Learning-Enabled NoC Design for Heterogeneous Manycore Systems

Abstract

As systems grow in specialization (e.g., domain specific architectures), we need the tools to handle the growing design space from increased heterogeneity and system sizes. In this paper, we investigate the specific challenges posed by heterogeneous systems on the NoC in two separate contexts: wireless- and 3D-enabled, formulate each as a separate multiobjective optimization problem, and present a machine learning based design space exploration technique, MOO-STAGE, to intelligently explore this growing design space.