Neuromorphic Applications Research Articles

Spatial-Temporal Hybrid Neural Network With Computing-in-Memory Architecture

Deep learning (DL) has gained unprecedented success in many real-world applications. However, DL poses difficulties for efficient hardware implementation due to the needs of a complex gradient-based learning algorithm and the required high memory bandwidth for synaptic weight storage, especially in today's data-intensive environment. Computing-in-memory (CIM) strategies have emerged as an alternative for realizing energy-efficient neuromorphic applications in silicon, reducing resources and energy required for neural computations. In this work, we exploit a CIM-based spatial-temporal hybrid neural network (STHNN) with a unique learning algorithm. To be specific, we integrate both multilayer perceptron and recurrent-based delay-dynamical system, making the network becomes linear separable while processing information in both spatial and temporal domains, better yet, reducing the memory bandwidth and hardware overhead through the CIM architecture. The prototype fabricated in 180 nm CMOS process is built of fully-analog components, yielding an average on-chip classification accuracy up to 86.9% on handprinted alphabet characters with a power consumption of 33 mW. Beyond that, through the handwritten digit database and the radio frequency fingerprinting dataset, software-based numerical evaluations offer 1.6 -to- 9.8 × and 1.9 -to- 4.4 × speedup, respectively, without significantly degrading its classification accuracy compared to the cutting-edge DL approaches.

The Role of the Internal Capacitance in Organic Memristive Device for Neuromorphic and Sensing Applications

AbstractOrganic electronics has recently emerged as a promising candidate for the emulation of brain‐like functionalities, especially at the device level. Among the proposed technologies, memristive devices have gained an increasing attention due to their non‐volatile behavior which makes them suitable for the implementation of artificial neuronal networks. However, most of them have an energy‐costly switching mechanism which limits the approach of brain like energy efficiency. Different from them, organic memristive devices (OMDs) have a narrow switching window and implement neuromorphic characteristics at voltages ≤ 1 V. Despite OMDs potentialities in bioinspired electronics, guidelines for the design of devices and materials are still missing. Here it is shown that the device capacitance represents a significant degree of freedom for targeting devices applications. It is also shown that a single OMD emulates activity dependent synaptic functions and neuronal temporal and spatial summation, taking advantage of its three‐terminals configuration. Interestingly, despite the neuromorphic applications, OMDs can also sense and amplify incoming signals on the basis of their capacitive and/or resistive values. This spectrum of applications, ranging from volatile to non‐volatile characteristics and from neuromorphic computing to bio signals sensing, sets the stage for the realization of integrated circuits for adaptive sensing.

Spinodal Decomposition-Driven Endurable Resistive Switching in Perovskite Oxides.

Common pursuits of developing nanometric logic and neuromorphic applications have motivated intensive research studies into low-dimensional resistive random-access memory (RRAM) materials. However, fabricating resistive switching medium with inherent stability and homogeneity still remains a bottleneck. Herein, we report a self-assembled uniform biphasic system, comprising low-resistance 3 nm-wide (Bi0.4,La0.6)FeO3-δ nanosheets coherently embedded in a high-resistance (Bi0.2,La0.8)FeO3-δ matrix, which were spinodally decomposed from an overall stoichiometry of the (Bi0.24,La0.76)FeO3-δ parent phase, as a promising nanocomposite to be a stable and endurable RRAM medium. The Bi-rich nanosheets accommodating high concentration of oxygen vacancies as corroborated by X-ray photoelectron spectroscopy and electron energy loss spectroscopy function as fast carrier channels, thus enabling an intrinsic electroforming-free character. Surficial electrical state and resistive switching properties are investigated using multimodal scanning probe microscopy techniques and macroscopic I-V measurements, showing high on/off ratio (∼103) and good endurance (up to 1.6 × 104 cycles). The established spinodal decomposition-driven phase-coexistence BLFO system demonstrates the merits of stability, uniformity, and endurability, which is promising for further application in RRAM devices.

TSSM: Three-State Switchable Memristor Model Based on Ag/TiOx Nanobelt/Ti Configuration

Memristive technologies are attractive due to their nonvolatility, high density, low power, nanoscale geometry, nonlinearity, binary/multiple memory capacity, and negative differential resistance. For memristive devices, a model corresponding with practical behavioral characteristics is highly favorable for the realization of its neuromorphic system and applications. In this paper, we propose a novel memristor model based on the Ag/TiOx nanobelt/Ti configuration, which can reflect three different states (i.e. original stage, transition stage, and resistive switching state) of the physical memristor with a satisfactory fitting precision (greater than 99.88%). Meanwhile, this work gives (1) an insight onto the electrical characteristics of the memristor model under different humidity conditions; (2) the influence of the water molecular concentration on the memristor behavior, which is of importance for the memristor fabrication and subsequent applications. For verification purposes, the proposed three-state switchable memristor is applied into the memristor-based logic implementation. The experimental results demonstrate that the constructed circuit is able to realize basic Boolean logic operations with fast response speed and high efficiency.

A Synaptic Transistor Based on Monolayer Monocrystalline‐MoS2 for Neuromorphic Applications

Ultrathin 2D semiconductor materials have attracted tremendous attention due to their potential application in neuromorphic computing, especially for electronic devices with high stability and low energy consumption. Two‐terminal memristors cannot extend the range of functionalities for complex neuromorphic applications due to their inherent single presynaptic input; thus three‐terminal synaptic transistors based on monolayer monocrystalline‐MoS2 films grown by chemical vapor deposition (CVD) are fabricated in this work. Importantly, there are two modes including drain and gate dual tunability available in nonvolatile memory functions. Both drain terminal and gate terminal can be used as presynaptic inputs to successfully simulate excitatory/inhibitory postsynaptic current (EPSC/IPSC), spike‐amplitude‐dependent plasticity (SADP), spike‐timing‐dependent plasticity (STDP), and long‐term plasticity (LTP) of biological synapses, providing multiple degrees of freedom for synaptic weight modulation. The conductance modulation of drain and gate terminals can be attributed to the migration and charges trapping/detrapping process of sulfur vacancies. Using monocrystalline‐MoS2 films, the device not only attains a lower operation voltage and a higher endurance (5000 cycles), but exhibit low energy consumption during the LTP process (≈3.75 pJ) with the gate terminal, indicating more flexibility and great potential of complex neuromorphic application.

Sputtered Oxide Thin-Film Transistors With Tunable Synaptic Spiking Behavior at 1 V

Indium-gallium-zinc-oxide (IGZO) thin-film transistors (TFTs) gated with sputtered silicon dioxide dielectric were fabricated. Under different sputtering pressures of silicon dioxide, the device is able to produce transfer curves from clockwise hysteresis to anticlockwise hysteresis, corresponding to different operation modes and thus adding complementary performance by the same material system and device structure. In the interface trapping mode, the transistor exhibited inhibitory synaptic spiking behavior induced by positive stimuli. In the electric double-layer coupling mode, positive stimuli led to excitatory synaptic spiking behavior, while negative stimuli resulted in inhibitory synaptic spiking behavior. All the transistor and synaptic spiking behaviors are conducted within ±1 V range. Enriched functionality and ultralow operation voltage make sputtered SiO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> -gated IGZO TFTs promising candidate for neuromorphic applications.

Performance Assessment of Amorphous HfO2-Based RRAM Devices for Neuromorphic Applications

The use of thin layers of amorphous hafnium oxide has been shown to be suitable for the manufacture of Resistive Random Access memories (RRAM). These memories are of great interest because of their simple structure and non-volatile character. In this work, it was studied the performance of the MIM structure that takes part of a 4 kbit memory array based on 1-transistor-1-resistance (1T1R) cells, in terms of memory maps and reading and writing loops. In each cell of the memory, a NMOS transistor is placed in series to a metal-insulator-metal (MIM) resistor formed by a TiN/a-HfO2/Ti/TiN structure. The 8 nm-thick HfO2 layers were grown by Atomic Layer Deposition (ALD); all metal films were deposited by magnetron sputtering. The measurements were carried out on MIM structures identical to those of the memory cell but with a larger area, manufactured on a different region of the chip. A previous study about the reduction of the programming pulse width in order to accomplish fast and low-energy switching operations as well as its influence in terms of endurance and data retention times has been published (Pérez et al. Electronics 9, p. 864, 2020). In addition, Logic-In-Memory (LIM) circuits based on this 4-kb RRAMs arrays have been analyzed in terms of reliability and energy consumption tradeoff (Zanotti et al. IEEE Trans. Electron. Devices 67, 10, p. 4611, 2020).In these MIM structures, the resistance value may decrease due to the creation of oxygen vacancy columns that act as conductive filaments between the two electrodes. This process is reversible and controllable, so that two well-defined states can be established. In addition, it is also possible to access intermediate states if appropriate dc bias waveforms are applied. In fact, by biasing with variable amplitude-triangular voltage signals, cumulative writing and erasing cycles can be performed. To obtain the memory maps, return-to-zero pulses rather than ramp voltages bias have to be used. Accordingly, the samples are biased by a train of square pulses with increasing amplitude; after each pulse the current value under a low bias value is measured. The memory map is built from the current values at 0.1 V measured immediately after applying each programming pulse. To carry out a detailed electrical characterization of these devices becomes essential to shed light on the physical nature of the mechanisms of creation and dissolution of the conducting filament, and ultimately to control the memory states. The controllable multilevel ability of these memory devices, due to their resistance transition in a gradual way, enables them to be used not only in the RAM memory field, but also in neuromorphic circuits based on the artificial synapse.It has been shown in HfO2-based RRAM devices that small signal parameters in a wide range of frequencies also show repetitiveness and intermediate states control, and allow to obtain memory maps testing the device at 0 V dc bias, hence without any static power consumption. (Castán et al. J. Appl. Phys. 124, 152101, 2018).In Fig. 1.a I-V cycles showing SET and RESET states are depicted. The blue-colored cycles are the most stable, whereas the very initial ones (red color) show some instability due to the fact that the filament formation process does not seem to have finished yet, and the last ones (green color) already exhibit aging effects.The memory map in Fig.1.b shows good window width values. SET transition is markedly abrupt compared to RESET transitions, which is much more gradual. This means that the best control can be obtained in the RESET transition.The accuracy of the intermediate current values control between SET and RESET is clearly shown in Figs. 1.c and 1.d. Both write and erase operations are carried out incrementally with a very precise control of the successive states through which the memory passes. The initial state must be well established by driving the memory to a full RESET (write operation) or to a full SET (erase operation). To achieve this, a negative or positive enough voltage, respectively, should be applied to the top electrode (with bottom electrode grounded).Similarly, an accurate control of admittance parameters was achieved during cumulative writing and erasing processes. Memory maps were also plotted for real (conductance, G) and imaginary (capacitance, C) components (Fig. 2).Moreover, endurance measurements made up to 6144 SET-RESET cycles exhibited excellent repetitiveness of current values in both states, as well as admittance parameters stability.To sum up, the electrical properties of these RRAM devices in terms of endurance and intermediate states control make them suitable for neuromorphic computing. Figure 1

CMOS back-end-of-line compatible ferroelectric tunnel junction devices

Ferroelectric tunnel junction devices based on ferroelectric thin films of solid solutions of hafnium dioxide can enable CMOS integration of ultra-low power ferroelectric devices with potential for memory and emerging computing schemes such as in-memory computing and neuromorphic applications. In this work, we present ferroelectric tunnel junctions based on Hf0.5Zr0.5O2 with materials and processes compatible with CMOS back-end-of-line integration. We show a device architecture based on W-Hf0.5Zr0.5O2-Al2O3-TiN stacks featuring low temperature annealing at 400° with performance comparable to those obtained with higher temperature annealing conditions.

Flexible Ta/TiO x /TaO x /Ru memristive synaptic devices on polyimide substrates

It is very urgent to build memristive synapses and even wearable devices to simulate the basic functions of biological synapses. The linear conductance modulation is the basis of analog memristor for neuromorphic computing. By optimizing the interface engineering wherein Ta/TiO x /TaO x /Ru was fabricated, all the memristor devices with different TiO x thickness showed electroforming-free property. The short-term and long-term plasticity in both potentiation and depression behaviors can be mimicked when TiO x was fixed at 25 nm. The presented memristive synapses simulated the stable paired-pulse facilitation and spike-timing dependent plasticity performance. The potentiation and depression in linearity and symmetry improved with the TiO x thickness increasing, which provides the feasibility for the application of artificial neural network. In addition, the device deposited on polyimide (PI) still exhibits the synaptic performance until the bending radii reaches 6 mm. By carefully tuning the interface engineering, this study can provide general revelation for continuous improvement of the memristive performance in neuromorphic applications.

Electrically connected spin-torque oscillators array for 2.4\u2009GHz WiFi band transmission and energy harvesting

The mutual synchronization of spin-torque oscillators (STOs) is critical for communication, energy harvesting and neuromorphic applications. Short range magnetic coupling-based synchronization has spatial restrictions (few µm), whereas the long-range electrical synchronization using vortex STOs has limited frequency responses in hundreds MHz (<500 MHz), restricting them for on-chip GHz-range applications. Here, we demonstrate electrical synchronization of four non-vortex uniformly-magnetized STOs using a single common current source in both parallel and series configurations at 2.4 GHz band, resolving the frequency-area quandary for designing STO based on-chip communication systems. Under injection locking, synchronized STOs demonstrate an excellent time-domain stability and substantially improved phase noise performance. By integrating the electrically connected eight STOs, we demonstrate the battery-free energy-harvesting system by utilizing the wireless radio-frequency energy to power electronic devices such as LEDs. Our results highlight the significance of electrical topology (series vs. parallel) while designing an on-chip STOs system.

An approach to emerging optical and optoelectronic applications based on NiO micro- and nanostructures

Abstract Nickel oxide (NiO) is one of the very few p-type semiconducting oxides, the study of which is gaining increasing attention in recent years due to its potential applicability in many emerging fields of technological research. Actually, a growing number of scientific works focus on NiO-based electrochromic devices, high-frequency spintronics, fuel cell electrodes, supercapacitors, photocatalyst, chemical/gas sensors, or magnetic devices, among others. However, less has been done so far in the development of NiO-based optical devices, a field in which this versatile transition metal oxide still lags in performance despite its potential applicability. This review could contribute with novelty and new forefront insights on NiO micro and nanostructures with promising applicability in optical and optoelectronic devices. As some examples, NiO lighting devices, optical microresonators, waveguides, optical limiters, and neuromorphic applications are reviewed and analyzed in this work. These emerging functionalities, together with some other recent developments based on NiO micro and nanostructures, can open a new field of research based on this p-type material which still remains scarcely explored from an optical perspective, and would pave the way to future research and scientific advances.

Performance Assessment of Amorphous HfO2-Based RRAM Devices for Neuromorphic Applications

The use of thin layers of amorphous hafnium oxide has been shown to be suitable for the manufacture of Resistive Random-Access memories (RRAM). These memories are of great interest because of their simple structure and non-volatile character. They are particularly appealing as they are good candidates for substituting flash memories. In this work, the performance of the MIM structure that takes part of a 4 kbit memory array based on 1-transistor-1-resistance (1T1R) cells was studied in terms of control of intermediate states and cycle durability. DC and small signal experiments were carried out in order to fully characterize the devices, which presented excellent multilevel capabilities and resistive-switching behavior.

New insight in the operation mechanism of Organic Memristive Devices: The role of PEO-based polyelectrolyte solute ions

The operation mechanism of polyaniline-based Organic Memristive Devices, synapse-like devices particularly important for neuromorphic applications, are further investigated to better clarify their principle of operation and in view of improving their performance, stability and reliability. In this framework, even though they have been studied extensively in the past, aspects of their operation mechanism require a better understanding. In earlier studies, an important role in terms of the promotion of device switching is attributed to lithium ions present in polyethylene oxide (PEO), which serves as the solid polyelectrolyte (SPE). In a recent work, the role of dopant ions as well as the material of the gate electrode have been assessed, suggesting that it is the chloride ions and silver, respectively, that are relevant for the system's memristive switching. The present paper further investigates the role of the solute ions in devices with various polyethylene oxide-based polyelectrolytes. In a series of experiments using a number of different polyelectrolyte dopant salts, we provide additional evidence that cations have no decisive effect on the memristive switching of the device, while the nature of the dopant anion, on the contrary, plays a crucial role. Moreover, giving an improved insight about the effects of the dopant in the SPE, this study shows that the polyelectrolyte is crucial for the performance of the device. To this end, it is shown that in addition to the hygroscopicity of the salt additive, it is quite important to consider also the lyotropicity of its ions to establish a favourable SPE structure. The results of our study provide a deeper understanding of the factors that cause the device's degradation and suggest novel approaches to improve stability, performance and endurance of the solid polyelectrolyte-based Organic Memristive Device. • Solid polyelectrolyte dopant cations play no significant role in OMD switching. • Lithium ions can be replaced by any other SPE dopant cation. • The nature of the dopant anion is crucial for the operation mechanism of OMD. • Hygroscopic and lyotropic salts provide favourable solid polyelectrolyte structure.

Resistive switching characteristics and theoretical simulation of a Pt/a-Ta2O5/TiN synaptic device for neuromorphic applications

Internet of things and big data demand the development of new techniques for memory devices going beyond conventional ways of memorizing and computing. In this work, we fabricated a Pt/a-Ta2O5/TiN resistive switching memory device and demonstrated its resistive and synaptic characteristics. Firstly, X-ray photoelectron spectroscopy (XPS) of a-Ta2O5/TiN analysis was conducted to determine elemental compositions of a-Ta2O5/TiN and TiON interfacial layer between a-Ta2O5 and TiN layer. Repetitive bipolar resistive switching was achieved by a set at a negative bias and a reset at a positive bias. Moreover, its biological potentiation and depression behaviors were well emulated by applying a repetitive pulse on the device. For deep understanding of this device’s properties based on materials, oxygen vacancies, and stack engineering, theoretical calculations were performed employing Vienna ab-initio simulation Package (VASP) code. All calculations were carried out using PBE and GGA+U method to obtain accurate results. Work function difference between electrodes provided a localized path for forming a Vo based conducting filament in a-Ta2O5. Iso-surface charge density plots confirmed the formation of intrinsic Vo based conducting filaments in a-Ta2O5. These conducting filaments became stronger with increasing concentration of Vos in a-Ta2O5. Integrated charge density, density of states (DOS), and potential line ups also confirmed that Vo was responsible for charge transportation in a-Ta2O5 based RRAM devices. Experimental and theoretical results confirmed the formation of TiON layer between a-Ta2O5 and active electrode (TiN), suggesting that the bipolar resistive switching phenomenon of the proposed device was based on oxygen vacancy (Vo).

Current-limiting amplifier for high speed measurement of resistive switching data.

Resistive switching devices, important for emerging memory and neuromorphic applications, face significant challenges related to the control of delicate filamentary states in the oxide material. As a device switches, its rapid conductivity change is involved in a positive feedback process that would lead to runaway destruction of the cell without current, voltage, or energy limitation. Typically, cells are directly patterned on MOS transistors to limit the current, but this approach is very restrictive as the necessary integration limits the materials available as well as the fabrication cycle time. In this article, we propose an external circuit to cycle resistive memory cells, capturing the full transfer curves while driving the cells in a way that suppresses runaway transitions. Using this circuit, we demonstrate the acquisition of 105 I, V loops per second without using on-wafer current limiting transistors. This setup brings voltage sweeping measurements to a relevant timescale for applications and enables many new experimental possibilities for device evaluation in a statistical context.

Emerging Fluorite‐ and Wurtzite‐Type Ferroelectrics: From (Hf,Zr)O2 to AlN and Related Materials

Emerging Fluorite‐ and Wurtzite‐Type Ferroelectrics: From (Hf,Zr)O₂ to AlN and Related Materials

Nanoscale wedge resistive-switching synaptic device and experimental verification of vector-matrix multiplication for hardware neuromorphic application

In this work, nanoscale wedge-structured silicon nitride (SiN x )-based resistive-switching random-access memory with data non-volatility and conductance graduality has been designed, fabricated, and characterized for its application in the hardware neuromorphic system. The process integration with full Si-processing-compatibility for constructing the unique wedge structure by which the electrostatic effects in the synaptic device operations are maximized is demonstrated. The learning behaviors of the fabricated synaptic devices are shown. In the end, vector-matrix multiplication is experimentally verified in the array level for application in more energy-efficient hardware-driven neuromorphic systems.

Ferroelectric Field Effect Transistors as a Synapse for Neuromorphic Application

In spite of the increasing use of machine learning techniques, in-memory computing and hardware have increased the interest to accelerate neural network operation. Henceforth, novel embedded nonvolatile memories (eNVMs) for highly scaled technology nodes, like ferroelectric field effect transistors (FeFETs), are heavily studied and very promising. Furthermore, inference and on-chip learning can be fostered by further eNVM technology options, such as multibit operation and linear switching. In this article, we present the advantages of hafnium oxide-based FeFETs for such purposes due to their basic three-terminal structure, which allows to selectively activate or deactivate selected devices as well as tune linearity and dynamic range for certain applications. Furthermore, we discuss the impact of the material properties of the ferroelectric layer, the interface layer thickness, and scaling on the device performance. Here, we demonstrate good device properties even for highly scaled devices ( $100\,\,nm \times 100$ nm).

Mimicking associative learning using an ion-trapping non-volatile synaptic organic electrochemical transistor

Associative learning, a critical learning principle to improve an individual’s adaptability, has been emulated by few organic electrochemical devices. However, complicated bias schemes, high write voltages, as well as process irreversibility hinder the further development of associative learning circuits. Here, by adopting a poly(3,4-ethylenedioxythiophene):tosylate/Polytetrahydrofuran composite as the active channel, we present a non-volatile organic electrochemical transistor that shows a write bias less than 0.8 V and retention time longer than 200 min without decoupling the write and read operations. By incorporating a pressure sensor and a photoresistor, a neuromorphic circuit is demonstrated with the ability to associate two physical inputs (light and pressure) instead of normally demonstrated electrical inputs in other associative learning circuits. To unravel the non-volatility of this material, ultraviolet-visible-near-infrared spectroscopy, X-ray photoelectron spectroscopy and grazing-incidence wide-angle X-ray scattering are used to characterize the oxidation level variation, compositional change, and the structural modulation of the poly(3,4-ethylenedioxythiophene):tosylate/Polytetrahydrofuran films in various conductance states. The implementation of the associative learning circuit as well as the understanding of the non-volatile material represent critical advances for organic electrochemical devices in neuromorphic applications.

Controlling Magnon Interaction by a Nanoscale Switch.

The ability to control and tune magnetic dissipation is a key concept of emergent spintronic technologies. Magnon scattering processes constitute a major dissipation channel in nanomagnets, redefine their response to spin torque, and hold the promise for manipulating magnetic states on the quantum level. Controlling these processes in nanomagnets, while being imperative for spintronic applications, has remained difficult to achieve. Here, we propose an approach for controlling magnon scattering by a switch that generates nonuniform magnetic field at nanoscale. We provide an experimental demonstration in magnetic tunnel junction nanodevices, consisting of a free layer and a synthetic antiferromagnet. By triggering the spin-flop transition in the synthetic antiferromagnet and utilizing its stray field, magnon interaction in the free layer is toggled. The results open up avenues for tuning nonlinearities in magnetic neuromorphic applications and for engineering coherent magnon coupling in hybrid quantum information technologies.