Mapping Deep Neural Networks on SpiNNaker2

EasyChair Preprint 3129

Abstract

SpiNNaker is an efficient many-core architecture for the real-time simulation of spiking neural networks. To also speed up deep neural networks (DNNs), the 2nd generation SpiNNaker2 will contain dedicated DNN accelerators in each processing element. When realizing large CNNs on SpiNNaker2, layers have to be split, mapped and scheduled onto 144 processing elements. We describe the underlying mapping procedure with optimized data reuse to achieve inference of VGG-16 and ResNet-50 models in tens of milliseconds.

Keyphrases: Deep Neural Networks, Neural algorithms and machine learning, SpiNNaker2, neuromorphic hardware

@booklet{EasyChair:3129,
  author    = {Florian Kelber and Binyi Wu and Bernhard Vogginger and Johannes Partzsch and Chen Liu and Marco Stolba and Christian Mayr},
  title     = {Mapping Deep Neural Networks on SpiNNaker2},
  howpublished = {EasyChair Preprint 3129},
  year      = {EasyChair, 2020}}