The inherent adversarial robustness of analog in-memory computing

A key challenge for deep neural network algorithms is their vulnerability to adversarial attacks. Inherently non-deterministic compute substrates, such as those based on analog in-memory computing, have been speculated to provide significant adversarial robustness when performing deep neural network inference. In this paper, we experimentally validate this conjecture for the first time on an analog in-memory computing chip based on phase change memory devices. We demonstrate higher adversarial robustness against different types of adversarial attacks when implementing an image classification network. Additional robustness is also observed when performing hardware-in-the-loop attacks, for which the attacker is assumed to have full access to the hardware. A careful study of the various noise sources indicate that a combination of stochastic noise sources (both recurrent and non-recurrent) are responsible for the adversarial robustness and that their type and magnitude disproportionately effects this property. Finally, it is demonstrated, via simulations, that when a much larger transformer network is used to implement a natural language processing task, additional robustness is still observed.

Nonlinear memristor model with exact solution allows for ex situ reservoir computing training and in situ inference

Memristive physical reservoir computing is a promising approach for solving data classification and temporal processing tasks. This method exploits the nonlinear dynamics of physical, low-power devices to achieve high-dimensional mapping of input signals. Ion-channel-based memristors, which operate with similar voltages, currents, and timescales as biological synapses, are promising due to their rich dynamics, especially for use in biological edge settings. Accurate modeling of their dynamics is essential for optimizing network hyperparameters ex situ to save time and energy. Here, a generalized sigmoidal growth model of ion-channel memristor conductance is presented and shown to be more accurate in predicting dynamics than linear or logistic models. Using the exact solution of the proposed sigmoidal model, the MNIST handwritten digit classification task is optimized and trained ex situ, then tested in situ with the same trained weights. This approach achieved an experimental testing accuracy of 90.6%.

Memristive Ion Dynamics to Enable Biorealistic Computing

Conventional artificial intelligence (AI) systems are facing bottlenecks due to the fundamental mismatches between AI models, which rely on parallel, in-memory, and dynamic computation, and traditional transistors, which have been designed and optimized for sequential logic operations. This calls for the development of novel computing units beyond transistors. Inspired by the high efficiency and adaptability of biological neural networks, computing systems mimicking the capabilities of biological structures are gaining more attention. Ion-based memristive devices (IMDs), owing to the intrinsic functional similarities to their biological counterparts, hold significant promise for implementing emerging neuromorphic learning and computing algorithms. In this article, we review the fundamental mechanisms of IMDs based on ion drift and diffusion to elucidate the origins of their diverse dynamics. We then examine how these mechanisms operate within different materials to enable IMDs with various types of switching behaviors, leading to a wide range of applications, from emulating biological components to realizing specialized computing requirements. Furthermore, we explore the potential for IMDs to be modified and tuned to achieve customized dynamics, which positions them as one of the most promising hardware candidates for executing bioinspired algorithms with unique specifications. Finally, we identify the challenges currently facing IMDs that hinder their widespread usage and highlight emerging research directions that could significantly benefit from incorporating IMDs.

Protected memristive implementations of cryptographic functions

Memristive technology mitigates the memory wall issue in von Neumann architectures by enabling in-memory data processing. Unlike traditional complementary metal-oxide semiconductor (CMOS) technology, memristors provide a new paradigm for implementing cryptographic functions and security considerations. While prior research explores memristors for cryptographic functions and side-channel attack vulnerabilities, our study uniquely addresses memristor-oriented countermeasures. We review different memristive crossbar configurations, implement a four-bit S-box cryptographic function, and analyse memristor-oriented hiding and masking techniques using a self-rectifying passive crossbar. Our findings confirm the efficacy of memristor-oriented hiding techniques but highlight limitations in memristor-oriented masked dual-rail pre-charge logic (MDPL) masking methods. Effective MDPL masking depends on specific power consumption conditions, i.e. the power profile of input data '01' and '10' are not clearly distinguishable from '00' and '11', which, however, are not satisfied across various memristive logic families. Despite passing t-tests, xor4Sbox with CRS-based MDPL masking failed stochastic approaches owing to power consumption differences. Our study prioritizes memristor-oriented countermeasures, advancing the understanding of challenges and opportunities in memristive technology for cryptographic functions.This article is part of the theme issue 'Emerging technologies for future secure computing platforms'.

Identifying and understanding the nonlinear behavior of memristive devices

Nonlinearity is a crucial characteristic for implementing hardware security primitives or neuromorphic computing systems. The main feature of all memristive devices is this nonlinear behavior observed in their current-voltage characteristics. To comprehend the nonlinear behavior, we have to understand the coexistence of resistive, capacitive, and inertia (virtual inductive) effects in these devices. These effects originate from corresponding physical and chemical processes in memristive devices. A physics-inspired compact model is employed to model and simulate interface-type RRAMs such as Au/BiFeO[Formula: see text]/Pt/Ti, Au/Nb[Formula: see text]O[Formula: see text]/Al[Formula: see text]O[Formula: see text]/Nb, while accounting for the modeling of capacitive and inertia effects. The simulated current-voltage characteristics align well with experimental data and accurately capture the non-zero crossing hysteresis generated by capacitive and inductive effects. This study examines the response of two devices to increasing frequencies, revealing a shift in their nonlinear behavior characterized by a reduced hysteresis range Fourier series analysis utilizing a sinusoidal input voltage of varying amplitudes and frequencies indicates harmonics or frequency components that considerably influence the functioning of RRAMs. Moreover, we propose and demonstrate the use of the frequency spectra as one of the fingerprints for memristive devices.

Two-Terminal Neuromorphic Devices for Spiking Neural Networks: Neurons, Synapses, and Array Integration

The ever-increasing volume of complex data poses significant challenges to conventional sequential global processing methods, highlighting their inherent limitations. This computational burden has catalyzed interest in neuromorphic computing, particularly within artificial neural networks (ANNs). In pursuit of advancing neuromorphic hardware, researchers are focusing on developing computation strategies and constructing high-density crossbar arrays utilizing history-dependent, multistate nonvolatile memories tailored for multiply-accumulate (MAC) operations. However, the real-time collection and processing of massive, dynamic data sets require an innovative computational paradigm akin to that of the human brain. Spiking neural networks (SNNs), representing the third generation of ANNs, are emerging as a promising solution for real-time spatiotemporal information processing due to their event-based spatiotemporal capabilities. The ideal hardware supporting SNN operations comprises artificial neurons, artificial synapses, and their integrated arrays. Currently, the structural complexity of SNNs and spike-based methodologies requires hardware components with biomimetic behaviors that are distinct from those of conventional memristors used in deep neural networks. These distinctive characteristics required for neuron and synapses devices pose significant challenges. Developing effective building blocks for SNNs, therefore, necessitates leveraging the intrinsic properties of the materials constituting each unit and overcoming the integration barriers. This review focuses on the progress toward memristor-based spiking neural network neuromorphic hardware, emphasizing the role of individual components such as memristor-based neurons, synapses, and array integration along with relevant biological insights. We aim to provide valuable perspectives to researchers working on the next generation of brain-like computing systems based on these foundational elements.

A Near-Infrared Retinomorphic Device with High Dimensionality Reservoir Expression

Physical reservoir-based reservoir computing (RC) systems for intelligent perception have recently gained attention because they require fewer computing resources. However, the system remains limited in infrared (IR) machine vision, including materials and physical reservoir expression power. Inspired by biological visual perception systems, the study proposes a near-infrared (NIR) retinomorphic device that simultaneously perceives and encodes narrow IR spectral information (at ≈980 nm). The proposed device, featuring core-shell upconversion nanoparticle/poly (3-hexylthiophene) (P3HT) nanocomposite channels, enables the absorption and conversion of NIR into high-energy photons to excite more photo carriers in P3HT. The photon-electron-coupled dynamics under the synergy of photovoltaic and photogating effects influence the nonlinearity and high dimensionality of the RC system under narrow-band NIR irradiation. The device also exhibits multilevel data storage capability (≥8 levels), excellent stability (≥2000 s), and durability (≥100 cycles). The system accurately identifies NIR static and dynamic handwritten digit images, achieving recognition accuracies of 91.13% and 90.07%, respectively. Thus, the device tackles intricate computations like solving second-order nonlinear dynamic equations with minimal errors (normalized mean squared error of 1.06 × 10⁻3 during prediction).

Dynamic memristor for physical reservoir computing

Reservoir computing (RC) has attracted considerable attention for its efficient handling of temporal signals and lower training costs. As a nonlinear dynamic system, RC can map low-dimensional inputs into high-dimensional spaces and implement classification using a simple linear readout layer. The memristor exhibits complex dynamic characteristics due to its internal physical processes, which renders them an ideal choice for the implementation of physical reservoir computing (PRC) systems. This review focuses on PRC systems based on memristors, explaining the resistive switching mechanism at the device level and emphasizing the tunability of their dynamic behavior. The development of memristor-based reservoir computing systems is highlighted, along with discussions on the challenges faced by this field and potential future research directions.

Neuromorphic Computing with Emerging Antiferromagnetic Ordering in Spin-Orbit Torque Devices

Field-free switching (FFS) and spin-orbit torque (SOT)-based neuromorphic characteristics were realized in a W/Pt/Co/NiO/Pt heterostructure with a perpendicular exchange bias (HEB) for brain-inspired neuromorphic computing (NC). Experimental results using NiO-based SOT devices guided the development of fully spin-based artificial synapses and sigmoidal neurons for implementation in a three-layer artificial neural network. This system achieved impressive accuracies of 91-96% when applied to the Modified National Institute of Standards and Technology (MNIST) image data set and 78.85-81.25% when applied to Fashion MNIST images, due presumably to the emergence of robust NiO antiferromagnetic (AFM) ordering. The emergence of AFM ordering favored the FFS with an enhanced HEB, which suppressed the memristivity and reduced the recognition accuracy. This indicates a trade-off between the requirements for solid-state memory and those required for brain-inspired NC devices. Nonetheless, our findings revealed opportunities by which the two technologies could be aligned via controllable exchange coupling.

Reconfigurable Ag/HfO2/NiO/Pt Memristors with Stable Synchronous Synaptic and Neuronal Functions for Renewable Homogeneous Neuromorphic Computing System

Artificial synapses and bionic neurons offer great potential in highly efficient computing paradigms. However, complex requirements for specific electronic devices in neuromorphic computing have made memristors face the challenge of process simplification and universality. Herein, reconfigurable Ag/HfO2/NiO/Pt memristors are designed for feasible switching between volatile and nonvolatile modes by compliance current controlled Ag filaments, which enables stable and reconfigurable synaptic and neuronal functions. A neuromorphic computing system effectively replicates the biological synaptic weight alteration and continuously accomplishes excitation and reset of artificial neurons, which consist of bionic synapses and artificial neurons based on isotype Ag/HfO2/NiO/Pt memristors. This reconfigurable electrical performance of the Ag/HfO2/NiO/Pt memristors takes advantage of simplified hardware design and delivers integrated circuits with high density, which exhibits great potency for future neural networks.

Unveiling transient current response in bilayer oxide-based physical reservoirs for time-series data analysis

Physical reservoirs employed to map time-series data and analyze extracted features have attracted interest owing to their low training cost and mitigated interconnection complexity. This study reports a physical reservoir based on a bilayer oxide-based dynamic memristor. The proposed device exhibits a nonlinear current response and short-term memory (STM), satisfying the requirements of reservoir computing (RC). These characteristics are validated using a compact model to account for resistive switching (RS) via the dynamic evolution of the internal state variable and the relocation of oxygen vacancies. Mathematically, the transient current response can be quantitatively described according to a simple set of equations to correlate the theoretical framework with experimental results. Furthermore, the device shows significant reliability and ability to distinguish 4-bit inputs and four diverse neural firing patterns. Therefore, this work shows the feasibility of implementing physical reservoirs in hardware and advances the understanding of the dynamic response.

Nanoelectronics Using Metal-Insulator Transition

Metal-insulator transition (MIT) coupled with an ultrafast, significant, and reversible resistive change in Mott insulators has attracted tremendous interest for investigation into next-generation electronic and optoelectronic devices, as well as a fundamental understanding of condensed matter systems. Although the mechanism of MIT in Mott insulators is still controversial, great efforts have been made to understand and modulate MIT behavior for various electronic and optoelectronic applications. In this review, recent progress in the field of nanoelectronics utilizing MIT is highlighted. A brief introduction to the physics of MIT and its underlying mechanisms is begun. After discussing the MIT behaviors of various Mott insulators, recent advances in the design and fabrication of nanoelectronics devices based on MIT, including memories, gas sensors, photodetectors, logic circuits, and artificial neural networks are described. Finally, an outlook on the development and future applications of nanoelectronics utilizing MIT is provided. This review can serve as an overview and a comprehensive understanding of the design of MIT-based nanoelectronics for future electronic and optoelectronic devices.

Artificial sensory system based on memristive devices

In the biological nervous system, the integration and cooperation of parallel system of receptors, neurons, and synapses allow efficient detection and processing of intricate and disordered external information. Such systems acquire and process environmental data in real-time, efficiently handling complex tasks with minimal energy consumption. Memristors can mimic typical biological receptors, neurons, and synapses by implementing key features of neuronal signal-processing functions such as selective adaption in receptors, leaky integrate-and-fire in neurons, and synaptic plasticity in synapses. External stimuli are sensitively detected and filtered by "artificial receptors," encoded into spike signals via "artificial neurons," and integrated and stored through "artificial synapses." The high operational speed, low power consumption, and superior scalability of memristive devices make their integration with high-performance sensors a promising approach for creating integrated artificial sensory systems. These integrated systems can extract useful data from a large volume of raw data, facilitating real-time detection and processing of environmental information. This review explores the recent advances in memristor-based artificial sensory systems. The authors begin with the requirements of artificial sensory elements and then present an in-depth review of such elements demonstrated by memristive devices. Finally, the major challenges and opportunities in the development of memristor-based artificial sensory systems are discussed.

Critical analysis for nonlinear oscillations by least square HPM

In this study, a novel adapted homotopy perturbation method (HPM) is used to treat the nonlinear phenomena of free vibration in a system with one degree of freedom. This adaptation involves the integration of HPM with a least-squares optimizer, resulting in a hybrid method called the least square homotopy perturbation method (LSHPM). The LSHPM is tested on various nonlinear problems documented in the existing literature. To evaluate the effectiveness of the proposed approach, the identified problems are also tackled using HPM and the MATLAB built-in function bvp5c, and then the results are compared with those obtained using LSHPM. In addition, a comparative analysis is carried out with the results of the AG method as found in the literature. The results show that LSHPM is a reliable and efficient method suitable for solving more complicated initial value problems in the fields of science and engineering.

Behavioral Modeling of Memristors under Harmonic Excitation

Memristors are devices built on the basis of fourth passive electrical elements in nanosystems. Because of the multitude of technologies used for memristor implementation, it is not always possible to obtain analytical models of memristors. This difficulty can be overcome using behavioral modeling, which is when mathematical models are constructed according to the input-output relationships on the input and output signals. For memristor modeling, piecewise neural and polynomial models with split signals are proposed. At harmonic input signals of memristors, this study suggests that split signals should be formed using a delay line. This method produces the minimum number of split signals and, as a result, simplifies behavioral models. Simplicity helps reduce the dimension of the nonlinear approximation problem solved in behavioral modeling. Based on the proposed method, the piecewise neural and polynomial models with harmonic input signals were constructed to approximate the transfer characteristic of the memristor, in which the current dynamics are described using the Bernoulli differential equation. It is shown that the piecewise neural model based on the feedforward network ensures higher modeling accuracy at almost the same complexity as the piecewise polynomial model.

Physical approach of a neuron model with memristive membranes

The membrane potential of a neuron is mainly controlled by the gradient distribution of electromagnetic field and concentration diversity between intracellular and extracellular ions. Without considering the thickness and material property, the electric characteristic of cell membrane is described by a capacitive variable and output voltage in an equivalent neural circuit. The flexible property of cell membrane enables controllability of endomembrane and outer membrane, and the capacitive properties and gradient field can be approached by double membranes connected by a memristor in an equivalent neural circuit. In this work, two capacitors connected by a memristor are used to mimic the physical property of two-layer membranes, and an inductive channel is added to the neural circuit. A biophysical neuron is obtained and the energy characteristic, dynamics, self-adaption is discussed, respectively. Coherence resonance and mode selection in adaptive way are detected under noisy excitation. The distribution of average energy function is effective to predict the appearance of coherence resonance. An adaptive law is proposed to control the capacitive parameters, and the controllability of cell membrane under external stimulus can be explained in theoretical way. The neuron with memristive membranes explains the self-adaptive mechanism of parameter changes and mode transition from energy viewpoint.

All-Solid-State Synaptic Transistors with Lithium-Ion-Based Electrolytes for Linear Weight Mapping and Update in Neuromorphic Computing Systems

Neuromorphic computing, an innovative technology inspired by the human brain, has attracted increasing attention as a promising technology for the development of artificial intelligence systems. This study proposes synaptic transistors with a Li1-xAlxTi2-x(PO4)3 (LATP) layer to analyze the conductance modulation linearity, which is essential for weight mapping and updating during on-chip learning processes. The high ionic conductivity of the LATP electrolyte provides a large hysteresis window and enables linear weight update in synaptic devices. The results demonstrate that optimizing the LATP layer thickness improves the conductance modulation and linearity of synaptic transistors during potentiation and degradation. A 20 nm-thick LATP layer results in the most nonlinear depression (αd = -6.59), whereas a 100 nm-thick LATP layer results in the smallest nonlinearity (αd = -2.22). Additionally, a device with the optimal 100 nm-thick LATP layer exhibits the highest average recognition accuracy of 94.8% and the smallest fluctuation, indicating that the linearity characteristics of a device play a crucial role in weight update during learning and can significantly affect the recognition accuracy.

Linear conductance update improvement of CMOS-compatible second-order memristors for fast and energy-efficient training of a neural network using a memristor crossbar array

Memristors are two-terminal memory devices that can change the conductance state and store analog values. Thanks to their simple structure, suitability for high-density integration, and non-volatile characteristics, memristors have been intensively studied as synapses in artificial neural network systems. Memristive synapses in neural networks have theoretically better energy efficiency compared with conventional von Neumann computing processors. However, memristor crossbar array-based neural networks usually suffer from low accuracy because of the non-ideal factors of memristors such as non-linearity and asymmetry, which prevent weights from being programmed to their targeted values. In this article, the improvement in linearity and symmetry of pulse update of a fully CMOS-compatible HfO2-based memristor is discussed, by using a second-order memristor effect with a heating pulse and a voltage divider composed of a series resistor and two diodes. We also demonstrate that the improved device characteristics enable energy-efficient and fast training of a memristor crossbar array-based neural network with high accuracy through a realistic model-based simulation. By improving the memristor device's linearity and symmetry, our results open up the possibility of a trainable memristor crossbar array-based neural network system that possesses great energy efficiency, high area efficiency, and high accuracy at the same time.

High Amplitude Spike Generator in Au Nanodot-Incorporated NbOx Mott Memristor

NbO_x-based Mott memristors exhibit fast threshold switching behaviors, making them suitable for spike generators in neuromorphic computing and stochastic clock generators in security devices. In these applications, a high output spike amplitude is necessary for threshold level control and accurate signal detection. Here, we propose a materialwise solution to obtain the high amplitude spikes by inserting Au nanodots into the NbO_x device. The Au nanodots enable increasing the threshold voltage by modulating the oxygen contents at the electrode-oxide interface, providing a higher ON current compared to nanodot-free NbO_x devices. Also, the reduction of the local switching region volume decreases the thermal capacitance of the system, allowing the maximum spike amplitude generation. Consequently, the Au nanodot incorporation increases the spike amplitude of the NbO_x device by 6 times, without any additional external circuit elements. The results are systematically supported by both a numerical model and a finite-element-method-based multiphysics model.

Highly Linear and Symmetric Synaptic Memtransistors Based on Polarization Switching in Two-Dimensional Ferroelectric Semiconductors

Brain-inspired neuromorphic computing hardware based on artificial synapses offers efficient solutions to perform computational tasks. However, the nonlinearity and asymmetry of synaptic weight updates in reported artificial synapses have impeded achieving high accuracy in neural networks. Here, this work develops a synaptic memtransistor based on polarization switching in a two-dimensional (2D) ferroelectric semiconductor (FES) of α-In₂ Se₃ for neuromorphic computing. The α-In₂ Se₃ memtransistor exhibits outstanding synaptic characteristics, including near-ideal linearity and symmetry and a large number of programmable conductance states, by taking the advantages of both memtransistor configuration and electrically configurable polarization states in the FES channel. As a result, the α-In₂ Se₃ memtransistor-type synapse reaches high accuracy of 97.76% for digit patterns recognition task in simulated artificial neural networks. This work opens new opportunities for using multiterminal FES memtransistors in advanced neuromorphic electronics.

Responsive Hydrogels Based on Triggered Click Reactions for Liver Cancer

Globally, liver cancer, which is one of the major cancers worldwide, has attracted the growing attention of technological researchers for its high mortality and limited treatment options. Hydrogels are soft 3D network materials containing a large number of hydrophilic monomers. By adding moieties such as nitrobenzyl groups to the network structure of a cross-linked nanocomposite hydrogel, the click reaction improves drug-release efficiency in vivo, which improves the survival rate and prolongs the survival time of liver cancer patients. The application of a nanocomposite hydrogel drug delivery system can not only enrich the drug concentration at the tumor site for a long time but also effectively prevents the distant metastasis of residual tumor cells. At present, a large number of researches have been working toward the construction of responsive nanocomposite hydrogel drug delivery systems, but there are few comprehensive articles to systematically summarize these discoveries. Here, this systematic review summarizes the synthesis methods and related applications of nanocomposite responsive hydrogels with actions to external or internal physiological stimuli. With different physical or chemical stimuli, the structural unit rearrangement and the controlled release of drugs can be used for responsive drug delivery in different states.

A non-linear two-dimensional float gate transistor as a lateral inhibitory synapse for retinal early visual processing

Synaptic transistors that accommodate concurrent signal transmission and learning in a neural network are attracting enormous interest for neuromorphic sensory processing. To remove redundant sensory information while keeping important features, artificial synaptic transistors with non-linear conductance are desired to apply filter processing to sensory inputs. Here, we report the realization of non-linear synapses using a two-dimensional van der Waals (vdW) heterostructure (MoS₂/h-BN/graphene) based float gate memory device, in which the semiconductor channel is tailored via a surface acceptor (ZnPc) for subthreshold operation. In addition to usual synaptic plasticity, the memory device exhibits highly non-linear conductance (rectification ratio >10⁶), allowing bidirectional yet only negative/inhibitory current to pass through. We demonstrate that in a lateral coupling network, such a float gate memory device resembles the key lateral inhibition function of horizontal cells for the formation of an ON-center/OFF-surround receptive field. When combined with synaptic plasticity, the lateral inhibition weights are further tunable to enable adjustable edge enhancement for early visual processing. Our results here hopefully open a new scheme toward early sensory perception via lateral inhibitory synaptic transistors.

Scalable Van der Waals Encapsulation by Inorganic Molecular Crystals

Encapsulation is critical for devices to guarantee their stability and reliability. It becomes an even more essential requirement for devices based on 2D materials with atomic thinness and far inferior stability compared to their bulk counterparts. Here a general van der Waals (vdW) encapsulation method for 2D materials using Sb₂ O₃ layer of inorganic molecular crystal fabricated via thermal evaporation deposition is reported. It is demonstrated that such a scalable encapsulation method not only maintains the intrinsic properties of typical air-susceptible 2D materials due to their vdW interactions but also remarkably improves their environmental stability. Specifically, the encapsulated black phosphorus (BP) exhibits greatly enhanced structural stability of over 80 days and more sustaining-electrical properties of 19 days, while the bare BP undergoes degradation within hours. Moreover, the encapsulation layer can be facilely removed by sublimation in vacuum without damaging the underlying materials. This scalable encapsulation method shows a promising pathway to effectively enhance the environmental stability of 2D materials, which may further boost their practical application in novel (opto)electronic devices.