Data traffic management in a reconfigurable Network-on-Chip for Dynamic Neural Networks
Résumé
Deep neural networks (DNNs) play an important role in modern applications. The growing need for their deployment on the edge led to the development of many low-power hardware architectures to accelerate inference. However, those architectures mostly target conventional static DNN models and do not implement features to directly address the challenges of dynamic models.
This paper introduces a methodology for data traffic management in a reconfigurable Network-on-Chip (NoC) for the implementation of a Dynamic Neural Network (DyNN).
Origine : Fichiers produits par l'(les) auteur(s)