Energy efficiency is a central issue in computing. In large-scale computing clusters, operational and cooling costs impose significant sustainability challenges. Embedded systems run increasingly complex, performance demanding workloads, making the well-known energy management policies inadequate. High power densities also cause high on-chip temperatures and large thermal variations, both of which degrade system reliability.
The research goal of this project is to demonstrate that 3D stacked systems, where multiple chips are vertically attached, can provide dramatic increases in energy efficiency. Realizing this ambitious energy efficiency goal requires novel analysis, design, and runtime management techniques. This project?s objective is creating these catalyst techniques to make 3D systems effective agents for attaining low-power, high-throughput computing in both embedded and large-scale computing domains. Specific research directions are: (1) developing a widely applicable methodology for jointly analyzing the performance, energy, and temperature characteristics of 3D multi-core systems; (2) designing runtime management policies to maximize performance under thermal or power constraints; (3) optimizing liquid-cooled 3D systems to push the performance bounds while maintaining reliable and low-energy operation.
A realizable target for this project in the next decade is a significant reduction in overall energy consumption for computing infrastructure, which translates into monetary savings and carbon footprint reduction. The project also has immediate impact on the system and software design practice. Relevance of the research topic to today?s fast growing computing paradigms such as embedded systems, high-performance computing, and cloud enables constructing a broad educational and outreach plan for K-12 and college students.
