Neural-like computing with populations of superparamagnetic basis functions

Self-assembled 3D Interconnected Magnetic Nanowire Networks for Neuromorphic Computing

Three-dimensional (3D) nanomagnetic systems offer promise toward implementing neuromorphic computing due to their intricate spin textures, magnetization dynamics, and nontrivial topology. However, the investigation of 3D nanomagnetic systems is often constrained by demanding fabrication and characterization requirements. Here, we present interconnected networks of self-assembled magnetic nanowires (NW) as a novel 3D platform with attractive characteristics for neuromorphic computing. The networks contain multiple unique transport pathways, each hosting discrete magnetization states. These pathways can be selectively addressed, and the magnetic state within them can be electrically controlled by applying current pulses. Consequently, the pathways can serve as synaptic weights, allowing for diverse programming by switching specific sections of the network using current pulses of varying magnitudes and durations. Additionally, unique features such as history-dependent magnetic state switching and interconnected transport paths are observed in these networks. These capabilities are leveraged to illustrate the potential of interconnected magnetic NW networks as reservoir layers in a neural network architecture, highlighting their promise as an efficient platform for neuromorphic computing.

Dynamic Control of Weight-Update Linearity in Magneto-Ionic Synapses

Multifunctional hardware technologies for neuromorphic computing are essential for replicating the complexity of biological neural systems, thereby improving the performance of artificial synapses and neurons. Integrating ionic and spintronic technologies offers new degrees of freedom to modulate synaptic potentiation and depression, introducing novel magnetic functionalities alongside the established ionic analogue behavior. We demonstrate that magneto-ionic devices can perform as synaptic elements with dynamically tunable depression linearity controlled by an external magnetic field, a functionality reminiscent of neuromodulation in biological systems. By applying magnetic fields we significantly reduce the nonlinearity of synaptic depression, transitioning from an exponential dependence to a linear response at higher fields. Neural network simulations reveal that this magnetically induced linearity enhancement improves learning accuracy across a wide range of learning rates, which is retained after the magnetic field is removed. These findings highlight the versatility and promise of magneto-ionic devices for developing tunable synaptic elements for neuromorphic hardware.

Neuromorphic overparameterisation and few-shot learning in multilayer physical neural networks

Physical neuromorphic computing, exploiting the complex dynamics of physical systems, has seen rapid advancements in sophistication and performance. Physical reservoir computing, a subset of neuromorphic computing, faces limitations due to its reliance on single systems. This constrains output dimensionality and dynamic range, limiting performance to a narrow range of tasks. Here, we engineer a suite of nanomagnetic array physical reservoirs and interconnect them in parallel and series to create a multilayer neural network architecture. The output of one reservoir is recorded, scaled and virtually fed as input to the next reservoir. This networked approach increases output dimensionality, internal dynamics and computational performance. We demonstrate that a physical neuromorphic system can achieve an overparameterised state, facilitating meta-learning on small training sets and yielding strong performance across a wide range of tasks. Our approach's efficacy is further demonstrated through few-shot learning, where the system rapidly adapts to new tasks.

Restricted Boltzmann Machines Implemented by Spin-Orbit Torque Magnetic Tunnel Junctions

Artificial intelligence has surged forward with the advent of generative models, which rely heavily on stochastic computing architectures enhanced by true random number generators with adjustable sampling probabilities. In this study, we develop spin-orbit torque magnetic tunnel junctions (SOT-MTJs), investigating their sigmoid-style switching probability as a function of the driving voltage. This feature proves to be ideally suited for stochastic computing algorithms such as the restricted Boltzmann machines (RBM) prevalent in pretraining processes. We exploit SOT-MTJs as both stochastic samplers and network nodes for RBMs, enabling the implementation of RBM-based neural networks to achieve recognition tasks for both handwritten and spoken digits. Moreover, we further harness the weights derived from the preceding image and speech training processes to facilitate cross-modal learning from speech to image generation. Our results clearly demonstrate that these SOT-MTJs are promising candidates for the development of hardware accelerators tailored for Boltzmann neural networks and other stochastic computing architectures.

Multiplex-free physical reservoir computing with an adaptive oscillator

Nonlinear oscillators can often be used as physical reservoir computers, in which the oscillator's dynamics simultaneously performs computation and stores information. Typically, the dynamic states are multiplexed in time, and then machine learning is used to unlock this stored information into a usable form. This time multiplexing is used to create virtual nodes, which are often necessary to capture enough information to perform different tasks, but this multiplexing procedure requires a relatively high sampling rate. Adaptive oscillators, which are a subset of nonlinear oscillators, have plastic states that learn and store information through their dynamics in a human readable form, without the need for machine learning. Highlighting this ability, adaptive oscillators have been used as analog frequency analyzers, robotic controllers, and energy harvesters. Here, adaptive oscillators are considered as a physical reservoir computer without the cumbersome time multiplexing procedure. With this multiplex-free physical reservoir computer architecture, the fundamental logic gates can be simultaneously calculated through dynamics without modifying the base oscillator.

Orbital multiferroicity in pentalayer rhombohedral graphene

Ferroic orders describe spontaneous polarization of spin, charge and lattice degrees of freedom in materials. Materials exhibiting multiple ferroic orders, known as multiferroics, have important parts in multifunctional electrical and magnetic device applications1-4. Two-dimensional materials with honeycomb lattices offer opportunities to engineer unconventional multiferroicity, in which the ferroic orders are driven purely by the orbital degrees of freedom and not by electron spin. These include ferro-valleytricity corresponding to the electron valley5 and ferro-orbital-magnetism6 supported by quantum geometric effects. These orbital multiferroics could offer strong valley-magnetic couplings and large responses to external fields-enabling device applications such as multiple-state memory elements and electric control of the valley and magnetic states. Here we report orbital multiferroicity in pentalayer rhombohedral graphene using low-temperature magneto-transport measurements. We observed anomalous Hall signals Rxy with an exceptionally large Hall angle (tanΘH > 0.6) and orbital magnetic hysteresis at hole doping. There are four such states with different valley polarizations and orbital magnetizations, forming a valley-magnetic quartet. By sweeping the gate electric field E, we observed a butterfly-shaped hysteresis of Rxy connecting the quartet. This hysteresis indicates a ferro-valleytronic order that couples to the composite field E · B (where B is the magnetic field), but not to the individual fields. Tuning E would switch each ferroic order independently and achieve non-volatile switching of them together. Our observations demonstrate a previously unknown type of multiferroics and point to electrically tunable ultralow-power valleytronic and magnetic devices.

Spiking Dynamics in Dual Free Layer Perpendicular Magnetic Tunnel Junctions

Spintronic devices have recently attracted a lot of attention in the field of unconventional computing due to their non-volatility for short- and long-term memory, nonlinear fast response, and relatively small footprint. Here we demonstrate experimentally how voltage driven magnetization dynamics of dual free layer perpendicular magnetic tunnel junctions can emulate spiking neurons in hardware. The output spiking rate was controlled by varying the dc bias voltage across the device. The field-free operation of this two-terminal device and its robustness against an externally applied magnetic field make it a suitable candidate to mimic the neuron response in a dense neural network. The small energy consumption of the device (4-16 pJ/spike) and its scalability are important benefits for embedded applications. This compact perpendicular magnetic tunnel junction structure could finally bring spiking neural networks to sub-100 nm size elements.

Emerging Iontronic Neural Devices for Neuromorphic Sensory Computing

Living organisms have a very mysterious and powerful sensory computing system based on ion activity. Interestingly, studies on iontronic devices in the past few years have proposed a promising platform for simulating the sensing and computing functions of living organisms, because: 1) iontronic devices can generate, store, and transmit a variety of signals by adjusting the concentration and spatiotemporal distribution of ions, which analogs to how the brain performs intelligent functions by alternating ion flux and polarization; 2) through ionic-electronic coupling, iontronic devices can bridge the biosystem with electronics and offer profound implications for soft electronics; 3) with the diversity of ions, iontronic devices can be designed to recognize specific ions or molecules by customizing the charge selectivity, and the ionic conductivity and capacitance can be adjusted to respond to external stimuli for a variety of sensing schemes, which can be more difficult for electron-based devices. This review provides a comprehensive overview of emerging neuromorphic sensory computing by iontronic devices, highlighting representative concepts of both low-level and high-level sensory computing and introducing important material and device breakthroughs. Moreover, iontronic devices as a means of neuromorphic sensing and computing are discussed regarding the pending challenges and future directions.

Artificial Neuronal Devices Based on Emerging Materials: Neuronal Dynamics and Applications

Artificial neuronal devices are critical building blocks of neuromorphic computing systems and currently the subject of intense research motivated by application needs from new computing technology and more realistic brain emulation. Researchers have proposed a range of device concepts that can mimic neuronal dynamics and functions. Although the switching physics and device structures of these artificial neurons are largely different, their behaviors can be described by several neuron models in a more unified manner. In this paper, the reports of artificial neuronal devices based on emerging volatile switching materials are reviewed from the perspective of the demonstrated neuron models, with a focus on the neuronal functions implemented in these devices and the exploitation of these functions for computational and sensing applications. Furthermore, the neuroscience inspirations and engineering methods to enrich the neuronal dynamics that remain to be implemented in artificial neuronal devices and networks toward realizing the full functionalities of biological neurons are discussed.

SCANet: Securing the Weights With Superparamagnetic-MTJ Crossbar Array Networks

Deep neural networks (DNNs) form a critical infrastructure supporting various systems, spanning from the iPhone neural engine to imaging satellites and drones. The design of these neural cores is often proprietary or a military secret. Nevertheless, they remain vulnerable to model replication attacks that seek to reverse engineer the network's synaptic weights. In this article, we propose SCANet (Superparamagnetic-MTJ Crossbar Array Networks), a novel defense mechanism against such model stealing attacks by utilizing the innate stochasticity in superparamagnets. When used as the synapse in DNNs, superparamagnetic magnetic tunnel junctions (s-MTJs) are shown to be significantly more secure than prior memristor-based solutions. The thermally induced telegraphic switching in the s-MTJs is robust and uncontrollable, thus thwarting the attackers from obtaining sensitive data from the network. Using a mixture of both superparamagnetic and conventional MTJs in the neural network (NN), the designer can optimize the time period between the weight updation and the power consumed by the system. Furthermore, we propose a modified NN architecture that can prevent replication attacks while minimizing power consumption. We investigate the effect of the number of layers in the deep network and the number of neurons in each layer on the sharpness of accuracy degradation when the network is under attack. We also explore the efficacy of SCANet in real-time scenarios, using a case study on object detection.

Hopf physical reservoir computer for reconfigurable sound recognition

The Hopf oscillator is a nonlinear oscillator that exhibits limit cycle motion. This reservoir computer utilizes the vibratory nature of the oscillator, which makes it an ideal candidate for reconfigurable sound recognition tasks. In this paper, the capabilities of the Hopf reservoir computer performing sound recognition are systematically demonstrated. This work shows that the Hopf reservoir computer can offer superior sound recognition accuracy compared to legacy approaches (e.g., a Mel spectrum + machine learning approach). More importantly, the Hopf reservoir computer operating as a sound recognition system does not require audio preprocessing and has a very simple setup while still offering a high degree of reconfigurability. These features pave the way of applying physical reservoir computing for sound recognition in low power edge devices.

The Road for 2D Semiconductors in the Silicon Age

Continued reduction in transistor size can improve the performance of silicon integrated circuits (ICs). However, as Moore's law approaches physical limits, high-performance growth in silicon ICs becomes unsustainable, due to challenges of scaling, energy efficiency, and memory limitations. The ultrathin layers, diverse band structures, unique electronic properties, and silicon-compatible processes of 2D materials create the potential to consistently drive advanced performance in ICs. Here, the potential of fusing 2D materials with silicon ICs to minimize the challenges in silicon ICs, and to create technologies beyond the von Neumann architecture, is presented, and the killer applications for 2D materials in logic and memory devices to ease scaling, energy efficiency bottlenecks, and memory dilemmas encountered in silicon ICs are discussed. The fusion of 2D materials allows the creation of all-in-one perception, memory, and computation technologies beyond the von Neumann architecture to enhance system efficiency and remove computing power bottlenecks. Progress on the 2D ICs demonstration is summarized, as well as the technical hurdles it faces in terms of wafer-scale heterostructure growth, transfer, and compatible integration with silicon ICs. Finally, the promising pathways and obstacles to the technological advances in ICs due to the integration of 2D materials with silicon are presented.

Almost Perfect Spin Filtering in Graphene-Based Magnetic Tunnel Junctions

We report on large spin-filtering effects in epitaxial graphene-based spin valves, strongly enhanced in our specific multilayer case. Our results were obtained by the effective association of chemical vapor deposited (CVD) multilayer graphene with a high quality epitaxial Ni(111) ferromagnetic spin source. We highlight that the Ni(111) spin source electrode crystallinity and metallic state are preserved and stabilized by multilayer graphene CVD growth. Complete nanometric spin valve junctions are fabricated using a local probe indentation process, and spin properties are extracted from the graphene-protected ferromagnetic electrode through the use of a reference Al₂O₃/Co spin analyzer. Strikingly, spin-transport measurements in these structures give rise to large negative tunnel magneto-resistance TMR = -160%, pointing to a particularly large spin polarization for the Ni(111)/Gr interface P_Ni/Gr, evaluated up to -98%. We then discuss an emerging physical picture of graphene-ferromagnet systems, sustained both by experimental data and ab initio calculations, intimately combining efficient spin filtering effects arising (i) from the bulk band structure of the graphene layers purifying the extracted spin direction, (ii) from the hybridization effects modulating the amplitude of spin polarized scattering states over the first few graphene layers at the interface, and (iii) from the epitaxial interfacial matching of the graphene layers with the spin-polarized Ni surface selecting well-defined spin polarized channels. Importantly, these main spin selection effects are shown to be either cooperating or competing, explaining why our transport results were not observed before. Overall, this study unveils a path to harness the full potential of low Resitance.Area (RA) graphene interfaces in efficient spin-based devices.

Local bifurcation with spin-transfer torque in superparamagnetic tunnel junctions

Modulation of the energy landscape by external perturbations governs various thermally-activated phenomena, described by the Arrhenius law. Thermal fluctuation of nanoscale magnetic tunnel junctions with spin-transfer torque (STT) shows promise for unconventional computing, whereas its rigorous representation, based on the Néel-Arrhenius law, has been controversial. In particular, the exponents for thermally-activated switching rate therein, have been inaccessible with conventional thermally-stable nanomagnets with decade-long retention time. Here we approach the Néel-Arrhenius law with STT utilising superparamagnetic tunnel junctions that have high sensitivity to external perturbations and determine the exponents through several independent measurements including homodyne-detected ferromagnetic resonance, nanosecond STT switching, and random telegraph noise. Furthermore, we show that the results are comprehensively described by a concept of local bifurcation observed in various physical systems. The findings demonstrate the capability of superparamagnetic tunnel junction as a useful tester for statistical physics as well as sophisticated engineering of probabilistic computing hardware with a rigorous mathematical foundation.

There has been much interest in using the probabilistic switching of magnetic tunnel junctions in unconventional computing, but to do so requires a detailed understanding of this switching. Here, Funatsu et al rigorously determine the switching exponents in superparamagnetic tunnel junctions.

Single skyrmion true random number generator using local dynamics and interaction between skyrmions

Magnetic skyrmions are of great interest to both fundamental research and applications in post-von-Neumann computing devices. The successful implementation of skyrmionic devices requires functionalities of skyrmions with effective controls. Here we show that the local dynamics of skyrmions, in contrast to the global dynamics of a skyrmion as a whole, can be introduced to provide effective functionalities for versatile computing. A single skyrmion interacting with local pinning centres under thermal effects can fluctuate in time and switch between a small-skyrmion and a large-skyrmion state, thereby serving as a robust true random number generator for probabilistic computing. Moreover, neighbouring skyrmions exhibit an anti-correlated coupling in their fluctuation dynamics. Both the switching probability and the dynamic coupling strength can be tuned by modifying the applied magnetic field and spin current. Our results could lead to progress in developing magnetic skyrmionic devices with high tunability and efficient controls.

Control of the stochastic response of magnetization dynamics in spin-torque oscillator through radio-frequency magnetic fields

Neuromorphic computing using spintronic devices, such as spin-torque oscillators (STOs), has been intensively studied for energy-efficient data processing. One of the critical issues in this application is stochasticity in magnetization dynamics, which limits the accuracy of computation. Such stochastic behavior, however, plays a key role in stochastic computing and machine learning. It is therefore important to develop methods for both suppressing and enhancing stochastic response in spintronic devices. We report on experimental investigations on control of stochastic quantity, such as the width of a distribution of transient time in magnetization dynamics in vortex-type STO. The spin-transfer effect can suppress stochasticity in transient dynamics from a non-oscillating to oscillating state, whereas an application of a radio-frequency magnetic field is effective in reducing stochasticity on the time evolution of the oscillating state.

Mutual control of stochastic switching for two electrically coupled superparamagnetic tunnel junctions

Superparamagnetic tunnel junctions (SMTJs) are promising sources for the randomness required by some compact and energy-efficient computing schemes. Coupling SMTJs gives rise to collective behavior that could be useful for cognitive computing. We use a simple linear electrical circuit to mutually couple two SMTJs through their stochastic electrical transitions. When one SMTJ makes a thermally induced transition, the voltage across both SMTJs changes, modifying the transition rates of both. This coupling leads to significant correlation between the states of the two devices. Using fits to a generalized Néel-Brown model for the individual thermally bistable magnetic devices, we can accurately reproduce the behavior of the coupled devices with a Markov model.

Towards peptide-based tunable multistate memristive materials

Development of new memristive hardware is a technological requirement towards widespread neuromorphic computing. Molecular spintronics seems to be a fertile field for the design and preparation of this hardware. Within molecular spintronics, recent results on metallopeptides demonstrating the interaction between paramagnetic ions and the chirality induced spin selectivity effect hold particular promise for developing fast (ns-μs) operation times. [R. Torres-Cavanillas et al., J. Am. Chem. Soc., 2020, DOI: 10.1021/jacs.0c07531]. Among the challenges in the field, a major highlight is the difficulty in modelling the spin dynamics in these complex systems, but at the same time the use of inexpensive methods has already allowed progress in that direction. Finally, we discuss the unique potential of biomolecules for the design of multistate memristors with a controlled- and indeed, programmable-nanostructure, allowing going beyond anything that is conceivable by employing conventional coordination chemistry.

Tunable Stochasticity in an Artificial Spin Network

Metamaterials present the possibility of artificially generating advanced functionalities through engineering of their internal structure. Artificial spin networks, in which a large number of nanoscale magnetic elements are coupled together, are promising metamaterial candidates that enable the control of collective magnetic behavior through tuning of the local interaction between elements. In this work, the motion of magnetic domain-walls in an artificial spin network leads to a tunable stochastic response of the metamaterial, which can be tailored through an external magnetic field and local lattice modifications. This type of tunable stochastic network produces a controllable random response exploiting intrinsic stochasticity within magnetic domain-wall motion at the nanoscale. An iconic demonstration used to illustrate the control of randomness is the Galton board. In this system, multiple balls fall into an array of pegs to generate a bell-shaped curve that can be modified via the array spacing or the tilt of the board. A nanoscale recreation of this experiment using an artificial spin network is employed to demonstrate tunable stochasticity. This type of tunable stochastic network opens new paths toward post-Von Neumann computing architectures such as Bayesian sensing or random neural networks, in which stochasticity is harnessed to efficiently perform complex computational tasks.

Nanosecond Random Telegraph Noise in In-Plane Magnetic Tunnel Junctions

We study the timescale of random telegraph noise (RTN) of nanomagnets in stochastic magnetic tunnel junctions (MTJs). From analytical and numerical calculations based on the Landau-Lifshitz-Gilbert and the Fokker-Planck equations, we reveal mechanisms governing the relaxation time of perpendicular easy-axis MTJs (p-MTJs) and in-plane easy-axis MTJs (i-MTJs), showing that i-MTJs can be made to have faster RTN. Superparamagnetic i-MTJs with small in-plane anisotropy and sizable perpendicular effective anisotropy show relaxation times down to 8 ns at negligible bias current, which is more than 5 orders of magnitude shorter than that of typical stochastic p-MTJs and about 100 times faster than the shortest time of i-MTJs reported so far. The findings give a new insight and foundation in developing stochastic MTJs for high-performance probabilistic computers.

Hardware Design for Autonomous Bayesian Networks

Directed acyclic graphs or Bayesian networks that are popular in many AI-related sectors for probabilistic inference and causal reasoning can be mapped to probabilistic circuits built out of probabilistic bits (p-bits), analogous to binary stochastic neurons of stochastic artificial neural networks. In order to satisfy standard statistical results, individual p-bits not only need to be updated sequentially but also in order from the parent to the child nodes, necessitating the use of sequencers in software implementations. In this article, we first use SPICE simulations to show that an autonomous hardware Bayesian network can operate correctly without any clocks or sequencers, but only if the individual p-bits are appropriately designed. We then present a simple behavioral model of the autonomous hardware illustrating the essential characteristics needed for correct sequencer-free operation. This model is also benchmarked against SPICE simulations and can be used to simulate large-scale networks. Our results could be useful in the design of hardware accelerators that use energy-efficient building blocks suited for low-level implementations of Bayesian networks. The autonomous massively parallel operation of our proposed stochastic hardware has biological relevance since neural dynamics in brain is also stochastic and autonomous by nature.

Magnonic Bending, Phase Shifting and Interferometry in a 2D Reconfigurable Nanodisk Crystal

Strongly interacting nanomagnetic systems are pivotal across next-generation technologies including reconfigurable magnonics and neuromorphic computation. Controlling magnetization states and local coupling between neighboring nanoelements allows vast reconfigurability and a host of associated functionalities. However, existing designs typically suffer from an inability to tailor interelement coupling post-fabrication and nanoelements restricted to a pair of Ising-like magnetization states. Here, we propose a class of reconfigurable magnonic crystals incorporating nanodisks as the functional element. Ferromagnetic nanodisks are crucially bistable in macrospin and vortex states, allowing interelement coupling to be selectively activated (macrospin) or deactivated (vortex). Through microstate engineering, we leverage the distinct coupling behaviors and magnonic band structures of bistable nanodisks to achieve reprogrammable magnonic waveguiding, bending, gating, and phase-shifting across a 2D network. The potential of nanodisk-based magnonics for wave-based computation is demonstrated via an all-magnon interferometer exhibiting XNOR logic functionality. Local microstate control is achieved here via topological magnetic writing using a magnetic force microscope tip.

Roadmap of spin-orbit torques

Spin-orbit torque (SOT) is an emerging technology that enables the efficient manipulation of spintronic devices. The initial processes of interest in SOTs involved electric fields, spin-orbit coupling, conduction electron spins and magnetization. More recently interest has grown to include a variety of other processes that include phonons, magnons, or heat. Over the past decade, many materials have been explored to achieve a larger SOT efficiency. Recently, holistic design to maximize the performance of SOT devices has extended material research from a nonmagnetic layer to a magnetic layer. The rapid development of SOT has spurred a variety of SOT-based applications. In this Roadmap paper, we first review the theories of SOTs by introducing the various mechanisms thought to generate or control SOTs, such as the spin Hall effect, the Rashba-Edelstein effect, the orbital Hall effect, thermal gradients, magnons, and strain effects. Then, we discuss the materials that enable these effects, including metals, metallic alloys, topological insulators, two-dimensional materials, and complex oxides. We also discuss the important roles in SOT devices of different types of magnetic layers, such as magnetic insulators, antiferromagnets, and ferrimagnets. Afterward, we discuss device applications utilizing SOTs. We discuss and compare three-terminal and two-terminal SOT-magnetoresistive random-access memories (MRAMs); we mention various schemes to eliminate the need for an external field. We provide technological application considerations for SOT-MRAM and give perspectives on SOT-based neuromorphic devices and circuits. In addition to SOT-MRAM, we present SOT-based spintronic terahertz generators, nano-oscillators, and domain wall and skyrmion racetrack memories. This paper aims to achieve a comprehensive review of SOT theory, materials, and applications, guiding future SOT development in both the academic and industrial sectors.

Roadmap on emerging hardware and technology for machine learning

Recent progress in artificial intelligence is largely attributed to the rapid development of machine learning, especially in the algorithm and neural network models. However, it is the performance of the hardware, in particular the energy efficiency of a computing system that sets the fundamental limit of the capability of machine learning. Data-centric computing requires a revolution in hardware systems, since traditional digital computers based on transistors and the von Neumann architecture were not purposely designed for neuromorphic computing. A hardware platform based on emerging devices and new architecture is the hope for future computing with dramatically improved throughput and energy efficiency. Building such a system, nevertheless, faces a number of challenges, ranging from materials selection, device optimization, circuit fabrication and system integration, to name a few. The aim of this Roadmap is to present a snapshot of emerging hardware technologies that are potentially beneficial for machine learning, providing the Nanotechnology readers with a perspective of challenges and opportunities in this burgeoning field.

Integration and Co-design of Memristive Devices and Algorithms for Artificial Intelligence

Memristive devices share remarkable similarities to biological synapses, dendrites, and neurons at both the physical mechanism level and unit functionality level, making the memristive approach to neuromorphic computing a promising technology for future artificial intelligence. However, these similarities do not directly transfer to the success of efficient computation without device and algorithm co-designs and optimizations. Contemporary deep learning algorithms demand the memristive artificial synapses to ideally possess analog weighting and linear weight-update behavior, requiring substantial device-level and circuit-level optimization. Such co-design and optimization have been the main focus of memristive neuromorphic engineering, which often abandons the “non-ideal” behaviors of memristive devices, although many of them resemble what have been observed in biological components. Novel brain-inspired algorithms are being proposed to utilize such behaviors as unique features to further enhance the efficiency and intelligence of neuromorphic computing, which calls for collaborations among electrical engineers, computing scientists, and neuroscientists.

Element-Specific Detection of Sub-Nanosecond Spin-Transfer Torque in a Nanomagnet Ensemble

Spin currents can exert spin-transfer torques on magnetic systems even in the limit of vanishingly small net magnetization, as recently shown for antiferromagnets. Here, we experimentally show that a spin-transfer torque is operative in a macroscopic ensemble of weakly interacting, randomly magnetized Co nanomagnets. We employ element- and time-resolved X-ray ferromagnetic resonance (XFMR) spectroscopy to directly detect subnanosecond dynamics of the Co nanomagnets, excited into precession with cone angle ≳0.003° by an oscillating spin current. XFMR measurements reveal that as the net moment of the ensemble decreases, the strength of the spin-transfer torque increases relative to those of magnetic field torques. Our findings point to spin-transfer torque as an effective way to manipulate the state of nanomagnet ensembles at subnanosecond time scales.

Electrically programmable probabilistic bit anti-correlator on a nanomagnetic platform

Execution of probabilistic computing algorithms require electrically programmable stochasticity to encode arbitrary probability functions and controlled stochastic interaction or correlation between probabilistic (p-) bits. The latter is implemented with complex electronic components leaving a large footprint on a chip and dissipating excessive amount of energy. Here, we show an elegant implementation with just two dipole-coupled magneto-tunneling junctions (MTJ), with magnetostrictive soft layers, fabricated on a piezoelectric film. The resistance states of the two MTJs (high or low) encode the p-bit values (1 or 0) in the two streams. The first MTJ is driven to a resistance state with desired probability via a current or voltage that generates spin transfer torque, while the second MTJ's resistance state is determined by dipole coupling with the first, thus correlating the second p-bit stream with the first. The effect of dipole coupling can be varied by generating local strain in the soft layer of the second MTJ with a local voltage (~ 0.2 V) and that varies the degree of anti-correlation between the resistance states of the two MTJs and hence between the two streams (from 0 to 100%). This paradigm generates the anti-correlation with "wireless" dipole coupling that consumes no footprint on a chip and dissipates no energy, and it controls the degree of anti-correlation with electrically generated strain that consumes minimal footprint and is extremely frugal in its use of energy. It can be extended to arbitrary number of bit streams. This realizes an "all-magnetic" platform for generating correlations or anti-correlations for probabilistic computing. It also implements a simple 2-node Bayesian network.

Memory devices and applications for in-memory computing

Traditional von Neumann computing systems involve separate processing and memory units. However, data movement is costly in terms of time and energy and this problem is aggravated by the recent explosive growth in highly data-centric applications related to artificial intelligence. This calls for a radical departure from the traditional systems and one such non-von Neumann computational approach is in-memory computing. Hereby certain computational tasks are performed in place in the memory itself by exploiting the physical attributes of the memory devices. Both charge-based and resistance-based memory devices are being explored for in-memory computing. In this Review, we provide a broad overview of the key computational primitives enabled by these memory devices as well as their applications spanning scientific computing, signal processing, optimization, machine learning, deep learning and stochastic computing.

Two-dimensional materials for next-generation computing technologies

Rapid digital technology advancement has resulted in a tremendous increase in computing tasks imposing stringent energy efficiency and area efficiency requirements on next-generation computing. To meet the growing data-driven demand, in-memory computing and transistor-based computing have emerged as potent technologies for the implementation of matrix and logic computing. However, to fulfil the future computing requirements new materials are urgently needed to complement the existing Si complementary metal-oxide-semiconductor technology and new technologies must be developed to enable further diversification of electronics and their applications. The abundance and rich variety of electronic properties of two-dimensional materials have endowed them with the potential to enhance computing energy efficiency while enabling continued device downscaling to a feature size below 5 nm. In this Review, from the perspective of matrix and logic computing, we discuss the opportunities, progress and challenges of integrating two-dimensional materials with in-memory computing and transistor-based computing technologies.

Immunity of nanoscale magnetic tunnel junctions with perpendicular magnetic anisotropy to ionizing radiation

Spin transfer torque magnetic random access memory (STT-MRAM) is a promising candidate for next generation memory as it is non-volatile, fast, and has unlimited endurance. Another important aspect of STT-MRAM is that its core component, the nanoscale magnetic tunneling junction (MTJ), is thought to be radiation hard, making it attractive for space and nuclear technology applications. However, studies on the effects of ionizing radiation on the STT-MRAM writing process are lacking for MTJs with perpendicular magnetic anisotropy (pMTJs) required for scalable applications. Particularly, the question of the impact of extreme total ionizing dose on perpendicular magnetic anisotropy, which plays a crucial role on thermal stability and critical writing current, remains open. Here we report measurements of the impact of high doses of gamma and neutron radiation on nanoscale pMTJs used in STT-MRAM. We characterize the tunneling magnetoresistance, the magnetic field switching, and the current-induced switching before and after irradiation. Our results demonstrate that all these key properties of nanoscale MTJs relevant to STT-MRAM applications are robust against ionizing radiation. Additionally, we perform experiments on thermally driven stochastic switching in the gamma ray environment. These results indicate that nanoscale MTJs are promising building blocks for radiation-hard non-von Neumann computing.

Double MgO-Based Perpendicular Magnetic Tunnel Junction for Artificial Neuron

A perpendicular spin transfer torque (p-STT)-based neuron was developed for a spiking neural network (SNN). It demonstrated the integration behavior of a typical neuron in an SNN; in particular, the integration behavior corresponding to magnetic resistance change gradually increased with the input spike number. This behavior occurred when the spin electron directions between double Co₂Fe₆B₂ free and pinned layers in the p-STT-based neuron were switched from parallel to antiparallel states. In addition, a neuron circuit for integrate-and-fire operation was proposed. Finally, pattern-recognition simulation was performed for a single-layer SNN.

LAO-NCS: Laser Assisted Spin Torque Nano Oscillator-Based Neuromorphic Computing System

Dealing with big data, especially the videos and images, is the biggest challenge of existing Von-Neumann machines while the human brain, benefiting from its massive parallel structure, is capable of processing the images and videos in a fraction of second. The most promising solution, which has been recently researched widely, is brain-inspired computers, so-called neuromorphic computing systems (NCS). The NCS overcomes the limitation of the word-at-a-time thinking of conventional computers benefiting from massive parallelism for data processing, similar to the brain. Recently, spintronic-based NCSs have shown the potential of implementation of low-power high-density NCSs, where neurons are implemented using magnetic tunnel junctions (MTJs) or spin torque nano-oscillators (STNOs) and memristors are used to mimic synaptic functionality. Although using STNOs as neuron requires lower energy in comparison to the MTJs, still there is a huge gap between the power consumption of spintronic-based NCSs and the brain due to high bias current needed for starting the oscillation with a detectable output power. In this manuscript, we propose a spintronic-based NCS (196 × 10) proof-of-concept where the power consumption of the NCS is reduced by assisting the STNO oscillation through a microwatt nanosecond laser pulse. The experimental results show the power consumption of the STNOs in the designed NCS is reduced by 55.3% by heating up the STNOs to 100°C. Moreover, the average power consumption of spintronic layer (STNOs and memristor array) is decreased by 54.9% at 100°C compared with room temperature. The total power consumption of the proposed laser assisted STNO-based NCS (LAO-NCS) at 100°C is improved by 40% in comparison to a typical STNO-based NCS at room temperature. Finally, the energy consumption of the LAO-NCA at 100°C is expected to reduce by 86% compared with a typical STNO-based NCS at the room temperature.

Memristive and CMOS Devices for Neuromorphic Computing

Neuromorphic computing has emerged as one of the most promising paradigms to overcome the limitations of von Neumann architecture of conventional digital processors. The aim of neuromorphic computing is to faithfully reproduce the computing processes in the human brain, thus paralleling its outstanding energy efficiency and compactness. Toward this goal, however, some major challenges have to be faced. Since the brain processes information by high-density neural networks with ultra-low power consumption, novel device concepts combining high scalability, low-power operation, and advanced computing functionality must be developed. This work provides an overview of the most promising device concepts in neuromorphic computing including complementary metal-oxide semiconductor (CMOS) and memristive technologies. First, the physics and operation of CMOS-based floating-gate memory devices in artificial neural networks will be addressed. Then, several memristive concepts will be reviewed and discussed for applications in deep neural network and spiking neural network architectures. Finally, the main technology challenges and perspectives of neuromorphic computing will be discussed.

Energy-efficient stochastic computing with superparamagnetic tunnel junctions

Superparamagnetic tunnel junctions (SMTJs) have emerged as a competitive, realistic nanotechnology to support novel forms of stochastic computation in CMOS-compatible platforms. One of their applications is to generate random bitstreams suitable for use in stochastic computing implementations. We describe a method for digitally programmable bitstream generation based on pre-charge sense amplifiers. This generator is significantly more energy efficient than SMTJ-based bitstream generators that tune probabilities with spin currents and a factor of two more efficient than related CMOS-based implementations. The true randomness of this bitstream generator allows us to use them as the fundamental units of a novel neural network architecture. To take advantage of the potential savings, we codesign the algorithm with the circuit, rather than directly transcribing a classical neural network into hardware. The flexibility of the neural network mathematics allows us to adapt the network to the explicitly energy efficient choices we make at the device level. The result is a convolutional neural network design operating at ≈ 150 nJ per inference with 97 % performance on MNIST-a factor of 1.4 to 7.7 improvement in energy efficiency over comparable proposals in the recent literature.

Neuromorphic Spintronics

Neuromorphic computing uses basic principles inspired by the brain to design circuits that perform artificial intelligence tasks with superior energy efficiency. Traditional approaches have been limited by the energy area of artificial neurons and synapses realized with conventional electronic devices. In recent years, multiple groups have demonstrated that spintronic nanodevices, which exploit the magnetic as well as electrical properties of electrons, can increase the energy efficiency and decrease the area of these circuits. Among the variety of spintronic devices that have been used, magnetic tunnel junctions play a prominent role because of their established compatibility with standard integrated circuits and their multifunctionality. Magnetic tunnel junctions can serve as synapses, storing connection weights, functioning as local, nonvolatile digital memory or as continuously varying resistances. As nano-oscillators, they can serve as neurons, emulating the oscillatory behavior of sets of biological neurons. As superparamagnets, they can do so by emulating the random spiking of biological neurons. Magnetic textures like domain walls or skyrmions can be configured to function as neurons through their non-linear dynamics. Several implementations of neuromorphic computing with spintronic devices demonstrate their promise in this context. Used as variable resistance synapses, magnetic tunnel junctions perform pattern recognition in an associative memory. As oscillators, they perform spoken digit recognition in reservoir computing and when coupled together, classification of signals. As superparamagnets, they perform population coding and probabilistic computing. Simulations demonstrate that arrays of nanomagnets and films of skyrmions can operate as components of neuromorphic computers. While these examples show the unique promise of spintronics in this field, there are several challenges to scaling up, including the efficiency of coupling between devices and the relatively low ratio of maximum to minimum resistances in the individual devices.

Band-Structure Spin-Filtering in Vertical Spin Valves Based on Chemical Vapor Deposited WS2

We report on spin transport in WS₂-based 2D-magnetic tunnel junctions (2D-MTJs), unveiling a band structure spin filtering effect specific to the transition metal dichalcogenides (TMDCs) family. WS₂ mono-, bi-, and trilayers are derived by a chemical vapor deposition process and further characterized by Raman spectroscopy, atomic force microscopy (AFM), and photoluminescence spectroscopy. The WS₂ layers are then integrated in complete Co/Al₂O₃/WS₂/Co MTJ hybrid spin-valve structures. We make use of a tunnel Co/Al₂O₃ spin analyzer to probe the extracted spin-polarized current from the WS₂/Co interface and its evolution as a function of WS₂ layer thicknesses. For monolayer WS₂, our technological approach enables the extraction of the largest spin signal reported for a TMDC-based spin valve, corresponding to a spin polarization of P_Co/WS2 = 12%. Interestingly, for bi- and trilayer WS₂, the spin signal is reversed, which indicates a switch in the mechanism of interfacial spin extraction. With the support of ab initio calculations, we propose a model to address the experimentally measured inversion of the spin polarization based on the change in the WS₂ band structure while going from monolayer (direct bandgap) to bilayer (indirect bandgap). These experiments illustrate the rich potential of the families of semiconducting 2D materials for the control of spin currents in 2D-MTJs.

Integer factorization using stochastic magnetic tunnel junctions

Conventional computers operate deterministically using strings of zeros and ones called bits to represent information in binary code. Despite the evolution of conventional computers into sophisticated machines, there are many classes of problems that they cannot efficiently address, including inference, invertible logic, sampling and optimization, leading to considerable interest in alternative computing schemes. Quantum computing, which uses qubits to represent a superposition of 0 and 1, is expected to perform these tasks efficiently^1-3. However, decoherence and the current requirement for cryogenic operation⁴, as well as the limited many-body interactions that can be implemented, pose considerable challenges. Probabilistic computing^1,5-7 is another unconventional computation scheme that shares similar concepts with quantum computing but is not limited by the above challenges. The key role is played by a probabilistic bit (a p-bit)-a robust, classical entity fluctuating in time between 0 and 1, which interacts with other p-bits in the same system using principles inspired by neural networks⁸. Here we present a proof-of-concept experiment for probabilistic computing using spintronics technology, and demonstrate integer factorization, an illustrative example of the optimization class of problems addressed by adiabatic⁹ and gated² quantum computing. Nanoscale magnetic tunnel junctions showing stochastic behaviour are developed by modifying market-ready magnetoresistive random-access memory technology^10,11 and are used to implement three-terminal p-bits that operate at room temperature. The p-bits are electrically connected to form a functional asynchronous network, to which a modified adiabatic quantum computing algorithm that implements three- and four-body interactions is applied. Factorization of integers up to 945 is demonstrated with this rudimentary asynchronous probabilistic computer using eight correlated p-bits, and the results show good agreement with theoretical predictions, thus providing a potentially scalable hardware approach to the difficult problems of optimization and sampling.

A Ferrite Synaptic Transistor with Topotactic Transformation

Hardware implementation of artificial synaptic devices that emulate the functions of biological synapses is inspired by the biological neuromorphic system and has drawn considerable interest. Here, a three-terminal ferrite synaptic device based on a topotactic phase transition between crystalline phases is presented. The electrolyte-gating-controlled topotactic phase transformation between brownmillerite SrFeO_2.5 and perovskite SrFeO_{3- δ} is confirmed from the examination of the crystal and electronic structure. A synaptic transistor with electrolyte-gated ferrite films by harnessing gate-controllable multilevel conduction states, which originate from many distinct oxygen-deficient perovskite structures of SrFeO_x induced by topotactic phase transformation, is successfully constructed. This three-terminal artificial synapse can mimic important synaptic functions, such as synaptic plasticity and spike-timing-dependent plasticity. Simulations of a neural network consisting of ferrite synaptic transistors indicate that the system offers high classification accuracy. These results provide insight into the potential application of advanced topotactic phase transformation materials for designing artificial synapses with high performance.

Magnetization reversal driven by low dimensional chaos in a nanoscale ferromagnet

Energy-efficient switching of magnetization is a central problem in nonvolatile magnetic storage and magnetic neuromorphic computing. In the past two decades, several efficient methods of magnetic switching were demonstrated including spin torque, magneto-electric, and microwave-assisted switching mechanisms. Here we experimentally show that low-dimensional magnetic chaos induced by alternating spin torque can strongly increase the rate of thermally-activated magnetic switching in a nanoscale ferromagnet. This mechanism exhibits a well-pronounced threshold character in spin torque amplitude and its efficiency increases with decreasing spin torque frequency. We present analytical and numerical calculations that quantitatively explain these experimental findings and reveal the key role played by low-dimensional magnetic chaos near saddle equilibria in enhancement of the switching rate. Our work unveils an important interplay between chaos and stochasticity in the energy assisted switching of magnetic nanosystems and paves the way towards improved energy efficiency of spin torque memory and logic.

Synaptic dynamics in complex self-assembled nanoparticle networks

We report a detailed study of neuromorphic switching behaviour in inherently complex percolating networks of self-assembled metal nanoparticles. We show that variation of the strength and duration of the electric field applied to this network of synapse-like atomic switches allows us to control the switching dynamics. Switching is observed for voltages above a well-defined threshold, with higher voltages leading to increased switching rates. We demonstrate two behavioral archetypes and show how the switching dynamics change as a function of duration and amplitude of the voltage stimulus. We show that the state of each synapse can influence the activity of the other synapses, leading to complex switching dynamics. We further demonstrate the influence of the morphology of the network on the measured device properties, and the constraints imposed by the overall network conductance. The correlated switching dynamics, device stability over long periods, and the simplicity of the device fabrication provide an attractive pathway to practical implementation of on-chip neuromorphic computing.

Spin-torque devices with hard axis initialization as Stochastic Binary Neurons

Employing the probabilistic nature of unstable nano-magnet switching has recently emerged as a path towards unconventional computational systems such as neuromorphic or Bayesian networks. In this letter, we demonstrate proof-of-concept stochastic binary operation using hard axis initialization of nano-magnets and control of their output state probability (activation function) by means of input currents. Our method provides a natural path towards addition of weighted inputs from various sources, mimicking the integration function of neurons. In our experiment, spin orbit torque (SOT) is employed to "drive" nano-magnets with perpendicular magnetic anisotropy (PMA) -to their metastable state, i.e. in-plane hard axis. Next, the probability of relaxing into one magnetization state (+mi) or the other (-mi) is controlled using an Oersted field generated by an electrically isolated current loop, which acts as a "charge" input to the device. The final state of the magnet is read out by the anomalous Hall effect (AHE), demonstrating that the magnetization can be probabilistically manipulated and output through charge currents, closing the loop from charge-to-spin and spin-to-charge conversion. Based on these building blocks, a two-node directed network is successfully demonstrated where the status of the second node is determined by the probabilistic output of the previous node and a weighted connection between them. We have also studied the effects of various magnetic properties, such as magnet size and anisotropic field on the stochastic operation of individual devices through Monte Carlo simulations of Landau Lifshitz Gilbert (LLG) equation. The three-terminal stochastic devices demonstrated here are a critical step towards building energy efficient spin based neural networks and show the potential for a new application space.

Overcoming device unreliability with continuous learning in a population coding based computing system

The brain, which uses redundancy and continuous learning to overcome the unreliability of its components, provides a promising path to building computing systems that are robust to the unreliability of their constituent nanodevices. In this work, we illustrate this path by a computing system based on population coding with magnetic tunnel junctions that implement both neurons and synaptic weights. We show that equipping such a system with continuous learning enables it to recover from the loss of neurons and makes it possible to use unreliable synaptic weights (i.e. low energy barrier magnetic memories). There is a tradeoff between power consumption and precision because low energy barrier memories consume less energy than high barrier ones. For a given precision, there is an optimal number of neurons and an optimal energy barrier for the weights that leads to minimum power consumption.