Brain-inspired Neuromorphic Computing Research Articles

Rational Design of Fluorinated 2D Polymer Film Based on Donor-Accepter Architecture toward Multilevel Memory Device for Neuromorphic Computing.

Fluorine-containing 2D polymer (F-2DP) film is a desired system to regulate the charge transport in organic electronics but rather rarely reports due to the limited fluorine-containing building blocks and difficulties in synthesis. Herein, a novel polar molecule with antiparallel columnar stacking is synthesized and further embedded into an F-2DP system to control over the crystallinity of F-2DP film through self-complementary π-electronic forces. The donor-accepter-accepter'-donor' (D-A-A'-D') structure regulates the charge transportation efficiently, inducing multilevel memory behavior through stepwise charge capture and transfer processes. Thus, the device exhibits ternary memory behavior with low threshold voltage (Vth1 of 1.1V, Vth2 of 2.0V), clearly distinguishable resistance states (1:102:104) and ternary yield (83%). Furthermore, the stepwise formation of the charge complex endows the device with a wider range to regulate the conductive state, which allows its application in brain-inspired neuromorphic computing. Modified National Institute of Standards and Technology recognition can reach an accuracy of 86%, showing great potential in neuromorphic computing applications in the post-Moore era.

Approaching the Ideal Linearity in Epitaxial Crystalline-Type Memristor by Controlling Filament Growth.

Brain-inspired neuromorphic computing has attracted widespread attention owing to its ability to perform parallel and energy-efficient computation. However, the synaptic weight of amorphous/polycrystalline oxide based memristor usually exhibits large nonlinear behavior with high asymmetry, which aggravates the complexity of peripheral circuit system. Controllable growth of conductive filaments is highly demanded for achieving the highly linear conductance modulation. However, the stochastic behavior of the filament growth in commonly used amorphous/polycrystalline oxide memristor makes it very challenging. Here, the epitaxially grown Hf0.5Zr0.5O2-based memristor with the linearity and symmetry approaching ideal case is reported. A layer of Cu nanoparticles is inserted into epitaxially grown Hf0.5Zr0.5O2 film to form the grain boundaries due to the breaking of the epitaxial growth. By combining with the local electric field enhancement, the growth of filament is confined in the grain boundaries due to the fact that the diffusion of oxygen vacancy in crystalline lattice is more difficult than that in the grain boundaries. Furthermore, the decimal operation and high-accuracy neural network are demonstrated by utilizing the highly linear and multi-level conductance modulation capacity. This method opens an avenue to control the filament growth for the application of resistance random access memory (RRAM) and neuromorphic computing.

Organic iontronic memristors for artificial synapses and bionic neuromorphic computing.

To tackle the current crisis of Moore's law, a sophisticated strategy entails the development of multistable memristors, bionic artificial synapses, logic circuits and brain-inspired neuromorphic computing. In comparison with conventional electronic systems, iontronic memristors offer greater potential for the manifestation of artificial intelligence and brain-machine interaction. Organic iontronic memristive materials (OIMs), which possess an organic backbone and exhibit stoichiometric ionic states, have emerged as pivotal contenders for the realization of high-performance bionic iontronic memristors. In this review, a comprehensive analysis of the progress and prospects of OIMs is presented, encompassing their inherent advantages, diverse types, synthesis methodologies, and wide-ranging applications in memristive devices. Predictably, the field of OIMs, as a rapidly developing research subject, presents an exciting opportunity for the development of highly efficient neuro-iontronic systems in areas such as in-sensor computing devices, artificial synapses, and human perception.

Solution-Processed Robust Multifunctional Memristor of 2D Layered Material Thin Film.

Memristors have gained significant attention recently due to their unique ability to exhibit functionalities for brain-inspired neuromorphic computing. Here, we demonstrate a high-performance multifunctional memristor using a thin film of liquid-phase exfoliated (LPE) 2D MoS2 pinched between two electrodes. Nanoscale inspection of a solution-processed MoS2 thin film using scanning electron and scanning probe microscopies revealed the high-quality and defect-free nature. Systematic current-voltage (I-V) characterizations depict a facile, nonvolatile resistive switching behavior of our 2D MoS2 thin film device with a current On/Off ratio of 103 and energy cost of only a few picojoules. Excellent performance metrics, including at least 103 cycle endurance, 104 s retention, and switching speed down to a few nanoseconds, reflect robust high-performance data storage capability. Charge carriers trapping and detrapping at the sulfur vacancy defect sites in MoS2 nanosheets mainly display the resistive switching property, supported by the impedance analysis and theoretical fitting results. Multifunctionality is leveraged through implementing two-input logic gate operations, edge computation, and crucial adaptive learning via a Pavlov's dogs experiment. Overall, our solution-processed MoS2 memristor has the potential for tremendous future opportunities in integrated circuits and different computing paradigms, including energy-efficient neuromorphic computing hardware in artificial intelligence.

Research on the Impact of Data Density on Memristor Crossbar Architectures in Neuromorphic Pattern Recognition.

Binary memristor crossbars have great potential for use in brain-inspired neuromorphic computing. The complementary crossbar array has been proposed to perform the Exclusive-NOR function for neuromorphic pattern recognition. The single crossbar obtained by shortening the Exclusive-NOR function has more advantages in terms of power consumption, area occupancy, and fault tolerance. In this paper, we present the impact of data density on the single memristor crossbar architecture for neuromorphic image recognition. The impact of data density on the single memristor architecture is mathematically derived from the reduced formula of the Exclusive-NOR function, and then verified via circuit simulation. The complementary and single crossbar architectures are tested by using ten 32 × 32 images with different data densities of 0.25, 0.5, and 0.75. The simulation results showed that the data density of images has a negative effect on the single memristor crossbar architecture while not affecting the complementary memristor crossbar architecture. The maximum output column current produced by the single memristor crossbar array decreases as data density decreases while the complementary memristor crossbar array architecture provides stable maximum output column currents. When recognizing images with data density as low as 0.25, the maximum output column currents of the single memristor crossbar architecture is reduced four-fold compared with the maximum currents from the complementary memristor crossbar architecture. This reduction causes the Winner-take-all circuit to work incorrectly and will reduce the recognition rate of the single memristor crossbar architecture. These simulation results show that the single memristor crossbar architecture has more advantages compared with the complementary crossbar architecture when the images do have not many different densities, and none of the images have very low densities. This work also indicates that the single crossbar architecture must be improved by adding a constant term to deal with images that have low data densities. These are valuable case studies for archiving the advantages of single memristor crossbar architecture in neuromorphic computing applications.

Two-dimensional materials memory devices with floating metal gate for neuromorphic applications

Emerging technologies such as neuromorphic computing and nonvolatile memories based on floating gate field-effect transistors (FETs) hold promise for addressing a wide range of artificial intelligence tasks. For example, neuromorphic computing seeks to emulate the human brain's functionality and employs a device that mimics the role of a synapse in the brain. However, achieving a high current ON/OFF ratio for the program and erase states of nonvolatile memory and neuromorphic computing device with a metal gate is necessary. This study demonstrates a multi-functional device based on heterostructures of transition metal dichalcogenides (TMDCs) with a metal floating gate. Five different channel materials (SnS2, WSe2, MoS2, WS2, and MoTe2) were employed, and hexagonal boron nitride (h-BN) was used as a tunneling layer. The study found that n-type SnS2 exhibits high endurance (15,000 cycles), good retention (2.4 × 105 s), and the highest current ON/OFF ratio (∼2.58 × 108) among the materials for the program and erase states. Moreover, the SnS2 device exhibits synaptic behavior and offers highly stable operation at room temperature. Furthermore, the device shows high linearity in both potentiation and depression, with good retention time and repeatable results with low cycle-to-cycle variations. Additionally, the study used an artificial neural network (ANN) for MNIST simulation of image recognition and achieved the highest accuracy of ∼92 % based on the SnS2 synaptic device experimental results. These findings pave the way for developing nonvolatile memory devices and their applications in brain-inspired neuromorphic computing and artificial intelligence systems.

Iontronic Nanopore Model for Artificial Neurons: The Requisites of Spiking.

Brain-inspired neuromorphic computing is currently being investigated for effective artificial intelligence (AI) systems. The development of artificial neurons and synapses is imperative to creating efficient computational biomimetic networks. Here we propose the minimal configuration of an effective iontronic spiking neuron based on a conical nanofluidic pore ionic diode. The conductance is composed of a Boltzmann open channel probability and a blocking inactivation function, forming the structure of a memristor. The presence of a negative resistance and the combination of activation-deactivation dynamics cause a Hopf bifurcation. Using the characteristic frequencies of small perturbation impedance spectroscopy, we discuss the conditions of spiking, in which the system enters a limit cycle oscillation. We arrive at the conclusion that an excitable neuron-like system can be made with a single active channel instead of the more complex combination of multiple channels that occurs in the Hodgkin-Huxley neuron model.

Biodegradable Oxide Neuromorphic Transistors for Neuromorphic Computing and Anxiety Disorder Emulation.

Brain-inspired neuromorphic computing and portable intelligent electronic products have received increasing attention. In the present work, nanocellulose-gated indium tin oxide neuromorphic transistors are fabricated. The device exhibits good electrical performance. Short-term synaptic plasticities were mimicked, including excitatory postsynaptic current, paired-pulse facilitation, and dynamic high-pass synaptic filtering. Interestingly, an effective linear synaptic weight updating strategy was adopted, resulting in an excellent recognition accuracy of ∼92.93% for the Modified National Institute of Standard and Technology database adopting a two-layer multilayer perceptron neural network. Moreover, with unique interfacial protonic coupling, anxiety disorder behavior was conceptually emulated, exhibiting "neurosensitization", "primary and secondary fear", and "fear-adrenaline secretion-exacerbated fear". Finally, the neuromorphic transistors could be dissolved in water, demonstrating potential in "green" electronics. These findings indicate that the proposed oxide neuromorphic transistors would have potential as implantable chips for nerve health diagnosis, neural prostheses, and brain-machine interfaces.

Light-stimulated low-power artificial synapse based on a single GaN nanowire for neuromorphic computing

The fast development of the brain-inspired neuromorphic computing system has ignited an urgent demand for artificial synapses with low power consumption. In this work, it is the first time a light-stimulated low-power synaptic device based on a single GaN nanowire has been demonstrated successfully. In such an artificial synaptic device, the incident light, the electrodes, and the light-generated carriers play the roles of action potential, presynaptic/postsynaptic membrane, and neurotransmitter in a biological synapse, respectively. Compared to those of other synaptic devices based on GaN materials, the energy consumption of the single-GaN-nanowire synaptic device can be reduced by more than 92%, reaching only 2.72×10−12 J. It is proposed that the oxygen element can contribute to the synaptic characteristics by taking the place of the nitrogen site. Moreover, it is found that the dynamic “learning−forgetting” performance of the artificial synapse can resemble the behavior of the human brain, where less time is required to relearn the missing information previously memorized and the memories can be strengthened after relearning. Based on the experimental conductance for long-term potentiation (LTP) and long-term depression (LTD), the simulated network can achieve a high recognition rate up to 90% after only three training epochs. Such few training times can reduce the energy consumption in the supervised learning processes substantially. Therefore, this work paves an effective way for developing single-nanowire-based synapses in the fields of artificial intelligence systems and neuromorphic computing technology requiring low-power consumption.

Leakage function in magnetic domain wall based artificial neuron using stray field

Recently, brain-inspired neuromorphic computing (NC) has been gaining traction as it is expected to be more power efficient and a more suitable platform for artificial intelligence. Artificial neurons and synapses are the main components of the NC architecture, and there have been many studies on artificial synapses. Experimental studies on artificial neurons that should exhibit the leaky integrate-and-fire properties are lacking due to the challenges in fabricating such a device. In this work, we have fabricated domain wall based devices consisting of (Co/Pt)n free and hard layers without interlayer exchange coupling, whereby the stray field from the hard layer triggers the automatic leakage function in the free layer. In addition, devices of smaller width were able to fully reset, showing the potential to scale down to smaller sizes. This experimental proof of concept provided evidence that the proposed neuron design has potential applications in NC. Further studies were performed via micromagnetic simulations to understand the role of the width of the device, thickness, and saturation magnetization of the hard layer.

A proposal for leaky integrate-and-fire neurons by domain walls in antiferromagnetic insulators

Brain-inspired neuromorphic computing is a promising path towards next generation analogue computers that are fundamentally different compared to the conventional von Neumann architecture. One model for neuromorphic computing that can mimic the human brain behavior are spiking neural networks (SNNs), of which one of the most successful is the leaky integrate-and-fire (LIF) model. Since conventional complementary metal-oxide-semiconductor (CMOS) devices are not meant for modelling neural networks and are energy inefficient in network applications, recently the focus shifted towards spintronic-based neural networks. In this work, using the advantage of antiferromagnetic insulators, we propose a non-volatile magnonic neuron that could be the building block of a LIF spiking neuronal network. In our proposal, an antiferromagnetic domain wall in the presence of a magnetic anisotropy gradient mimics a biological neuron with leaky, integrating, and firing properties. This single neuron is controlled by polarized antiferromagnetic magnons, activated by either a magnetic field pulse or a spin transfer torque mechanism, and has properties similar to biological neurons, namely latency, refraction, bursting and inhibition. We argue that this proposed single neuron, based on antiferromagnetic domain walls, is faster and has more functionalities compared to previously proposed neurons based on ferromagnetic systems.

Nanowire-based synaptic devices for neuromorphic computing

The traditional von Neumann structure computers cannot meet the demands of high-speed big data processing; therefore, neuromorphic computing has received a lot of interest in recent years. Brain-inspired neuromorphic computing has the advantages of low power consumption, high speed and high accuracy. In human brains, the data transmission and processing are realized through synapses. Artificial synaptic devices can be adopted to mimic the biological synaptic functionalities. Nanowire (NW) is an important building block for nanoelectronics and optoelectronics, and many efforts have been made to promote the application of NW-based synaptic devices for neuromorphic computing. Here, we will introduce the current progress of NW-based synaptic memristors and synaptic transistors. The applications of NW-based synaptic devices for neuromorphic computing will be discussed. The challenges faced by NW-based synaptic devices will be proposed. We hope this perspective will be beneficial for the application of NW-based synaptic devices in neuromorphic systems.

Improvement of polarization switching in ferroelectric transistor by interface trap reduction for brain-inspired artificial synapses

Numerous devices have been studied to accomplish a brain-inspired neuromorphic computing system; however, their characteristics require improvement to enhance operation accuracy of neuromorphic systems. Here, we propose a ferroelectric field-effect transistor based on Hf0.5Zr0.5O2 for a high-performance artificial synaptic device by reducing interface trap density. Using polarization in the Hf0.5Zr0.5O2 ferroelectric layer, threshold voltage is precisely controlled to achieve synaptic characteristics. However, synaptic characteristics are degraded because interface traps cause threshold voltage shift in the direction opposite to the threshold voltage shift by polarization. Therefore, oxygen plasma treatment was performed to minimize degradation by interface traps. Thus, outstanding synaptic characteristics including the linearity of the weight update, the max/min weight ratio, and the number of weights were observed. Furthermore, high repeatability and long-term plasticity was observed at low operating power. Therefore, our study demonstrated a promising method for implementing a high-accuracy, low-power neuromorphic system with progressed artificial synaptic devices.

Uniform resistive switching and highly stable synaptic characteristics of HfOx sandwiched TaOx-based memristor for neuromorphic system

Emerging nanoscale devices, including memristors, have been extensively studied to implement biological synaptic functions such as learning and plasticity, which are the fundamental building blocks of brain-inspired neuromorphic computing. The memristor with analog switching ability exhibits linear tuning of weight during neural network training is a desirable synaptic device behavior. The importance of inserting a HfOx sandwiched layer in a TaOx/HfOx/TaOx memristor is to achieve analog set/reset operation along with improved spatial/temporal switching uniformity. The optimal resistive switching (RS) behavior can be attributed to asymmetric oxygen vacancy distribution in the stacked structure leading to the formation of an hourglass-shaped conductive filament. Furthermore, confining filament formation/rapture in the narrow fixed region displays superior endurance characteristics (dc cycles >2000 and ac cycles > 106) and uniform resistive switching with the set (reset) voltage variation constrained to 1.8 (2.9) %. Paired-pulse facilitation (PPF), a form of short-term synaptic plasticity is stimulated to replicate bio-synapse behavior. The stable long-term potentiation (LTP) and depression (LTD) behavior for more than 1000 epochs (>105 pulses) with excellent symmetry and linearity is achieved with 50 ns voltage pulse stimulation. The pattern recognition accuracy of 93% was achieved for an image of size 10 × 10 pixels after 13 epochs by deploying 100 synapses in Hopfield Neural Network (HNN) simulation. This comprehensive study demonstrates that the HfOx-inserted TaOx memristor has tremendous potential for application in future neuromorphic computing.

Reconfigurable 2D-ferroelectric platform for neuromorphic computing

To meet the requirement of data-intensive computing in the data-explosive era, brain-inspired neuromorphic computing have been widely investigated for the last decade. However, incompatible preparation processes severely hinder the cointegration of synaptic and neuronal devices in a single chip, which limited the energy-efficiency and scalability. Therefore, developing a reconfigurable device including synaptic and neuronal functions in a single chip with same homotypic materials and structures is highly desired. Based on the room-temperature out-of-plane and in-plane intercorrelated polarization effect of 2D α-In2Se3, we designed a reconfigurable hardware platform, which can switch from continuously modulated conductance for emulating synapse to spiking behavior for mimicking neuron. More crucially, we demonstrate the application of such proof-of-concept reconfigurable 2D ferroelectric devices on a spiking neural network with an accuracy of 95.8% and self-adaptive grow-when required network with an accuracy of 85% by dynamically shrinking its nodes by 72%, which exhibits more powerful learning ability and efficiency than the static neural network.

Synaptic Characteristic of Hafnia-Based Ferroelectric Tunnel Junction Device for Neuromorphic Computing Application.

Owing to the 4th Industrial Revolution, the amount of unstructured data, such as voice and video data, is rapidly increasing. Brain-inspired neuromorphic computing is a new computing method that can efficiently and parallelly process rapidly increasing data. Among artificial neural networks that mimic the structure of the brain, the spiking neural network (SNN) is a network that imitates the information-processing method of biological neural networks. Recently, memristors have attracted attention as synaptic devices for neuromorphic computing systems. Among them, the ferroelectric doped-HfO2-based ferroelectric tunnel junction (FTJ) is considered as a strong candidate for synaptic devices due to its advantages, such as complementary metal-oxide-semiconductor device/process compatibility, a simple two-terminal structure, and low power consumption. However, research on the spiking operations of FTJ devices for SNN applications is lacking. In this study, the implementation of long-term depression and potentiation as the spike timing-dependent plasticity (STDP) rule in the FTJ device was successful. Based on the measured data, a CrossSim simulator was used to simulate the classification of handwriting images. With a high accuracy of 95.79% for the Mixed National Institute of Standards and Technology (MNIST) dataset, the simulation results demonstrate that our device is capable of differentiating between handwritten images. This suggests that our FTJ device can be used as a synaptic device for implementing an SNN.

Ultralow-Power Vertical Transistors for Multilevel Decoding Modes.

Organic field-effect transistors with parallel transmission and learning functions are of interest in the development of brain-inspired neuromorphic computing. However, the poor performance and high power consumption are the two main issues limiting their practical applications. Herein, an ultralow-power vertical transistor is demonstrated based on transition-metal carbides/nitrides (MXene) and organic single crystal. The transistor exhibits a high JON of 16.6mA cm-2 and a high JON /JOFF ratio of 9.12 × 105 under an ultralow working voltage of -1mV. Furthermore, it can successfully simulate the functions of biological synapse under electrical modulation along with consuming only 8.7 aJ of power per spike. It also permits multilevel information decoding modes with a significant gap between the readable time of professionals and nonprofessionals, producing a high signal-to-noise ratio up to 114.15dB. This work encourages the use of vertical transistors and organic single crystal in decoding information and advances the development of low-power neuromorphic systems.

Optical synaptic devices with ultra-low power consumption for neuromorphic computing

Brain-inspired neuromorphic computing, featured by parallel computing, is considered as one of the most energy-efficient and time-saving architectures for massive data computing. However, photonic synapse, one of the key components, is still suffering high power consumption, potentially limiting its applications in artificial neural system. In this study, we present a BP/CdS heterostructure-based artificial photonic synapse with ultra-low power consumption. The device shows remarkable negative light response with maximum responsivity up to 4.1 × 108 A W−1 at VD = 0.5 V and light power intensity of 0.16 μW cm−2 (1.78 × 108 A W−1 on average), which further enables artificial synaptic applications with average power consumption as low as 4.78 fJ for each training process, representing the lowest among the reported results. Finally, a fully-connected optoelectronic neural network (FONN) is simulated with maximum image recognition accuracy up to 94.1%. This study provides new concept towards the designing of energy-efficient artificial photonic synapse and shows great potential in high-performance neuromorphic vision systems.

Nanostructured CuAlO2@ZnO optoelectronic device for artificial synaptic applications

Developing a neuromorphic system beyond conventional von Neumann’s architecture is necessary for future artificial intelligence applications and data-intensive computation. Optoelectronic synapses, like biological synapses, are extremely comparable in structure and operating mechanism and can learn directly from the light stimulus, making them a suitable technology for brain-inspired neuromorphic computing. Here, by constructing a core–shell structure of CuAlO2@ZnO oxide heterojunction, we have successfully fabricated the optoelectronic synaptic devices structured as ITO/CuAlO2@ZnO/ITO-glass that can be effectively motivated by light signal at visual light range. Moreover, there is an ultra-high gain of photocurrent generated in the device, resulting in an excitatory postsynaptic current (EPSC) of µA level even at a lower light intensity. In addition, the device could mimic multiple synaptic plasticities at the micro-level such as EPSC, paired-pulse facilitation (PPF), short-term plasticity (STP), spike-number-dependent plasticity (SNDP), and “advanced behaviors” of human-like innate learning and memory functions. The results presented here may serve as a new benchmark for the use of optoelectronic synaptic devices in artificial intelligence.

Excitatory postsynaptic current model for synaptic thin-film transistors

Synaptic devices that mimic biological neurons have attracted much attention for brain-inspired neuromorphic computing. Especially, synaptic thin-film transistors (TFTs) have emerged with simultaneous signal processing and information storage advantages. However, the analysis of excitatory postsynaptic current (EPSC) relies on an empirical model such as a serial RC circuit, which limits a systematic and in-depth study of synaptic devices in terms of material and electrical properties. Herein, the single-pulse-driven synaptic EPSC (SPSE) model, including capacitive effect and information of the synaptic window, is analytically proposed. The SPSE model can simulate EPSC of synaptic devices at given TFT-operating conditions. EPSC with the SPSE model can be characterized with quantified parameters for the capacitive effects and the synaptic windows, which also depend on the electrical condition applied to TFTs. Various kinds of synaptic-TFTs with different gate insulators (e.g., SiO2 and ion-gel) are used to confirm the performance of the SPSE model. For example, the SPSE model can capture the long-term robustness of ion-gel-based TFTs with specific quantified parameters. In addition, the SPSE model enables the estimation of energy consumption, which can potentially be leveraged to compare the energy cost of EPSC fairly. The SPSE model can provide a guideline to understand the physical properties of synaptic TFTs.