YANG Wei-ke, HE Guang-hui, JING Nai-feng. Design and Implementation of CNN Acceleration Module Based on Rocket-Chip Open Source Processor[J]. 2018, 35(4): 17-21.
YANG Wei-ke, HE Guang-hui, JING Nai-feng. Design and Implementation of CNN Acceleration Module Based on Rocket-Chip Open Source Processor[J]. 2018, 35(4): 17-21.DOI:
Based on the RISC-V open source instruction set and Rocket-Chip open source processor
we propose a convolution neural network acceleration module based on Eyeriss structure and form a complete system with the processor connection. The accelerator structure greatly reduces the time consumption of the convolution layer by performing data reuse in the lateral (convolution core weight)
longitudinal (output calculation result) and oblique (input image). In addition
the use of the UCB developed Chisel3 language
on the one hand to generate the system C language simulator for debugging
on the other hand generate Verilog for integrated and layout and routing. Through the LeNet-5 handwritten digital identification network and MNIST data set
we achieved satisfactory results on speed and accuracy.
关键词
Keywords
references
陆志坚. 基于FPGA的卷积神经网络并行结构研究[D]. 哈尔滨: 哈尔滨工程大学, 2013.
Asanovic K, Avizienis R, Bachrach J, et al. The rocket chip generator[R]. EECS Department, University of California, Berkeley, 2016.
Chen Y H, Krishna T, Emer J S, et al. Eyeriss: an energy-efficient reconfigurable accelerator for deep convolutional neural networks[J]. IEEE Journal of Solid-State Circuits, 2017, 52(1): 127-138. DOI: 127
Waterman A, Lee Y, Patterson D A, et al. The risc-v instruction set manual, volume i: Base user-level isa[J]. Eecs Department, 2011, 7(9): 475 DOI: 475
Chen Y H, Emer J, Sze V. Eyeriss: A spatial architecture for energy-efficient dataflow for convolutional neural networks[C]//Computer Architecture (ISCA), 2016 ACM/IEEE 43rd Annual International Symposium on. Seoul, South Korea, IEEE, 2016: 367-379.
LeCun Y, Bottou L, Bengio Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324. DOI: 2278