Exploration of Neural Network-based Joint Method of Mapping and Wavelength Assignment in Optical Network-on-Chip (ONoC)
Hui Li,Yuxiang Niu,Feiyang Liu,Hui LI,Yuxiang NIU,Feiyang LIU
DOI: https://doi.org/10.1364/jocn.489564
IF: 4.508
2023-07-05
Journal of Optical Communications and Networking
Abstract:Optical network-on-chip (ONoC) is promising to provide higher bandwidth and lower latency, compared with the traditional electrical interconnects at either chip-scale or wafer-scale. There is research on the impact of mapping or wavelength assignment on reliability in ONoC. However, mapping and wavelength assignment have an interactive influence on each other, pushing a necessity of research on the joint method. In addition, there are various ways to realize the joint method, which have an influence on the reliability and thus the power efficiency. In this paper, we propose a neural-network-based iterative joint method of mapping and wavelength assignment. Compared to the methods without considering the interactive influence, the proposed iterative joint method based on the continuous Hopfield neural network provides a worst-case optical signal-to-noise ratio (OSNR WC ) improvement of at least 61% under the considered applications. Compared to the simultaneous joint method and two-step joint method, the proposed iterative joint method obtains an OSNR WC improvement of at least 17.9% and 64.6%, respectively, under the considered applications. Thanks to the improvement of OSNR, the laser power is reduced by 87.9% by using our method of wavelength assignment, compared to the random method of wavelength assignment.
computer science, information systems,telecommunications, hardware & architecture,optics