Prediction of instruction SDC vulnerability in routing algorithms based on graph convolutional network

The transient faults caused by high-energy particles in the space radiation environment impact the reliability of routing algorithms in network communication mechanisms, where SDC (Silent Data Corruption) poses a significant threat to space-based network systems due to its insidious nature. Given that existing SDC fault tolerance methods do not incorporate the characteristics of space-based network communication mechanisms and require time-consuming fault injections, this paper proposes a method for strengthening the SDC vulnerability of network routing algorithms based on GCN (Graph Convolutional Networks). First, the impact of transient faults on routing algorithms is analyzed, and a fault model of the routing algorithm is constructed. Next, the instruction features and instruction dependencies of the routing algorithm are extracted to construct an instruction dependency graph. Then, a GCN-based instruction SDC vulnerability prediction model is constructed and trained to predict the SDC vulnerability of the routing algorithm instructions. Finally, the routing algorithm is reinforced to enable self-detection and fault tolerance of SDC. Experimental results show that, compared to existing methods, the proposed method achieves higher accuracy in SDC vulnerability prediction and detection rates without requiring large-scale fault injection.