Neural Network-Based Prediction of Optical Input Power for Reliability-Aware Green Communication in Silicon Photonic Interconnects on Chip

Silicon photonic interconnects on chip is a promising technology for manycore systems, with the characteristics of high bandwidth and low latency. During the communication process from the source to the destination, the optical signal encounters signal losses and crosstalk, which would have a negative influence on the reliability at the destination and in turn impact the power efficiency. To achieve green communication in silicon photonic interconnects on chip, we propose a reliability-aware prediction method of optical input power based on neural network. To conquer two types of errors, model error correction and environment error correction are proposed. The results show that the proposed method can save power and fulfill the requirements of communication reliability at the same time. Compared to the common method of power setting in state-of-art work, the optical input power is saved by 83.64% in average by applying our proposed prediction method, while satisfying the reliability requirements.