Memristive devices have proven themselves as synaptic elements with rich internal dynamics and stochasticity for bio-inspired neuromorphic computing systems. Memristors based on parylene are of special interest due to their promising memristive properties, biocompatibility and ease of production. However, their synaptic behavior has not yet been fully demonstrated. In addition, the previously proposed model of resistive switching (RS) of these memristors was of a qualitative character and therefore important switching nuances turned out to be hidden. In this paper, a phenomenological model of resistive switching in parylene memristors is developed, based on the electromigration of metal cations from the top electrode, taking into account the stochastic nature of the RS process and conductivity of the parylene gap between the filament and the electrode, which determines various resistive states (plasticity) of the structures. The model is confirmed by a good correspondence between the calculated and measured current-voltage characteristics of the memristors. We also demonstrate various forms of bio-inspired plasticity of the structures, such as paired pulse facilitation/depression, long-term potentiation/depression and spike timing/amplitude/width/rate dependent plasticity. In the case of rate-based plasticity it resembles the theoretically and experimentally thoroughly studied BCM plasticity rule. The results obtained show the possibility of using such structures in the development of next-generation neuromorphic computing systems with promising calculating and learning capabilities.

A manufacturing sequence for the formation of controlled memory in the 2T1R architecture using Si MOSFETs (metal-oxide-semiconductor field-effect transistor) and memristor crossbars based on thin layers of CoFeB-LiNbO3 nanocomposite and a-LiNbO3 is developed. It is shown that when using Cu top electrodes in memristors, relatively small resistive-switching voltages and currents (2.5 V and 0.4 mA) are achieved, which are suitable for creating a memory microchip within the MPW technology service ( https://mpw.miet.ru/ ).

MLP-Mixer neuromorphic network based on nanocomposite memristive synapses has been developed for efficient and robust classification of images.

The structural, electrical, magnetic, magneto-optical properties and magnetoresistance of {[(Co40Fe40B20)(34)(SiO2)(66)]/[ZnO]}(n) multilayer structures, where n = 50 is the number of bilayers (Co40Fe40B20)(34)(SiO2)(66) nanocomposite and ZnO have been studied. The thicknesses of (Co40Fe40B20)(34)(SiO2)(66) nanocomposite layers as well as ZnO spacers were varied in a wide range. The samples were synthesized by ion-beam sputtering onto glass ceramic substrates. The (Co40Fe40B20)(34)(SiO2)(66) composite have an amorphous structure and the semiconductor ZnO interlayers have a hexagonal crystalline structure with the p63mc symmetry group. The nanocomposite layers containing a ferromagnetic component far from the percolation threshold are in a superparamagnetic state. The presented in the paper data of magnetization, magneto-optical transverse Kerr effect and magnetoresistance indicates that long-range ferromagnetic order does not form down to 77 K both for references ZnO films and studied multilayers with thin and thick ZnO interlayers. An increase in the magneto-optical signal in multilayers compared to references (Co40Fe40B20)(34)(SiO2)(66) composite films has been detected at 1.2 eV. The magnetoresistance of {[(Co40Fe40B20)(34)(SiO2)(66)]/[ZnO]}(n) multilayers with thick (>32 nm) ZnO interlayers is lower than in reference (Co40Fe40B20)(34)(SiO2)(66) nanocomposite, while at thin ZnO interlayers magnetoresistance is significantly higher and reaches 12 % at temperatures of 77 & Kcy;. Possible mechanisms of ferromagnetic and antiferromagnetic ordering, enhancement of the magneto-optical response and magnetoresistance in {[(Co40Fe40B20)(34)(SiO2)(66)]/[ZnO]}(n) multilayer nanostructures are discussed.

The use of memristors as modulators of synaptic connections is a promising direction in the development of neuromorphic computing systems (NCS), including those that use reinforcement learning. To implement the latter, spike-timing-dependent plasticity (STDP), depending on the time of arrival of pulses with dopamine-like modulation, can be used. Using an example of a memristor array based on a nanocomposite (Co–Fe–B) x (LiNbO 3 ) 100– x the possibility of changing the conductivity of memristor devices according to the STDP rules with dopamine-like modulation is studied, and the variation in the characteristics of the array memristors from cycle to cycle (C2C) and from device to device (D2D) is assessed. It is established that the D2D variation, compared to the C2C variation, has a greater impact on the STDP window, which must be taken into account when modeling and creating neural networks capable of reinforcement learning to solve complex cognitive tasks.

А compact phenomenological model is proposed to describe the plasticity of memristive structures based on nanolayers of LiNbO3 and (Co–Fe–B)x(LiNbO3)100–x composite, taking into account the features of resistive switching of the structures and hopping electron transport in amorphous LiNbO3. The model well describes the current–voltage characteristics of memristors in a crossbar array, and the effective microscopic parameters found in the fitting of the current—voltage characteristics make it possible to predict the result of changes in the conductivity according to spike-timing-dependent plasticity (STDP), and in addition, the dependence of the STDP window on the initial conductivity of the memristor. The results obtained can be used in the development of algorithms for training spiking neuromorphic computing systems and identifying memristive STDP, which is effective for their implementation.

Films of metal-insulator nanogranular composites M_xD_100-x with different compositions and atomic percentage of metal and dielectric phases (M = Fe, Co, Ni, CoFeB; D = Al_2O_3, SiO_2, ZrO_2; x ≈ 15-60 at. are investigated by electron magnetic resonance in a wide range of frequencies (f = 7-37 GHz) and temperatures (T = 4.2-360 K). At concentrations of the metallic ferromagnetic phase below the percolation threshold, the experimental spectra, besides the conventional ferromagnetic resonance signal, demonstrate an additional absorption peak characterized by a double effective g-factor g ≈ 4. The appearance of such a peak in the resonance spectra and its unusual properties are explained in the framework of the quantum mechanical "giant spin" model by the excitation of "forbidden" ("double quantum") transitions in magnetic nanogranules with a change of the spin projection Δ m = ±2.

Many applications, including learning neuromorphic computing systems, require the analogue or, at least, multilevel resistive switching (RS) of memristor devices. In metal–insulator–metal (MIM) structures it is often associated with the creation of a large number of conductive channels, or filaments, but in the majority of cases, especially for the vacancy mechanism of RS, the direct observation of many filaments is difficult or impossible. In this work, a model of the explicit filament conductivity distribution for a nanocomposite (NC) LiNbO 3 -based memristor, that nonetheless may be applicable to other types of MIM memristors with hopping conductivity over dangling bonds in an amorphous insulator layer, is derived from physical principles. This model, in addition to indirect evidence from magnetic and capacitance measurements, reaffirms the multifilamentary RS mechanism by the approximation of experimental data on synaptic plasticity, i.e., a change in the conductivity of a memristor by a pair of pulses applied to it. The results of this research offer opportunities to study RS in MIM structures at the macroscopic level using model filament distributions and substantiates the possibility of successfully applying the considered NC memristors as synaptic elements of neuromorphic networks.

Films of metal-insulator nanogranular composites M x D 100 – x with different composition and percentage of metal and dielectric phases (M = Fe, Co, CoFeB; D = Al 2 O 3 , SiO 2 , LiNbO 3 ; x ≈ 15–70 at %) are investigated by magnetic resonance in a wide range of frequencies ( f = 7–37 GHz) and temperatures ( T = 4.2–360 K). In addition to the usual ferromagnetic resonance signal from an array of nanogranules, the experimental spectra contain an additional absorption peak, which we associate with the electron paramagnetic resonance (EPR) of Fe and Co ions dispersed in the insulating space between the granules. In contrast to the traditional EPR of Fe and Co ions in weakly doped non-magnetic matrices, the observed peak demonstrates a number of unusual properties, which we explain by the presence of magnetic interactions between ions and granules.

Изучен скейлинг в поведении сопротивления аномального эффекта Холла ρAHE от продольногоρ в нанокомпозитах (CoFeB)x (LiNbO3)100-x с низкой концентрацией диспергированных атомов Co и Fe (Nd ~ 4 · 1020 см-3) в аморфной матрице LiNbO3. Исследования выполнены ниже порога перколяции (xp ≈ 49 ат. %) в диапазоне x ≈ 40-48 ат. %, в котором наблюдается логарифмический закон в температурной зависимости проводимости σ ∝ ln T (x ≈ 44-48 ат. %), переходящий в закон "1/2" ln σ ∝ -(T0/T )1/2 при x ≈ 40-42 ат. %, характерный для со-туннельных процессов переноса в нанокомпозитах. Обнаружено, что в скейлинговой зависимости ρAHE/x ∝ [ρ(x)]n степень n ≈ 0.24 с точностью 5 % совпадает с n в аналогичной зависимости для НК на базе иной матрицы (CoFeB)x(Al2O3)100-x c высоким содержанием Nd ~ 1021-1022 см-3, а также с n в параметрической зависимости ρAHE ∝ [ρ(T )]n для образцов с наименьшими x ≈ 40 ат. %. Обнаруженные особенности связываются с коррелированным изменением вероятности со-туннельных переходов в совокупности из более 3-х центров под действием спин-орбитального взаимодействия. Не исключается также возможность проявления туннельного аномального эффекта Холла барьерного типа на интерфейсах гранул.

Films of metal-insulator nanogranular composites MxD100 – x with different composition and percentage of metal and dielectric phases (M = Fe, Co, CoFeB; D = Al2O3, SiO2, LiNbO3; x ≈ 15–70 at %) are investigated by magnetic resonance in a wide range of frequencies (f = 7–37 GHz) and temperatures (T = 4.2–360 K). In addition to the usual ferromagnetic resonance signal from an array of nanogranules, the experimental spectra contain an additional absorption peak, which we associate with the electron paramagnetic resonance (EPR) of Fe and Co ions dispersed in the insulating space between the granules. In contrast to the traditional EPR of Fe and Co ions in weakly doped non-magnetic matrices, the observed peak demonstrates a number of unusual properties, which we explain by the presence of magnetic interactions between ions and granules.

A scaling behavior of the anomalous Hall effect resistivity ρAHE versus the longitudinal resistivity ρ in (CoFeB)x(LiNbO3)100 − x nanocomposites with a low content of dispersed Co and Fe atoms (Nd 4 × 1020 cm−3) in an amorphous LiNbO3 matrix is studied in the range x ≈ 40–48 at

Magnetooptical spectroscopy is an effective method for studying the magnetic microstructure of homogeneous and heterogeneous magnets. This review is devoted to analysis of numerous factors affecting the intensity and spectral dependence of a magnetooptical signal of the equatorial Kerr effect in nanocomposites “ferromagnetic metal–dielectric” in the visible and near infrared spectral regions. Examples of the influence of the metal concentration, nanoparticle size and shape, the substrate, the material of the dielectric, the amorphization of grains, the deposition method, and other factors on the magnetooptical spectrum are considered. The differences in the magnetooptical spectra for the superparamagnetic, superferromagnetic, and ferromagnetic states are demonstrated. It is noted that in the presence of fractions with different field dependences of the magnetization in a nanocomposite, the magnetooptical signal is not proportional to the total magnetization. Examples of enhancement and sign inversion of the magnetooptical signal in nanocomposites are considered. The possibility of the description of magnetooptical spectra using the methods of the effective medium (the Bruggeman method and the Maxwell–Garnett symmetrized approximation) is discussed.

Memristor-based neuromorphic computing systems (NСSs) provide a fast, high computational and energy efficient approach to neural network (NN) training and solving cognitive problems (pattern recognition, big data processing, prediction, etc.) [1]. Memristors could be organized in large crossbar arrays to perform vector-matrix multiplication (VMM) in a natural one-step way by the weighted electrical current summation (according to the Ohm’s and Kirchhoff’s laws) [1]. In contrast, being the most massively parallel operation in NN learning and inference, VMM is extremely time- and energy-expensive in traditional von Neumann architectures. Owing to this difference, memristor-based NCSs are of high interest. Memristors have already been successfully implemented for diverse NCS realizations, and such schemes as multi-layer perceptron (MLP) [2], long short-term memory and others have been demonstrated. Most of these NCSs are usually trained by various types of gradient descent learning algorithm, the hardware realization of which is challenging due to unreliable cycle-to-cycle (c2c) and device-to-device (d2d) variations of memristive devices. Several approaches have been proposed to partially mitigate these problems, including reservoir computing [3] and fine feature engineering [4]. The general idea of such approaches is to reduce the number of required weights (i.e. memristors) compared with fully connected NNs. In this respect, such novel architectures as convolutional NN (CNN) and MLP-mixer are of high interest as they provide significant weight reduction without classification efficiency drop. Although CNN based on memristors was already demonstrated, different aspects of its realization (such as hybrid hardware-software co-design) have yet to be studied. MLP-mixer was realized only in software. Therefore, in this work we have studied the possibility of hardware realization of CNN and MLP-mixer networks based on crossbar arrays of memristors. For this purpose, we studied (Co-Fe-B)x(LiNbO3)100−x nanocomposite (CFB-LNO NC) memristors, which operate through a multifilamentary resistive switching (RS) mechanism, demonstrate high endurance, long retention and possess multilevel RS [5]. Crossbar array of memristors was fabricated using laser photolithography for patterning electrode buses and ion-beam sputtering on the original facility for active layer deposition (~10 nm thick LiNbO3 and ~290 nm thick CFB-LNO NC with x ≈10–25 at.%). Details of the fabrication process could be found elsewhere [5]. I-V curves of the fabricated memristors showed small c2c and d2d variations, plasticity with 16 different resistive states and endurance of more than 105 cycles. Using the nanocomposite based crossbar arrays, we implemented a hybrid CNN, consisting of a hardware feature extractor with one/two kernels and a software classifier. Additionally, we have demonstrated in simulation that the usage of the memristors under study in the accurately adapted MLP-Mixer architecture results in high classification accuracy that is resilient to memristive variations and stuck devices.

A strongly nonmonotonic temperature dependence of the magnetoresistance in (CoFeB)x(LiNbOy)100 – x film nanocomposites (x ≈ 40–48 at

The field dependences of the electrical conductivity of Pt/diamond-like carbon (DLC)/Pt structures based on thin layers of high-resistivity DLC are studied. It is shown that the nonohmic behavior of the conductance of structures is described by the Frenkel–Poole formula and is related to correlated distribution of charges under conditions of their percolation hopping transport between low-resistance DLC regions.

Nanocomposites (CoFeB)x(LiNbO3)100 – x with x = 17–48 at % have been synthesized by ion beam sputtering of a composite target comprised of Co40Fe40B20 and LiNbO3 onto silicon substrates, and the tran-sitions from the superparamagnetic state to the superferromagnetic and ferromagnetic states with an increase in the concentration of the magnetic component are studied by magneto-optical methods. The magneto-optical properties have been investigated in the geometry of the equatorial (transverse) Kerr effect (TKE). Magneto-optical spectra are recorded in the range of 0.5–4.0 eV in fields up to 2.5 kOe at 20–300 K, field and temperature dependences of the TKE at certain wavelengths are obtained, and the domain structure during magnetization reversal is visualized using a magneto-optical Kerr microscope. It is shown that the sample with x = 17 at % is superparamagnetic at temperatures above the blocking temperature (about 30 K). The interaction between the granules is considerable already at x = 20 at %, the transition to the superferro-magnetic state occurs at x ≈ 32–36 at %, and the transition to the ferromagnetic state occurs at x ≈ 44 at %near the metal–dielectric transition, i.e., at a concentration below the percolation transport threshold.

We suggested a compact behavioral model of a nanocomposite memristor (Co 40 Fe 40 B 20 ) x (LiNbO 3 ) 100 – x , which quantitatively describes the dynamics of changes in the conductivity of laboratory samples, and also implements the mechanisms of the finite storage time of resistive states and the spread in switching voltages from cycle to cycle and from device to device. The possibility of implementing a pulsed neural network with synaptic memristor connections based on this model is shown.

This report presents the current state of affairs in the implementation of artificial intelligence hardware accelerators based on practically successful neural network algorithms of the first and second generations based on formal artificial neural networks (ANNs). The shortcomings of existing solutions are noted and ways to overcome them using analog neuromorphic architectures are outlined. The latter are created on the principles of the structuring and functioning of a living nervous system, using artificial neurons and models of synaptic contacts - the so–called memristors, electrically rewritable nanoscale elements of non-volatile memory [1-3]. With the use of these elements, it is possible to significantly increase the performance and energy efficiency of algorithm accelerators based on the ANNs [4-6], as well as the formation of promising computing systems based on bioplausible 3rd generation neural network algorithms - Spiking Neural Networks (SNNs) [7-9]. The original method of substantiating the optimal rules for local tuning SNNs with frequency encoding and the possibility of their implementation in the form of the Spike-Timing-Dependent Plasicity (STDP) are discussed [10]. The results of SNN learning stability to a variability of analog memristors, as well as the use of noise as a constructive factor in the fine-tuning and maintenance of SNN memristive weights are demonstrated [7, 11]. Also, approaches to the implementation of local plasticity rules with dopamine-like modulation as a type of SNN reinforcement learning are discussed. The latter is necessary for the formation of imitative "needs" of an agent in the process of its autonomous functioning [12, 13, 14]. The first results on the creation of a prototype of a memristive implantable neuroprosthesis of the motor activity are considered [15, 16]. Finally, possible hardware solutions for both neuronal elements and synaptic connections based on suitable memristive devices are demonstrated. The concept and first results on the creation of an analog neuromorphic computing system based on the above components are presented. Thus, an attempt is made to systematize the existing and original methods of implementing energy-efficient and compact analog neuromorphic computing systems for real-time and life-learning artificial intelligence.

Convolutional neural networks (CNNs) have been widely used in image recognition and processing tasks. Memristor-based CNNs accumulate the advantages of emerging memristive devices, such as nanometer critical dimensions, low power consumption, and functional similarity to biological synapses. Most studies on memristor-based CNNs use either software models of memristors for simulation analysis or full hardware CNN realization. Here, we propose a hybrid CNN, consisting of a hardware fixed pre-trained and explainable feature extractor and a trainable software classifier. The hardware part was realized on passive crossbar arrays of memristors based on nanocomposite (Co-Fe-B)x(LiNbO3)100−x structures. The constructed 2-kernel CNN was able to classify the binarized Fashion-MNIST dataset with ~ 84% accuracy. The performance of the hybrid CNN is comparable to the other reported memristor-based systems, while the number of trainable parameters for the hybrid CNN is substantially lower. Moreover, the hybrid CNN is robust to the variations in the memristive characteristics: dispersion of 20% leads to only a 3% accuracy decrease. The obtained results pave the way for the efficient and reliable realization of neural networks based on partially unreliable analog elements.