Search

Controlling the contact properties of a copper (Cu) electrode is crucial for enhancing the performance of InSnZnO (ITZO) thin-film transistors (TFTs) in high-speed applications. This is due to the inherently low resistance-capacitance product constant of Cu. High diffusivity of Cu, when integrated into ITZO, poses a major challenge as it leads to the formation of detrimental electron trap states, resulting in performance degradation of ITZO TFTs. A self-assembled monolayer (SAM) serves a dual purpose, effectively functioning as a diffusion barrier (DB) for Cu and a passivation layer. The study showcases the efficacy of vapor-phase SAM with suitable functional groups in enhancing the electrical properties of ITZO TFTs, with SAMs featuring -SH groups being particularly effective. The impressive electrical performance of SAMs with -SH groups makes them highly effective as Cu DBs and passivation layers. These versatile SAMs exhibit potential as a great choice for a thin copper DB in upcoming electronic technologies. © 1963-2012 IEEE.

The parameter server architecture has shown promising performance advantages when handling deep learning (DL) applications. One crucial issue in this regard is the presence of stragglers, which significantly retards DL training progress. Previous solutions for solving stragglers may not fully exploit the computation resource of the cluster as evidenced by our experiments, especially in the heterogeneous environment. This motivates us to design a heterogeneity-aware parameter server paradigm that addresses stragglers and accelerates DL training from the perspective of computation parallelism. We introduce a novel methodology named straggler projection to give a comprehensive inspection of stragglers and reveal practical guidelines to solve this problem in two aspects: (1) controlling each worker's training speed via elastic training parallelism control and (2) transferring blocked tasks from stragglers to pioneers to fully utilize the computation resource. Following these guidelines, we propose the abstraction of parallelism as an infrastructure and design the Elastic-Parallelism Synchronous Parallel (EPSP) algorithm to handle distributed training and parameter synchronization, supporting both enforced- A nd slack-synchronization schemes. The whole idea has been implemented into a prototype called ${\sf Falcon}$Falcon which effectively accelerates the DL training speed with the presence of stragglers. Evaluation under various benchmarks with baseline comparison demonstrates the superiority of our system. Specifically, ${\sf Falcon}$Falcon reduces the training convergence time, by up to 61.83, 55.19, 38.92, and 23.68 percent shorter than FlexRR, Sync-opt, ConSGD, and DynSGD, respectively. © 1968-2012 IEEE.

The multi-interface networks are efficient infrastructures to deploy distributed Deep Learning (DL) tasks as the model gradients generated by each worker can be exchanged to others via different links in parallel. Although this decentralized parameter synchronization mechanism can reduce the time of gradient exchange, building a high-performance distributed DL architecture still requires the balance of communication efficiency and computational utilization, i.e., addressing the issues of traffic burst, data consistency, and programming convenience. To achieve this goal, we intend to asynchronously exchange gradient pieces without the central control in multi-interface networks. We propose the Piece-level Gradient Exchange and Multi-interface Collective Communication to handle parameter synchronization and traffic transmission, respectively. Specifically, we design the gradient sketch approach based on 8-bit uniform quantization to compress gradient tensors and introduce the colayer abstraction to better handle gradient partition, exchange and pipelining. Also, we provide general programming interfaces to capture the synchronization semantics and build the Gradient Exchange Index (GEI) data structures to make our approach online applicable. We implement our algorithms into a prototype system called ${\sf Canary}$Canary by using PyTorch-1.4.0. Experiments conducted in Alibaba Cloud demonstrate that ${\sf Canary}$Canary reduces 56.28 percent traffic on average and completes the training by up to 1.61x, 2.28x, and 2.84x faster than BML, Ako on PyTorch, and PS on TensorFlow, respectively. © 1990-2012 IEEE.

This installment of Computer's series highlighting the work published in IEEE Computer Society journals comes from IEEE Transactions on Computers. © 2021 Springer. All rights reserved.

In MENs, the task offloading services are expected to free MUs from complex tasks, which contributes to reduce the service latency and thus improves the QoS. Although the well-proven cooperative task offloading proposals effectively address such requirements by encouraging the cooperation among MUs and edges, the QoE from MUs' perspective still remains low. Therefore, exploring the QoE-based offloading mechanism for better serving MUs in MENs is an urgent task. In this article, by reshaping the conventional offloading process, we explore a novel cooperative offloading mechanism in order to improve QoE. Our contributions can be identified in three aspects: the task-hub construction optimizes traditional MENs, which supports efficient MU management and low-latency communication; the task preprocessing thoroughly refines the task inputs according to the priority and redundancy, resulting in higher QoE; and the task scheduling with Deep Reinforcement Learning (DRL) named Double Dueling Deterministic Policy Gradient (Double DDPG) makes rational offloading policy that guarantees minimal service latency. Experimental results show that our proposed approach achieves much higher QoE performance than the existing schemes. © 2002-2012 IEEE.

In mobile edge networks (MENs), big data caching services are expected to provide mobile users with better quality of experience (QoE) than normal scenarios. However, the increasing types of sensors and devices are producing an explosion of big data. Extracting valuable contents for caching is becoming a vital issue for the satisfaction of QoE. Therefore, it is urgent to propose some rational strategies to improve QoE, which is the major challenge for content-centric caching. This article introduces a novel big data architecture consisting of data management units for content extraction and caching decision, improving quality of service and ensuring QoE. Then a caching strategy is proposed to improve QoE, including three parts: (1) the caching location decision, which means the method of deploying caching nodes to make them closer to users; (2) caching capacity assessment, which aims to seek suitable contents to match the capacity of caching nodes; and (3) caching priority choice, which leads to contents being cached according to their priority to meet user demands. With this architecture and strategy, we particularly use a caching algorithm based on deep reinforcement learning to achieve lower cost for intelligent caching. Experimental results indicate that our schemes achieve higher QoE than existing algorithms. © 1986-2012 IEEE.