CCASM: A Computation- and Communication-Aware Scheduling and Mapping Algorithm for NoC-Based DNN Accelerators

Network-on-Chip-based (NoC-based) Deep Neural Network (DNN) accelerators have been proposed to enable flexible data transmission. However, mapping a DNN model to NoC efficiently can be challenging because of the different communication and computation patterns of each layer. The performance will be degraded if computation is halted at a core by the data-dependent communication. To address this cha...