Reliability-Aware DQN-based Optimization Method of Initial Mapping Solutions in Optical Network-on- Chip (ONoC)

Optical Network-on-Chip (ONoC) provides a promising solution for on-chip communication with higher bandwidth and lower latency, compared to traditional Electrical Network-on-Chip (ENoC). However, the reliability challenges (such as insertion loss and crosstalk) exist in ONoC. As for this issue, reliability-aware mapping can be considered. This work proposes a reliability-aware DQN-based optimization method of initial mapping solutions in ONoC, to generate better initial mapping solutions. This optimization method of initial mapping solutions can be combined with existing mapping algorithms, further improving the mapping performance. The evaluation results show that for a given core graph and network topology, our proposed method generates better solutions compared to the traditional random method. Also, the proposed method is helpful for finding the optimal mapping solution more efficiently when combined with the existing mapping algorithms.