Modeling Adversarial Strategies in GPU Side-Channel Attacks: A Game-Theoretic Perspective

The use of Graphics Processing Units (GPUs) in safety-critical applications is increasing; yet, their highly parallel designs provide weaknesses that facilitate side-channel attacks, enabling the theft of confidential information. Prior research has shown that time, contention, power, and access pattern side-channels are viable against GPU workloads. Nonetheless, their current protections remain insufficient to comprehensively safeguard real-world shader operations. A novel game-theoretical framework is proposed to model and evaluate the intricate dynamics between attackers and defenders in GPU side-channel assaults. The interactions are reformulated as a two-player, non-cooperative game, and the best tactics for both players have been established under various payment models and danger situations. Experimental results obtained from commercial GPUs corroborate that the suggested technique effectively represents hostile dynamics in reality, necessitating the development of resilient countermeasures. This project seeks to bridge the gap between theoretical security analysis and practical GPU protection methods. We provide a robust foundation for designing appropriate and secure GPU architectures for continuously changing side channel threadings. Human-centric approaches to driving work emphasise the need of considering human elements like as perception, judgement, and decision-making to effectively assess adversary techniques in cybersecurity. The game theoretical model offers a structured approach for forecasting probable assault vectors, assessing defence methods, and formulating effective counters customised to the adversary.