Game-Theoretic Modeling of Adversarial Strategies in GPU Side-Channel Attacks

Abstract

Graphics Processing Units (GPUs) used in safety-critical applications are growing, but their massively parallel architectures introduce vulnerabilities that allow side-channel attacks to steal secret information. Previous studies have established that the timing, contention, power, and access pattern side-channel are feasible against GPU workloads. However, their existing defences are still inadequate to protect real-world shader executions systematically. A new and inventive game-theoretical methodology is suggested to simulate and appraise the complex interplay between attackers and defenders in GPU side-channel attacks. The interactions are recast as a two-player, non-cooperative game, and the optimal strategies of both players have been determined under different payoff models and threat scenarios. Results from experiments conducted on commercial GPUs further validate that the proposed method accurately depicts adversarial dynamics in real life, urging the creation of robust countermeasures. This research aims to close the gap between theoretical security analysis and pragmatic GPU defence mechanisms. We give a rigid base to design suitable and safe GPU architectures for constantly evolving side channel threadings. Driving work through humans follows a human-centred point of view, insisting that thinking about human factors like perception, judgement, and decision-making is essential to analyse adversarial strategies in the cybersecurity domain. The game theoretical model provides a systematic framework for predicting the most likely attack vectors, evaluating defence strategies, and developing robust countermeasures tailored to the adversarial environment.