Neuromorphic Behavior Research Articles

Control of the neuromorphic learning behavior based on the aggregation of thiol-protected Ag-Ag2S core–shell nanoparticles

The neuromorphic learning switching behavior was investigated in electric devices that were constructed based on the aggregation of thiol-stabilized Ag/Ag2S core–shell nanoparticles (NPs). The NPs were synthesized using the two-phase modified Brust–Schiffrin procedure and exhibited Ag–S and Ag–S–R bonding states at the surfaces of Ag NPs. The memristive behavior of such a device at room temperature under ambient pressure, which can be used to emulate the functions of biological synapses with long and short memories, is achieved. The importance of the Ag2S and the thiol layer at the surface of Ag NPs for the demonstration of learning behavior, such as the potentiation and depression of synapses in the human brain is explained.

Fabricating Molybdenum Disulfide Memristors

Memristors with simple two-terminal structures have great potential for use in high-density memory and neuromorphic devices. Many neuromorphic behaviors can be imitated using a two-terminal device....

CMOS Compatible Hf0.5Zr0.5O2 Ferroelectric Tunnel Junctions for Neuromorphic Devices

While machine learning algorithms are becoming more and more elaborate, their underlying artificial neural networks most often still rely on the binary von Neumann computer architecture. However, artificial neural networks access their full potential when combined with gradually switchable artificial synapses. Herein, complementary metal oxide semiconductor‐compatible Hf0.5Zr0.5O2 ferroelectric tunnel junctions fabricated by radio‐frequency magnetron sputtering are used as artificial synapses. On a single synapse level, their neuromorphic behavior is quantitatively investigated with spike‐timing‐dependent plasticity. It is found that the learning rate of the synapses mainly depends on the voltage amplitude of the applied stimulus. The experimental findings are well reproduced with simulations based on the nucleation‐limited‐switching model.

Parylene Based Memristive Devices with Multilevel Resistive Switching for Neuromorphic Applications

In this paper, the resistive switching and neuromorphic behaviour of memristive devices based on parylene, a polymer both low-cost and safe for the human body, is comprehensively studied. The Metal/Parylene/ITO sandwich structures were prepared by means of the standard gas phase surface polymerization method with different top active metal electrodes (Ag, Al, Cu or Ti of ~500 nm thickness). These organic memristive devices exhibit excellent performance: low switching voltage (down to 1 V), large OFF/ON resistance ratio (up to 104), retention (≥104 s) and high multilevel resistance switching (at least 16 stable resistive states in the case of Cu electrodes). We have experimentally shown that parylene-based memristive elements can be trained by a biologically inspired spike-timing-dependent plasticity (STDP) mechanism. The obtained results have been used to implement a simple neuromorphic network model of classical conditioning. The described advantages allow considering parylene-based organic memristors as prospective devices for hardware realization of spiking artificial neuron networks capable of supervised and unsupervised learning and suitable for biomedical applications.

Neuromorphic switching behavior in multi-stacking composed of Pt/graphene oxide/Ag2S/Ag

The present work reports a modification in the design of the previously known Ag/Ag2S neuron-inspired devices by the addition of a modulating layer of graphene oxide (GO). The Ag2S/GO structures were fabricated with a simple yet effective method consisting of the sulfurization of high-purity silver wires dip-coated in GO solution. The structure was observed to maintain the switching characteristics of Ag2S, but also to exhibit tunneling OFF state currents modulated by the GO thin film layer. This result was also confirmed through the direct observation of the Ag2S/GO surface with a scanning electron microscope. The notable volatility-based neuromorphic properties observed in Ag2S were enhanced in the presence of GO, and a switching model is proposed considering the present experimental results. The ultra-thin insulating nature of GO emphasizes its potential modulating capabilities, and the results suggest that GO may be an important addition for future electrochemical resistive switching based neuron-inspired devices.

Memristor in a Reservoir System-Experimental Evidence for High-Level Computing and Neuromorphic Behavior of PbI2.

Lead halides in an asymmetric layered structure form memristive devices which are controlled by the electronic structure of the PbX2|metal interface. In this paper, we explain the mechanism that stands behind the I- V pinched hysteresis loop of the device and shortly present its synaptic-like plasticity (spike-timing-dependent plasticity and spike-rate-dependent plasticity) and nonvolatile memory effects. This memristive element was incorporated into a reservoir system, in particular, the echo-state network with delayed feedback, which exhibits brain-like recurrent behavior and demonstrates metaplasticity as one of the available learning mechanisms. It can serve as a classification system that classifies input signals according to their amplitude.

Recent Developments and Perspectives for Memristive Devices Based on Metal Oxide Nanowires

AbstractMemristive devices are considered one of the most promising candidates to overcome technological limitations for realizing next‐generation memories, logic applications, and neuromorphic systems in the modern nanoelectronics and information technology. Despite the continuous efforts, understanding the resistive switching mechanism underlying memristive/neuromorphic behavior still represents a challenge. Metal oxide nanowire‐based memristors appear suitable model systems for a deeper understanding of the involved physical/chemical phenomena due the possibility for localizing and investigating the switching mechanism. In practical aspects, nanowire‐based devices can be grown using a bottom‐up approach, thus being considered reliable candidates for going beyond the current scaling limitations of the top‐down approach by the standard lithography. In addition, taking into advantage of the high surface‐to‐volume ratio of these nanostructures, new device functionalities can be achieved by exploiting surface effects. In the literature, a variety of nanowire‐based devices such as single nanowires, nanowire arrays, and networks are reported to exhibit memristive behavior, explained by different switching mechanisms. This work provides a comparative review and a comprehensive analysis of device performances and physical phenomena responsible for memristive and neuromorphic behavior in such nanostructures. The analysis of the state‐of‐art of memristor devices based on nanowires and nanorods represent a milestone toward the development of nanowire‐based artificial neural networks.

An electronic synapse memristor device with conductance linearity using quantized conduction for neuroinspired computing

An electrochemical metallization memristor based on Zr0.5Hf0.5O2film and an active Cu electrode with quantum conductance and neuromorphic behavior has been reported in this work.

Synaptic dynamics in complex self-assembled nanoparticle networks.

We report a detailed study of neuromorphic switching behaviour in inherently complex percolating networks of self-assembled metal nanoparticles. We show that variation of the strength and duration of the electric field applied to this network of synapse-like atomic switches allows us to control the switching dynamics. Switching is observed for voltages above a well-defined threshold, with higher voltages leading to increased switching rates. We demonstrate two behavioral archetypes and show how the switching dynamics change as a function of duration and amplitude of the voltage stimulus. We show that the state of each synapse can influence the activity of the other synapses, leading to complex switching dynamics. We further demonstrate the influence of the morphology of the network on the measured device properties, and the constraints imposed by the overall network conductance. The correlated switching dynamics, device stability over long periods, and the simplicity of the device fabrication provide an attractive pathway to practical implementation of on-chip neuromorphic computing.

Mimicking synaptic functionality with an InAs nanowire phototransistor

We demonstrate a nanowire (NW) phototransistor with synaptic behavior based on inherent persistent photoconductivity. The device is comprised of a single crystalline InAs NW, covered by a native indium oxide layer acting as the photogating layer (PGL). In the negative photoresponse range, the device mimics synaptic neuromorphic behaviors of short-term plasticity, long-term plasticity (LTP), and paired-pulse facilitation. Moreover, the transition from short-term to LTP is observed as the stimulus intensity increases, behaving in accord with the feature of cooperativity. The synaptic behaviors of the device are attributed to the photo-generated electrons trapped/detrapped in the PGL. This NW-based photonic synaptic device would find promising applications in neuromorphic systems and networks.

Bidirectionally-trigged 2D MoS2 synapse through coplanar-gate electric-double-layer polymer coupling for neuromorphic complementary spatiotemporal learning

Abstract Synapse is not merely the connection of two adjacent neurons in the human brain but also the fundamental structure for information transmission in an ever-changing pattern. Therefore, the hardware implementation of artificial synapses/neurons with intriguing spatiotemporal-processed functions is of great significance for brain-like intelligent electronics. Herein, bidirectionally-trigged coplanar-gate electric-double-layer two-dimensional (2D) MoS2 neuromorphic transistors with complementary spatiotemporal learning abilities are successfully demonstrated through the laterally-coupled and proton-conducting poly (vinyl alcohol) electrolytes. The fundamental bidirectional neuromorphic behaviors such as excitatory and inhibitory postsynaptic current (EPSC, IPSC), excitatory and inhibitory paired-pulse facilitation (E-PPF, I-PPF), and bidirectional frequency-dependent dynamic filtering from high-pass to low-pass transitions were successfully mimicked. Most importantly, for the first time, with the multi-coplanar-gates as inputs, the complementary spatiotemporal learning and anti-spatiotemporal learning, and bidirectional tunable spiking logic dynamics were experimentally demonstrated. Our results could provide a promising opportunity for adding the intelligent complementary spatiotemporal-learning functions in brain-like 2D nanoscale bioelectronics.

Coplanar Multigate MoS2 Electric-Double-Layer Transistors for Neuromorphic Visual Recognition.

Spatial coordinate and visual orientation recognition in cortical cells play important roles in the visual system. Herein, spatiotemporally processed visual neurons are mimicked by a facile coplanar multigate two-dimensional (2D) MoS2 electric-double-layer transistor with proton-conducting poly(vinyl alcohol) electrolytes as laterally coupled gate dielectrics. Fundamental neuromorphic behaviors, e.g., excitatory postsynaptic current and paired-pulse facilitation, were successfully mimicked. For the first time, a proof-of-principle artificial visual neural network system for mimicking spatiotemporal coordinate and orientation recognition was experimentally demonstrated in such devices. The experimental results provide a promising opportunity for adding intelligent spatiotemporally-processed functions in emerging brain-like neuromorphic nanoelectronics.

Flexible Neuromorphic Architectures Based on Self-Supported Multiterminal Organic Transistors.

Because of the fast expansion of artificial intelligence, development and applications of neuromorphic systems attract extensive interest. In this paper, a highly interconnected neuromorphic architecture (HINA) based on flexible self-supported multiterminal organic transistors is proposed. Au electrodes, poly(3-hexylthiophene) active channels, and ion-conducting membranes were combined to fabricate organic neuromorphic devices. Especially, freestanding ion-conducting membranes were used as gate dielectrics as well as support substrates. Basic neuromorphic behavior and four forms of spike-timing-dependent plasticity were emulated. The fabricated neuromorphic device showed excellent electrical stability and mechanical flexibility after 1000 bends. Most importantly, the device structure is interconnected in a way similar to the neural architecture of the human brain and realizes not only the structure of the multigate but also characteristics of the global gate. Dynamic processes of memorizing and forgetting were incorporated into the global gate matrix simulation. Pavlov's learning rule was also simulated by taking advantage of the multigate array. Realization of HINAs would open a new path for flexible and sophisticated neural networks.

(Invited) Self Healing Interconnects for Reliable Flexible Electronics

The recent decade has seen a surge in research and development on flexible and wearable electronic systems. These systems consist of thin film transistor integrated circuits or off-shelf intergrated circuit chips fabricated and packaged onto a mechanically flexible substrate. Mechanically flexible electronics offers a host of functional advantages resulting in technologies such as foldable displays, x-ray imagers conformable to the human body, spherical image sensors offering wide angle view etc. However these advantages are accompanied by a number of reliability problems of which one major problem is the reliability of interconnects. Interconnects on flexible and wearable devices are prone to significant structural and electrical stress due to mechanical flexing, electro static discharge etc. and are subject to open circuit failure [1]. Interconnects have been conventionally made more robust to open circuit failure by several passive approaches such as the use of geometry (eg. meandered interconnects stretch more) or novel materials (eg. conductive elastomers) [2]. However, despite these precautions, open faults are expected to occur and there isn’t much that can be done once they do occur. Active techniques on the other hand offer online repair of the interconnect fault [3]-[7]. We propose an active technique for self healing open faults in interconnects by the use of a dispersion of electrically conductive particles in an insulating fluid [4]-[7]. This dispersion is contained and isolated over each interconnect. Upon the occurrence of an open fault in a current carrying interconnect, the field appearing across the open gap polarizes the conductive particles in the dispersion. The polarized particles experience dipole-dipole attractive forces and eventually chain up to create a bridge across the gap thereby healing the fault (seen in (a) in the figure). This mechanism is truly self healing as it turns on and shuts off by itself without external interference. The occurrence of the fault creates the electric field that triggers the mechanism. After the repair of the fault, the electric field across the open gap disappears thereby stopping the mechanism. The heal effectiveness is determined by the dispersion concentration, geometry of particles used, viscosity and permittivity of the fluid and the electric field across the open gap. We discuss the methods of implementation with circuits on flex and rigid substrates (seen in (b) in the figure) and the impact of circuit parameters on the heal time and heal impedance (current carrying capacity) in real world implementation. The limitations and structural stability of the heal are also discussed. In this talk we not only discuss the mechanism of self healing, but also some remarkable properties shown by the heal. Firstly, the heal shows active re-routability i.e. it can be re-oriented from one set of electrodes to another thereby permitting a relay like operation [6] (seen in (c) in the figure). Secondly, the heal shows a neuromorphic behaviour where it ‘remembers’ a connection between routes, ‘forgets’ the connection if programmed, and ‘re-learns’ the connection faster once re-engaged [6]. Finally, the heal shows high stretchability that makes it very suited to electronics on elastomers. Reference: Sambandan, R. B. Apte, W. S. Wong, R. Lujan, M. Young, B. Russo, S. Ready, and R. A. Street, IEEE J Display Tech., 5, 27, (2009)Dang, V. Vinciguerra , L. Lorenzelli and R. Dahiya, IOP Flex. Print. Electron. 2 013003 (2017)A. Odom, M. M. Caruso, A. D. Finke, A. M. Prokup, J. A. Ritchey, J. H. Leonard, S. R. White, N. R. Sottos and J. S. Moore, Adv. Funct. Mater. 20, 1721 (2010).J. Blaiszik, S. L. B. Kramer, M. E. Grady, D. A. Mcllroy, J. S. Moore, N. R. Sottos, and S. R. White, Adv. Mater. 24, 398 (2012).Sambandan, IEEE Trans. Elec. Dev. 59, 1773 (2012).Nair, K. Raghunandan, V. Yaswant, S. Pillai and S. Sambandan, Appl. Phys. Lett. 106 123103 (2015).Yaswanth, A. Kumar and S. Sambandan, Appl. Phys. Letts. 109 024101 (2016). Figure 1

Unsupervised Learning of Polychronous Wavefront Computation Configurations for Pattern Recognition

Polychronous Wavefront Computation (PWC) provides a potentially simple model for large scale implementation of spiking neural networks and deep learning. Although the definition of predefined pattern recognition configurations has been demonstrated, dynamic organization of configurations from examples remains a difficult problem. This paper explores the hypothesis that a properly arranged field of PWC transponders with neuromorphic behaviors can self-organize into recognition configurations based on training examples. The PWC transponders used are augmented with a position learning algorithm based on spike-timing-dependent plasticity, suppression of non-specific transponders using a stimulation fatigue approach and deactivation of unused transponders using potentiation decay. The paper provides the results of initial research demonstrating that pattern recognition configurations can be learned if the initial density and distribution of transponders is properly selected with respect to the learning behavior parameters. The effectiveness of the learning process can be improved by encoding layering information in the wavefronts to focus the transponder activations. The results define a means for PWC transponders to self-organize into recognition configurations providing a basis for development of more complex configurations and deep learning applications.

(Invited) Resistance Switching in Silicon-Rich Silica: Electronic, Structural and Photonic Perspectives

We present results from a study of resistance switching in silicon-rich silica – ie silicon dioxide containing an excess of silicon. We demonstrate that changes in resistance are the result of field-driven movement of oxygen, and that this can result in large-scale changes in oxide structure and stoichiometry. While resistance switching in oxides has been studied for a number of years as a route to non-volatile electronic memories, the optical response of such systems has remained largely unknown. Here we demonstrate that simple two-terminal MIM devices can be switched between different resistance states using a combination of electrical bias and optical stimulation. This opens up intriguing possibilities for novel classes of sensors. We also demonstrate neuromorphic behaviour in our devices. Our electrical measurements show that, under the right conditions, the devices can be made to emulate some of the important behaviours of biological neurons – plasticity, for example. Taken together with the light sensitivity of these resistance switches, this opens up further possibilities for the development of light-triggered neural networks for such applications as pattern recognition and image classification.

Multi-gate organic neuron transistors for spatiotemporal information processing

Due to similar transmission characteristics of biological synaptic activities, neuromorphic behaviors simulated by organic electrochemical transistors (OECTs) is of great interest. In this letter, the fabrication and performance of multi-gate poly(3-hexylthiophene) (P3HT) OECTs with ion-gel gating are reported. The neuromorphic behaviors, such as dendrite correlated excitatory postsynaptic current (EPSC), paired pulse facilitation, and modulation, were simulated in the OECTs. These behaviors were observed to depend on the degree of temporal correlation and distance between the in-plane-gate and the channel. More importantly, by using dendritic integration from two different gates, spatiotemporally correlated outputs were also emulated. The spatial orientations of the input pulse are defined, and changing the orientation will result in a change in the EPSC amplitude. Our results provide a way to construct spatiotemporally neural network based on multi-gate OECTs.

Nanosecond Programming of SiOx Based Resistive RAM: Simplified Digital Programmers and Analog Neuromorphic Type Behavior

Typical non-volatile flash memory technology has been approaching its scalability limit and showing slow access time, and high power dissipation. Meanwhile, filamentary resistive RAM (ReRAM) offers scalability, low-voltage nanosecond writing and therefore a low power profile; programming techniques also allow for beyond-binary applications. Here we stimulate our intrinsic switching TiN/SiOx/TiN (x<2) ReRAM devices using nanosecond pulses and look at aspects of their electrical response. Simplified programmers appear possible thanks to dynamic impedance mismatch between the programmer and the device under test; identical pulse sequences switch devices between two or more states. Such mismatch also allows us to track the electroforming process and confirm it occurs on the nanosecond timescale. Additionally, pulses of appropriate timing allow devices to enter temporary states. Behavior reminiscent of neuronal synapses (potentiation, depression and short-term memory) is therefore observed in our devices showing great potential for integration into novel parallel hardware neural networks.

Associative Enhancement of Time Correlated Response to Heterogeneous Stimuli in a Neuromorphic Nanowire Device

In spite of the strong interest in brain‐like or neuromorphic computation, relatively few devices have emerged whose neuromorphic behavior is embedded in the hardware itself and not reliant on external programming of synaptic weights. Here, a neuromorphic device is described based on a TiO2 nanowire that exhibits an associative memory response to the time correlation between voltage and optical stimuli. Memristive characteristics are also observed with current–voltage sweeps showing hysteresis loops and continuum resistance levels. The nanowire device responds to heterogeneous voltage and optical pulse stimuli with spike‐like neuromorphic outputs. Moreover, uncorrelated pulses produce a weak response, consistent with the interaction of coincident pulses with adsorbed and bulk oxygen in the surface depletion region, leading to a nonlinear enhancement in conductance. The strength of this learned enhancement depends on both the time correlation and the number of pulse stimuli, consistent with spike timing dependent plasticity. The nanowire devices presented have neural synapse‐like properties that could serve as a building block for neuromorphic computation.

Neuromorphic behavior in percolating nanoparticle films.

We show that the complex connectivity of percolating networks of nanoparticles provides a natural solid-state system in which bottom-up assembly provides a route to realization of neuromorphic behavior. Below the percolation threshold the networks comprise groups of particles separated by tunnel gaps; an applied voltage causes atomic scale wires to form in the gaps, and we show that the avalanche of switching events that occurs is similar to potentiation in biological neural systems. We characterize the level of potentiation in the percolating system as a function of the surface coverage of nanoparticles and other experimentally relevant variables, and compare our results with those from biological systems. The complex percolating structure and the electric field driven switching mechanism provide several potential advantages in comparison to previously reported solid-state neuromorphic systems.