Multi-vector Monitoring, Detecting and Classifying GPU Side-Channel Attack Vectors on a Secure GPU Execution Framework

Graphics Processing Units (GPUs) are increasingly deployed in security-sensitive domains, but their parallel architectures introduce vulnerabilities enabling side-channel attacks to steal sensitive data. Recent works have shown feasible timing, contention, power, and access pattern side-channel attacks against cryptographic implementations, randomness protections, and other workloads on GPUs, highlighting risks of intellectual property theft, cryptographic key extraction, and spying. However, existing academic defences remain inadequate in systematically protecting real-world shader execution. This research aims to design, develop, and evaluate a secure GPU shader execution framework incorporating runtime monitoring, anomaly detection, and response mechanisms for commercial GPUs. The methodology involves implementing and assessing the framework against simulated side-channel attacks on shader benchmarks; this can help secure emerging GPU adoption in sectors like finance, healthcare, and government against rising data breaches.