Communication cost in terms of energy consumption and network latency, along with the dynamic allocation time of the tasks and their execution time, is the primary design concern of a run-time mapping and scheduling strategy that may significantly affect the overall performance of an application. The present work proposes a dynamic mapping and scheduling algorithm based on the Cube-Tree-Hybrid (CTH) topology. The CTH network can integrate a large number of IP cores under significantly low network diameter, as opposed to mesh. Moreover, the design of CTH comes with considerable path diversity. For the above properties, mapping on CTH becomes simpler than mesh, which successfully minimizes the communication hop distance amongst various tasks of an application. Extensive experimentation has been done across several synthetic and real application workloads. Compared to the prevalent mesh-based mapping techniques, the proposed algorithm achieves a minimum gain of 56% on communication latency, whereas it is 36% in total power consumption. Minimum 14% improvement on applications' deadline satisfaction is achieved.

The proposed work demonstrates a novel flow control mechanism that brings the flexibility to route every flit of a packet independently, alleviating the network congestion more efficiently in Virtual Channel (VC) based NoC routing. A 100% buffer utilization is achieved by putting an end to the overhead of VC arbitration. The modified router design comes with a 20% area and 4.73% energy saving on the generic one. Packet Latency improvement of 35 -78 %, 27 - 68%, and 45 - 62% is observed over the mesh, torus, and flattened butterfly topologies respectively under uniform random traffic. The average packet drop rate is reduced by a minimum of 28% across the above topologies.