Self-Templating Copolymerization to Produce Robust Conductive Nanocoatings Based on Conjugated Polymer Brushes with Implementable Memristive Characteristics

An effective synthesis of conductive polymer brushes, i.e., self-templating surface-initiated copolymerization (ST-SICP), is developed. It proceeds through copolymerization of pendant thiophene groups in the precursor multimonomer poly(3-methylthienyl methacrylate) (PMTM) brushes with free 3-methylthiophene (3MT) monomers leading to PMTM-co-P3MT brushes. This approach leads to improved conformational freedom of generated conjugated poly(thiophene)-based chains and their higher share in the brushes with respect to conjugation of pendant thiophene groups only. As a result, best performing conjugated PMTM-co-P3MT brushes demonstrate high ohmic conductivity in both out-of-plane and in-plane direction. Furthermore, thanks to the covalent anchoring as well as intra- and intermolecular connections, highly stable and mechanically robust nanocoatings are produced which can survive mechanical cleaning and long-term storage under ambient conditions. Grafting of ionic poly(sodium 4-styrenesulfonate) (PSSNa) in between PMTM-co-P3MT chains brings new properties to such binary mixed brushes that can operate as thin-film memristive coating with switchable conductance. It is worth mentioning that the crucial synthetic steps, i.e., grafting of precursor PMTM brushes by surface-initiated organocatalyzed atom transfer radical polymerization (SI-O-ATRP) and PSSNa chains by surface-initiated photoiniferter-mediated polymerization (SI-PIMP) are conducted under ambient conditions using only microliter volumes of reagents providing methodology that can be considered for use beyond the laboratory scale.

Infrared Nanoimaging of Hydrogenated Perovskite Nickelate Memristive Devices

Solid-state devices made from correlated oxides, such as perovskite nickelates, are promising for neuromorphic computing by mimicking biological synaptic function. However, comprehending dopant action at the nanoscale poses a formidable challenge to understanding the elementary mechanisms involved. Here, we perform operando infrared nanoimaging of hydrogen-doped correlated perovskite, neodymium nickel oxide (H-NdNiO3, H-NNO), devices and reveal how an applied field perturbs dopant distribution at the nanoscale. This perturbation leads to stripe phases of varying conductivity perpendicular to the applied field, which define the macroscale electrical characteristics of the devices. Hyperspectral nano-FTIR imaging in conjunction with density functional theory calculations unveils a real-space map of multiple vibrational states of H-NNO associated with OH stretching modes and their dependence on the dopant concentration. Moreover, the localization of excess charges induces an out-of-plane lattice expansion in NNO which was confirmed by in situ X-ray diffraction and creates a strain that acts as a barrier against further diffusion. Our results and the techniques presented here hold great potential for the rapidly growing field of memristors and neuromorphic devices wherein nanoscale ion motion is fundamentally responsible for function.

Mechanism and Impact of Bipolar Current Voltage Asymmetry in Computational Phase-Change Memory

Nanoscale resistive memory devices are being explored for neuromorphic and in-memory computing. However, non-ideal device characteristics of read noise and resistance drift pose significant challenges to the achievable computational precision. Here, it is shown that there is an additional non-ideality that can impact computational precision, namely the bias-polarity-dependent current flow. Using phase-change memory (PCM) as a model system, it is shown that this "current-voltage" non-ideality arises both from the material and geometrical properties of the devices. Further, we discuss the detrimental effects of such bipolar asymmetry on in-memory matrix-vector multiply (MVM) operations and provide a scheme to compensate for it.

Memristive Neural Networks for Predicting Seizure Activity

The aim of the study is to assess the possibilities of predicting epileptiform activity using the neuronal activity data recorded from the hippocampus and medial entorhinal cortex of mice with chronic epileptiform activity. To reach this goal, a deep artificial neural network (ANN) has been developed and its implementation based on memristive devices has been demonstrated.

Materials and Methods

The biological part of the investigation. Young healthy outbred CD1 mice were used in our study. They were divided into two groups: control (n=6) and the group with induced chronic epileptiform activity (n=6). Local field potentials (LFP) were recorded from the hippocampus and medial entorhinal cortex of the mice of both groups to register neuronal activity. The LFP recordings were used for deep ANN training. Epileptiform activity in mice was modeled by intraperitoneal injection of pilocarpine (280 mg/kg). LFP were recorded in the awake mice a month after the induction of epileptiform activity.Mathematical part of the investigation. A deep long short-term memory (LSTM) ANN capable of predicting biological signals of neuronal activity in mice has been developed. The ANN implementation is based on memristive devices, which are described by the equations of the redox processes running in the memristive thin metal-oxide-metal films, e.g., Au/ZrO2(Y)/TiN/Ti and Au/SiO2(Y)/TiN/Ti. In order to train the developed ANN to predict epileptiform activity, a supervised learning algorithm was used, which allowed us to adjust the network parameters and train LSTM on the described recordings of neuronal activity.

Results

After training on the LFP recordings from the hippocampus and medial entorhinal cortex of the mice with chronic epileptiform activity, the proposed deep ANN has demonstrated high values of evaluation metric (root-mean-square error, RMSE) and successfully predicted epileptiform activity shortly before its occurrence (40 ms). The results of the numerical experiments have shown that the RMSE value of 0.019 was reached, which indicates the efficacy of proposed approach. The accuracy of epileptiform activity prediction 40 ms before its occurrence is a significant result and shows the potential of the developed neural network architecture.

Conclusion

The proposed deep ANN can be used to predict pathological neuronal activity including epileptic seizure (focal) activity in mice before its actual occurrence. Besides, it can be applied for building a long-term prognosis of the disease course based on the LFP data. Thus, the proposed ANN based on memristive devices represents a novel approach to the prediction and analysis of pathological neuronal activity possessing a potential for improving the diagnosis and prognostication of epileptic seizures and other diseases associated with neuronal activity.

Oxygen vacancy dynamics in Pt/TiOx/TaOy/Pt memristors: exchange with the environment and internal electromigration

Memristors are expected to be one of the key building blocks for the development of new bio-inspired nanoelectronics. Memristive effects in transition metal oxides are usually linked to the electromigration at the nanoscale of charged oxygen vacancies (OV). In this paper we address, for Pt/TiO_x/TaO_y/Pt devices, the exchange of OV between the device and the environment upon the application of electrical stress. From a combination of experiments and theoretical simulations we determine that both TiO_xand TaO_ylayers oxidize, via environmental oxygen uptake, during the electroforming process. Once the memristive effect is stabilized (post-forming behavior) our results suggest that oxygen exchange with the environment is suppressed and the OV dynamics that drives the memristive behavior is restricted to an internal electromigration between TiO_xand TaO_ylayers. Our work provides relevant information for the design of reliable binary oxide memristive devices.

Experimental and Modeling Study of Metal-Insulator Interfaces to Control the Electronic Transport in Single Nanowire Memristive Devices

Memristive devices relying on redox-based resistive switching mechanisms represent promising candidates for the development of novel computing paradigms beyond von Neumann architecture. Recent advancements in understanding physicochemical phenomena underlying resistive switching have shed new light on the importance of an appropriate selection of material properties required to optimize the performance of devices. However, despite great attention has been devoted to unveiling the role of doping concentration, impurity type, adsorbed moisture, and catalytic activity at the interfaces, specific studies concerning the effect of the counter electrode in regulating the electronic flow in memristive cells are scarce. In this work, the influence of the metal-insulator Schottky interfaces in electrochemical metallization memory (ECM) memristive cell model systems based on single-crystalline ZnO nanowires (NWs) is investigated following a combined experimental and modeling approach. By comparing and simulating the electrical characteristics of single NW devices with different contact configurations and by considering Ag and Pt electrodes as representative of electrochemically active and inert electrodes, respectively, we highlight the importance of an appropriate choice of electrode materials by taking into account the Schottky barrier height and interface chemistry at the metal-insulator interfaces. In particular, we show that a clever choice of metal-insulator interfaces allows to reshape the hysteretic conduction characteristics of the device and to increase the device performance by tuning its resistance window. These results obtained from single NW-based devices provide new insights into the selection criteria for materials and interfaces in connection with the design of advanced ECM cells.

Controlled Formation of Conduction Channels in Memristive Devices Observed by X-ray Multimodal Imaging

Neuromorphic computing provides a means for achieving faster and more energy efficient computations than conventional digital computers for artificial intelligence (AI). However, its current accuracy is generally less than the dominant software-based AI. The key to improving accuracy is to reduce the intrinsic randomness of memristive devices, emulating synapses in the brain for neuromorphic computing. Here using a planar device as a model system, the controlled formation of conduction channels is achieved with high oxygen vacancy concentrations through the design of sharp protrusions in the electrode gap, as observed by X-ray multimodal imaging of both oxygen stoichiometry and crystallinity. Classical molecular dynamics simulations confirm that the controlled formation of conduction channels arises from confinement of the electric field, yielding a reproducible spatial distribution of oxygen vacancies across switching cycles. This work demonstrates an effective route to control the otherwise random electroforming process by electrode design, facilitating the development of more accurate memristive devices for neuromorphic computing.

Quantum Conductance in Memristive Devices: Fundamentals, Developments, and Applications

Quantum effects in novel functional materials and new device concepts represent a potential breakthrough for the development of new information processing technologies based on quantum phenomena. Among the emerging technologies, memristive elements that exhibit resistive switching, which relies on the electrochemical formation/rupture of conductive nanofilaments, exhibit quantum conductance effects at room temperature. Despite the underlying resistive switching mechanism having been exploited for the realization of next-generation memories and neuromorphic computing architectures, the potentialities of quantum effects in memristive devices are still rather unexplored. Here, a comprehensive review on memristive quantum devices, where quantum conductance effects can be observed by coupling ionics with electronics, is presented. Fundamental electrochemical and physicochemical phenomena underlying device functionalities are introduced, together with fundamentals of electronic ballistic conduction transport in nanofilaments. Quantum conductance effects including quantum mode splitting, stability, and random telegraph noise are analyzed, reporting experimental techniques and challenges of nanoscale metrology for the characterization of memristive phenomena. Finally, potential applications and future perspectives are envisioned, discussing how memristive devices with controllable atomic-sized conductive filaments can represent not only suitable platforms for the investigation of quantum phenomena but also promising building blocks for the realization of integrated quantum systems working in air at room temperature.

Bipolar Resistive Switching Behavior of PVP-GQD/HfOx/ITO/Graphene Hybrid Flexible Resistive Random Access Memory

We have investigated highly flexible memristive devices using reduced graphene oxide (RGO) nanosheet nanocomposites with an embedded GQD Layer. Resistive switching behavior of poly (4-vinylphenol):graphene quantum dot (PVP:GQD) composite and HfOx hybrid bilayer was explored for developing flexible resistive random access memory (RRAM) devices. A composite active layer was designed based on graphene quantum dots, which is a low-dimensional structure, and a heterogeneous active layer of graphene quantum dots was applied to the interfacial defect structure to overcome the limitations. Increasing to 0.3–0.6 wt % PVP-GQD, V_f changed from 2.27–2.74 V. When negative deflection is applied to the lower electrode, electrons travel through the HfOx/ITO interface. In addition, as the PVP-GQD concentration increased, the depth of the interfacial defect decreased, and confirmed the repetition of appropriate electrical properties through Al and HfOx/ITO. The low interfacial defects help electrophoresis of Al⁺ ions to the PVP GQD layer and the HfOx thin film. A local electric field increase occurred, resulting in the breakage of the conductive filament in the defect.

Emerging Artificial Neuron Devices for Probabilistic Computing

In recent decades, artificial intelligence has been successively employed in the fields of finance, commerce, and other industries. However, imitating high-level brain functions, such as imagination and inference, pose several challenges as they are relevant to a particular type of noise in a biological neuron network. Probabilistic computing algorithms based on restricted Boltzmann machine and Bayesian inference that use silicon electronics have progressed significantly in terms of mimicking probabilistic inference. However, the quasi-random noise generated from additional circuits or algorithms presents a major challenge for silicon electronics to realize the true stochasticity of biological neuron systems. Artificial neurons based on emerging devices, such as memristors and ferroelectric field-effect transistors with inherent stochasticity can produce uncertain non-linear output spikes, which may be the key to make machine learning closer to the human brain. In this article, we present a comprehensive review of the recent advances in the emerging stochastic artificial neurons (SANs) in terms of probabilistic computing. We briefly introduce the biological neurons, neuron models, and silicon neurons before presenting the detailed working mechanisms of various SANs. Finally, the merits and demerits of silicon-based and emerging neurons are discussed, and the outlook for SANs is presented.

Exploring Area-Dependent Pr0.7Ca0.3MnO3-Based Memristive Devices as Synapses in Spiking and Artificial Neural Networks

Memristive devices are novel electronic devices, which resistance can be tuned by an external voltage in a non-volatile way. Due to their analog resistive switching behavior, they are considered to emulate the behavior of synapses in neuronal networks. In this work, we investigate memristive devices based on the field-driven redox process between the p-conducting Pr0.7Ca0.3MnO3 (PCMO) and different tunnel barriers, namely, Al2O3, Ta2O5, and WO3. In contrast to the more common filamentary-type switching devices, the resistance range of these area-dependent switching devices can be adapted to the requirements of the surrounding circuit. We investigate the impact of the tunnel barrier layer on the switching performance including area scaling of the current and variability. Best performance with respect to the resistance window and the variability is observed for PCMO with a native Al2O3 tunnel oxide. For all different layer stacks, we demonstrate a spike timing dependent plasticity like behavior of the investigated PCMO cells. Furthermore, we can also tune the resistance in an analog fashion by repeated switching the device with voltage pulses of the same amplitude and polarity. Both measurements resemble the plasticity of biological synapses. We investigate in detail the impact of different pulse heights and pulse lengths on the shape of the stepwise SET and RESET curves. We use these measurements as input for the simulation of training and inference in a multilayer perceptron for pattern recognition, to show the use of PCMO-based ReRAM devices as weights in artificial neural networks which are trained by gradient descent methods. Based on this, we identify certain trends for the impact of the applied voltages and pulse length on the resulting shape of the measured curves and on the learning rate and accuracy of the multilayer perceptron.

Adaptive Extreme Edge Computing for Wearable Devices

Wearable devices are a fast-growing technology with impact on personal healthcare for both society and economy. Due to the widespread of sensors in pervasive and distributed networks, power consumption, processing speed, and system adaptation are vital in future smart wearable devices. The visioning and forecasting of how to bring computation to the edge in smart sensors have already begun, with an aspiration to provide adaptive extreme edge computing. Here, we provide a holistic view of hardware and theoretical solutions toward smart wearable devices that can provide guidance to research in this pervasive computing era. We propose various solutions for biologically plausible models for continual learning in neuromorphic computing technologies for wearable sensors. To envision this concept, we provide a systematic outline in which prospective low power and low latency scenarios of wearable sensors in neuromorphic platforms are expected. We successively describe vital potential landscapes of neuromorphic processors exploiting complementary metal-oxide semiconductors (CMOS) and emerging memory technologies (e.g., memristive devices). Furthermore, we evaluate the requirements for edge computing within wearable devices in terms of footprint, power consumption, latency, and data size. We additionally investigate the challenges beyond neuromorphic computing hardware, algorithms and devices that could impede enhancement of adaptive edge computing in smart wearable devices.

Non-linear Memristive Synaptic Dynamics for Efficient Unsupervised Learning in Spiking Neural Networks

Spiking neural networks (SNNs) are a computational tool in which the information is coded into spikes, as in some parts of the brain, differently from conventional neural networks (NNs) that compute over real-numbers. Therefore, SNNs can implement intelligent information extraction in real-time at the edge of data acquisition and correspond to a complementary solution to conventional NNs working for cloud-computing. Both NN classes face hardware constraints due to limited computing parallelism and separation of logic and memory. Emerging memory devices, like resistive switching memories, phase change memories, or memristive devices in general are strong candidates to remove these hurdles for NN applications. The well-established training procedures of conventional NNs helped in defining the desiderata for memristive device dynamics implementing synaptic units. The generally agreed requirements are a linear evolution of memristive conductance upon stimulation with train of identical pulses and a symmetric conductance change for conductance increase and decrease. Conversely, little work has been done to understand the main properties of memristive devices supporting efficient SNN operation. The reason lies in the lack of a background theory for their training. As a consequence, requirements for NNs have been taken as a reference to develop memristive devices for SNNs. In the present work, we show that, for efficient CMOS/memristive SNNs, the requirements for synaptic memristive dynamics are very different from the needs of a conventional NN. System-level simulations of a SNN trained to classify hand-written digit images through a spike timing dependent plasticity protocol are performed considering various linear and non-linear plausible synaptic memristive dynamics. We consider memristive dynamics bounded by artificial hard conductance values and limited by the natural dynamics evolution toward asymptotic values (soft-boundaries). We quantitatively analyze the impact of resolution and non-linearity properties of the synapses on the network training and classification performance. Finally, we demonstrate that the non-linear synapses with hard boundary values enable higher classification performance and realize the best trade-off between classification accuracy and required training time. With reference to the obtained results, we discuss how memristive devices with non-linear dynamics constitute a technologically convenient solution for the development of on-line SNN training.

Water-Mediated Ionic Migration in Memristive Nanowires with a Tunable Resistive Switching Mechanism

Memristive devices based on electrochemical resistive switching effects have been proposed as promising candidates for in-memory computing and for the realization of artificial neural networks. Despite great efforts toward understanding the nanoionic processes underlying resistive switching phenomena, comprehension of the effect of competing redox processes on device functionalities from the materials perspective still represents a challenge. In this work, we experimentally and theoretically investigate the concurring reactions of silver and moisture and their impact on the electronic properties of a single-crystalline ZnO nanowire (NW). A decrease in electronic conductivity due to surface adsorption of moisture is observed, whereas, at the same time, water molecules reduce the energy barrier for Ag+ ion migration on the NW surface, facilitating the conductive filament formation. By controlling the relative humidity, the ratio of intrinsic electronic conductivity and surface ionic conductivity can be tuned to modulate the device performance. The results achieved on a single-crystalline memristive model system shed new light on the dual nature of the mechanism of how moisture affects resistive switching behavior in memristive devices.

Mixed-Precision Deep Learning Based on Computational Memory

Deep neural networks (DNNs) have revolutionized the field of artificial intelligence and have achieved unprecedented success in cognitive tasks such as image and speech recognition. Training of large DNNs, however, is computationally intensive and this has motivated the search for novel computing architectures targeting this application. A computational memory unit with nanoscale resistive memory devices organized in crossbar arrays could store the synaptic weights in their conductance states and perform the expensive weighted summations in place in a non-von Neumann manner. However, updating the conductance states in a reliable manner during the weight update process is a fundamental challenge that limits the training accuracy of such an implementation. Here, we propose a mixed-precision architecture that combines a computational memory unit performing the weighted summations and imprecise conductance updates with a digital processing unit that accumulates the weight updates in high precision. A combined hardware/software training experiment of a multilayer perceptron based on the proposed architecture using a phase-change memory (PCM) array achieves 97.73% test accuracy on the task of classifying handwritten digits (based on the MNIST dataset), within 0.6% of the software baseline. The architecture is further evaluated using accurate behavioral models of PCM on a wide class of networks, namely convolutional neural networks, long-short-term-memory networks, and generative-adversarial networks. Accuracies comparable to those of floating-point implementations are achieved without being constrained by the non-idealities associated with the PCM devices. A system-level study demonstrates 172 × improvement in energy efficiency of the architecture when used for training a multilayer perceptron compared with a dedicated fully digital 32-bit implementation.

Spatial Properties of STDP in a Self-Learning Spiking Neural Network Enable Controlling a Mobile Robot

Development of spiking neural networks (SNNs) controlling mobile robots is one of the modern challenges in computational neuroscience and artificial intelligence. Such networks, being replicas of biological ones, are expected to have a higher computational potential than traditional artificial neural networks (ANNs). The critical problem is in the design of robust learning algorithms aimed at building a “living computer” based on SNNs. Here, we propose a simple SNN equipped with a Hebbian rule in the form of spike-timing-dependent plasticity (STDP). The SNN implements associative learning by exploiting the spatial properties of STDP. We show that a LEGO robot controlled by the SNN can exhibit classical and operant conditioning. Competition of spike-conducting pathways in the SNN plays a fundamental role in establishing associations of neural connections. It replaces the irrelevant associations by new ones in response to a change in stimuli. Thus, the robot gets the ability to relearn when the environment changes. The proposed SNN and the stimulation protocol can be further enhanced and tested in developing neuronal cultures, and also admit the use of memristive devices for hardware implementation.

Memristors for Neuromorphic Circuits and Artificial Intelligence Applications

Artificial Intelligence has found many applications in the last decade due to increased computing power. Artificial Neural Networks are inspired in the brain structure and consist in the interconnection of artificial neurons through artificial synapses in the so-called Deep Neural Networks (DNNs). Training these systems requires huge amounts of data and, after the network is trained, it can recognize unforeseen data and provide useful information. As far as the training is concerned, we can distinguish between supervised and unsupervised learning. The former requires labelled data and is based on the iterative minimization of the output error using the stochastic gradient descent method followed by the recalculation of the strength of the synaptic connections (weights) with the backpropagation algorithm. On the other hand, unsupervised learning does not require data labeling and it is not based on explicit output error minimization. Conventional ANNs can function with supervised learning algorithms (perceptrons, multi-layer perceptrons, convolutional networks, etc.) but also with unsupervised learning rules (Kohonen networks, self-organizing maps, etc.). Besides, another type of neural networks are the so-called Spiking Neural Networks (SNNs) in which learning takes place through the superposition of voltage spikes launched by the neurons. Their behavior is much closer to the brain functioning mechanisms they can be used with supervised and unsupervised learning rules. Since learning and inference is based on short voltage spikes, energy efficiency improves substantially. Up to this moment, all these ANNs (spiking and conventional) have been implemented as software tools running on conventional computing units based on the von Neumann architecture. However, this approach reaches important limits due to the required computing power, physical size and energy consumption. This is particularly true for applications at the edge of the internet. Thus, there is an increasing interest in developing AI tools directly implemented in hardware for this type of applications. The first hardware demonstrations have been based on Complementary Metal-Oxide-Semiconductor (CMOS) circuits and specific communication protocols. However, to further increase training speed andenergy efficiency while reducing the system size, the combination of CMOS neuron circuits with memristor synapses is now being explored. It has also been pointed out that the short time non-volatility of some memristors may even allow fabricating purely memristive ANNs. The memristor is a new device (first demonstrated in solid-state in 2008) which behaves as a resistor with memory and which has been shown to have potentiation and depression properties similar to those of biological synapses. In this Special Issue, we explore the state of the art of neuromorphic circuits implementing neural networks with memristors for AI applications.

Memristive and CMOS Devices for Neuromorphic Computing

Neuromorphic computing has emerged as one of the most promising paradigms to overcome the limitations of von Neumann architecture of conventional digital processors. The aim of neuromorphic computing is to faithfully reproduce the computing processes in the human brain, thus paralleling its outstanding energy efficiency and compactness. Toward this goal, however, some major challenges have to be faced. Since the brain processes information by high-density neural networks with ultra-low power consumption, novel device concepts combining high scalability, low-power operation, and advanced computing functionality must be developed. This work provides an overview of the most promising device concepts in neuromorphic computing including complementary metal-oxide semiconductor (CMOS) and memristive technologies. First, the physics and operation of CMOS-based floating-gate memory devices in artificial neural networks will be addressed. Then, several memristive concepts will be reviewed and discussed for applications in deep neural network and spiking neural network architectures. Finally, the main technology challenges and perspectives of neuromorphic computing will be discussed.

Quasi-Hodgkin-Huxley Neurons with Leaky Integrate-and-Fire Functions Physically Realized with Memristive Devices

Artificial neurons with functions such as leaky integrate-and-fire (LIF) and spike output are essential for brain-inspired computation with high efficiency. However, previously implemented artificial neurons, e.g., Hodgkin-Huxley (HH) neurons, integrate-and-fire (IF) neurons, and LIF neurons, only achieve partial functionality of a biological neuron. In this work, quasi-HH neurons with leaky integrate-and-fire functions are physically demonstrated with a volatile memristive device, W/WO₃ /poly(3,4-ethylenedioxythiophene): polystyrene sulfonate/Pt. The resistive switching behavior of the device can be attributed to the migration of protons, unlike the migration of oxygen ions normally involved in oxide-based memristors. With multifunctions similar to their biological counterparts, quasi-HH neurons are advantageous over the reported HH and LIF neurons, demonstrating their potential for neuromorphic computing applications.

Au Nanoparticles as Template for Defect Formation in Memristive SrTiO₃ Thin Films

We investigated the possibility of tuning the local switching properties of memristive crystalline SrTiO 3 thin films by inserting nanoscale defect nucleation centers. For that purpose, we employed chemically-synthesized Au nanoparticles deposited on 0.5 wt%-Nb-doped SrTiO 3 single crystal substrates as a defect formation template for the subsequent growth of SrTiO 3 . We studied in detail the resulting microstructure and the local conducting and switching properties of the SrTiO 3 thin films. We revealed that the Au nanoparticles floated to the SrTiO 3 surface during growth, leaving behind a distorted thin film region in their vicinity. By employing conductive-tip atomic force microscopy, these distorted SrTiO 3 regions are identified as sites of preferential resistive switching. These findings can be attributed to the enhanced oxygen exchange reaction at the surface in these defective regions.

Quantitative Observation of Threshold Defect Behavior in Memristive Devices with Operando X-ray Microscopy

Memristive devices are an emerging technology that enables both rich interdisciplinary science and novel device functionalities, such as nonvolatile memories and nanoionics-based synaptic electronics. Recent work has shown that the reproducibility and variability of the devices depend sensitively on the defect structures created during electroforming as well as their continued evolution under dynamic electric fields. However, a fundamental principle guiding the material design of defect structures is still lacking due to the difficulty in understanding dynamic defect behavior under different resistance states. Here, we unravel the existence of threshold behavior by studying model, single-crystal devices: resistive switching requires that the pristine oxygen vacancy concentration reside near a critical value. Theoretical calculations show that the threshold oxygen vacancy concentration lies at the boundary for both electronic and atomic phase transitions. Through operando, multimodal X-ray imaging, we show that field tuning of the local oxygen vacancy concentration below or above the threshold value is responsible for switching between different electrical states. These results provide a general strategy for designing functional defect structures around threshold concentrations to create dynamic, field-controlled phases for memristive devices.

Subfilamentary Networks Cause Cycle-to-Cycle Variability in Memristive Devices

A major obstacle for the implementation of redox-based memristive memory or logic technology is the large cycle-to-cycle and device-to-device variability. Here, we use spectromicroscopic photoemission threshold analysis and operando XAS analysis to experimentally investigate the microscopic origin of the variability. We find that some devices exhibit variations in the shape of the conductive filament or in the oxygen vacancy distribution at and around the filament. In other cases, even the location of the active filament changes from one cycle to the next. We propose that both effects originate from the coexistence of multiple (sub)filaments and that the active, current-carrying filament may change from cycle to cycle. These findings account for the observed variability in device performance and represent the scientific basis, rather than prior purely empirical engineering approaches, for developing stable memristive devices.

Interfacial Metal-Oxide Interactions in Resistive Switching Memories

Metal oxides are commonly used as electrolytes for redox-based resistive switching memories. In most cases, non-noble metals are directly deposited as ohmic electrodes. We demonstrate that irrespective of bulk thermodynamics predictions an intermediate oxide film a few nanometers in thickness is always formed at the metal/insulator interface, and this layer significantly contributes to the development of reliable switching characteristics. We have tested metal electrodes and metal oxides mostly used for memristive devices, that is, Ta, Hf, and Ti and Ta₂O₅, HfO₂, and SiO₂. Intermediate oxide layers are always formed at the interfaces, whereas only the rate of the electrode oxidation depends on the oxygen affinity of the metal and the chemical stability of the oxide matrix. Device failure is associated with complete transition of short-range order to a more disordered main matrix structure.

Anomalous Resistance Hysteresis in Oxide ReRAM: Oxygen Evolution and Reincorporation Revealed by In Situ TEM

The control and rational design of redox-based memristive devices, which are highly attractive candidates for next-generation nonvolatile memory and logic applications, is complicated by competing and poorly understood switching mechanisms, which can result in two coexisting resistance hystereses that have opposite voltage polarity. These competing processes can be defined as regular and anomalous resistive switching. Despite significant characterization efforts, the complex nanoscale redox processes that drive anomalous resistive switching and their implications for current transport remain poorly understood. Here, lateral and vertical mapping of O vacancy concentrations is used during the operation of such devices in situ in an aberration corrected transmission electron microscope to explain the anomalous switching mechanism. It is found that an increase (decrease) in the overall O vacancy concentration within the device after positive (negative) biasing of the Schottky-type electrode is associated with the electrocatalytic release and reincorporation of oxygen at the electrode/oxide interface and is responsible for the resistance change. This fundamental insight presents a novel perspective on resistive switching processes and opens up new technological opportunities for the implementation of memristive devices, as anomalous switching can now be suppressed selectively or used deliberately to achieve the desirable so-called deep Reset.

Double-Barrier Memristive Devices for Unsupervised Learning and Pattern Recognition

The use of interface-based resistive switching devices for neuromorphic computing is investigated. In a combined experimental and numerical study, the important device parameters and their impact on a neuromorphic pattern recognition system are studied. The memristive cells consist of a layer sequence Al/Al2O3/Nb x O y /Au and are fabricated on a 4-inch wafer. The key functional ingredients of the devices are a 1.3 nm thick Al2O3 tunnel barrier and a 2.5 mm thick Nb x O y memristive layer. Voltage pulse measurements are used to study the electrical conditions for the emulation of synaptic functionality of single cells for later use in a recognition system. The results are evaluated and modeled in the framework of the plasticity model of Ziegler et al. Based on this model, which is matched to experimental data from 84 individual devices, the network performance with regard to yield, reliability, and variability is investigated numerically. As the network model, a computing scheme for pattern recognition and unsupervised learning based on the work of Querlioz et al. (2011), Sheridan et al. (2014), Zahari et al. (2015) is employed. This is a two-layer feedforward network with a crossbar array of memristive devices, leaky integrate-and-fire output neurons including a winner-takes-all strategy, and a stochastic coding scheme for the input pattern. As input pattern, the full data set of digits from the MNIST database is used. The numerical investigation indicates that the experimentally obtained yield, reliability, and variability of the memristive cells are suitable for such a network. Furthermore, evidence is presented that their strong I-V non-linearity might avoid the need for selector devices in crossbar array structures.

Acceleration of Deep Neural Network Training with Resistive Cross-Point Devices: Design Considerations

In recent years, deep neural networks (DNN) have demonstrated significant business impact in large scale analysis and classification tasks such as speech recognition, visual object detection, pattern extraction, etc. Training of large DNNs, however, is universally considered as time consuming and computationally intensive task that demands datacenter-scale computational resources recruited for many days. Here we propose a concept of resistive processing unit (RPU) devices that can potentially accelerate DNN training by orders of magnitude while using much less power. The proposed RPU device can store and update the weight values locally thus minimizing data movement during training and allowing to fully exploit the locality and the parallelism of the training algorithm. We evaluate the effect of various RPU device features/non-idealities and system parameters on performance in order to derive the device and system level specifications for implementation of an accelerator chip for DNN training in a realistic CMOS-compatible technology. For large DNNs with about 1 billion weights this massively parallel RPU architecture can achieve acceleration factors of 30, 000 × compared to state-of-the-art microprocessors while providing power efficiency of 84, 000 GigaOps∕s∕W. Problems that currently require days of training on a datacenter-size cluster with thousands of machines can be addressed within hours on a single RPU accelerator. A system consisting of a cluster of RPU accelerators will be able to tackle Big Data problems with trillions of parameters that is impossible to address today like, for example, natural speech recognition and translation between all world languages, real-time analytics on large streams of business and scientific data, integration, and analysis of multimodal sensory data flows from a massive number of IoT (Internet of Things) sensors.

Synaptic Metaplasticity Realized in Oxide Memristive Devices

Metaplasticity, a higher order of synaptic plasticity, as well as a key issue in neuroscience, is realized with artificial synapses based on a WO3 thin film, and the activity-dependent metaplastic responses of the artificial synapses, such as spike-timing-dependent plasticity, are systematically investigated. This work has significant implications in neuromorphic computation.

A memristive spiking neuron with firing rate coding

Perception, decisions, and sensations are all encoded into trains of action potentials in the brain. The relation between stimulus strength and all-or-nothing spiking of neurons is widely believed to be the basis of this coding. This initiated the development of spiking neuron models; one of today's most powerful conceptual tool for the analysis and emulation of neural dynamics. The success of electronic circuit models and their physical realization within silicon field-effect transistor circuits lead to elegant technical approaches. Recently, the spectrum of electronic devices for neural computing has been extended by memristive devices, mainly used to emulate static synaptic functionality. Their capabilities for emulations of neural activity were recently demonstrated using a memristive neuristor circuit, while a memristive neuron circuit has so far been elusive. Here, a spiking neuron model is experimentally realized in a compact circuit comprising memristive and memcapacitive devices based on the strongly correlated electron material vanadium dioxide (VO₂) and on the chemical electromigration cell Ag/TiO_2−x/Al. The circuit can emulate dynamical spiking patterns in response to an external stimulus including adaptation, which is at the heart of firing rate coding as first observed by E.D. Adrian in 1926.