Enabling Emerging, Heterogeneous Memory Systems

Abstract

Optimizing data storage and data movement remain critical roadblocks to overall computing performance and efficiency. These barriers are due to a convergence of motivating factors that begins with the memory wall and limitations of well-established memory technologies and design paradigms, and is compounded by increasingly data-intensive applications. This shift can and should be answered by many research thrusts to address the density and efficiency of memory systems, including technological heterogeneity in on-chip memory, increased specialization and cross-computing-stack choices in system and device design, and re-thinking the relationship and co-location of memory and compute resources. This thesis proposes and evaluates both specific solutions and broad design methodologies, uncovering enormous potential for increased memory system efficiency while unlocking numerous doors for further exploration and innovation.