This project will develop a pre-silicon framework for evaluating arithmetic reliability in scientific computing hardware. The project will help hardware designers determine which arithmetic blocks are most vulnerable to silent data corruption, what protections cost in area and power, and which circuit-code combinations offer the best trade-offs for exascale and extreme-environment computing.
2026
Reliable Arithmetic for Scientific Computing: A Circuit-Code Co-Design Framework for Exascale and Extreme Environments
Framework overview. (Left) Arithmetic datapaths constitute a growing fraction of chip area, increasing vulnerability to radiation and aging. (Center) Our hybrid formal-statistical framework evaluates error detectability of combined circuit and code schemes before tape-out. (Right) Target applications span scientific computing, space, and high-reliability systems.
Other Researchers
Nathaniel Bleier, Assistant Professor, Computer Science and Engineering Division
Dennis Sylvester, Professor, Electrical and Computer Engineering Division