Many-core system-on-chips, together with their established communication infrastructures, Networks-on-Chip (NoC), are growing in complexity, which encourages the integration of third-party components to simplify and accelerate production processes. However, this also adversely exposes the surface for attacks through the injection of hardware Trojans. This work addresses active attacks on NoCs and focuses on the integrity and availability of transmitted data. In particular, we consider the modification and/or dropping of data during transmission as active attacks that might be performed by malicious routers. To mitigate the impact of such active attacks, we propose two lightweight solutions that respect the performance constraints of NoCs. Assuming the presence of symmetric keys, these approaches combine lightweight authentication codes for integrity protection with network coding for increased efficiency and robustness. The proposed solutions prevent undetected modifications and significantly increase availability through a reliable detection of attacks. The efficiency of these solutions is investigated in different scenarios using cycle-accurate simulations and the area overhead is analyzed relative to state-of-the-art many-core system. The results demonstrate that one authentication scheme with network coding protects the integrity of data to a low residual error of at 0.2 attack probability with an area overhead of . For faster and more flexible evaluation, an analytical approach is developed which is validated against the cycle-accurate simulations. The analytical approach is more than 1000× faster while having a maximum estimation error of . Moreover, the analytical model provides a deeper insight into the system’s behavior. For example, it reveals which factors influence the performance parameters.
Any data transmission over multiple hops requires the use of routing algorithms to find a path from the sender to the receiver(s). Supporting secure data transmissions further requires the selected path to be trustworthy, i.e., not under the control of an attacker. Since attacks can occur at any time during data transmission, the current state of the network must be considered when selecting trustworthy paths. In this paper, we introduce a novel metric – called Locally Evaluated Trust (LET) – for local trust evaluation. The LET metric can be used by adaptive routing algorithms to select a trustworthy path. To evaluate our approach, we simulate different routing algorithms and compare their performance in the presence of one or more malicious nodes in a two-dimensional torus mesh. Our results show that the LET metric allows adaptive routing algorithms to effectively identify and circumvent malicious nodes.