Search

Resistive random-access-memory (RRAM)-based computing-in-memory (CIM) chips could overcome the von Neumann bottleneck and drastically improve energy efficiency for artificial intelligence (AI) applications. However, realizing their scalability necessitates the realization of higher-density integration, calling for cross-layer innovations from RRAM device optimization and unit cell design to integration strategies.

Skyrmions, a type of topological spin texture, have garnered interest for use in spintronic devices. Typically, these devices necessitate moving the skyrmions via applied currents. Here, Yang et al demonstrate the driving of skyrmions by surface acoustic waves.

Random two-dimensional arrays of carbon nanotubes, which are self-assembled via ion-exchange chemistry, can be used to create cryptographic keys by determining the connection yield and switching type of the nanotube devices.

Heterostructures consisting of ferromagnets and heavy metals have become a focus of interest because their strong spin–orbit coupling allows for efficient current-induced magnetization switching phenomena. Now, a magnetically doped topological insulator bilayer is shown to display a range of appealing characteristics for current-induced magnetization switching, including a significantly enhanced efficiency.

High-performance carbon nanotube thin-film transistors and complementary circuits can be fabricated on flexible substrates, including ring oscillators that have a stage delay of only 5.7 ns.

The authors demonstrate voltage-controlled multiferroic magnon torque in BiFeO₃ heterostructures, enabling reconfigurable logic-in-memory devices. This work highlights potential for low-power, scalable magnonics in room-temperature computing.

A recent study demonstrates the potential of using in-memory computing architecture for implementing large language models for an improved computational efficiency in both time and energy while maintaining a high accuracy.

Dynamic and non-volatile memristors can be used to create hardware-based reservoir and readout layers in artificial neural networks, providing a fully analogue signal processing chain for efficient data classification.

A memristor-based architecture for federated learning can implement compute-in-memory technology for encryption and decryption computations, a physical unclonable function for key generation and a true random number generator for error polynomial generation all within the same memristor array and peripheral circuits.

A neuromorphic decoder for brain–computer interfaces that is based on a 128k-cell memristor chip can offer software-equivalent decoding performance and can adapt to changing brain signals.

A memristor-based artificial dendrite enables the neural network to perform high-accuracy computation tasks with reduced power consumption.

An analogue–digital unified compute-in-memory architecture can offer native support for floating-point-based complex regression tasks, providing improved accuracy and energy efficiency compared with pure analogue compute-in-memory systems.

This study reports a fully integrated 128 × 8 optoelectronic memristor array with Si complementary metal–oxide–semiconductor circuits, featuring configurable multi-mode functionality. It demonstrates diversified in-sensor computing tasks and consumes 20 times less energy than GPUs.

The practicality of memristor-based computation-in-memory (CIM) is limited by the specific hardware design and the manual parameters tuning process. Here, the authors develop a full-stack CIM system with both hardware and software design for improved flexibility and efficiency.

The proposed edge detection based on ferroelectric field effect transistor does not rely on conventional convolution operation, realizing no-accuracy-loss, low-power (~10 fJ/per operation) and analogue-to-digital converter (ADC)-free edge computing.

Functional devices based on sliding ferroelectrics remain elusive. This work demonstrates the rewritable, non-volatile memory devices at room-temperature with two-dimensional sliding ferroelectric rhombohedral-stacked bilayer MoS₂. The device shows overall good performance and can be made flexible.

This study introduces an in-memory deep Bayesian active learning framework that uses the stochastic properties of memristors for in situ probabilistic computations. This framework can greatly improve the efficiency and speed of artificial intelligence learning tasks, as demonstrated with a robot skill-learning task.

High-performance, CMOS-compatible ring-oscillators based on carbon nanotubes are fabricated by scalable and fully manufacturable processes.

Dendritic computing is a promising approach to enhance the processing capability of artificial neural networks. Here, the authors report the development of a neurotransistor based on a vertical dual-gate electrolyte-gated transistor with short-term memory characteristics, a 30 nm channel length, a low read power of ~3.16 fW and read energy of ~30 fJ for dendritic computing.

Designing efficient 3D artificial neural networks chip remains a challenge. Here, the authors report a M3D-LIME chip with monolithic three-dimensional integration of hybrid memory architecture based on resistive random-access memory, which achieves a high classification accuracy of 96% in one-shot learning task while exhibiting 18.3× higher energy efficiency than GPU.

Image reconstruction algorithms raise critical challenges in massive data processing for medical diagnosis. Here, the authors propose a solution to significantly accelerate medical image reconstruction on memristor arrays, showing 79× faster speed and 153× higher energy efficiency than state-of-the-art graphics processing unit.

Designing efficient multistate resistive switching devices is promising for neuromorphic computing. Here, the authors demonstrate a reversible hydrogenation in WO₃ thin films at room temperature with an electrically-biased scanning probe. The associated insulator to metal transition offers the opportunity to precisely control multistate conductivity at nanoscale.

The stochastic features of memristors make them suitable for computation and probabilistic sampling; however, implementing these properties in hardware is extremely challenging. Lin et al. introduce an approach that leverages the cycle-to-cycle read variability of memristors as a physical random variable for in situ, real-time random number generation, and demonstrate it on a risk-sensitive reinforcement learning task.

Voltage control of magnetism in ferromagnetic semiconductor is appealing for spintronic applications, which is yet hindered by compound formation and low Curie temperature. Here, Nie et al. report electric-field control of ferromagnetism in Mn_xGe_1−xnanomeshes with a Curie temperature above 400 K and controllable giant magnetoresistance.

Sound localization is one of the many learning tasks accomplished by the brain based on the binaural signals of the ears. Here, Wu et al demonstrate in-situ learning of sound localization function using a memristor array, with dramatic improvements in energy efficiency.

Reservoir computing has demonstrated high-level performance, however efficient hardware implementations demand an architecture with minimum system complexity. The authors propose a rotating neuron-based architecture for physically implementing all-analog resource efficient reservoir computing system.

Designing efficient neuromorphic systems for complex temporal tasks remains a challenge. Zhong et al. develop a parallel memristor-based reservoir computing system capable of tuning critical parameters, achieving classification accuracy of 99.6% in spoken-digit recognition and time-series prediction error of 0.046 in the Hénon map.

A fully hardware-based memristor convolutional neural network using a hybrid training method achieves an energy efficiency more than two orders of magnitude greater than that of graphics-processing units.

Designing energy efficient and high performance brain-machine interfaces with millions of recording electrodes for in-situ analysis remains a challenge. Here, the authors develop a memristor-based neural signal analysis system capable of filtering and identifying epilepsy-related brain activities with an accuracy of 93.46%.

Spin–orbit torques in a geometrically asymmetric device made from a perpendicularly magnetized ferromagnet can switch its magnetization without the assistance of an applied magnetic field.

Electric field control of spin–orbit torque and magnetization switching can be achieved in a Cr-doped topological insulator thin film incorporated in a field-effect transistor structure, promising gate-controlled spintronic applications.

Controlling the magnetic properties of a materials system by electric means can lead to efficient electronic and memory devices. Now, for the first time, the control of ferromagnetism by the application of an electric voltage is demonstrated in germanium quantum dots for temperatures up to 100 K.

The convergence of spintronics and traditional semiconductor technology marks a critical juncture in the evolution of computing architectures, for which the development of foundation cells become indispensable. Here we discuss the current landscape of spintronics and propose a holistic co-design methodology to integrate spintronic devices into silicon platforms.

This Review examines the development of electrical reservoir computing, considering the architectures, physical nodes, and input and output layers of the approach, as well as performance benchmarks and the competitiveness of different implementations.

This Review Article examines the development of neuro-inspired computing chips and their key benchmarking metrics, providing a co-design tool chain and proposing a roadmap for future large-scale chips.