We Show that pair hopping through a Kondo singlet give rise to a negative Josephson coupling. Thus, our calculation supports the existance of staggered pair correlations in the strong coupling limit of a one dimensional Kondo lattice.

Negative Josephson coupling in the Kondo strong coupling limit

DLC (Diamond-Like Carbon) coatings feature high hardness, good wear resistance and low friction. Applying DLC coatings to titanium alloys using different coating techniques can improve the surface properties. However, in many cases, this hard coating fails under mechanical loading due to the mechanical incompatibility of the coating and substrate. Therefore, an intermediate layer providing improved load support is necessary to prevent the failure of the DLC film. For this purpose, the Micro Arc Oxidation (MAO) and Closed Field Unbalance Magnetron Sputtering (CFUBMS) coating techniques were used to deposit duplex MAO/DLC coating on Ti6Al4V (Grade 5) alloy. The surface topography, morphology, crystallographic structure and thickness of the duplex coatings were determined using Scanning Electron Microscopy (SEM), XRD and RAMAN. The hardness was measured using a nano-hardness tester. The hardness, adhesion to the substrate and load-bearing capacity of DLC increased due to the intermediate layer of TiO 2 deposited by MAO because it decreased the mechanical incompatibility between the soft substrate and hard coating. The duplex coating using the MAO/CFUBMS process was effective in maintaining a convenient, low, stable coefficient of friction and improving the tribological behaviour of the Ti6Al4V substrate. It was observed that the duplex coatings deposited on the Ti6Al4V substrate indicate a better tribological performance than the DLC or MAO coatings on the Ti6Al4V alloy substrate.

Wear and adhesion resistance of duplex coatings deposited on Ti6Al4V alloy using MAO and CFUBMS

The amorphous carbon coatings of a-C and a-C:H type were deposited by pulsed magnetron sputtering in argon and argon/acetylene atmosphere, respectively. The deposition rate, chemical composition, structure and mechanical properties of these coatings were studied as a function of acetylene flow rate. The adding of acetylene to working atmosphere caused increase of deposition rate and hydrogen content in coatings, and at the same time decrease in their hardness. The friction and wear behaviour of a-C and a-C:H coatings in ambient air are highly dependent on kind of counterparts material. The “true” friction coefficients of a-C and a-C:H coatings sliding against a-C and a-C:H coatings, respectively, are similar in values (0.06–0.08) and wear rates are similar too. Significantly higher friction coefficients (0.2–0.3) and wear rates were observed for both a-C and a-C:H coatings sliding against 100Cr6 steel. The lowest friction coefficients (0.02–0.04) and wear rates were obtained for a-C and a-C:H coatings sliding against Alumina counterpart.

Preparation and characterisation of a-C and a-C:H coatings deposited by pulsed magnetron sputtering

Hydrogenated diamond-like carbon films were deposited by reactive Ar/CH4 high-power impulse magnetron sputtering. The deposition rate increased from 0.25 µm/h in the case of without CH4 to about 1.6 µm/h at a CH4 fraction of 50%. On the other hand, the hardness of the deposited films markedly decreased from 17 to 5 GPa with the increase in CH4 fraction. Film properties were characterized by X-ray photoelectron spectroscopy and Raman spectroscopy. C 1s peak broadening with increasing CH4 fraction was observed, indicating a slight increase in sp3 content. In Raman spectra, two, very broad overlapping G and D bands extending between 1200 and 1700 cm−1 were observed. The G-peak position shifted to lower wave numbers and the ratio of the D to G bands decreased at CH4 fractions lower than 5%, beyond which the G-peak position shifted to higher wave numbers and the ratio increased.

Formation of hydrogenated diamond-like carbon films by reactive Ar/CH4 high-power impulse magnetron sputtering

Effects of Ar ion bombardment by unbalanced magnetron sputtering on mechanical and thermal properties of Ti-Cu-Zr-Ni-Hf-Si thin film metallic glass

Structures and mechanical properties of diamond like carbon films prepared by closed-field unbalanced magnetron sputtering


 Graph Neural Networks (GNNs) are receiving a spotlight as a powerful tool that can effectively serve various inference tasks on graph structured data. As the size of real-world graphs continues to scale, the GNN training system faces a scalability challenge. Distributed training is a popular approach to address this challenge by scaling out CPU nodes. However, not much attention has been paid to
 disk-based
 GNN training, which can scale up the single-node system in a more cost-effective manner by leveraging high-performance storage devices like NVMe SSDs. We observe that the data movement between the main memory and the disk is the primary bottleneck in the SSD-based training system, and that the conventional GNN training pipeline is sub-optimal without taking this overhead into account. Thus, we propose Ginex, the first SSD-based GNN training system that can process billion-scale graph datasets on a single machine. Inspired by the inspector-executor execution model in compiler optimization, Ginex restructures the GNN training pipeline by separating
 sample
 and
 gather
 stages. This separation enables Ginex to realize a provably optimal replacement algorithm, known as
 Belady's algorithm
 , for caching feature vectors in memory, which account for the dominant portion of I/O accesses. According to our evaluation with four billion-scale graph datasets and two GNN models, Ginex achieves 2.11X higher training throughput on average (2.67X at maximum) than the SSD-extended PyTorch Geometric.


Ginex

Transformer architecture is one of the most remarkable recent breakthroughs in neural networks, achieving state-of-the-art (SOTA) performance on various natural language processing (NLP) and computer vision tasks. Self-attention is the key enabling operation for transformer-based models. However, its quadratic computational complexity to the sequence length makes this operation the major performance bottleneck for those models. Thus, we propose a novel self-attention accelerator that skips most of the computation by utilizing an approximate candidate selection algorithm. Implemented in a 40nm CMOS technology, our 5.64 mm2 chip operates at 100–600 MHz consuming 48.3-685 mW to achieve the energy and area efficiency of 0.354-5.61 TOPS/W and 239 GOPS/mm2, respectively.

Jae W. Lee

Papers

Structures and mechanical properties of diamond like carbon films prepared by closed-field unbalanced magnetron sputtering

Ginex

A 40nm 5.6TOPS/W 239GOPS/mm2 Self-Attention Processor with Sign Random Projection-based Approximation