Techniques that can cost-effectively tolerate transient faults, also known as soft errors, play an essential role in enhancing the reliability of future billion-transistor microprocessors, built using nano-scale devices and advanced integration technologies. Today's design methodologies optimize microprocessor soft-error reliability using approaches within a single-layer (e.g. device, circuit, architecture). Moreover, existing soft-error tolerant mechanisms largely ignore the impact and effect of emerging process technologies (e.g. design parameter variation, 3D vertical integration). An important and urgent research task is to explore novel, cross-layer approaches that will allow microarchitects and circuit designers to cost-effectively mitigate the deleterious impact of soft errors in light of rapidly increasing design complexity and continuously advancing process technologies. This research addresses the above challenge by (1) investigating combined circuit and microarchitecture techniques to cost-effectively mitigate microprocessor soft error susceptibility; (2) studying the impact of process variation on soft error robustness and exploring techniques that can significantly enhance microarchitecture soft error reliability under design parameter variation; and (3) characterizing soft-error behavior in 3D die stacked chips and developing methods to effectively mitigate soft error vulnerability hot-spots in 3D microarchitecture designs, while maintaining desirable reliability/thermal tradeoffs. The outcome of this research will provide a cross-layer solution to tolerate soft erros in future microprocessors. The PI will use the concepts, tools, techniques and other results of this research to introduce graduate and undergraduate students to the nature of soft errors, their impact on software execution and methods to mitigate their impact under emerging process technologies. The developed tools will be used to expose students to quantitative research methods and introduce them to fundamental techniques in reliability evaluation. The PI will actively recruit graduate and undergraduate students from minorities and under-represented groups for this project.
