A Selective Packet Discard Technique for Efficient Deadlock Recovery in Networks-on-Chip

The concept of Networks-on-Chip (NoCs) was introduced to provide scalable on-chip interconnection infrastructures required for today's increasingly complex Systems-on-Chip. However, NoCs suffer from the problem of deadlocks which is generally approached by deadlock avoidance. As an alternative, this dissertation investigates deadlock recovery techniques to approach message dependent deadlocks. Thereby, the focus lies on regressive recovery in the context of fault tolerance. Furthermore, hierarchical network architectures are studied and optimized for shared resource access.