Search

A neuromorphic sensing system combining a flexible piezoelectric sensor array, event-triggered circuitry and a memristive chip enables energy-efficient, low-latency edge processing of event-based data through analogue in-memory computing.

A cellular neural network that incorporates transistors and multilevel non-volatile memristor synapses can be used for efficient image and video processing.

Layered black phosphorous has gained significant attention in the 2D materials community, and dynamical control of its bandgap is key to enable novel applications. Here, the authors demonstrate continuous electrical bandgap tuning using moderate displacement fields.

Radiofrequency switches based on micro-electromechanical systems or phase-change materials are large and require high voltages. Here, the authors demonstrate a nanoscale radiofrequency switch based on a memristive device with actuation voltages as low as 0.4 V and a typical cutoff frequency of 35 THz.

A hafnium oxide memristor crossbar array integrated with transistors can provide a provable key destruction scheme in which unique physical fingerprints are extracted by comparing the conductance of neighbouring memristors, and can only be revealed if a digital key stored on the same array is erased.

A radiofrequency signal processing system that uses a multicore memristive system-on-a-chip can perform analogue discrete Fourier transformation, in-phase and quadrature demodulation and analogue neural network tasks for radiofrequency transmitter identification and anomaly detection.

A nanofin structure enables a working memristor array with 2-nm feature size and 4.5-Tbit per square inch packing density.

Borrowing the operating principles of a battery, a three-terminal organic switch has been developed on a flexible plastic substrate. The device consumes very little power and can be used as an artificial synapse for brain-inspired computing.

Xiong et al. report two scalable in-sensor visual processing arrays based on dual-gate silicon photodiodes, parallelizing the temporal and spatial information analysis. The bipolar analog output captures the amplitude of event-driven light changes, facilitating the classification of dynamic motions and static images.

Memristor-based neural networks hold promise for neuromorphic computing, yet large-scale experimental execution remains difficult. Here, Xia et al. create a multi-layer memristor neural network with in-situ machine learning and achieve competitive image classification accuracy on a standard dataset.

A memristor-based annealing system that uses an analogue neuromorphic architecture based on a Hopfield neural network can solve non-deterministic polynomial (NP)-hard max-cut problems in an approach that is potentially more efficient than current quantum, optical and digital approaches.

Chips with 256 × 256 memristor arrays that were monolithically integrated on complementary metal–oxide–semiconductor (CMOS) circuits in a commercial foundry achieved 2,048 conductance levels in individual memristors.

The n- and p-type channel characteristics of WSe₂ are exploited to implement multiply–accumulate and activation functions simultaneously in an in-memory computing core.

Memristors are key structural units of complex memory and computing systems, yet most currently available memristors are based on materials that are not compatible with silicon technology. Here, the authors demonstrate a CMOS-compatible, self-rectifying memristor and arrays entirely based on p-Si/SiO₂/n-Si.

A seemingly disordered network of nanowires governed by thermodynamics is used as the physical ‘reservoir’ in a memristive implementation of reservoir computing to process spatiotemporal information.

Deep neural networks are increasingly popular in data-intensive applications, but are power-hungry. New types of computer chips that are suited to the task of deep learning, such as memristor arrays where data handling and computing take place within the same unit, are required. A well-used deep learning model called long short-term memory, which can handle temporal sequential data analysis, is now implemented in a memristor crossbar array, promising an energy-efficient and low-footprint deep learning platform.

Though artificial sensory systems based on electronic devices have been realized, further transformation of data into spikes is required for neural network optimization. Here, based on NbO_x Mott memristors, the authors report artificial spiking afferent nerves for accessing spiking systems and demonstrate spiking mechanoreceptor systems.

A three-dimensional circuit composed of eight layers of monolithically integrated memristive devices is built and used to implement complex neural networks, demonstrating accurate MNIST classification and effective edge detection in videos.

Though memristors can potentially emulate neuron and synapse functionality, useful signal energy is lost to Joule heating. Here, the authors demonstrate neuro-transistors with a pseudo-memcapacitive gate that actively process signals via energy-efficient capacitively-coupled neural networks.

Memristive devices can provide energy-efficient neural network implementations, but they must be tailored to suit different network architectures. Wang et al. develop a trainable weight-sharing mechanism for memristor-based CNNs and ConvLSTMs, achieving a 75% reduction in weights without compromising accuracy.

The development of humanoid robots with artificial intelligence calls for smart solutions for tactile sensing systems that respond to dynamic changes in the environment. Here, Yoon et al. emulate non-adaption and sensitization function of a nociceptor—a sensory neuron—using diffusive oxide-based memristors.

Memristor crossbars with array sizes of up to 128 × 64 cells are capable of analogue vector-matrix multiplication and can be used for signal processing, image compression and convolutional filtering.

Leaky integrate-and-fire artificial neurons based on diffusive memristors enable unsupervised weight updates of drift-memristor synapses in an integrated convolutional neural network capable of pattern recognition.

Memristors can switch between high and low electrical-resistance states, but the switching behaviour can be unpredictable. Here, the authors harness this unpredictability to develop a memristor-based true random number generator that uses the stochastic delay time of threshold switching

A reinforcement learning algorithm can be implemented on a hybrid analogue–digital platform based on memristive arrays for parallel and energy-efficient in situ training.

Calcium ions play a vital role in enabling synaptic plasticity in biological systems. The dynamic behaviour of these ions has now been emulated in a metal atom diffusion-based memristor.

This Review summarizes latest advancements in memristor-based hardware accelerators, an energy-efficient solution for computing-intensive artificial intelligence algorithms, covering crossbar arrays, peripheral circuits, architectures and software–hardware co-designs. It analyses challenges and pathways for the transition of memristor technology to commercial products.

Memristor as the fourth basic element of electric circuits has drawn substantial attention for developing future computing technologies. Sun et al. report the progress and the challenges facing researchers on understanding memristive switching, and advocate continuous studies using a synergistic approach.

Resistive switching materials enable novel, in-memory information processing, which may resolve the von Neumann bottleneck. This Review focuses on how the switching mechanisms and the resultant electrical properties lead to various computing applications.

Memristive devices show great potential as artificial synapses and neurons, yet brain-inspired computing can be realized only by integrating a large number of these devices into reliable arrays. This Review discusses the challenges in the integration and use in computation of large-scale memristive neural networks.