Alperen Cakin

Communication-Cost-Aware Task-to-Core Mapping for 2D Concentrated-Mesh NoC-Based Neural Network Accelerators

Task placement strongly affects the on-chip communication overhead of Network-on-Chip (NoC)-based deep-neural network (DNN) accelerators. This work addresses task-to-core mapping on two-dimensional (2D) Concentrated Mesh (CMesh) NoCs with the explicit objective of minimizing communication cost, defined as the traffic-w...