An Area- and Energy-Efficient Spiking Neural Network With Spike-Time-Dependent Plasticity Realized With SRAM Processing-in-Memory Macro and On-Chip Unsupervised Learning.

In this article, we present a spiking neural network (SNN) based on both SRAM processing-in-memory (PIM) macro and on-chip unsupervised learning with Spike-Time-Dependent Plasticity (STDP). Co-design of algorithm and hardware for hardware-friendly SNN and efficient STDP-based learning methodology is used to improve area and energy efficiency. The proposed macro utilizes charge sharing of capacitors to perform fully parallel Reconfigurable Multi-bit PIM Multiply-Accumulate (RMPMA) operations. A thermometer-coded Programmable High-precision PIM Threshold Generator (PHPTG) is designed to achieve low differential non-linearity (DNL) and high linearity. In the macro, each column of PIM cells and a comparator act as a neuron to accumulate membrane potential and fire spikes. A simplified Winner Takes All (WTA) mechanism is used in the proposed hardware-friendly architecture. By combining the hardware-friendly STDP algorithm as well as the parallel Word Lines (WLs) and Processing Bit Lines (PBLs), we realize unsupervised learning and recognize the Modified National Institute of Standards and Technology (MNIST) dataset. The chip for the hardware implementation was fabricated with a 55 nm CMOS process. The measurement shows that the chip achieves a learning efficiency of 0.47 nJ/pixel, with a learning energy efficiency of 70.38 TOPS/W. This work paves a pathway for the on-chip learning algorithm in PIM with lower power consumption and fewer hardware resources.

Braille recognition by E-skin system based on binary memristive neural network.

Braille system is widely used worldwide for communication by visually impaired people. However, there are still some visually impaired people who are unable to learn Braille system due to various factors, such as the age (too young or too old), brain damage, etc. A wearable and low-cost Braille recognition system may substantially help these people recognize Braille or assist them in Braille learning. In this work, we fabricated polydimethylsiloxane (PDMS)-based flexible pressure sensors to construct an electronic skin (E-skin) for the application of Braille recognition. The E-skin mimics human touch sensing function for collecting Braille information. Braille recognition is realized with a neural network based on memristors. We utilize a binary neural network algorithm with only two bias layers and three fully connected layers. Such neural network design remarkably reduces the calculation burden and, thus, the system cost. Experiments show that the system can achieve a recognition accuracy of up to 91.25%. This work demonstrates the possibility of realizing a wearable and low-cost Braille recognition system and a Braille learning-assistance system.

A dynamic AES cryptosystem based on memristive neural network.

This paper proposes an advanced encryption standard (AES) cryptosystem based on memristive neural network. A memristive chaotic neural network is constructed by using the nonlinear characteristics of a memristor. A chaotic sequence, which is sensitive to initial values and has good random characteristics, is used as the initial key of AES grouping to realize "one-time-one-secret" dynamic encryption. In addition, the Rivest-Shamir-Adleman (RSA) algorithm is applied to encrypt the initial values of the parameters of the memristive neural network. The results show that the proposed algorithm has higher security, a larger key space and stronger robustness than conventional AES. The proposed algorithm can effectively resist initial key-fixed and exhaustive attacks. Furthermore, the impact of device variability on the memristive neural network is analyzed, and a circuit architecture is proposed.

Sharp selective scattering of red, green, and blue light achieved via gain material's loss compensation.

Frequency-selective scattering of light can be achieved by metallic nanoparticle's localized surface plasmon resonance (LSPR). And this property may find an application in a transparent projection screen: ideally, specially designed metallic nanoparticles dispersed in a transparent matrix only selectively scatter red, green and blue light and transmit the visible light of other colors. However, optical absorption and surface dispersion of a metallic nanoparticle, whose size is comparable or smaller than mean free path of electrons in the constituent material, degenerate the desired performance by broadening the resonance peak width (i.e., decreasing frequency-selectivity) and decreasing light scattering intensity. In this work, it is shown that the problem can be solved by introducing gain material. Numerical simulations are performed on nanostructures based on silver (Ag), gold (Au) or aluminum (Al) with or without gain material, to examine the effect of gain material and to search for suitable structures for sharp selective scattering of red, green and blue light. And it is found that introducing gain material greatly improves performance of the structures based on Ag or Au except the structures based on Al. The most suitable structures for sharp selective scattering of red, green and blue light are, respectively, found to be the core-shell structures of silica/Au (core/shell), silica/Ag and Ag/silica, all with gain material.

Effect of Surface Scattering of Electrons on Ratios of Optical Absorption and Scattering to Extinction of Gold Nanoshell.

Gold nanoshell's high light scattering and absorption at its resonance wavelength have found applications in biomedical imaging and photothermal therapy. However, at nanoscale, metallic material's dielectric function is affected by nanoparticle's size, mainly via a mechanism called surface scattering of conduction electrons. In this work, the effect of surface scattering of electrons on the ratios of optical absorption and scattering to extinction (which is the sum of the absorption and scattering) of gold nanoshell is investigated. Simulation results for several shell thicknesses are compared. It is found that the electrons' surface scattering increases the optical absorption ratio, and the thinner the shell thickness, the larger the increase in the difference of the absorption ratio between the situations with and without the surface scattering considered. The increase of absorption ratio is then verified by comparing simulation results to experimental measurements for three nanoshells. The parameters of the simulations to fit the experimental measurements show that the damping of conduction electrons in metallic shell geometry is larger than that predicted by the billiard scattering model.

Handwritten-Digit Recognition by Hybrid Convolutional Neural Network based on HfO2 Memristive Spiking-Neuron.

Although there is a huge progress in complementary-metal-oxide-semiconductor (CMOS) technology, construction of an artificial neural network using CMOS technology to realize the functionality comparable with that of human cerebral cortex containing 1010-1011 neurons is still of great challenge. Recently, phase change memristor neuron has been proposed to realize a human-brain level neural network operating at a high speed while consuming a small amount of power and having a high integration density. Although memristor neuron can be scaled down to nanometer, integration of 1010-1011 neurons still faces many problems in circuit complexity, chip area, power consumption, etc. In this work, we propose a CMOS compatible HfO2 memristor neuron that can be well integrated with silicon circuits. A hybrid Convolutional Neural Network (CNN) based on the HfO2 memristor neuron is proposed and constructed. In the hybrid CNN, one memristive neuron can behave as multiple physical neurons based on the Time Division Multiplexing Access (TDMA) technique. Handwritten digit recognition is demonstrated in the hybrid CNN with a memristive neuron acting as 784 physical neurons. This work paves the way towards substantially shrinking the amount of neurons required in hardware and realization of more complex or even human cerebral cortex level memristive neural networks.

A MoS2-based coplanar neuron transistor for logic applications.

The human brain is an extremely complex system of 1010-1011 neurons. To construct brain-like neuromorphic hardware, the neuron unit should be implemented effectively. Here, we report a neuron transistor based on a MoS2 flake, which has summation and threshold functions similar to biological neurons and may act as a basic neuron unit in neuromorphic hardware. The neuron transistor is composed of a floating gate and two control gates. A heavily doped silicon substrate serves as the floating gate, while the two control gates are capacitively coupled with the floating gate. The neuron transistor can be well controlled by the two control gates individually or simultaneously. The drain current can be modulated by the input voltages at the control gates. While the current response of the neuron transistor has a large dependence on the magnitude of the input signal, it shows little dependence on the frequency of the input signal. To demonstrate the potential neuromorphic application of the neuron transistor, functions including abacus-like function, AND logic and OR logic are realized in the neuron transistor.

Highly spectrum-selective ultraviolet photodetector based on p-NiO/n-IGZO thin film heterojunction structure.

Ultraviolet photodetector with p-n heterojunction is fabricated by magnetron sputtering deposition of n-type indium gallium zinc oxide (n-IGZO) and p-type nickel oxide (p-NiO) thin films on ITO glass. The performance of the photodetector is largely affected by the conductivity of the p-NiO thin film, which can be controlled by varying the oxygen partial pressure during the deposition of the p-NiO thin film. A highly spectrum-selective ultraviolet photodetector has been achieved with the p-NiO layer with a high conductivity. The results can be explained in terms of the "optically-filtering" function of the NiO layer.

Associative memory realized by a reconfigurable memristive Hopfield neural network.

Although synaptic behaviours of memristors have been widely demonstrated, implementation of an even simple artificial neural network is still a great challenge. In this work, we demonstrate the associative memory on the basis of a memristive Hopfield network. Different patterns can be stored into the memristive Hopfield network by tuning the resistance of the memristors, and the pre-stored patterns can be successfully retrieved directly or through some associative intermediate states, being analogous to the associative memory behaviour. Both single-associative memory and multi-associative memories can be realized with the memristive Hopfield network.

Evolution of dielectric function of Al-doped ZnO thin films with thermal annealing: effect of band gap expansion and free-electron absorption.

Evolution of dielectric function of Al-doped ZnO (AZO) thin films with annealing temperature is observed. It is shown that the evolution is due to the changes in both the band gap and the free-electron absorption as a result of the change of free-electron concentration of the AZO thin films. The change of the electron concentration could be attributed to the activation of Al dopant and the creation/annihilation of the donor-like defects like oxygen vacancy in the thin films caused by annealing.

Influence of localized surface plasmon resonance and free electrons on the optical properties of ultrathin Au films: a study of the aggregation effect.

The contributions of localized surface plasmon resonance (LSPR) and Drude (free electrons) absorption to the complex dielectric function of ultrathin Au films were investigated with spectroscopic ellipsometry. When the Au film thickness is thinner than ~10 nm, Au nanoparticles (NPs) are formed as a result of the discontinuity in the films, leading to the emergence of LSPR of Au NPs; and the LSPR exhibits a splitting when the films thinner than ~8 nm, which could be attributed to the near-field coupling of the Au NPs and/or the inhomogeneous polarizations of the Au NPs. On the other hand, the delocalization of electrons in Au NPs due to the aggregation of Au NPs in a thicker film leads to an increase in the free-electron absorption and a suppression of the LSPR.

Effects of free electrons and quantum confinement in ultrathin ZnO films: a comparison between undoped and Al-doped ZnO.

Band gaps and exciton binding energies of undoped and Al-doped ZnO thin films were determined from optical absorption measurement based on the Elliott's exciton absorption theory. As compared to the undoped films, the doped films exhibit a band gap expansion and a reduction in the exciton binding energies due to the free electron screening effect, which suppresses the excitonic absorption and results in a blue shift of the absorption edge. The undoped and doped films show the same quantum size dependence, i.e. both the exciton binding energies and band gap energies increase with decreasing grain size of the oxides.

Potent cardioprotection from ischemia-reperfusion injury by a two-domain fusion protein comprising annexin V and Kunitz protease inhibitor.

BACKGROUND: Considerable evidence suggests that coagulation proteases (tissue factor [TF]/activated factor VII [FVIIa]/FXa/thrombin) and their target protease activated receptors (PAR-1/PAR-2) play important roles in myocardial ischemia-reperfusion (I-R) injury. We hypothesized that localized inhibition of TF/FVIIa on the membrane surfaces of ischemic cells could effectively block coagulation cascade and subsequent PAR-1/PAR-2 cell signaling, thereby protecting the myocardium from I-R injury. OBJECTIVES: We recently developed an annexin V-Kunitz inhibitor fusion protein (ANV-6L15) that could specifically bind to anionic phospholipids on the membrane surfaces of apoptotic cells and efficiently inhibit the membrane-anchored TF/FVIIa. In this study, we investigated the cardioprotective effect of ANV-6L15 in a rat cardiac I-R model in comparison with that of hirudin. METHODS: Left coronary artery occlusion was maintained for 45 min followed by 4 h of reperfusion in anesthetized Sprague-Dawley rats. One minute before or 2 min after coronary ligation, rats received an intravenous bolus injection of ANV-6L15 (2.5-250 µg kg(-1) ), vehicle, or hirudin via bolus injection and continuous infusion. RESULTS AND CONCLUSIONS: ANV-6L15 dose-dependently reduced infarct size by up to 87% and decreased plasma levels of cardiac troponin I, tumor necrosis factor-α, and soluble intercellular adhesion molecule-1, by up to 97%, 96%, and 66%, respectively, with little impact on the coagulation parameters. ANV-6L15 also ameliorated hemodynamic derangements, attenuated neutrophil infiltration and reduced Terminal deoxynucleotidyl transferase dUTP nick end labeling-positive apoptotic cardiomyocytes. Hirudin was less efficacious even at supraclinical dose. ANV-6L15 confers exceptionally potent cardioprotection and is a promising drug candidate for the prevention of myocardial I-R injury.

Size-suppressed dielectrics of Ge nanocrystals: skin-deep quantum entrapment.

Although the dielectric behavior of nanostructured semiconductors has been intensively investigated, the physics behind observations remains disputed with possible mechanisms such as quantum confinement and dangling bond polarization. Here we show that theoretical reproduction of the measured dielectric suppression of Ge nanocrystals asserts that the dielectric suppression originates from the shorter and stronger bonds at the skin-deep surface, the associated local densification and quantum entrapment of energy. Coordination-imperfection induced local quantum entrapment perturbs the Hamiltonian that determines the band gap and hence, the process of electron polarization consequently.

Ultraviolet random lasing action from highly disordered n-AlN/p-GaN heterojunction.

Room-temperature random lasing is achieved from an n-AlN/p-GaN heterojunction. The highly disordered n-AlN layer, which was deposited on p-GaN:Mg layer via radio frequency magnetron sputtering, acts as a scattering medium to sustain coherent optical feedback. The p-GaN:Mg layer grown on sapphire provides optical amplification to the scattered light propagating along the heterojunction. Hence, lasing peaks of line width less than 0.4 nm are emerged from the emission spectra at round 370 nm for the heterojunction under forward bias larger than 5.1 V. Lasing characteristics of the heterojunction are in agreement with the behavior of random lasers.

Laterally-current-injected light-emitting diodes based on nanocrystalline-Si/SiO2 superlattice.

Laterally electrically-pumped Si light-emitting diodes (LEDs) based on truncated nanocrystalline-Si (nc-Si)/SiO2 quantum wells are fabricated with complementary-metal-semiconductor-oxide (CMOS) process. Visible electroluminescence (EL) can be observed under a reverse bias larger than ~6 V. The light emission would probably originate from the spontaneous hot-carrier relaxations within the conduction and the valance bands when the device is sufficiently reverse-biased. The EL spectral profile is found to be modulated by varying structure parameters of the interdigitated finger electrodes. Up to ~20 times EL intensity enhancement is achieved as compared to vertical-current-injection LED prepared using the same material system. Based on the lateral-current-injection scheme, a Si/SiO2 MQW LED with Fabry-Perot (FP) microcavity and an on-chip waveguided LED that emits at 1.55-µm are proposed.

A two-terminal write-once-read-many-times-memory device based on an aluminum nitride thin film containing Al nanocrystals.

A switching from a high-conduction state to a low-conduction state occurs in an AIN thin film containing Al nanocrystals (nc-Al) when the nc-Al is charged with electrons. The switching is explained in terms of breaking of the conductive percolation paths of the nc-AI as a result of the charging of the nc-Al. A write-once-read many times-memory (WORM) device is demonstrated based on this phenomenon. The device can be switched by charging the nc-Al with a voltage of +10 V for 100 ms, yielding a current ratio of the two memory states of more than 300 at the reading voltage of 1 V. The charged state (i.e., the low-conduction state) remains unchanged after more than 1 x 106 read cycles, and its retention time is predicted to be more than 10 years.

Profile uniformity of overlapped oxide dots induced by atomic force microscopy.

The technique for assembling a uniform oxide line by overlapping a series of nanosized oxide dots induced by atomic force microscopy is analytically and experimentally investigated. In addition to the normal continuous (static) pulses, the oxide growth rates under various discontinuous (modulated) pluses are studied to quantify the overlapping effect under multiple pulses used by the assembling technique. In the analysis of the assembling technique, the superposition principle is used to predict the assembled profiles and to define the uniformity criteria. Experiments have been performed to demonstrate the analytical prediction, including the threshold or minimum pitch for forming uniform lines, and the onset pitch for the overlapping effect to be considered. Indeed, by following the uniformity criteria developed, uniform and reliable oxide lines can be obtained by overlapping oxide dots.

Charge storage behaviors of Ge nanocrystals embedded in SiO2 for the application in non-volatile memory devices.

Ge nanocrystals distributed in the SiO2 of metal-oxide-semiconductor structure are synthesized by low-energy Ge ion implantation with various energies and doses. Their charge storage behaviors are influenced by both the ion implantation dose and energy. The larger flatband voltage shift achieved by increasing either the implantation dose or energy is explained by the locations and concentration of the charge trapping sites. The smaller charge loss achieved by decreasing the implantation dose or increasing the implantation energy is explained by the co-existence of the charge leakage to the gate electrode and the lateral charge loss to the adjacent Ge nanocrystals.