Automatically learning micro-architectural models

Understanding the timing behavior of modern CPUs is crucial for optimizing code and for ensuring timing-related security and safety properties. Unfortunately, the timing behavior of today’s high-performance processors depends on subtle and poorly documented details of their micro-architecture, which has triggered laborious efforts from researchers to build models of different components.

Goals

This project aims at constructing techniques and tools for automatically inferring and learning high-level models of micro-architectural components (like caches and prefetchers) directly from timing measurements.

Avatar
Marco Guarnieri
Assistant professor

Publications

. Flushgeist: Cache Leaks from Beyond the Flush. Technical report, 2020.

PDF Project Extended version (arXiv)