A Customized NoC Architecture to Enable Highly Localized Computing-on-the-Move DNN Dataflow

The ever-increasing computation complexity of fast-growing Deep Neural Networks (DNNs) has requested new computing paradigms to overcome the memory wall in conventional Von Neumann architectures. emerging Computing-In-Memory (CIM) architecture been a promising candidate accelerate neural network computing. However, data movement between CIM arrays may still dominate total power consumption designs. This brief proposes flexible processor named Domino and “Computing-On-the-Move” (COM) dataflow, enable stream local access significantly reduce energy. Meanwhile, employs customized distributed instruction scheduling within Network-on-Chip (NoC) implement inter-memory attain mapping flexibility. evaluation with prevailing DNN models shows that achieves 1.77-to- $2.37\times $ efficiency over several state-of-the-art accelerators improves throughput by 1.28-to- notation="LaTeX">$13.16\times .