SHF: Medium: Collaborative Research: Ultra-Responsive Architectures for Mobile Platforms

传统的微处理器主要是为可持续性能而设计的；它们基本上可以无限期地以接近峰值的性能运行，直到它们的能源耗尽。在电池和冷却受限的环境中，如移动设备，持续功率(以及相应的峰值性能)必须限制在最多几瓦特，以便设备可以仅使用被动换热来散热。然而，许多交互式移动应用程序(如书写/语音识别或基于图像的搜索)要求响应用户输入进行密集计算，从而产生了对新的超响应操作机制的需求：系统不应在假设持续运行的情况下限制峰值功率，而应利用热存储来实现短暂的计算突发，这种突发大大超过了可持续的热极限，而不会过热。这个项目研究了一种被称为“计算冲刺”的方法，其核心本质是以不可持续的速度进行计算，但只是短暂的，这样温度就不会达到不安全的水平。这个项目的目标是解决通过计算冲刺来实现超响应的建筑、热、电气和软件障碍。该项目尤其探索了：(1)通过并行性(激活数十个备用功能单元/核心)和电压提升(超速驱动核心以实现单线程性能)进行冲刺的架构和存储系统；(2)改进热响应行为以实现更长时间和更高强度冲刺的热设计；(3)移动优化的电子设计，其提供稳定的电源电压，尽管电流激增一个数量级或更多；以及(4)明确管理有限的热预算并预测和准备冲刺期间所需数据的软件机制。该项目包括制造一个实验试验台，以接近未来多核可移动设备的计算、热和电性能，并提供冲刺实践经验。项目影响包括：(1)开发和推进提高移动设备响应能力的技术；(2)将这些发现整合到新的跨部门计算机系统设计课程中。

Conventional microprocessors are designed primarily for sustainedperformance; they can operate at near-peak performance essentiallyindefinitely until their energy source is exhausted. In battery andcooling constrained environments such as mobile devices, sustained power(and, consequently, peak performance) must be limited to at most a fewwatts so that the device can dissipate heat using only passiveconvection. However, many interactive mobile applications (such ashandwriting/speech recognition or image-based search) instead call forbursts of intense computation in response to user input, creating theneed for a new ultra-responsive operating regime: rather than limit peakpower assuming sustained operation, systems should instead exploit heatstorage to enable brief computation bursts that greatly exceedsustainable thermal limits without overheating. This projectinvestigates an approach called "computational sprinting", the centralessence of which is to compute at unsustainable rates but only brieflyso that temperatures do not reach unsafe levels.The goal of this project is to address the architectural, thermal,electrical, and software barriers to ultra-responsiveness viacomputational sprinting. In particular, the project explores: (1)architectures and memory systems that sprint via parallelism (activatingtens of reserve functional units/cores) and voltage boosting(overdriving cores for single-thread performance) while facilitatingfast burst onsets; (2) thermal designs that improve thermal responsebehavior to enable longer and more intense sprints; (3) mobile-optimized electrical designs that provide stable supply voltagesdespite current surges of an order-of-magnitude or more; and (4) software mechanisms that explicitly manage limited thermal budgets and anticipate and stage data needed during the sprint. The projectincludes the fabrication of an experimental testbed to approximate thecomputational, thermal and electrical capabilities of a future many-coremobile device and to provide practical experience with sprinting.Project impacts include (1) developing and advancing techniques forimproving the responsiveness of mobile devices and (2) integrating thediscoveries into a new cross-departmental course on computer systemdesign.