SHF: Small: Novel Architecture Energy Harvesting for Sustainable Spot Cooling and Energy Management

点击翻译按钮获取中文摘要

Increased power dissipation in computing devices has led to a sharp rise in thermal hot spots on computer chips, creating a vicious cycle ? higher temperatures bring higher leakage power, and higher power dissipation increases temperature, thereby leading to thermal avalanches. To reduce the additional power dissipation and reliability concerns caused by high temperature, current heat management approaches apply cooling mechanisms to remove heat aggressively as well as devise dynamic management techniques that avoid thermal emergencies by slowing down heat generation of processors. However, current trends to squeeze more computing power, e.g., in the form of large data centers or mobile devices, stand in direct conflict to our ability to slow down demand for more energy. The shrinking of transistor sizes further exacerbates the problem of reduced energy efficiency. The solution proposed in this work is anticipated to not only reduce cooling expenses and ambient temperatures, but also increase energy utilization, device lifetime, and physical space utilization. The technology developed here can be applied to a broad range of computing devices, large or small. If the research is successful, it has the potential of having a significant economic benefit as well as a significant, positive impact on the environment. This is because performance improvement and power reduction of processors under thermal constraints will have a direct impact upon the cooling costs of huge data warehouses such as those of Google, Yahoo, Amazon, etc. Data centers in the US consume many tens of billion kWh of electricity and generate about nearly a billion metric tons of carbon dioxide. Even if this project resulted in a 5% improvement in the energy consumption of a modern high performance processor and therefore, in the millions of such processors housed in data centers, that itself could reduce the amount of carbon dioxide released into the atmosphere per year, and realize ten of millions of dollars in energy cost savings. Furthermore, this energy harvesting research requires cross-disciplinary engagement in areas such as material engineering, VLSI architecture, system architecture, and mechanical engineering and will attract a diverse set of student researchers. Overall, the engineering and scientific contributions will also have important societal impacts, including the broadening of ASU?s engineering curriculum, the engagement of graduate and undergraduate students in research activities, the potential of creating high-school or middle-school scientific projects, and the increased representation of target underrepresented minorities in science and engineering.This project addresses the heat management problem using an innovative approach ? rather than removing heat or slowing down heat generation, the proposed work transforms the waste heat into reusable energy for new applications such as self-powered spot-cooling. The main objective of this project is to design and implement a novel architectural framework to create the mechanisms, policies, and system support that allow waste heat generated by computing devices to be harvested efficiently, to achieve better energy utilization efficiency. This will be achieved by exploiting the thermal characteristics of modern computing nodes and by leveraging thermoelectric and pyroelectric energy harvesting materials. By leveraging the thermoelectric and pyroelectric effects at the architectural level, the varying spatial and temporal thermal gradients from computations are exploited to transform processor waste heat (that otherwise dissipates) into reusable energy. A novel application is also proposed that uses the newly introduced energy in the form of a self-sustaining cooling system for processors. This work evaluates the applicability of energy harvesting materials by considering the intricate electrical properties of the materials and heterogeneous temperature distribution of the components on a processor. The proposed methodology is generic and can be readily adopted with commercially-available thermoelectric and pyroelectric energy harvesting materials. Nonetheless, with breakthroughs in the energy conversion efficiency of the materials, the proposed framework could be applied directly with a further improved degree of harvested energy, leading to even higher system energy efficiency.