A Single Chip Multiprocessor Intelligent Anti Attack Method Based on Deep Reinforcement Learning
Yiyong Lei, Yanzhang Gu, Xianzhi Yang, Jiayi Lin, Chun Xu · 2025
With the advancement of semiconductor manufacturing technology, hardware components such as the number of cores, cache structure, and on-chip network topology of single chip multiprocessor continue to evolve. The cache hierarchy has also become more complex, expanding from simple first level cache to multi-level cache with increasing capacity. A single, persistent attack and defense strategy is difficult to meet the requirements of single chip multiprocessor attack prevention. Therefore, a single chip multiprocessor intelligent anti attack method based on deep reinforcement learning is proposed. Extract feature data of single chip multiprocessor attacks through convolutional neural networks and select the most effective feature set. Select the proximal policy optimization algorithm as the deep reinforcement learning method, construct an intelligent anti attack value function, and evaluate the corresponding value function based on the anti attack strategy. Optimizing anti attack strategies by learning the behavior patterns and characteristics of attackers until the training of deep reinforcement learning models reaches a convergence state, ultimately obtaining the optimal anti attack strategy. Deep reinforcement learning utilizes near end policy optimization algorithms to construct an intelligent anti attack value function, which can evaluate the corresponding value function based on different anti attack strategies and continuously optimize the anti attack strategies. Through experimental verification, it has been found that the proposed deep reinforcement learning model has fast convergence speed and low network loss. The F1 value of detecting malicious attack behavior using this model is high, and the packet loss robustness is excellent, reflecting the superiority of the proposed method in the field of single chip multiprocessor security protection.