Spectre Attacks Detection in RISC-V Processors Using Hardware-Level Machine Learning Models


ŞİRİN M., GÜVEN E. Y.

10th European Interdisciplinary Cybersecurity Conference, EICC 2026, Darmstadt, Almanya, 17 - 18 Haziran 2026, cilt.3012 CCIS, ss.201-211, (Tam Metin Bildiri)

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Cilt numarası: 3012 CCIS
  • Doi Numarası: 10.1007/978-3-032-28957-5_12
  • Basıldığı Şehir: Darmstadt
  • Basıldığı Ülke: Almanya
  • Sayfa Sayıları: ss.201-211
  • Anahtar Kelimeler: Attack Detection, Hardware Attacks, Hardware Performance Counters, Spectre Detection
  • İstanbul Üniversitesi-Cerrahpaşa Adresli: Evet

Özet

Modern processors used features such as speculative execution and branch prediction to enhance performance; however, these mechanisms leave microarchitectural traces when mispredictions occur, enabling attacks such as Spectre. The Spectre attack has had a significant impact on the hardware security domain and has emerged as a vulnerability affecting almost all contemporary processors. To mitigate its effects, various methods have been developed to detect and defend against Spectre attacks. Owing to their capability to observe hardware-level events, hardware performance counters (HPC) stand out as a valuable resource for detecting Spectre attacks. In this study, we investigate the detection of Spectre-v2 attacks using HPC-based machine learning models. With our models trained with data collected on RISC-V architecture, we achieved Spectre attack detection with over 99% accuracy. Our findings demonstrate that traces of Spectre-v2 are distinguishable within HPC data and that HPC-based models are effective in attack detection.