Sawtooth: Investigating solutions to memory and storage challenges

The overall goals of power consumption and performance drive the investigation of energy-efficient memory devices and architectures. Experts expect these new memory architectures to have deep memory hierarchies with drastically diverse technical characteristics including capacity, latency, bandwidth, reliability, durability, and persistence. We already see this trend in contemporary heterogeneous architectures, in the near term, the expected DOE CORAL systems at ORNL and ANL will have at least three memory types within a node. The community expects these trends to continue into the foreseeable future. This joint project between Oak Ridge National Laboratory (ORNL) and Micron Technology investigates these challenges with synergistic research tasks in (1) memory devices, (2) memory architectures including deep memory hierarchies, (3) portable and efficient programming systems for complex memory hierarchies, (4) compute-in-memory (CIM) designs, (5) programming systems for CIM, (6) architectures and programming systems for nonvolatile memory (NVM), and (7) application codesign including memory-centric workload analysis in scientific, deep learning, and data-intensive applications in collaboration with domain and software experts.