High-bandwidth, high-capacity, low-power memory systems

In large systems, scale is determined by the memory needed: users of supercomputers choose the number of nodes based on the amount of DRAM they will receive; administrators in data centers and enterprise computing run the largest workload possible before paging makes performance unacceptable. These systems are not compute-bound; they are memory-bound. This talk will discuss several of the recent solutions that our group has helped to develop, including flash-based main memory systems and Micron's Hybrid Memory Cube DRAM.