SHF: Small: Variability-Aware System-Level Power Management in Multi-Processor Systems

随着纳米级CMOS器件和VLSI互连特性的可变性水平的增加以及VLSI电路工作条件的持续不确定性，在工艺、电压和温度变化以及电流应力、器件老化和互连磨损现象下实现电子系统的功率效率和高性能已成为一项艰巨但至关重要的任务。该方案解决了系统级动态电源管理（DPM）的问题，这些系统是由纳米级CMOS技术制造的，并且在系统的生命周期内运行在各种不同的条件下。这些系统受到工艺变化水平增加的极大影响，这些变化通常表现为设备和互连特性中的变异性和磨损/老化效应的内在（随机）或系统来源，以及通常表现为不确定性来源的广泛变化的工作量和温度波动。在系统一级，这种可变性和不确定性开始破坏传统DPM方法的有效性。因此，我们开发具有以下独特特征和能力的可变性感知、减少不确定性的DPM方法的数学基础和实际应用是至关重要的：利用基于可变性敏感、部分可观察马尔可夫决策模型和闭环反馈控制理论的两层随机建模框架，能有效应对系统的可变性，有效降低系统关键参数的不确定性。该框架还允许自学习（自适应）策略优化方法，以及具有多个奖励和成本率的多管理系统，以同时优化系统能耗和性能。成功地克服了这个项目所提出的挑战，将为一台典型的服务器节省大量的能源。这项研究的其他影响包括开发了一个新的强大的数学框架，用于复杂和大型系统的资源管理，该框架可以处理多个代理，多个奖励和成本率以及折扣因素，同时考虑可变性的影响，同时通过测量和抽样减少不确定性的影响。具有闭环反馈控制的随机决策框架也可用于解决多核处理器系统中的动态热控制、并发DPM和任务调度、考虑整个系统？S能效，高效节能的电力输送网络设计。如果成功，该方法可以产生一个实用的随机优化框架，用于处理许多重要问题，从提高电子系统的能源效率（从而降低操作成本）到由大量服务器/存储元素组成的数据中心。教育、推广和培训项目包括新课程；招收代表性不足的学生；本科生科研实习机会；以及针对高中生的初级学者项目。

With the increasing levels of variability in the characteristics of nanoscale CMOS devices and VLSI interconnects and continued uncertainty in the operating conditions of VLSI circuits, achieving power efficiency and high performance in electronic systems under process, voltage, and temperature variations as well as current stress, device aging, and interconnect wear-out phenomena has become a daunting, yet vital, task. This proposal tackles the problem of system-level dynamic power management (DPM) in systems which are manufactured in nanoscale CMOS technologies and are operated under widely varying conditions over the lifetime of the system. Such systems are greatly affected by increasing levels of process variations typically materializing as intrinsic (random) or systematic sources of variability and wearout/aging effects in device and interconnect characteristics, and widely varying workloads and temperature fluctuations usually appearing as sources of uncertainty. At the system level this variability and uncertainty is beginning to undermine the effectiveness of traditional DPM approaches. It is thus critically important that we develop the mathematical basis and practical applications of a variability-aware, uncertainty-reducing DPM approach with the following unique features and capabilities: Utilization of a two-tier stochastic modeling framework based on the theories of variability-sensitive, partially observable Markovian Decision Model and closed-loop feedback control theory, which can efficiently cope with variability and effectively reduce uncertainty in key system parameters. The framework also allows for self-learning (adaptive) policy optimization approaches, and multi-manager systems with multiple reward and cost rates for simultaneous optimization of the system energy consumption and performance.Successfully overcoming the challenges addressed by this project will result in significant energy savings for a typical server. Other impacts of this research includes the development of a new and powerful mathematical framework for resource management in complex and large systems that can deal with multiple-agents, multiple reward and cost rates and discount factors while accounting for effects of variability and simultaneously reducing the impact of uncertainty through measurements and sampling. The stochastic decision making framework with closed loop feedback control is also useful for solving a variety of other problems including dynamic thermal control, concurrent DPM and task scheduling in multi-core processor systems, consideration of total system?s energy efficiency, energy-efficient power delivery network design. If successful, the approach can result in a practical stochastic optimization framework for handling many important problems, ranging from energy efficiency improvement (and hence reduction in cost of operation) for electronics systems to data centers comprised of a large number of server/storage elements. Education, Outreach, and Training Programs include new curricula; recruiting under-represented students; research internship opportunities for undergraduates; and a Junior Scholars program for high school students.