CSR: Small: Collaborative Research: Dynamic Reconfiguration for Adaptive Computing in Heterogeneous SoCs

由于技术的进步和对复杂应用日益增长的需求，嵌入式计算系统的复杂性正在以快速的速度增加。各种各样的此类系统采用多核架构，以满足在严格的功率和热量限制下的性能需求。自适应应用的动态变化是提高系统整体性能、能源效率和可靠性的关键，虽然动态自适应(重构)技术受到了各个领域的广泛关注，但它们并没有应用于实时系统芯片(SoC)设计中。这是因为实时系统可能由具有软或硬截止日期的关键任务组成，而动态重新配置开销可能会对关键任务产生不利影响。错过最后期限可能会导致这些系统的性能下降，甚至出现灾难性故障。提出的研究将开发自动化工具和技术来解决在实时多任务系统中开发动态重构的三个基本挑战：i)缺乏能够协同地利用计算、通信和存储的动态重构优势的综合框架；ii)在存在动态重构技术的情况下缺乏有效的实时调度技术；iii)缺乏考虑各种设计和优化约束的动态重构。该研究不是着眼于某一优化目标或特定领域，而是综合考虑计算、通信和存储的协同效应，在不违反实时约束的情况下优化整体性能、能量和可靠性，该项目将在几个方面产生重大的更广泛的影响。所提出的方法可以实现广泛的节能和低成本的嵌入式系统。能源消耗的降低直接关系到电池供电移动设备运行寿命的显著提高。同样，温度的降低提高了系统的可靠性。具有动态重构能力的嵌入式系统将产生两大影响。一个是为公众提供低成本和高质量的日常电器，另一个是提高生物医学和军事设备等安全关键设备的效率。

The complexity of embedded computing systems is increasing at a rapid rate due to both advances in technology and the increasing demand for complex applications. A wide variety of these systems employ multi-core architectures to keep up with the demand for performance under stringent power and thermal constraints. Adaptation to the dynamic changes in the application is critical to improve overall performance, energy efficiency as well as reliability.Although dynamic adaptation (reconfiguration) techniques have received considerable attention from various domains, they are not employed in real-time System-on-Chip (SoC) designs. This is due to the fact that real-time systems may consist of critical tasks with soft or hard deadlines, and the dynamic reconfiguration overhead may adversely affect the critical tasks. Missing deadlines may lead to performance degradation or even catastrophic failures in these systems. The proposed research will develop automated tools and techniques to address three fundamental challenges in exploiting dynamic reconfigurations in real-time multitasking systems: i) lack of a comprehensive framework that can synergistically exploit the advantages of dynamic reconfiguration of computation, communication as well as storage, ii) lack of efficient real-time scheduling techniques in the presence of dynamic reconfigurations techniques, iii) lack of dynamic reconfiguration that considers a wide variety of design and optimization constraints. Rather than focusing on one optimization objective or specific domain, this research takes a comprehensive approach by considering synergistic effects of computation, communication as well as storage to optimize overall performance, energy and reliability without violating real-time constraints.This project will make significant broader impact in several fronts. The proposed approaches can enable a wide range of energy-efficient and low-cost embedded systems. Reduction in energy consumption directly relates to significant improvements in operating lifetime for battery powered mobile devices. Similarly, reduction in temperature improves system reliability. Embedded systems with dynamic reconfiguration capabilities will have two major impacts. One being low-cost and high-quality everyday appliances for the public, and the other is improved efficiency of safety-critical devices such as biomedical and military equipment.