Differentiated Memory Systems

Catalog description

Memory is the major bottleneck in computers today. 20 years ago, the practical limits of clock speeds forced processors to move to multicore. The bandwidth and capacity limits of low-latency, random-access memory are catalyzing a similarly large shift towards a heterogeneous collection of different memory technologies, optimized for different uses. These Differentiated Memory Systems will require us to revisit and re-examine computing at all levels, from the design of new memory technologies to high-level algorithms. This course is a research seminar that seeks to answer the following three questions: (1) what new memory technologies will we need, how do we compose them in systems, and what caching and consistency mechanisms do they need? (2) how will low-level software present heterogeneous memories to applications and manage them as a resource? (3) how can applications and algorithms guide the composition of heterogeneous memories, and how will they evolve to best use them? Students will read and discuss published research papers and complete an original research project. Open to PhD and masters students as well as advanced undergraduate students.