Abstract
Three-dimensional Network-on-Chip (3D NoC) is gaining popularity among designers due to its scalability, higher bandwidth, fault tolerance, and reliability. However, stacking of multiple dies leads to increase in power density and accumulation of heat flux, thereby causing severe thermal problems. Unequal traffic distribution across the chip and higher power density result in higher on-chip temperatures. This leads to performance degradation, increase in leakage power, and circuit failure. In this paper, we have proposed a thermal-aware adaptive routing algorithm called fast multi-objective thermal aware routing (FMoTAR) to optimize the thermal profile of 3D NoCs. The effectiveness of any adaptive routing algorithm is correlated to the selection strategy used for packet forwarding. In the proposed technique, routing function returns multiple admissible output channels. A selection function is used to select the output channel to which a packet will be forwarded. For faster convergence of finding minimal path from source to destination, a bidirectional search is carried out in FMoTAR. Analysis of the simulation results shows that on-chip traffic is more uniform in FMoTAR. This leads to reduction in the peak on-chip temperature. Moreover, the average propagation delay is lesser in FMoTAR in comparison to existing thermal-aware routing techniques.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call