Advances in memristor based artificial neuron fabrication-materials,models,and applications

Spiking neural network(SNN),widely known as the third-generation neural network,has been frequently investigated due to its excellent spatiotemporal information processing capability,high biological plausibility,and low energy consumption characteristics.Analogous to the working mechanism of human brain,the SNN system transmits information through the spiking action of neurons.Therefore,artificial neurons are critical building blocks for constructing SNN in hardware.Memristors are drawing growing attention due to low consumption,high speed,and nonlinearity characteristics,which are recently introduced to mimic the functions of biological neurons.Researchers have proposed multifarious memristive materials including organic materials,inorganic materials,or even two-dimensional materials.Taking advantage of the unique electrical behavior of these materials,several neuron models are successfully implemented,such as Hodgkin–Huxley model,leaky integrate-and-fire model and integrate-and-fire model.In this review,the recent reports of artificial neurons based on memristive devices are discussed.In addition,we highlight the models and applications through combining artificial neuronal devices with sensors or other electronic devices.Finally,the future challenges and outlooks of memristor-based artificial neurons are discussed,and the development of hardware implementation of brain-like intelligence system based on SNN is also prospected.

High-performance artificial neurons based on Ag/MXene/GST/Pt threshold switching memristors

Threshold switching(TS) memristors can be used as artificial neurons in neuromorphic systems due to their continuous conductance modulation, scalable and energy-efficient properties. In this paper, we propose a low power artificial neuron based on the Ag/MXene/GST/Pt device with excellent TS characteristics, including a low set voltage(0.38 V)and current(200 nA), an extremely steep slope(< 0.1 m V/dec), and a relatively large off/on ratio(> 10^(3)). Besides, the characteristics of integrate and fire neurons that are indispensable for spiking neural networks have been experimentally demonstrated. Finally, its memristive mechanism is interpreted through the first-principles calculation depending on the electrochemical metallization effect.

FPGA implementation of bit-stream neuron and perceptron based on sigma delta modulation

To solve the excessive huge scale problem of the traditional multi-bit digital artificial neural network（ANN） hardware implementation methods,a bit-stream ANN hardware implementation method based on sigma delta（ΣΔ） modulation is presented.The bit-stream adder,multiplier,threshold function unit and fully digital ΣΔ modulator are implemented in a field programmable gate array（FPGA）,and these bit-stream arithmetical units are employed to build the bit-stream artificial neuron.The function of the bit-stream artificial neuron is verified through the realization of the logic function and a linear classifier.The bit-stream perceptron based on the bit-stream artificial neuron with the pre-processed structure is proved to have the ability of nonlinear classification.The FPGA resource utilization of the bit-stream artificial neuron shows that the bit-stream ANN hardware implementation method can significantly reduce the demand of the ANN hardware resources.

Improving Dendritic Neuron Model With Dynamic Scale-Free Network-Based Differential Evolution

Some recent research reports that a dendritic neuron model(DNM)can achieve better performance than traditional artificial neuron networks(ANNs)on classification,prediction,and other problems when its parameters are well-tuned by a learning algorithm.However,the back-propagation algorithm(BP),as a mostly used learning algorithm,intrinsically suffers from defects of slow convergence and easily dropping into local minima.Therefore,more and more research adopts non-BP learning algorithms to train ANNs.In this paper,a dynamic scale-free network-based differential evolution(DSNDE)is developed by considering the demands of convergent speed and the ability to jump out of local minima.The performance of a DSNDE trained DNM is tested on 14 benchmark datasets and a photovoltaic power forecasting problem.Nine meta-heuristic algorithms are applied into comparison,including the champion of the 2017 IEEE Congress on Evolutionary Computation(CEC2017)benchmark competition effective butterfly optimizer with covariance matrix adapted retreat phase(EBOwithCMAR).The experimental results reveal that DSNDE achieves better performance than its peers.

Prognostic role of artificial intelligence among patients with hepatocellular cancer:A systematic review

BACKGROUND Prediction of survival after the treatment of hepatocellular carcinoma(HCC)has been widely investigated,yet remains inadequate.The application of artificial intelligence(AI)is emerging as a valid adjunct to traditional statistics due to the ability to process vast amounts of data and find hidden interconnections between variables.AI and deep learning are increasingly employed in several topics of liver cancer research,including diagnosis,pathology,and prognosis.AIM To assess the role of AI in the prediction of survival following HCC treatment.METHODS A web-based literature search was performed according to the Preferred Reporting Items for Systemic Reviews and Meta-Analysis guidelines using the keywords“artificial intelligence”,“deep learning”and“hepatocellular carcinoma”(and synonyms).The specific research question was formulated following the patient(patients with HCC),intervention(evaluation of HCC treatment using AI),comparison(evaluation without using AI),and outcome(patient death and/or tumor recurrence)structure.English language articles were retrieved,screened,and reviewed by the authors.The quality of the papers was assessed using the Risk of Bias In Non-randomized Studies of Interventions tool.Data were extracted and collected in a database.RESULTS Among the 598 articles screened,nine papers met the inclusion criteria,six of which had low-risk rates of bias.Eight articles were published in the last decade;all came from eastern countries.Patient sample size was extremely heterogenous(n=11-22926).AI methodologies employed included artificial neural networks(ANN)in six studies,as well as support vector machine,artificial plant optimization,and peritumoral radiomics in the remaining three studies.All the studies testing the role of ANN compared the performance of ANN with traditional statistics.Training cohorts were used to train the neural networks that were then applied to validation cohorts.In all cases,the AI models demonstrated superior predictive performance compared with traditional statistics with significantly improved areas under the curve.CONCLUSION AI applied to survival prediction after HCC treatment provided enhanced accuracy compared with conventional linear systems of analysis.Improved transferability and reproducibility will facilitate the widespread use of AI methodologies.

Voltage-controllable magnetic skyrmion dynamics for spiking neuron device applications

Voltage-controlled magnetic skyrmions have attracted special attention because they satisfy the requirements for well-controlled high-efficiency and energy saving for future skyrmion-based neuron device applications.In this work,we propose a compact leaky-integrate-fire(LIF)spiking neuron device by using the voltage-driven skyrmion dynamics in a multiferroic nanodisk structure.The skyrmion dynamics is controlled by well tailoring voltage-induced piezostrains,where the skyrmion radius can be effectively modulated by applying the piezostrain pulses.Like the biological neuron,the proposed skyrmionic neuron will accumulate a membrane potential as skyrmion radius is varied by inputting the continuous piezostrain spikes,and the skyrmion radius will return to the initial state in the absence of piezostrain.Therefore,this skyrmion radius-based membrane potential will reach a definite threshold value by the strain stimuli and then reset by removing the stimuli.Such the LIF neuronal functionality and the behaviors of the proposed skyrmionic neuron device are elucidated through the micromagnetic simulation studies.Our results may benefit the utilization of skyrmionic neuron for constructing the future energy-efficient and voltage-tunable spiking neural networks.

Bit-stream linear artificial neural networks based on Sigma-delta modulation

To solve the excessive huge scale problem of the traditional multi-bit digital artificial neural networks （ANN） hardware implementation methods, a bit-stream ANN construction method based on direct sigma-delta （Z-A） signal processing is presented. The bit-stream adder, multiplier and fully digital X-A modulator used in the bit-stream linear ANN are implemented in a field programmable gate array （FPGA）. A bit-stream linear ANN based on these bit-stream modules is presented and implemented. To verify the function and performance of the bit-stream linear ANN, the bit-stream adaptive predictor and the bit-stream adaptive noise cancellation system are presented. The predicted result of the bit-stream adaptive predictor is very close to the desired signal. Also, the bit-stream adaptive noise cancellation system removes the electric power noise effectively.

Applications of artificial intelligence technology to wastewater treatment fields in China

Current applications of artificial intelligence technology to wastewater treatment in China are summarized. Wastewater treatment plants use expert system mainly in the operation decision-making and fault diagnosis of system operation, use artificial neuron network for system modeling, water quality forecast and soft measure, and use fuzzy control technology for the intelligence control of wastewater treatment process. Finally, the main problems in applying artificial intelligence technology to wastewater treatment in China are analyzed.

Swarm Optimization and Machine Learning for Android Malware Detection

Malware Security Intelligence constitutes the analysis of applications and their associated metadata for possible security threats.Application Programming Interfaces(API)calls contain valuable information that can help with malware identification.The malware analysis with reduced feature space helps for the efficient identification of malware.The goal of this research is to find the most informative features of API calls to improve the android malware detection accuracy.Three swarm optimization methods,viz.,Ant Lion Optimization(ALO),Cuckoo Search Optimization(CSO),and Firefly Optimization(FO)are applied to API calls using auto-encoders for identification of most influential features.The nature-inspired wrapperbased algorithms are evaluated using well-known Machine Learning(ML)classifiers such as Linear Regression(LR),Decision Tree(DT),Random Forest(RF),K-Nearest Neighbor(KNN)&SupportVector Machine(SVM).A hybrid Artificial Neuronal Classifier(ANC)is proposed for improving the classification of android malware.The experimental results yielded an accuracy of 98.87%with just seven features out of hundred API call features,i.e.,a massive 93%of data optimization.

Illumination of parameter contributions on uneven break： phenomenon in underground stoping mines

One of the most serious conundrum facing the stope production in underground metalliferous mining is uneven break （UB： unplanned dilution and ore-loss）. Although the UB has a huge economic fallout to the entire mining process, it is practically unavoidable due to the complex causing mechanism. In this study, the contribution of ten major UB causative parameters ha,; been scrutinised based on a published UB predicting artificial neuron network （ANN） model to put UB under the engineering management. Two typical ANN sensitivity analysis methods, i.e., connection weight algorithm （CWA） and profile method （PM） have been applied. As a result of CWA and PM applications, adjusted Qrate （AQ） revealed as the most influential parameter to UB with contribution of 22,40% in CWA and 20,48% in PM respectively. The findings of this study can be used as an important reference in stope design, production, and reconciliation stages on underground stoping mine.

Data Fusion Method for Manufacturing Measurement

A data fusion method of online multisensors is prop os ed in this paper based on artificial neuron. First, the dynamic data fusion mode l on artificial neuron is built. Then the calibration of data fusion is discusse d with self-adaptive weighing technique. Finally performance of the method is d emonstrated by an online vibration measurement case. The results show that the f used data are more stable, sensitive, accurate, reliable than that of single sen sor data.

Spin torque oscillator based on magnetic tunnel junction with MgO cap layer for radio-frequency-oriented neuromorphic computing

Recently,it has been proposed that spin torque oscillators(STOs)and spin torque diodes could be used as artificial neurons and synapses to directly process microwave signals,which could lower latency and power consumption greatly.However,one critical challenge is to make the microwave emission frequency of the STO stay constant with a varying input current.In this work,we study the microwave emission characteristics of STOs based on magnetic tunnel junction with MgO cap layer.By applying a small magnetic field,we realize the invariability of the microwave emission frequency of the STO,making it qualified to act as artificial neuron.Furthermore,we have simulated an artificial neural network using STO neuron to recognize the handwritten digits in the Mixed National Institute of Standards and Technology database,and obtained a high accuracy of 92.28%.Our work paves the way for the development of radio-frequency-oriented neuromorphic computing systems.

Application of Neural Networks for Monitoring Mechanical Defects of Rotating Machines

Good monitoring of the deterioration in rotating machinery can result in reduced maintenance costs by minimizing the loss of production due to the number of machine breakdown and decreasing in the number of spare parts. In the present paper, a prognostic method based on recurrent neural networks is applied to forecast the rate of machine deterioration. Promising results have been obtained through the application of this method to the prediction of vibration based fault trends of an auxiliary gearbox of a power generation plant. This method evaluates also the seriousness of damage caused by faults.

Versatile SrFeO_(x) for memristive neurons and synapses

Spiking neural network(SNN)consisting of memristor-based artificial neurons and synapses has emerged as a compact and energy-efficient hardware solution for spatiotemporal information processing.However,it is challenging to develop memristive neurons and synapses based on the same material system because the required resistive switching(RS)characteristics are different.Here,it is shown that SrFeO_(x)(SFO),an intriguing material system exhibiting topotactic phase transformation between insulating brownmillerite(BM)SrFeO_(2).5 phase and conductive perovskite(PV)SrFeO_(3) phase,can be engineered into both neuronal and synaptic devices.Using a BM-SFO single layer as the RS medium,the Au/BM-SFO/SrRuO_(3)(SRO)memristor exhibits nonvolatile RS behavior originating from the formation/rupture of PV-SFO filaments in the BM-SFO matrix.By contrast,using a PV-SFO(matrix)/BM-SFO(interfacial layer)bilayer as the RS medium,the Au/PV-SFO/BM-SFO/SRO memristor exhibits volatile RS behavior originating from the interfacial BM-PV phase transformation.Synaptic and neuronal characteristics are further demonstrated in the Au/BM-SFO/SRO and Au/PV-SFO/BM-SFO/SRO memristors,respectively.Using the SFO-based synapses and neurons,fully memristive SNNs are constructed by simulation,which show good performance on unsupervised image recognition.Our study suggests that SFO is a versatile material platform on which both neuronal and synaptic devices can be developed for constructing fully memristive SNNs.

Memory-centric neuromorphic computing for unstructured data processing

The unstructured data such as visual information,natural language,and human behaviors opens up a wide array of opportunities in the field of artificial intelligence(Al).The memory-centric neuromorphic computing(MNC)has been proposed for the efficient processing of unstructured data,bypassing the von Neumann bottleneck of current computing architecture.The development of MNC would provide massively parallel processing of unstructured data,realizing the cognitive Al in edge and wearable systems.In this review,recent advances in memory-centric neuromorphic devices are discussed in terms of emerging nonvolatile memories,volatile switches,synaptic plasticity,neuronal models,and memristive neural network.

Soft Sensors for Property‑Controlled Multi‑Stage Press Hardening of 22MnB5

In multi-stage press hardening,the product properties are determined by the thermo-mechanical history during the sequence of heat treatment and forming steps.To measure these properties and finally to control them by feedback,two soft sensors are developed in this work.The press hardening of 22MnB5 sheet material in a progressive die,where the material is first rapidly austenitized,then pre-cooled,stretch-formed,and finally die bent,serves as the framework for the development of these sensors.To provide feedback on the temporal and spatial temperature distribution,a soft sensor based on a model derived from the Dynamic mode decomposition(DMD)is presented.The model is extended to a parametric DMD and combined with a Kalman filter to estimate the temperature(-distribution)as a function of all process-relevant control vari-ables.The soft sensor can estimate the temperature distribution based on local thermocouple measurements with an error of less than 10°C during the process-relevant time steps.For the online prediction of the final microstructure,an artificial neural network(ANN)-based microstructure soft sensor is developed.As part of this,a transferable framework for deriving input parameters for the ANN based on the process route in multi-stage press hardening is presented,along with a method for developing a training database using a 1-element model implemented with LS-Dyna and utilizing the material model Mat248(PHS_BMW).The developed ANN-based microstructure soft sensor can predict the final microstructure for specific regions of the formed and hardened sheet in a time span of far less than 1 s with a maximum deviation of a phase fraction of 1.8%to a reference simulation.

Triboelectric nanogenerator for neuromorphic electronics

Building the brain-inspired neural network computing system based neuromorphic electronics is an effective approach to break the von Neumann bottleneck on the hardware level and realize the information processing with high efficiency and low energy consumption in this big data explosion age.Triboelectric nanogenerator(TENG)has two functions of sensing and energy conversion,which promote the application as sensor and/or power supply in self-powered neuromorphic electronics for data storage and biological synapse/neuron behaviors mimicking.This article highlights the relevant works of TENGs for memory devices,artificial synapses and artificial neurons,performs a systematic comparison,and puts forward the future research possibilities and challenges,with the hope of attracting more researchers into this field and promoting the development of TENG based neuromorphic electronics.