Federal Grant Award Summary

The University of California, Los Angeles received an $800,000 Project Grant award from the National Science Foundation (NSF) Division of Information and Intelligent Systems under the Computer and Information Science and Engineering (CISE) program (CFDA 47.070), effective May 1, 2026 through April 30, 2030. This collaborative research project delivers a formal foundation and system support infrastructure for detecting silent computation errors (SCEs) in scientific high-performance computing (HPC) applications. The research addresses the critical challenge of microprocessor reliability degradation in large-scale HPC deployments, where SCEs can corrupt computational results without user detection. The project produces novel end-to-end, modular, and verifiable SCE detection runtime monitoring systems that reduce overhead compared to existing approaches while providing formal mathematical guarantees of detection correctness.

The core deliverables include formal foundations for reasoning about CPU SCEs with mechanized proofs validating detection mechanism correctness, control-path abstractions enabling modular error detection across application fragments, and runtime monitoring systems optimized for efficiency across HPC environments. By decomposing HPC applications into fragments with special properties amenable to monitoring, the research enables both overhead reduction and provably correct error detection. These outputs directly support scientific discovery by enabling researchers and HPC practitioners to trust their computational results and draw robust conclusions from large-scale scientific computing initiatives.