Transistor size is gradually decreasing. This fact enables fabrication of chips consisting of multiple heterogeneous components. In the past, components were sharing a bus to communicate with each other. However, such a bus structure is optimal only for a certain configuration (number of components, traffic patterns between components.). To cope with this problem bus-based communication architecture is replaced by new and more efficient approach, i.e. Networks-on-Chip (NoC). The benefit of a NoC lies in its regularity; it can be easily extended if more components shall be placed onto the chip. However, together with the great benefit of scaling of microchip technology problems arise. The probability of permanent and transient faults is increasing, thus countermeasures are required to ensure reliable communication among the network components. For this reason, Fault Tolerance (FT) is considered to be a necessity in current MPSoC designs.

In this thesis a FT End-to-End protocol is proposed, which ensures reliable communication in presence of transient and permanent faults. The protocol mechanisms are divided in two parts. The first part deals with recovery of corrupted and lost packets due to transient faults. Certain parameters are investigated to reduce the protocol overhead and minimize the latency in case of faults. The second part describes a new approach to locate permanent faults that occur during runtime by means of relaxation of End-to-End protocol. In this way it is possible to ensure a reliable communication with a minimum performance of degradation.