Search

Nowadays, field programmable gate arrays (FPGAs) have been widely adopted to serve as accelerators in artificial intelligence and big data related applications. Since the static random access memory (SRAM)-based FPGA is suffering from limited density and high leakage power, nonvolatile FPGAs have been proposed, where SRAM is replaced with emerging nonvolatile memories (NVMs). Multilevel cell (MLC), which can store multiple bits within one memory cell, further improves the density of nonvolatile FPGAs and shows great potential to enable large on-chip memory. However, it suffers from limited lifetime. In this article, we propose a wear leveling scheme to improve lifetime of MLC-based nonvolatile FPGAs. Instead of generating a series of configuration files for runtime reconfiguration, we propose to identify write-heavy MLC regions and dynamically transform them to durable single-level cell (SLC) mode. Specifically, we propose three modules: 1) pertaining to write behavior monitor; 2) approximate cost calculator; and 3) mode transformation manager to achieve adaptive mode transformations. We consider FPGA features to design these modules, which is different from implementations for MLC-SLC transformation in CPU architecture. Evaluation shows that the proposed scheme can improve lifetime for MLC nonvolatile FPGAs by $6.03\times $ , at cost of 12.5% storage overhead. [ABSTRACT FROM AUTHOR]

Traditional operations, e.g. graph edit distance (GED), are no longer suitable for processing the massive quantities of graph-structured data now available, due to their irregular structures and high computational complexities. With the advent of graph neural networks (GNNs), the problems of graph representation and graph similarity search have drawn particular attention in the field of computer vision. However, GNNs have been less studied for efficient and fast retrieval after graph representation. To represent graph-based data, and maintain fast retrieval while doing so, we introduce an efficient hash model with graph neural networks (HGNN) for a newly designed task (i.e. fast graph-based data retrieval). Due to its flexibility, HGNN can be implemented in both an unsupervised and supervised manner. Specifically, by adopting a graph neural network and hash learning algorithms, HGNN can effectively learn a similarity-preserving graph representation and compute pair-wise similarity or provide classification via low-dimensional compact hash codes. To the best of our knowledge, our model is the first to address graph hashing representation in the Hamming space. Our experimental results reach comparable prediction accuracy to full-precision methods and can even outperform traditional models in some cases. In real-world applications, using hash codes can greatly benefit systems with smaller memory capacities and accelerate the retrieval speed of graph-structured data. Hence, we believe the proposed HGNN has great potential in further research. [ABSTRACT FROM AUTHOR]

A novel constant power charging strategy is proposed for LCC resonant capacitor charging power supply (CCPS) in this article, which combines the advantages of discontinuous current mode (DCM) and continuous current mode (CCM) to increase the charging speed of capacitor and the utilization of input grid capacity. In order to implement the multimode constant power control (CPC) strategy, a steady-state model of LCC series-parallel resonant converter is built by state plane analysis, and the analytical expressions for output voltage, average output current, and switching frequency are derived. By the analytical expressions, this control strategy directly determines the relationship of the output voltage and the switch frequency, so the frequency control curve of the pulse frequency modulation could be realized easily. Accordingly, the converter operating along the control curve can take advantage of high average and low peak current of CCM, small switching loss of DCM. Finally, the accuracy of LCC state plane model and the effectiveness of multimode CPC are verified by the simulation and a 2-kW CCPS experiment. [ABSTRACT FROM AUTHOR]

Land surface temperature (LST) is an important parameter for measuring the water–heat balance on the Earth's surface. Remote sensing technology provides a unique way to monitor LSTs over large spatial areas. Thermal infrared (TIR)-derived LST has a high spatial resolution and accuracy, but missing values caused by clouds hinder applications of this method. Passive microwave radiation can penetrate clouds, but the data have a relatively lower spatial resolution and accuracy. The Bayesian maximum entropy (BME) method was used for blending the Moderate Resolution Imaging Spectroradiometer (MODIS) LST and Advanced Microwave Scanning Radiometer-Earth Observing System (AMSR-E) LST to produce a spatially complete, accurate, and high-spatial-resolution LST over the Tibetan Plateau (TP) and the Heihe River Basin (HRB). The accuracy of the BME method was validated using the adjusted soil temperature collected from the two verification regions. The root-mean-square errors (RMSEs) are less than 3.54 and 4.89 K over the relatively flat verification region during nighttime and daytime, respectively. In another rugged terrain verification region, the RMSE is less than 3 K during nighttime and ranges from 4.2 to 8.29 K during daytime. The RMSE of the blended LST is significantly better than that of a recent study of the AMSR-E LST retrieved during nighttime, and the RMSE is comparable to the AMSR-E LST retrieved during daytime. The BME method was adopted to integrate the MODIS LST and AMSR-E LST acquired on November 30, 2010 over the TP and HRB. The spatial completeness of the blended LST reached 100%, and the blended LST spatial pattern was generally consistent with the spatial patterns of the MODIS LST and AMSR-E LST. This paper demonstrates the utility of generating all-weather regional LSTs using the BME method from TIR LST and microwave LST products. [ABSTRACT FROM AUTHOR]

Mid-infrared erbium (Er3+)-doped ZrF4–BaF2–LaF3–AlF3–NaF fiber laser mode-locked by multilayer graphene saturable absorber was demonstrated. Mode-locked pulses at 2.8 \mu \textm with an average output power of 18 mW at a repetition rate of 25.4 MHz, corresponding to a pulse energy of 0.7 nJ, were obtained. The pulsewidth was measured to be $\sim 42$ ps by a home-made autocorrelator. Our experiment has validated the mode-locking capability of graphene in the 3- \mu \textm wavelength region. [ABSTRACT FROM PUBLISHER]

Protrusions are generated with random distribution on the contacts surfaces of Vacuum Circuit Breaker (VCB) after arc interruption. They will degrade the vacuum insulation level by Joule Heating Vaporization while withstanding rated voltage. In order to investigate the behavior of protrusions in VCB after interruption, the point-to-plane insulation structure is constructed in the vacuum chamber. And, the pre-discharge phenomena are observed using high speed video camera. It is observed that successive current pulse cascades before the explosion of the point electrode, which results in craters on the surface of the point electrode. Through the analysis of the images, the speed of the particle is estimated to be less than 125m/s. This research also shows the vaporization is non-uniformly distributed in some channels around the point electrode, which can be the possible reason for the later discharge in VCB. Through the computation of the Fowler-Nordheim (F-N) Curve, two patterns of field emission are revealed on the F-N plot, which are the rising and falling pattern. In the voltage rising pattern, the field enhancement factor is 11 times larger than that in the voltage falling pattern; however, the effective field emission area is only 1/1000 of the voltage falling pattern. Moreover, the two patterns are enclosed with the non-field emission half cycle to form a cycle on F-N plot. It is concluded that the point electrode is under periodic conditioning process with the observation of the cycle, which is considered to be the key reason for late discharge in VCB. [ABSTRACT FROM AUTHOR]