Research On Optimization Of 3D Optical Network-on-Chip Based On Temperature Sensitivity

In recent years,Network-on-Chip(NoC)based on electrical interconnection has widely used in multi-core interconnection systems due to its efficient communication mode and flexible scalability.However,the development of multi-core processors is limited by the quantum tunneling effect and high energy consumption caused in the traditional NoC.Optical Network-on-Chip(ONoC)based on Optical interconnection gradually becomes an alternative to traditional NoC due to its high transmission bandwidth,low energy consumption and strong scalability.However,the resonant wavelength of microring resonator of silicon optical device in ONoC will shift when affected by temperature,resulting in the reduction of signal-to-noise ratio during transmission and reliability of ONoC operation.Aiming at the effect of temperature on the performance of ONoC,IP core mapping,wavelength allocating,and communication routing are studied.Firstly,the temperature sensitivity of ONoC core devices is studied and analyzed,and edge first IP-core mapping for temperature equalization is proposed.Based on the symmetry of the network,the algorithm extracts the set of asymmetric nodes in the network and sorts them in descending order according to the Manhattan distance of the network hotspots.When start to mapping,the IP cores of the task are mapped to the set of asymmetric point in order of task quantity and task complexity,so that IP cores can be adaptively mapped to network nodes according to the network temperature.Simulation results show that compared with random mapping,ant colony mapping and particle swarm mapping,the proposed mapping is superior in both mapping speed and network energy consumption.Secondly,the crosstalk noise generated by Wavelength Division Multiplexing(WDM)in ONoC is analyzed,and the effect of temperature on Wavelength allocation crosstalk noise is discussed,and a Wavelength allocation method based on improved ant colony algorithm is proposed.After the communication path between IP cores is established,the number and wave spacing of wavelength allocated by each pair of communication nodes are searched according to the complexity of communication between cores and the temperature distribution of the network,and the wavelength is allocated to the communication nodes according to the objective function of the minimum crosstalk.Experimental results show that the proposed wavelength allocation strategy reduces the energy consumption of crosstalk noise by 25% on average compared with the traditional wavelength allocation strategy under different mapping models and network temperature distribution.Finally,the communication routing of ONoC is studied,and it is found that the traditional deterministic routing has the problems of single path planning and uneven channel load.This problem will lead to congestion in the local channel of the network and increase the delay of the network,especially in the case of uneven network temperature distribution,this problem will become more prominent.Therefore,aiming at the problem of communication path optimization in ONoC,this paper proposes a communication routing algorithm based on political optimizer with refraction learning.After getting the result of IP cores mapping and wavelength allocation between IP cores,the routing algorithm adjust the communication path between IP cores according to the channel load and temperature distribution of the network,and obtains the optimal routing result of SNR.The simulation results show that the SNR of the proposed routing algorithm is 7% and 6% higher than that of the dimensional-order and the minimum-loss route,respectively,which proves the effectiveness of the proposed routing algorithm for SNR optimization.