coloring are demonstrated with SNNs and ONNs, respectively, using semiempirical simulations. This structure increases the oscillator integration density significantly, enabling complex tasks with a large number of oscillators. Moreover, it can enhance the computational speed of neural networks due to its rapid switching speed.

Leaky FinFET for Reservoir Computing with Temporal Signal Processing.

Reservoir computing can greatly reduce the hardware and training costs of recurrent neural networks with temporal data processing. To implement reservoir computing in a hardware form, physical reservoirs transforming sequential inputs into a high-dimensional feature space are necessary. In this work, a physical reservoir with a leaky fin-shaped field-effect transistor (L-FinFET) is demonstrated by the positive use of a short-term memory property arising from the absence of an energy barrier to suppress the tunneling current. Nevertheless, the L-FinFET reservoir does not lose its multiple memory states. The L-FinFET reservoir consumes very low power when encoding temporal inputs because the gate serves as an enabler of the write operation, even in the off-state, due to its physical insulation from the channel. In addition, the small footprint area arising from the scalability of the FinFET due to its multiple-gate structure is advantageous for reducing the chip size. After the experimental proof of 4-bit reservoir operations with 16 states for temporal signal processing, handwritten digits in the Modified National Institute of Standards and Technology dataset are classified by reservoir computing.

Artificial Multisensory Neuron with a Single Transistor for Multimodal Perception through Hybrid Visual and Thermal Sensing.

An artificial multisensory device applicable to in-sensor computing is demonstrated with a single-transistor neuron (1T-neuron) for multimodal perception. It simultaneously receives two sensing signals from visual and thermal stimuli. The 1T-neuron transforms these signals into electrical signals in the form of spiking and then fires them for a spiking neural network at the same time. This feature makes it feasible to realize input neurons for multimodal sensing. Visual and thermal sensing is achieved due to the inherent optical and thermal behaviors of the 1T-neuron. To demonstrate a neuromorphic multimodal sensing system with the artificial multisensory 1T-neuron, fingerprint recognition, widely used for biometric security, is implemented. Owing to the simultaneous sensing of heat as well as light, the proposed fingerprint recognition system composed of multisensory 1T-neurons not only identifies a genuine pattern but also judges whether or not it is forged.