Abstract
In contrast to conventional computing-centric applications, CNNs are known to be both computationally and memory intensive. The emerging Processing-in-Memory (PIM) alleviates the memory bottleneck by integrating both processing elements and memory into a 3D-stacked architecture. Although this architecture can offer near-data processing to reduce data movement, memory is still a limiting factor of the entire system. We observe that an unsolved key challenge is how to efficiently allocate convolutions to 3D-stacked PIM to combine the advantages of both neural and computational processing.
Funding Information
  • National Natural Science Foundation of China (61502309)
  • Natural Science Foundation of Guangdong Province (2016A030313045, 2017B030314073)
  • Shenzhen Science and Technology Foundation (JCYJ20150529164656096, JCYJ20170302153955969)
  • Natural Science Foundation of SZU (803/000026080154, 827-000073)
  • State Key Laboratory of Computer Architecture, Institute of Computing Technology, Chinese Academy of Sciences (CARCH201608)