Mixed National Institute Of Standards And Technology Research Articles

Neuromorphic Computing of Optoelectronic Artificial BFCO/AZO Heterostructure Memristors Synapses.

Compared with purely electrical neuromorphic devices, those stimulated by optical signals have gained increasing attention due to their realistic sensory simulation. In this work, an optoelectronic neuromorphic device based on a photoelectric memristor with a Bi2FeCrO6/Al-doped ZnO (BFCO/AZO) heterostructure is fabricated that can respond to both electrical and optical signals and successfully simulate a variety of synaptic behaviors, such as STP, LTP, and PPF. In addition, the photomemory mechanism was identified by analyzing the energy band structures of AZO and BFCO. A convolutional neural network (CNN) architecture for pattern classification at the Mixed National Institute of Standards and Technology (MNIST) was used and improved the recognition accuracy of the MNIST and Fashion-MNIST datasets to 95.21% and 74.19%, respectively, by implementing an improved stochastic adaptive algorithm. These results provide a feasible approach for future implementation of optoelectronic synapses.

Coupled Ferroelectric-Photonic Memory in a Retinomorphic Hardware for In-Sensor Computing.

The development of all-in-one devices for artificial visual systems offers an attractive solution in terms of energy efficiency and real-time processing speed. In recent years, the proliferation of smart sensors in the growth of Internet-of-Things (IoT) has led to the increasing importance of in-sensor computing technology, which places computational power at the edge of the data-flow architecture. In this study, a prototype visual sensor inspired by the human retina is proposed, which integrates ferroelectricity and photosensitivity in two-dimensional (2D) α-In2Se3 material. This device mimics the functions of photoreceptors and amacrine cells in the retina, performing optical reception and memory computation functions through the use of electrical switching polarization in the channel. The gate-tunable linearity of excitatory and inhibitory functions in photon-induced short-term plasticity enables to encode and classify 12000 images in the Mixed National Institute of Standards and Technology (MNIST) dataset with remarkable accuracy, achieving ≈94%. Additionally, in-sensor convolution image processing through a network of phototransistors, with five convolutional kernels electrically pre-programmed into the transistors is demonstrated. The convoluted photocurrent matrices undergo straightforward arithmetic calculations to produce edge and feature-enhanced scenarios. The findings demonstrate the potential of ferroelectric α-In2Se3 for highly compact and efficient retinomorphic hardware implementation, regardless of ambipolar transport in the channel.

Spectrum Analysis for Fully Connected Neural Networks.

This article studies the meaning of parameters of fully connected neural networks with single hidden layer from the perspective of spectrum. Under the constraints of numerical range, the corresponding relationship between parameters and the spectrum of network function can be established by the Fourier series coefficients of the activation function, which is truncated and periodically extended. This work is substantiated on the Mixed National Institute of Standards and Technology (MNIST) handwritten dataset and two illustrative examples with certain spectra. The simulations complete the conversion between spectrum and parameters with high precision and give the significance of hidden nodes to the spectrum of network function. Some algorithms derived from these properties, such as the parameter initialization method using spectrum and the pruning method by sorting amplification weights, are also presented to introduce how spectrum analysis affects neural network decision-making. Thus, spectrum analysis has great potential in network interpretation.

Threshold-type memristor-based crossbar array design and its application in handwritten digit recognition

Neuromorphic computing simulates the operation of biological brain function for information processing and can potentially solve the bottleneck of the Von Neumann architecture. Inspired by the real characteristics of physical memristive devices, we propose a threshold-type nonlinear voltage-controlled memristor mathematical model which is used to design a novel memristor-based crossbar array. The presented crossbar array can simulate the synaptic weight in real number field rather than only positive number field. Theoretical analysis and simulation results of a 2×2 image inversion operation validate the feasibility of the proposed crossbar array and the necessary training and inference functions. Finally, the presented crossbar array is used to construct the neural network and then applied in the handwritten digit recognition. The Mixed National Institute of Standards and Technology (MNIST) database is adopted to train this neural network and it achieves a satisfactory accuracy.

Single-shot Fresnel incoherent correlation holography via deep learning based phase-shifting technology.

Fresnel incoherent correlation holography (FINCH) realizes non-scanning three-dimension (3D) images using spatial incoherent illumination, but it requires phase-shifting technology to remove the disturbance of the DC term and twin term that appears in the reconstruction field, thus increasing the complexity of the experiment and limits the real-time performance of FINCH. Here, we propose a single-shot Fresnel incoherent correlation holography via deep learning based phase-shifting (FINCH/DLPS) method to realize rapid and high-precision image reconstruction using only a collected interferogram. A phase-shifting network is designed to implement the phase-shifting operation of FINCH. The trained network can conveniently predict two interferograms with the phase shift of 2/3 π and 4/3 π from one input interferogram. Using the conventional three-step phase-shifting algorithm, we can conveniently remove the DC term and twin term of the FINCH reconstruction and obtain high-precision reconstruction through the back propagation algorithm. The Mixed National Institute of Standards and Technology (MNIST) dataset is used to verify the feasibility of the proposed method through experiments. In the test with the MNIST dataset, the reconstruction results demonstrate that in addition to high-precision reconstruction, the proposed FINCH/DLPS method also can effectively retain the 3D information by calibrating the back propagation distance in the case of reducing the complexity of the experiment, further indicating the feasibility and superiority of the proposed FINCH/DLPS method.

Experimental quantum end-to-end learning on a superconducting processor

Machine learning can be enhanced by a quantum computer via its inherent quantum parallelism. In the pursuit of quantum advantages for machine learning with noisy intermediate-scale quantum devices, it was proposed that the learning model can be designed in an end-to-end fashion, i.e., the quantum ansatz is parameterized by directly manipulable control pulses without circuit design and compilation. Such gate-free models are hardware friendly and can fully exploit limited quantum resources. Here, we report the experimental realization of quantum end-to-end machine learning on a superconducting processor. The trained model can achieve 98% recognition accuracy for two handwritten digits (via two qubits) and 89% for four digits (via three qubits) in the MNIST (Mixed National Institute of Standards and Technology) database. The experimental results exhibit the great potential of quantum end-to-end learning for resolving complex real-world tasks when more qubits are available.

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.

Synaptic and resistive switching behaviors in NiO/Cu2O heterojunction memristor for bioinspired neuromorphic computing

Artificial neural network-based computing prospectively overcomes the von Neumann bottleneck of conventional computers and significantly improves computational efficiency, which shows a wide range of application prospects. Here the NiO/Cu2O memristor is fabricated by magnetron sputtering, which enables functions that emulate biological synapses, such as short/long plasticity, paired-pulse facilitation, and spike timing-dependent plasticity, etc. Furthermore, a artificial neural network based on synaptic weight modulation was presented at the Mixed National Institute of Standards and Technology (MNIST) with recognition accuracy of up to 96.84 % on average, and the device proved able to simulate an array of trainable memristors for image information processing. The results demonstrate the potential of artificial synapses in artificial intelligence systems that incorporate neuromorphological computations and synaptic neural functions.

Flexible and Compatible Synaptic Transistor Based on Electrospun In2O3 Nanofibers

Emulating synaptic behavior using the three-terminal and ion-coupled transistors is considered as a promising strategy for the realization of neuromorphic devices. Meanwhile, one-dimensional (1-D) nanostructures show great potential for the fabrication of flexible electronic devices. In this work, 1-D indium oxide (In <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> O <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sub> ) nanofibers were fabricated by the electrospinning (ES), and the flexible and compatible synaptic transistor based on In <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> O <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sub> nanofibers was integrated on polyimide (PI) substrate. The synaptic behaviors, such as the short-term plasticity (STP) and long-term plasticity (LTP), are simulated by the synaptic transistor with chitosan solution as the gate dielectric. The fabricated synaptic transistor exhibits good biocompatibility and environmental compatibility, and the pattern recognition rate of the simulated device array by the Mixed National Institute of Standards and Technology (MNIST) database of the handwritten digits is confirmed to be as high as 92%.

Low Energy and Analog Memristor Enabled by Regulation of Ru ion Motion for High Precision Neuromorphic Computing

AbstractMobile species and matrix materials in ion motion‐mediated memristors predominantly determine the switching characteristics and device performance. As a result of exploring a new type of mobile species, a Ru ion‐mediated electrochemical metallization‐like memristor with an amorphous oxide matrix is recently suggested to achieve a low switching current, voltage, and good retention simultaneously. Although the ion migration of Ru in the oxide matrix is previously confirmed, no in‐depth study on how the crystallinity of the oxide matrix influences the Ru ion motion and switching characteristics has not been reported. Therefore, in this study, the crystallinity‐dependent resistive switching behavior of the Pt/HfO2/Ru structure device is investigated. With the crystallized HfO2 layer, the preferred Ru ion migration through the grain boundaries occurs owing to the enhanced ion mobility, resulting in a high switching current (≈100 µA) with continuous metallic Ru conducting filaments. The discontinuous conducting filaments with amorphous HfO2 exhibit a low switching current. In addition, highly linear and symmetric conductance modulation properties are achieved, and over 91.5% accuracy in the Mixed National Institute of Standards and Technology (MNIST) pattern recognition test is demonstrated.

Synaptic behaviors in flexible Au/WOx/Pt/mica memristor for neuromorphic computing system

Neuromorphic computing, composed of artificial synapses and neural network algorithms, is expected to replace the traditional von Neumann computer architecture to build the next-generation intelligent systems due to its more energy-efficient features. In this work, the flexible Au/WOx/Pt/Mica memristor with simple structure is fabricated by RF magnetron sputtering, and the highly adjustable resistance states, the function of biological synapses and neurons in different states, such as short/long-term plasticity, paired-pulse facilitation, and spike-time-dependent plasticity, were demonstrated in flexible WOx memristor. Furthermore, we established a convolutional neural networks (CNNs) architecture for the Mixed National Institute of Standards and Technology (MNIST) pattern categorization and demonstrated that the recognition performance is comparable to that of a software-based neural network. These results provide a feasible approach for the realization of flexible neuromorphic computing systems in the future.

A neural network architecture optimizer based on DARTS and generative adversarial learning

Neural network architecture search automatically configures a set of network architectures according to the targeted rules. Thus, it relieves the human-dependent effort and repetitive resources consumption for designing neural network architectures and makes the task of finding the optimum network architecture with better performance much more accessible. Network architecture search methods based on differentiable architecture search (DARTS), however, introduces parameter redundancy. To address this issue, this work presents a novel method for optimizing network architectures that combines DARTS with generative adversarial learning (GAL). We first find the module structures utilizing the DARTS algorithm. Afterwards, the retrieved modules are stacked to derive the initial neural network architecture. Next, the GAL is used to prune some branches of the initial neural network, thereby obtaining the final neural network architecture. The proposed DARTS-GAL method re-optimizes the network architecture searched by DARTS to simplify the network connection and reduce network parameters without compromising network performance. Experimental results on benchmark datasets, i.e., Mixed National Institute of Standards and Technology (MNIST), FashionMNIST, Canadian Institute for Advanced Research10 (CIFAR10), Canadian Institute for Advanced Research100 (CIAFR100), Cats vs Dogs, and voiceprint recognition datasets, indicate that the test accuracies of the DARTS-GAL are higher than those of the DARTS in the majority of the cases. In particular, the proposed solution exhibits an improvement in accuracy by 7.35% on CIFAR10 compared with DARTS, attaining the state-of-the-art result of 99.60%. Additionally, the number of network parameters derived by the DARTS-GAL is significantly lower than that by the DARTS method, with a pruning rate of 62.3% at the highest case.

Memristive Quantized Neural Networks: A Novel Approach to Accelerate Deep Learning On-Chip.

Existing deep neural networks (DNNs) are computationally expensive and memory intensive, which hinder their further deployment in novel nanoscale devices and applications with lower memory resources or strict latency requirements. In this paper, a novel approach to accelerate on-chip learning systems using memristive quantized neural networks (M-QNNs) is presented. A real problem of multilevel memristive synaptic weights due to device-to-device (D2D) and cycle-to-cycle (C2C) variations is considered. Different levels of Gaussian noise are added to the memristive model during each adjustment. Another method of using memristors with binary states to build M-QNNs is presented, which suffers from fewer D2D and C2C variations compared with using multilevel memristors. Furthermore, methods of solving the sneak path issues in the memristive crossbar arrays are proposed. The M-QNN approach is evaluated on two image classification datasets, that is, ten-digit number and handwritten images of mixed National Institute of Standards and Technology (MNIST). In addition, input images with different levels of zero-mean Gaussian noise are tested to verify the robustness of the proposed method. Another highlight of the proposed method is that it can significantly reduce computational time and memory during the process of image recognition.

A New Concept of Multiple Neural Networks Structure Using Convex Combination

In this article, a new concept of convex-combined multiple neural networks (NNs) structure is proposed. This new approach uses the collective information from multiple NNs to train the model. Based on both theoretical and experimental analyses, the new approach is shown to achieve faster training convergence with a similar or even better test accuracy than a conventional NN structure. Two experiments are conducted to demonstrate the performance of our new structure: the first one is a semantic frame parsing task for spoken language understanding (SLU) on the Airline Travel Information System (ATIS) data set and the other is a handwritten digit recognition task on the Mixed National Institute of Standards and Technology (MNIST) data set. We test this new structure using both the recurrent NN and convolutional NNs through these two tasks. The results of both experiments demonstrate a 4× - 8× faster training speed with better or similar performance by using this new concept.

Efficient and Robust Spike-Driven Deep Convolutional Neural Networks Based on NOR Flash Computing Array

In this article, we propose an efficient and robust spike-driven convolutional neural network (SCNN) based on the NOR flash computing array (NFCA), which is mapped by the pretrained convolutional neural network with the same structure. The spike-driven system eliminates the additional analog-to-digital/digital-to-analog (AD/DA) conversion in the NFCA-based CNN. To study the performance of the hardware implementation, an NFCA-based SCNN for the recognition of the Mixed National Institute of Standards and Technology (MNIST) data set is simulated. Simulation results illustrate that the system achieves 97.94% accuracy with the computing speed of $1 \times 10^{6}$ frame per second (fps). Compared with the typical mixed-signal NFCA-based CNN, the NFCA-based SCNN saves 97% area and 56% energy consumption. Moreover, the NFCA-based SCNN demonstrates great robustness to 30% image noise with less than 2% accuracy loss. The impact of random telegraph noise (RTN) is also greatly reduced in which less than 1% accuracy decrease can be achieved at the 32-nm technology node.

Reproducible Ultrathin Ferroelectric Domain Switching for High-Performance Neuromorphic Computing.

Neuromorphic computing consisting of artificial synapses and neural network algorithms provides a promising approach for overcoming the inherent limitations of current computing architecture. Developments in electronic devices that can accurately mimic the synaptic plasticity of biological synapses, have promoted the research boom of neuromorphic computing. It is reported that robust ferroelectric tunnel junctions can be employed to design high-performance electronic synapses. These devices show an excellent memristor function with many reproducible states (≈200) through gradual ferroelectric domain switching. Both short- and long-term plasticity can be emulated by finely tuning the applied pulse parameters in the electronic synapse. The analog conductance switching exhibits high linearity and symmetry with small switching variations. A simulated artificial neural network with supervised learning built from these synaptic devices exhibited high classification accuracy (96.4%) for the Mixed National Institute of Standards and Technology (MNIST) handwritten recognition data set.

Improving the Accuracy and Hardware Efficiency of Neural Networks Using Approximate Multipliers

Improving the accuracy of a neural network (NN) usually requires using larger hardware that consumes more energy. However, the error tolerance of NNs and their applications allow approximate computing techniques to be applied to reduce implementation costs. Given that multiplication is the most resource-intensive and power-hungry operation in NNs, more economical approximate multipliers (AMs) can significantly reduce hardware costs. In this article, we show that using AMs can also improve the NN accuracy by introducing noise. We consider two categories of AMs: 1) deliberately designed and 2) Cartesian genetic programing (CGP)-based AMs. The exact multipliers in two representative NNs, a multilayer perceptron (MLP) and a convolutional NN (CNN), are replaced with approximate designs to evaluate their effect on the classification accuracy of the Mixed National Institute of Standards and Technology (MNIST) and Street View House Numbers (SVHN) data sets, respectively. Interestingly, up to 0.63% improvement in the classification accuracy is achieved with reductions of 71.45% and 61.55% in the energy consumption and area, respectively. Finally, the features in an AM are identified that tend to make one design outperform others with respect to NN accuracy. Those features are then used to train a predictor that indicates how well an AM is likely to work in an NN.

Max-margin multi-scale convolutional factor analysis model with application to image classification

This paper presents a max-margin multi-scale convolutional factor analysis (MMCFA) model, which explores the strongly discriminative principle of max-margin learning to improve the classification performance of multi-scale convolutional factor analysis (CFA) model with application to image data classification. Compared with the traditional factor analysis (FA) model, the CFA model can maintain the spatial correlation among the image pixels in two-dimensional space and capture the structural information from images via convolution kernels. Moreover, to extract multi-level features, multi-scale convolution kernels are adopted to capture richer features at different scales of images. Since the unsupervised model may not offer discriminative factors for the classification task, it is expected to introduce the supervised information to the multi-scale CFA model when supervised information is available. To deal with it, a latent variable support vector machine (LVSVM) is linked to the factors learned from multi-scale CFA model, yielding max-margin discrimination, as the classification criterion in the feature space in our proposed model. The multi-scale CFA model and LVSVM learn parameters jointly in a united framework via the Gibbs inference. Experimental results on mixed national institute of standards and technology (MNIST) dataset, Fashion-MNIST dataset, SVHN dataset and measured synthetic aperture radar (SAR) images show that the learned convolution kernels and factors can describe data information well and the proposed model has excellent classification performance.

Studying on Improved Spiking Neural Network in Handwritten Digital Recognition

In this paper, an unsupervised learning algorithm of spiking neural network (SNN) by using biologically plausible mechanisms is presented for achieving handwritten digital recognition. Firstly, the spike-timing-dependent plasticity (STDP) model was established based on pre- and postsynaptic trace learning rule to determine the connection between neurons. Secondly, genetic algorithm (GA) was used to optimize the initial presynaptic weights and axon delay in the neural network. Finally, the Mixed National Institute of Standards and Technology (MNIST) dataset was trained and tested. Experimental results show that the proposed method can effectively improve the recognition rate of handwritten numbers and realize unsupervised learning of handwritten digital recognition. The accuracy in the MNIST benchmark test is increased by using this improved unsupervised SNN learning scheme. The computational complexity is greatly reduced; therefore, the calculation speed is increased. Simulation result is better than the implementation of the previous unsupervised SNNs.

Affine invariant fusion feature extraction based on geometry descriptor and BIT for object recognition

It is difficult to recognise an image with affine transformation due to viewing angle and distance variations. Therefore, affine invariant feature extraction is a valuable technology in the field of image recognition. Inspired by bio-visual mechanism, an affine invariant for object recognition method based on a fusion feature framework is proposed in this study, which employs geometry descriptor and double biologically inspired transformation (DBIT). First, a shape feature of interest detector is adopted to detect contour features. Then, the area estimation of affine region detector is utilised to construct area ratio feature vectors. Second, an orientation edge detector is built to highlight the edges of different directions. On this basis, local space frequency detector is adopted to measure the spatial frequency at each direction and interval, which converts the output map into DBIT feature vectors. A weighted fusion strategy is performed based on Pearson correlation distance to fuse the geometry feature and DBIT feature. Some tests for Alphanumeric, Coil-100 MPEG-7, Mixed National Institute of Standards and Technology (MNIST) and Olivetti Research Laboratory face images database (ORL) database remain highly stable recognition accuracy, even when the shear factor is between -0.5 and + 0.5. The experiment results show the authors' proposed approach has a nice performance in feature invariance, selectivity and recognition accuracy.