Hyper Scale Research Articles

Attaining quantitatively fewer defects in close-packed InGaZnO synthesized using atomic layer deposition

Owing to their uniformity across large areas, low-temperature processability, and low off-current characteristics, oxide semiconductors demonstrate significant potential for integration into displays, memories, and logic devices. In the hyper scaling era, atomic layer deposition (ALD) is well suited for downsizing device fabrication, leveraging self-limiting chemical reactions to provide nanoscale control, and conformal coating on substrates with high aspect ratios. Hence, in this study, we fabricated highly oriented c-axis-aligned crystalline (CAAC) indium-gallium-zinc oxide (IGZO) thin films using plasma-enhanced ALD (PEALD), introducing an additional thermodynamic driving force for crystallization. A comparative study was conducted on the properties of CAAC-IGZO in correspondence to conventional sputter-based IGZO. In particular, photo-induced current transient spectroscopy (PICTS) is employed to map the electronic structure (density of states (DOS)) of the films, which quantitatively confirmed fewer defects in CAAC-IGZO. Fewer defects result in highly stable CAAC-IGZO transistors with positive bias stress threshold (PBTS) at 0.03 V and negative bias stress threshold (NBTS) at - 0.06 V. Thus, this research on thermally and electrically stable CAAC-IGZO synthesized at low temperatures through PEALD offers insights into the processing and material perspectives for the application of oxide semiconductors in 3D integration.

Aligned Intra Prediction and Hyper Scale Decoder Under Multistage Context Model for JPEG AI

Aligned Intra Prediction and Hyper Scale Decoder Under Multistage Context Model for JPEG AI

HFOS$$_L$$: hyper scale fast optical switch-based data center network with L-level sub-network

HFOS$$_L$$: hyper scale fast optical switch-based data center network with L-level sub-network

Clock Synchronized Transmission of 51.2 GBd Optical Packets for Optically Switched Data Center Interconnects

Optical switching has attracted significant attention in recent research on data center networks (DCNs) as it is a promising viable route for the further scaling of hyper scale data centers, so that DCNs can keep pace with the rapid growth of machine-to-machine traffic. It has been shown that optical clock synchronization enables sub-nanosecond clock and data recovery time and is crucial to high performance optically switched DCN. Moreover, the interconnect data rate is expected to increase from the current 100 Gb/s per fiber to scale to 800 Gb/s and beyond, requiring high baud rate signaling at >50 GBd. Thus, future optically switched DCN should support >50 GBd data transmission with optical clock synchronization. Here, we demonstrate the clock-synchronized transmission of 128-byte optical packets at 51.2 GBd and study the impact of reference clock phase noise on system performance, focusing on the tolerance to the clock phase misalignment that affects the system scalability and reliability. By comparing the tolerable sampling clock phase offsets using different reference clocks, we show that a clock phase offset window of about 8 ps could be achieved with a <0.2ps source clock. Furthermore, we model and numerically study the de-correlation of clock phase noise. This allows the total jitter to be estimated, and thereby, the estimation of the transmission performance for future generations of high baud rate, clock synchronized DC interconnects.

Bayesian inference for adaptive low rank and sparse matrix estimation

Abstract Low rank and sparse matrix estimation has been attracting significant interest in recent years. Generally, such a problem is modeled by imposing the l1-norm to pursuit a sparse and low rank matrix decomposition. However, the l1-norm is only a conservative sparse regularizer which leads to over-penalty. To remedy this issue, this paper presents an adaptive regularizer learning strategy to provide advanced low rank solution and avoid over-penalty. The new method is termed ARLLR. In the Bayesian inference, the prior distribution of the singular values is assumed to be Laplacian with hyper scale parameters. With the help of full Maximize A Posterior (MAP), we learn the optimal scale parameters by revealing its correlation to the inherent variables. We indicate that the adaptively estimated regularizer corresponds to the log function and the global minimum is given for the proposed non-convex problem. Furthermore, by employing the adaptive regularizer on the sparse part, a double log regularized low rank and sparse matrix decomposition model which is denoted by ARLLRE, is proposed. The ADMM algorithm is utilized to solve the ARLLRE problem, and the convergence of the algorithm is proved. In experiment, we use ARLLR for image denoising and ARLLRE for foreground and background extraction, respectively. Experimental results show that ARLLR enhances image denoising performance compared with the state-of-the-art image denoising algorithms in both quantity value and visual quality. Meanwhile, ARLLRE delivers excellent results in foreground and background extraction.

2.5D FPGA-HBM Integration Challenges

Abstract FPGA partitioning and high density integration using interposer based 2.5D stacked silicon interconnect technology (SSIT) has been the pioneering work at Xilinx for several years enabling advanced applications in high performance computing, networking, hyper scale data center and cloud services etc. With the insatiable demand for acceleration workloads, FPGAs need to be coupled with in package memory to enable higher bandwidth, lower power and smaller form factor architecture. 3D stacking based high bandwidth memory (HBM) has paved the way to realize such applications providing 10X higher bandwidth, 4X lower power vs DDR4. This paper provides an overview of unique challenges involved with 2.5D FPGA-HBM SSIT integration from design, process/package development, test and reliability point of view

UNLIMITED, BLURRED LIMITS IN A BORDERLESS WORLD

ABSTRACT This study explores the difficulties of the unlimited, a world which everyday is more connected to imprecise limits, instability and unclear visibility. The unlimited is the realm of the vague and blurred. The objective of this paper is to analyze the driving forces that are dramatically transforming the business environment and proposing consequences on how value is created, and show their huge impact on actual firm’s industrial sectors, business models, and products/services. It is made a reference to the disrupting technologies that are known today and are abruptly changing the future and the executive’s role. It states that traditional strategic and analytical tools are diminishing their impact and effectiveness, and that old paradigms must be revised. New factors as money, global cities and speed are having tremendous consequences on businesses and society. Technology and digitalization propose instant interactions, data analysis and hyper scaling, and also new optimized lives, complex tasks resolution and decision-making improvements through opening and interconnecting new fields. Innovation creates new wealth, advancement and growth, and it is the basis for new generation products which referrer to modularity, platforms, global resources and one-consumer-experience-at-a-time for interactive and collaborative experiences, services, processes and business models. The study was exploratory and descriptive, and a qualitative methodology was used. Its design was not experimental and transversal, as the information was collected at a given moment in time.

Hyperscaling-violating Lifshitz solutions in cubic gravity

Considering the cubic theory of gravity which has been constructed in \cite{myers}, we study the existence of Lifshitz and hyper scaling violating Lifshitz solutions. We firstly extend the black hole solution of \cite{myers} and find that such extended solutions are valid for any value of dynamical exponent $z$. Next, we examine the existence of the AdS black hole solution with non-zero hyperscaling-violating exponent $\theta$ and general dynamical exponent $z$. We find that the solutions do exist only for $\theta=0,3$ in 4 dimension and $\theta=0,4$ in 5 dimension. In particular, when $\theta=3(4)$ in $4(5)$ dimension, we have Lifshitz and Schwarzschild-AdS black hole solution solution for $z=\{0,1\}$.

Critical phenomena without “hyper scaling”: How is the finite-size scaling analysis of Monte Carlo data affected?

The finite size scaling analysis of Monte Carlo data is discussed for two models for which hyperscaling is violated: (i) the random field Ising model (using a model for a colloid-polymer mixture in a random matrix as a representative) (ii) The Ising bi-pyramid in computing surface fields.