SHF: Small: Physically Adaptive Computing and its Applications

随着集成电路（IC）缩小到纳米范围，温度变化和辐射引起的软错误等不可预测的物理现象开始影响计算机的准确性和可靠性。为了解决这个问题，本项目将研究物理自适应计算（PAC），这是一种使IC能够在线感知内部物理变化并自动重新配置其结构以减轻这些变化的负面影响的新方法。这项工作将利用现场可编程门阵列（FPGA）和纳米PLA等IC类型的灵活性。它还将建立在密歇根大学最近对FPGA应用的研究基础上，以及分析概率电路行为的快速方法。 一个主要目标是获得PAC的深入了解和它在纳米计算机设计中的作用。该项目将研究虚拟传感技术，使用片上传感器来检测和推断各种物理参数。它将寻求更具成本效益的方法，利用可重新编程的电路类型来实现快速、动态的自适应方案。最后，它将研究分析方法，以表征计算精度和关键物理参数，特别是能源使用之间的基本权衡。 理论研究将通过软件仿真和基于FPGA的硬件实验进行补充和验证。预期的结果应该是广泛的兴趣，以计算机工程师和科学家在科隆，以及电路设计师和设计工具开发人员在美国微电子工业和新兴的纳米技术领域。该项目的一个关键目标是支持研究生的培训，他们将直接参与研究，作为他们的MS的一部分。和博士密歇根大学的课程。 该研究所最近的研究生包括妇女和少数民族，将作出特别努力，吸引更多这样的学生到这一新兴的研究领域。

As integrated circuits (ICs) shrink to the nanometer range, unpredictable physical phenomena such as temperature changes and radiation-induced soft errors are starting to affect the accuracy and reliability of computers. To address this problem, this project will investigate physically adaptive computing (PAC), a new way to enable an IC to sense internal physical changes on-line and automatically reconfigure its structure to mitigate the negative effects of these changes. The work will exploit the flexibility of IC types like field-programmable gate arrays (FPGAs) and nanoPLAs. It will also build on recent research at Michigan into FPGA applications, as well as fast methods to analyze probabilistic circuit behavior. A major goal is to obtain a deep understanding of PAC and its role in nanoscale computer design. The project will study virtual sensing techniques that use on-chip sensors to detect and infer a wide range of physical parameters. It will seek more cost-effective ways to exploit reprogrammable circuit types to implement fast, on-the-fly adaption schemes. Finally, it will investigate analytic methods to characterize fundamental trade-offs between computational accuracy and key physical parameters, especially energy usage. The theoretical research will be complemented and validated by software simulations and hardware experiments using FPGA-based equipment. The expected results should be of broad interest to computer engineers and scientists in academe, as well as circuit designers and design tool developers in the U.S. microelectronics industry and in the emerging nanotechnology field. A key goal of the project is to support the training of graduate students, who will participate directly in the research as part of their M.S. and Ph.D. programs at the University of Michigan. Recent graduate students of the PI have included women and minorities, and a special effort will be made to attract more such students to this emerging research area.