UC Merced

Speculative execution enhances performance but introduces security risks by leaving observable microarchitectural side effects that attackers can exploit. Spectre-class attacks manipulate speculative execution to leak sensitive data, and since the discovery of Spectre V1 and V2, numerous new variants have emerged, bypassing existing defenses and expanding the attack surface. To counter these threats, researchers have proposed mitigation techniques to constrain speculative execution and detection methods to identify Spectre exploits. Mitigation techniques often introduce significant performance overhead or require hardware changes. Runtime detection has gained growing attention as a practical alternative, using hardware performance counters (HPCs) and machine learning (ML) models to identify anomalies during speculative execution.

This survey systematically explores recent Spectre variants and analyzes Spectre runtime detection, categorizing techniques into ML-based, deep learning-based, and hybrid approaches. We evaluate Spectre variant coverage and highlight dataset limitations that impact detection effectiveness. Our findings provide insights into current challenges and outline future research directions to enhance adaptive and scalable runtime Spectre detection methods.