Search

Graph neural networks (GNNs) have emerged as a popular strategy for handling non-Euclidean data due to their state-of-the-art performance. However, most of the current GNN model designs mainly focus on task accuracy, lacking in considering hardware resources limitation and real-time requirements of edge application scenarios. Comprehensive profiling of typical GNN models indicates that their execution characteristics are significantly affected across different computing platforms, which demands hardware awareness for efficient GNN designs. In this work, HGNAS is proposed as the first Hardware-aware Graph Neural Architecture Search framework targeting resource constraint edge devices. By decoupling the GNN paradigm, HGNAS constructs a fine-grained design space and leverages an efficient multi-stage search strategy to explore optimal architectures within a few GPU hours. Moreover, HGNAS achieves hardware awareness during the GNN architecture design by leveraging a hardware performance predictor, which could balance the GNN model accuracy and efficiency corresponding to the characteristics of targeted devices. Experimental results show that HGNAS can achieve about $10.6\times$ speedup and $88.2\%$ peak memory reduction with a negligible accuracy loss compared to DGCNN on various edge devices, including Nvidia RTX3080, Jetson TX2, Intel i7-8700K and Raspberry Pi 3B+., Comment: Accepted by DAC'23

Additional file 1. Fig. S1: The effect of miR-132-3p in MSC (passage 2) proliferation and senescence. Cell proliferation of MSC was determined by Cell Counting Kit 8 (CCK-8) assay. The graph shows data coming from MSCCon, MSCmiR-132-3p, and MSCantagomiR-132-3p groups (A). Representative images and summary data showing the senescence of MSC in MSCCon, MSCmiR-132-3p, and MSCantagomiR-132-3p groups. *p＜0.05. (PDF 2433 kb)

We report coupling of excitons in monolayers of molybdenum disulphide to their mirror image in an underlying gold substrate. Excitons at the direct band gap are little affected by the substrate whereas strongly bound C-excitons associated with a van-Hove singularity change drastically. On quartz substrates only one C-exciton is visible (in the blue) but on gold substrates a strong red-shifted extra resonance in the green is seen. Exciton coupling to its image leads to formation of a “mirror biexciton” with enhanced binding energy. Estimates of this energy shift in an emitter-gold system match experiments well. The absorption spectrum of MoS2 on gold thus resembles a bilayer of MoS2 which has been created by optical coupling. Additional top-mirrors produce an “optical bulk.”

Y.S. acknowledges the support from the National Natural Science Foundation of China (grant no. 61874074), Science and Technology Project of Shenzhen (JCYJ20170817101100705), and the (Key) Project of Department of Education of Guangdong Province (grant no. 2016KZDXM008). H.L. acknowledges the support from Natural Science Foundation of SZU (grant no. 2017011) and the support from Science and Technology Project of Shenzhen (JCYJ20170817100111548). This project was supported by Shenzhen Peacock Plan (grant no. KQTD2016053112042971) and the postgraduate innovation development fund project of Shenzhen University (grant no. 315-0000470527).

We report the band alignment parameters of the GaN/single-layer (SL) MoS2 heterostructure where the GaN thin layer is grown by molecular beam epitaxy on CVD deposited SL-MoS2/c-sapphire. We confirm that the MoS2 is an SL by measuring the separation and position of room temperature micro-Raman E1 2g and A1g modes, absorbance, and micro-photoluminescence bandgap studies. This is in good agreement with HRTEM cross-sectional analysis. The determination of band offset parameters at the GaN/SL-MoS2 heterojunction is carried out by high-resolution X-ray photoelectron spectroscopy accompanying with electronic bandgap values of SL-MoS2 and GaN. The valence band and conduction band offset values are, respectively, measured to be 1.8660.08 and 0.5660.1 eV with type II band alignment. The determination of these unprecedented band offset parameters opens up a way to integrate 3D group III nitride materials with 2D transition metal dichalcogenide layers for designing and modeling of their heterojunction based electronic and photonic devices.

This is an open access article published under a Creative Commons Attribution (CC-BY) License.-- et al., Nanometer-sized gaps between plasmonically coupled adjacent metal nanoparticles enclose extremely localized optical fields, which are strongly enhanced. This enables the dynamic investigation of nanoscopic amounts of material in the gap using optical interrogation. Here we use impinging light to directly tune the optical resonances inside the plasmonic nanocavity formed between single gold nanoparticles and a gold surface, filled with only yoctograms of semiconductor. The gold faces are separated by either monolayers of molybdenum disulfide (MoS2) or two-unit-cell thick cadmium selenide (CdSe) nanoplatelets. This extreme confinement produces modes with 100-fold compressed wavelength, which are exquisitely sensitive to morphology. Infrared scattering spectroscopy reveals how such nanoparticle-on-mirror modes directly trace atomic-scale changes in real time. Instabilities observed in the facets are crucial for applications such as heat-assisted magnetic recording that demand long-lifetime nanoscale plasmonic structures, but the spectral sensitivity also allows directly tracking photochemical reactions in these 2-dimensional solids., This work was supported by the UK EPSRC grants EP/G060649/1, EP/L027151/1, Defence Science and Technology Laboratory (DSTL), and ERC grant 320503 LINASS. C.T. and J.A. acknowledge financial support from Project FIS2013-41184-P from MINECO, ETORTEK 2014-15 of the Basque Department of Industry and IT756-13 from the Basque consolidated groups.

The self-organization of water in metallic surfaces is detemined by the competition between inter-molecular and metal-substrate interactions. An additional complication is given by the possibility that significant fraction of water molecules undergo partial dissociation. The wetting of Cu(ll0) has been recently reported. At low temperature and coverage, water forms one-dimensional arrays that clear indicate the competition of different interactions mentioned above. Carrasco et al. identified this structure with a wire formed by side-sharing pentagonal rings fomed by intact molecules. We present here a combined experimental and theoretical study that challenges this interpretation. Our STM images show structures different that those reported by Carrasco et al. We have compared our experimental STM images with a series of theoretical calculations of different linear structures based on pentagonal and hexagonal motifs. In this presentation we discuss different structures that produce simulated STM images in good correspondence with experiment. The most stable ones are based on hexagonal motifs with some dissociated molecules.