A Thermal-aware Power Optimization And Simunation Approach For Optical NoCs

For future chip multiprocessor systems,more and more processor cores would be integrated on a single chip.Optical networks-on-chip(NoC)have been proposed as an emerging on-chip interconnect architecture because of its high bandwidth density,low power consumption,and low latency.However,due to the thermo-optic effect,lasers as well as wavelength-selective optical switches suffer from temperature-dependent wavelength shift.The wavelength shift will cause significant additional optical power loss in optical NoCs.Considering the on-chip temperature variations,the thermal sensitivity has been one of the main bottlenecks in optical NoC designs.In order to tackle the thermal issue,firstly we analyzed and modeled the thermal effects in optical NoCs,from the device level to the network level.Then,we proposed a learning-based thermal-aware adaptive routing algorithm for loss optimization in optical NoCs in presence of on-chip temperature variations.The proposed routing algorithm can find an optimal path with the minimum optical loss.Finally,in order to verify our proposed thermal-aware power optimization and simulation approach,we developed a thermal-network co-simulation platform for optical NoCs.The co-simulation platform can simulate the on-chip temperature distribution,and the real-time on-chip temperature distribution is fed back to the network simulator.We applied the proposed thermal-aware adaptive routing method to an 8x8 mesh-based optical NoC under several temperature scenarios,and compared with the XY routing,the Odd-Even routing and the West-First routing.Simulation results show that it is able to find optimal paths with the minimum optical loss.