Analog memristive synapse based on topotactic phase transition for high-performance neuromorphic computing and neural network pruning

Ionic Diffusive Nanomemristors with Dendritic Competition and Cooperation Functions for Ultralow Voltage Neuromorphic Computing

Realization of dendric signal processing in the human brain is of great significance for spatiotemporal neuromorphic engineering. Here, we proposed an ionic dendrite device with multichannel communication, which could realize synaptic behaviors even under an ultralow action potential of 80 mV. The device not only could simulate one-to-one information transfer of axons but also achieve a many-to-one modulation mode of dendrites. By the adjustment of two presynapses, Pavlov's dog conditioning experiment was learned successfully. Furthermore, the device also could emulate the biological synaptic competition and synaptic cooperation phenomenon through the comodulation of three presynapses, which are crucial for artificial neural network (ANN) implementation. Finally, an ANN was further constructed to realize highly efficient and anti-interference recognition of fashion patterns. By introducing the cooperative device, synaptic weight updates could be improved for higher linearity and larger dynamic regulation range in neuromorphic computing, resulting in higher recognition accuracy and efficiency. Such an artificial dendric device has great application prospects in the processing of more complex information and the construction of an ANN system with more functions.

Stochastic neuro-fuzzy system implemented in memristor crossbar arrays

Neuro-symbolic artificial intelligence has garnered considerable attention amid increasing industry demands for high-performance neural networks that are interpretable and adaptable to previously unknown problem domains with minimal reconfiguration. However, implementing neuro-symbolic hardware is challenging due to the complexity in symbolic knowledge representation and calculation. We experimentally demonstrated a memristor-based neuro-fuzzy hardware based on TiN/TaOx/HfOx/TiN chips that is superior to its silicon-based counterpart in terms of throughput and energy efficiency by using array topological structure for knowledge representation and physical laws for computing. Intrinsic memristor variability is fully exploited to increase robustness in knowledge representation. A hybrid in situ training strategy is proposed for error minimizing in training. The hardware adapts easier to a previously unknown environment, achieving ~6.6 times faster convergence and ~6 times lower error than deep learning. The hardware energy efficiency is over two orders of magnitude greater than field-programmable gate arrays. This research greatly extends the capability of memristor-based neuromorphic computing systems in artificial intelligence.

Synergistically Modulating Conductive Filaments in Ion-Based Memristors for Enhanced Analog In-Memory Computing

Memristors offer a promising solution to address the performance and energy challenges faced by conventional von Neumann computer systems. Yet, stochastic ion migration in conductive filament often leads to an undesired performance tradeoff between memory window, retention, and endurance. Herein, a robust memristor based on oxygen-rich SnO2 nanoflowers switching medium, enabled by seed-mediated wet chemistry, to overcome the ion migration issue for enhanced analog in-memory computing is reported. Notably, the interplay between the oxygen vacancy (Vo) and Ag ions (Ag+ ) in the Ag/SnO2 /p++ -Si memristor can efficiently modulate the formation and abruption of conductive filaments, thereby resulting in a high on/off ratio (>106), long memory retention (10-year extrapolation), and low switching variability (SV = 6.85%). Multiple synaptic functions, such as paired-pulse facilitation, long-term potentiation/depression, and spike-time dependent plasticity, are demonstrated. Finally, facilitated by the symmetric analog weight updating and multiple conductance states, a high image recognition accuracy of ≥ 91.39% is achieved, substantiating its feasibility for analog in-memory computing. This study highlights the significance of synergistically modulating conductive filaments in optimizing performance trade-offs, balancing memory window, retention, and endurance, which demonstrates techniques for regulating ion migration, rendering them a promising approach for enabling cutting-edge neuromorphic applications.

Highly efficient and fast modulation of magnetic coupling interaction in the SrCoO2.5/La0.7Ca0.3MnO3 heterostructure

Effective manipulation of magnetic properties in transition-metal oxides is one of the crucial issues for the application of materials. Up to now, most investigations have focused on electrolyte-based ionic control, which is limited by the slow speed. In this work, the interfacial coupling of the SrCoO2.5/La0.7Ca0.3MnO3 (LCMO) bilayer is effectively modulated with fast response time. After being treated with diluted acetic acid, the bilayer changes from antiferromagnetic/ferromagnetic (AFM/FM) coupling to FM/FM coupling and the Curie temperature is also effectively increased. Meanwhile, the corresponding electric transport properties are modulated within a very short time. Combined with the structure characterization and X-ray absorption measurements, we find that the top SrCoO2.5 layer is changed from the antiferromagnetic insulator to the ferromagnetic metal phase, which is attributed to the formation of the active oxygen species due to the reaction between the protons in the acid and the SrCoO2.5 layer. The bottom LCMO layer remains unchanged during this process. The response time of the bilayer with the acid treatment method is more than an order of magnitude faster than other methods. It is expected that this acid treatment method may open more possibilities for manipulating the magnetic and electric properties in oxide-based devices.

Strain-Induced Uphill Hydrogen Distribution in Perovskite Oxide Films

Incorporating hydrogen into transition-metal oxides (TMOs) provides a facile and powerful way to manipulate the performances of TMOs, and thus numerous efforts have been invested in developing hydrogenation methods and exploring the property modulation via hydrogen doping. However, the distribution of hydrogen ions, which is a key factor in determining the physicochemical properties on a microscopic scale, has not been clearly illustrated. Here, focusing on prototypical perovskite oxide (NdNiO3 and La0.67Sr0.33MnO3) epitaxial films, we find that hydrogen distribution exhibits an anomalous "uphill" feature (against the concentration gradient) under tensile strain, namely, the proton concentration enhances upon getting farther from the hydrogen source. Distinctly, under a compressive strain state, hydrogen shows a normal distribution without uphill features. The epitaxial strain significantly influences the chemical lattice coupling and the energy profile as a function of the hydrogen doping position, thus dominating the hydrogen distribution. Furthermore, the strain-(H+) distribution relationship is maintained in different hydrogenation methods (metal-alkali treatment) which is first applied to perovskite oxides. The discovery of strain-dependent hydrogen distribution in oxides provides insights into tailoring the magnetoelectric and energy-conversion functionalities of TMOs via strain engineering.

Filament-free memristors for computing

Memristors have attracted increasing attention due to their tremendous potential to accelerate data-centric computing systems. The dynamic reconfiguration of memristive devices in response to external electrical stimuli can provide highly desirable novel functionalities for computing applications when compared with conventional complementary-metal-oxide-semiconductor (CMOS)-based devices. Those most intensively studied and extensively reviewed memristors in the literature so far have been filamentary type memristors, which typically exhibit a relatively large variability from device to device and from switching cycle to cycle. On the other hand, filament-free switching memristors have shown a better uniformity and attractive dynamical properties, which can enable a variety of new computing paradigms but have rarely been reviewed. In this article, a wide range of filament-free switching memristors and their corresponding computing applications are reviewed. Various junction structures, switching properties, and switching principles of filament-free memristors are surveyed and discussed. Furthermore, we introduce recent advances in different computing schemes and their demonstrations based on non-filamentary memristors. This Review aims to present valuable insights and guidelines regarding the key computational primitives and implementations enabled by these filament-free switching memristors.

Analog Memory and Synaptic Plasticity in an InGaZnO-Based Memristor by Modifying Intrinsic Oxygen Vacancies

This study focuses on InGaZnO-based synaptic devices fabricated using reactive radiofrequency sputtering deposition with highly uniform and reliable multilevel memory states. Electron trapping and trap generation behaviors were examined based on current compliance adjustments and constant voltage stressing on the ITO/InGaZnO/ITO memristor. Using O2 + N2 plasma treatment resulted in stable and consistent cycle-to-cycle memory switching with an average memory window of ~95.3. Multilevel resistance states ranging from 0.68 to 140.7 kΩ were achieved by controlling the VRESET within the range of -1.4 to -1.8 V. The modulation of synaptic weight for short-term plasticity was simulated by applying voltage pulses with increasing amplitudes after the formation of a weak conductive filament. To emulate several synaptic behaviors in InGaZnO-based memristors, variations in the pulse interval were used for paired-pulse facilitation and pulse frequency-dependent spike rate-dependent plasticity. Long-term potentiation and depression are also observed after strong conductive filaments form at higher current compliance in the switching layer. Hence, the ITO/InGaZnO/ITO memristor holds promise for high-performance synaptic device applications.

Uniform multilevel switching and synaptic properties in RF-sputtered InGaZnO-based memristor treated with oxygen plasma

Bipolar gradual resistive switching was investigated in ITO/InGaZnO/ITO resistive switching devices. Controlled intrinsic oxygen vacancy formation inside the switching layer enabled the establishment of a stable multilevel memory state, allowing for RESET voltage control and non-degradable data endurance. The ITO/InGaZnO interface governs the migration of oxygen ions and redox reactions within the switching layer. Voltage-stress-induced electron trapping and oxygen vacancy formation were observed before conductive filament electroforming. This device mimicked biological synapses, demonstrating short- and long-term potentiation and depression through electrical pulse sequences. Modulation of post-synaptic currents and pulse frequency-dependent short-term potentiation were successfully emulated in the InGaZnO-based artificial synapse. The ITO/InGaZnO/ITO memristor exhibited spike-amplitude-dependent plasticity, spike-rate-dependent plasticity, and potentiation-depression synaptic learning with low energy consumption, making it a promising candidate for large-scale integration.

Research Progress on the Application of Topological Phase Transition Materials in the Field of Memristor and Neuromorphic Computing

Topological phase transition materials have strong coupling between their charge, spin orbitals, and lattice structure, which makes them have good electrical and magnetic properties, leading to promising applications in the fields of memristive devices. The smaller Gibbs free energy difference between the topological phases, the stable oxygen vacancy ordered structure, and the reversible topological phase transition promote the memristive effect, which is more conducive to its application in information storage, information processing, information calculation, and other related fields. In particular, extracting the current resistance or conductance of the two-terminal memristor to convert to the weight of the synapse in the neural network can simulate the behavior of biological synapses in their structure and function. In addition, in order to improve the performance of memristors and better apply them to neuromorphic computing, methods such as ion doping, electrode selection, interface modulation, and preparation process control have been demonstrated in memristors based on topological phase transition materials. At present, it is considered an effective method to obtain a unique resistive switching behavior by improving the process of preparing functional layers, regulating the crystal phase of topological phase transition materials, and constructing interface barrier-dependent devices. In this review, we systematically expound the resistance switching mechanism, resistance switching performance regulation, and neuromorphic computing of topological phase transition memristors, and provide some suggestions for the challenges faced by the development of the next generation of non-volatile memory and brain-like neuromorphic devices based on topological phase transition materials.

Intermittent Defect Fluctuations in Oxide Heterostructures

The heterogeneous nature, local presence, and dynamic evolution of defects typically govern the ionic and electronic properties of a wide variety of functional materials. While the last 50 years have seen considerable efforts into development of new methods to identify the nature of defects in complex materials, such as the perovskite oxides, very little is known about defect dynamics and their influence on the functionality of a material. Here, the discovery of the intermittent behavior of point defects (oxygen vacancies) in oxide heterostructures employing X-ray photon correlation spectroscopy is reported. Local fluctuations between two ordered phases in strained SrCoOx with different degrees of stability of the oxygen vacancies are observed. Ab-initio-informed phase-field modeling reveals that fluctuations between the competing ordered phases are modulated by the oxygen ion/vacancy interaction energy and epitaxial strain. The results demonstrate how defect dynamics, evidenced by measurement and modeling of their temporal fluctuations, give rise to stochastic properties that now can be fully characterized using coherent X-rays, coupled for the first time to multiscale modeling in functional complex oxide heterostructures. The study and its findings open new avenues for engineering the dynamical response of functional materials used in neuromorphic and electrochemical applications.

Complex Oxides for Brain-Inspired Computing: A Review

The fields of brain-inspired computing, robotics, and, more broadly, artificial intelligence (AI) seek to implement knowledge gleaned from the natural world into human-designed electronics and machines. In this review, the opportunities presented by complex oxides, a class of electronic ceramic materials whose properties can be elegantly tuned by doping, electron interactions, and a variety of external stimuli near room temperature, are discussed. The review begins with a discussion of natural intelligence at the elementary level in the nervous system, followed by collective intelligence and learning at the animal colony level mediated by social interactions. An important aspect highlighted is the vast spatial and temporal scales involved in learning and memory. The focus then turns to collective phenomena, such as metal-to-insulator transitions (MITs), ferroelectricity, and related examples, to highlight recent demonstrations of artificial neurons, synapses, and circuits and their learning. First-principles theoretical treatments of the electronic structure, and in situ synchrotron spectroscopy of operating devices are then discussed. The implementation of the experimental characteristics into neural networks and algorithm design is then revewed. Finally, outstanding materials challenges that require a microscopic understanding of the physical mechanisms, which will be essential for advancing the frontiers of neuromorphic computing, are highlighted.

Emerging Iontronic Neural Devices for Neuromorphic Sensory Computing

Living organisms have a very mysterious and powerful sensory computing system based on ion activity. Interestingly, studies on iontronic devices in the past few years have proposed a promising platform for simulating the sensing and computing functions of living organisms, because: 1) iontronic devices can generate, store, and transmit a variety of signals by adjusting the concentration and spatiotemporal distribution of ions, which analogs to how the brain performs intelligent functions by alternating ion flux and polarization; 2) through ionic-electronic coupling, iontronic devices can bridge the biosystem with electronics and offer profound implications for soft electronics; 3) with the diversity of ions, iontronic devices can be designed to recognize specific ions or molecules by customizing the charge selectivity, and the ionic conductivity and capacitance can be adjusted to respond to external stimuli for a variety of sensing schemes, which can be more difficult for electron-based devices. This review provides a comprehensive overview of emerging neuromorphic sensory computing by iontronic devices, highlighting representative concepts of both low-level and high-level sensory computing and introducing important material and device breakthroughs. Moreover, iontronic devices as a means of neuromorphic sensing and computing are discussed regarding the pending challenges and future directions.

Intensive harmonized synapses with amorphous Cu2O-based memristors using ultrafine Cu nanoparticle sublayers formed via atomically controlled electrochemical pulse deposition

Resistive random-access memory (RRAM) devices have significant advantages for neuromorphic computing but have fatal problems of uncontrollability and abrupt resistive switching behaviors degrading their synaptic performance. In this paper, we propose the electrochemical design of an active Cu2O layer containing a strategic sublayer of ultrafine Cu nanoparticles (U-Cu NPs) to form uniformly dispersed conducting filaments, which can effectively improve the reliability for analog switching of RRAM-based neuromorphic computing. The electrochemical pulse deposited (EPD) U-Cu NPs are linked to the bottom electrode through a semi-conductive path within the bottom Cu2O layer, since the EPD is preferentially carried out on the conductive sites. All Cu2O films with U-Cu NPs are developed in situ in the single electrolyte bath without any pause. The proposed U-Cu NPs can concentrate the external electric field and can generate conductive filament paths for analog resistive switching. The applied electric field was uniformly spread to U-Cu NPs at the center of the active layer and displays resistive switching behavior via multiple conductive filaments. This shows a strong harmony between the resistance-switching characteristics and the analog operation of the active layer. This RRAM device shows outstanding gradual analog switching, great linearity, dynamic range, endurance, precision, speed, and retention characteristics simultaneously and adequately for neuromorphic computing by realizing multiple weak filament-type operation.

Achieving reinforcement learning in a three-active-terminal neuromorphic device based on a 2D vdW ferroelectric material

Currently, for most three-terminal neuromorphic devices, only the gate terminal is active. The inadequate modes and freedom of modulation in such devices greatly hinder the implementation of complex neural behaviors and brain-like thinking strategies in hardware systems. Taking advantage of the unique feature of co-existing in-plane (IP) and out-of-plane (OOP) ferroelectricity in two-dimensional (2D) ferroelectric α-In2Se3, we construct a three-active-terminal neuromorphic device where any terminal can modulate the conductance state. Based on the co-operation mode, controlling food intake as a complex nervous system-level behavior is achieved to carry out positive and negative feedback. Specifically, reinforcement learning as a brain-like thinking strategy is implemented due to the coupling between polarizations in different directions. Compared to the single modulation mode, the chance of the agent successfully obtaining the reward in the Markov decision process is increased from 68% to 82% under the co-operation mode through the coupling effect between IP and OOP ferroelectricity in 2D α-In2Se3 layers. Our work demonstrates the practicability of three-active-terminal neuromorphic devices in handling complex tasks and advances a significant step towards implementing brain-like learning strategies based on neuromorphic devices for dealing with real-world challenges.

A high linearity and multilevel polymer-based conductive-bridging memristor for artificial synapses

Conductive-bridging memristors based on a metal ion redox mechanism have good application potential in future neuromorphic computing nanodevices owing to their high resistance switch ratio, fast operating speed, low power consumption and small size. Conductive-bridging memristor devices rely on the redox reaction of metal ions in the dielectric layer to form metal conductive filaments to control the conductance state. However, the migration of metal ions is uncontrollable by the applied bias, resulting in the random generation of conductive filaments, and the conductance state is difficult to accurately control. Herein, we report an organic polymer carboxylated chitosan-based memristor doped with a small amount of the conductive polymer PEDOT:PSS to improve the polymer ionic conductivity and regulate the redox of metal ions. The resulting device exhibits uniform conductive filaments during device operation, more than 100 and non-volatile conductance states with a ∼1 V range, and linear conductance regulation. Moreover, simulation using handwritten digital datasets shows that the recognition accuracy of the carboxylated chitosan-doped PEDOT:PSS memristor array can reach 93%. This work provides a path to facilitate the performance of metal ion-based memristors in artificial synapses.

Highly Reliable Textile-Type Memristor by Designing Aligned Nanochannels

Information-processing devices are the core components of modern electronics. Integrating them into textiles is the indispensable demand for electronic textiles to form close-loop functional systems. Memristors with crossbar configuration are regarded as promising building blocks to design woven information-processing devices that seamlessly unify with textiles. However, the memristors always suffer from severe temporal and spatial variations due to the random growth of conductive filaments during filamentary switching processes. Here, inspired by the ion nanochannels across synaptic membranes, a highly reliable textile-type memristor made of Pt/CuZnS memristive fiber with aligned nanochannels, showing small set voltage variation (<5.6%) under ultralow set voltage (≈0.089 V), high on/off ratio (≈106 ), and low power consumption (0.1 nW), is reported. Experimental evidence indicate that nanochannels with abundant active S defects can anchor silver ions and confine their migrations to form orderly and efficient conductive filaments. Such memristive performances enable the resultant textile-type memristor array to have high device-to-device uniformity and process complex physiological data like brainwave signals with high recognition accuracy (95%). The textile-type memristor arrays are mechanically durable to withstand hundreds of bending and sliding deformations, and seamlessly unified with sensing, power-supplying, and displaying textiles/fibers to form all-textile integrated electronic systems for new generation human-machine interactions.

Nanoscale multistate resistive switching in WO3 through scanning probe induced proton evolution

Multistate resistive switching device emerges as a promising electronic unit for energy-efficient neuromorphic computing. Electric-field induced topotactic phase transition with ionic evolution represents an important pathway for this purpose, which, however, faces significant challenges in device scaling. This work demonstrates a convenient scanning-probe-induced proton evolution within WO₃, driving a reversible insulator-to-metal transition (IMT) at nanoscale. Specifically, the Pt-coated scanning probe serves as an efficient hydrogen catalysis probe, leading to a hydrogen spillover across the nano junction between the probe and sample surface. A positively biased voltage drives protons into the sample, while a negative voltage extracts protons out, giving rise to a reversible manipulation on hydrogenation-induced electron doping, accompanied by a dramatic resistive switching. The precise control of the scanning probe offers the opportunity to manipulate the local conductivity at nanoscale, which is further visualized through a printed portrait encoded by local conductivity. Notably, multistate resistive switching is successfully demonstrated via successive set and reset processes. Our work highlights the probe-induced hydrogen evolution as a new direction to engineer memristor at nanoscale.

Two-Terminal Lithium-Mediated Artificial Synapses with Enhanced Weight Modulation for Feasible Hardware Neural Networks

Recently, artificial synapses involving an electrochemical reaction of Li-ion have been attributed to have remarkable synaptic properties. Three-terminal synaptic transistors utilizing Li-ion intercalation exhibits reliable synaptic characteristics by exploiting the advantage of non-distributed weight updates owing to stable ion migrations. However, the three-terminal configurations with large and complex structures impede the crossbar array implementation required for hardware neuromorphic systems. Meanwhile, achieving adequate synaptic performances through effective Li-ion intercalation in vertical two-terminal synaptic devices for array integration remains challenging. Here, two-terminal Au/LixCoO2/Pt artificial synapses are proposed with the potential for practical implementation of hardware neural networks. The Au/LixCoO2/Pt devices demonstrated extraordinary neuromorphic behaviors based on a progressive dearth of Li in LixCoO2 films. The intercalation and deintercalation of Li-ion inside the films are precisely controlled over the weight control spike, resulting in improved weight control functionality. Various types of synaptic plasticity were imitated and assessed in terms of key factors such as nonlinearity, symmetricity, and dynamic range. Notably, the LixCoO2-based neuromorphic system outperformed three-terminal synaptic transistors in simulations of convolutional neural networks and multilayer perceptrons due to the high linearity and low programming error. These impressive performances suggest the vertical two-terminal Au/LixCoO2/Pt artificial synapses as promising candidates for hardware neural networks.

Applications of Graphene in Five Senses, Nervous System, and Artificial Muscles

Graphene remains of great interest in biomedical applications because of biocompatibility. Diseases relating to human senses interfere with life satisfaction and happiness. Therefore, the restoration by artificial organs or sensory devices may bring a bright future by the recovery of senses in patients. In this review, we update the most recent progress in graphene based sensors for mimicking human senses such as artificial retina for image sensors, artificial eardrums, gas sensors, chemical sensors, and tactile sensors. The brain-like processors are discussed based on conventional transistors as well as memristor related neuromorphic computing. The brain-machine interface is introduced for providing a single pathway. Besides, the artificial muscles based on graphene are summarized in the means of actuators in order to react to the physical world. Future opportunities remain for elevating the performances of human-like sensors and their clinical applications.

Phase Transition Dynamics in a Complex Oxide Heterostructure

Understanding the behavior of defects in the complex oxides is key to controlling myriad ionic and electronic properties in these multifunctional materials. The observation of defect dynamics, however, requires a unique probe-one sensitive to the configuration of defects as well as its time evolution. Here, we present measurements of oxygen vacancy ordering in epitaxial thin films of SrCoO_{x} and the brownmillerite-perovskite phase transition employing x-ray photon correlation spectroscopy. These and associated synchrotron measurements and theory calculations reveal the close interaction between the kinetics and the dynamics of the phase transition, showing how spatial and temporal fluctuations of heterointerface evolve during the transformation process. The energetics of the transition are correlated with the behavior of oxygen vacancies, and the dimensionality of the transformation is shown to depend strongly on whether the phase is undergoing oxidation or reduction. The experimental and theoretical methods described here are broadly applicable to in situ measurements of dynamic phase behavior and demonstrate how coherence may be employed for novel studies of the complex oxides as enabled by the arrival of fourth-generation hard x-ray coherent light sources.

Nanostructured perovskites for nonvolatile memory devices

Perovskite materials have driven tremendous advances in constructing electronic devices owing to their low cost, facile synthesis, outstanding electric and optoelectronic properties, flexible dimensionality engineering, and so on. Particularly, emerging nonvolatile memory devices (eNVMs) based on perovskites give birth to numerous traditional paradigm terminators in the fields of storage and computation. Despite significant exploration efforts being devoted to perovskite-based high-density storage and neuromorphic electronic devices, research studies on materials' dimensionality that has dominant effects on perovskite electronics' performances are paid little attention; therefore, a review from the point of view of structural morphologies of perovskites is essential for constructing perovskite-based devices. Here, recent advances of perovskite-based eNVMs (memristors and field-effect-transistors) are reviewed in terms of the dimensionality of perovskite materials and their potentialities in storage or neuromorphic computing. The corresponding material preparation methods, device structures, working mechanisms, and unique features are showcased and evaluated in detail. Furthermore, a broad spectrum of advanced technologies (e.g., hardware-based neural networks, in-sensor computing, logic operation, physical unclonable functions, and true random number generator), which are successfully achieved for perovskite-based electronics, are investigated. It is obvious that this review will provide benchmarks for designing high-quality perovskite-based electronics for application in storage, neuromorphic computing, artificial intelligence, information security, etc.

Resistive Switching Memristor: On the Direct Observation of Physical Nature of Parameter Variability

Ion-based memristive switching has attracted widespread attention from industries owing to its outstanding advantages in storage and neuromorphic computing. Major issues for achieving brain-inspired computation of highly functional memory in redox-based ion devices are relatively large variability in their operating parameters and limited cycling endurance. In some devices, volatile and nonvolatile operations often replace each other without changing operating conditions. To address these issues, it is important to observe directly what is happening in repeated operations. Herein, we use a planar device that enables direct capturing of microscopic behaviors in the nucleation and growth of metal whiskers under repeated switching to verify the microscopic origin of the large parameter variability. We report direct observations that reveal the physical origin for the large cycle-to-cycle and device-to-device variability in memristive switching, which was achieved using planar polymer atomic switches with a gap >1 μm. We find that the deposition location of metal atoms is closely related to the crystallinity of the ion transport layer (solid polymer electrolyte, SPE). The filament variability (shape, position, quantity, etc.) during different cycles and devices is indeed the main reason for the observed variability in the operating characteristics. The results shed unique light on the complexity of the operation of the ion device, that is, the evolution of the dielectric layer and metal filament must be considered.