Detecting hardware attacks in RISC-V with on-chip performance counters

Abstract

Poster presentado en HiPEAC 2026 (Cracovia)

Malware is evolving faster than traditional defenses. Signature-based tools often miss new polymorphic and metamorphic variants, leaving modern systems exposed. Embedded platforms add further constraints, demanding lightweight and autonomous protection. Processor Hardware Performance Counters (HPCs) reveal microarchitectural footprints of malicious activity. When combined with Machine Learning, they enable real-time, low-overhead malware detection—even on resource-constrained devices. Across five evaluated ML models, a Random Forest trained on five key HPCs achieves over 99.5% accuracy and detects nearly all unseen attacks, demonstrating strong precision and generalization.

Funded by the European Union. Project number: 101093062.

Preprint