CPS: Synergy: Collaborative Research: Thermal-Aware Management of Cyber-Physical Systems

网络物理系统中的处理器越来越多地用于必须在恶劣环境条件和可能导致芯片高温的计算工作负载下运行的应用中。例子包括汽车、机器人、飞机和宇宙飞船。高温会加速芯片的老化，从而增加瞬态和永久故障率。当前处理这个问题的方法主要是关闭处理器核心，或者在看到某些部分超过给定的温度阈值时大幅降低处理器速度。然而，这种通过/失败方法忽略了以下事实：(a)处理器由于高温而加速老化，即使这些温度低于阈值；(b)虽然最后期限是实时任务的约束，以保持受控设备在允许的状态空间中，但如果降低电压或频率（以冷却芯片），实际控制器响应时间将增加，这是决定受控设备性能的因素。现有的方法也不能利用控制器可靠性（受其温度历史影响）和设备性能之间的权衡。这个项目解决了这些问题。正在设计负载成形算法来管理热应力，同时确保适当的控制质量水平。这些动作包括任务迁移、改变执行速度、选择控制功能的替代算法或软件实现，以及过早地终止迭代任务的可选部分。该项目的验证平台包括汽车和无人机。这些平台的选择是基于它们对社会的重要性和它们带来的重大技术挑战。随着CPS在我们的生活和商业中变得越来越重要，这个项目将使CPS控制器更加可靠和/或经济，具有广泛的潜在社会和经济影响。与通用汽车公司的合作促进了新技术向工业的过渡。该项目包括在高中、本科和研究生课程中向学生介绍计算机热控制的活动。

Processors in cyber-physical systems are increasingly being used in applications where they must operate in harsh ambient conditions and a computational workload which can lead to high chip temperatures. Examples include cars, robots, aircraft and spacecraft. High operating temperatures accelerate the aging of the chips, thus increasing transient and permanent failure rates. Current ways to deal with this mostly turn off the processor core or drastically slow it down when some part of it is seen to exceed a given temperature threshold. However, this pass/fail approach ignores the fact that (a) processors experience accelerated aging due to high temperatures, even if these are below the threshold, and (b) while deadlines are a constraint for real-time tasks to keep the controlled plant in the allowed state space, the actual controller response times that will increase if the voltage or frequency is lowered (to cool down the chip) are what determine the controlled plant performance. Existing approaches also fail to exploit the tradeoff between controller reliability (affected by its temperature history) and the performance of the plant. This project addresses these issues. Load-shaping algorithms are being devised to manage thermal stresses while ensuring appropriate levels of control quality. Such actions include task migration, changing execution speed, selecting an alternative algorithm or software implementation of control functions, and terminating prematurely optional portions of iterative tasks. Validation platforms for this project include automobiles and unmanned aerial vehicles. These platforms have been chosen based on both their importance to society and the significant technical challenges they pose.With CPS becoming ever more important in our lives and businesses, this project which will make CPS controllers more reliable and/or economical has broad potential social and economic impacts. Collaboration with General Motors promotes transition of the new technology to industry. The project includes activities to introduce students to thermal control in computing, in courses spanning high-school, undergraduate and graduate curricula.