The Implementation of LeNet-5 with NVDLA on RISC-V SoC
Authors:
- 1 October 2019
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 39-42
Abstract
NVDLA (NVIDIA Deep Learning Accelerator) is an open-source CNN (Convolutional Neural Network) accelerator and has broad application prospects. The implement of Lenet-5 based on NVDLA is presented in this paper, which makes the recognition of handwritten numeral more efficient. Under the condition of no compiler for NVDLA is available so far, the function of registers and the working principles of NVDLA were explored. The LeNet-5 runs on NVDLA successfully through configuring the NVDLA registers. This work could also make a base for the implementation of other deep learning networks, such as AlexNet and VGG. The performance of the accelerator is estimated, and the results of the experiment indicate that Rocket SoC + NVDLA is able to achieve higher CNN processing efficiency, up to 4647x times than RISC-V core.Keywords
This publication has 6 references indexed in Scilit:
- Integrating NVIDIA Deep Learning Accelerator (NVDLA) with RISC-V SoC on FireSim2019
- CoNNA – Compressed CNN Hardware Accelerator2018
- Eyeriss: An Energy-Efficient Reconfigurable Accelerator for Deep Convolutional Neural NetworksIEEE Journal of Solid-State Circuits, 2016
- EIEACM SIGARCH Computer Architecture News, 2016
- DianNaoACM SIGPLAN Notices, 2014
- Gradient-based learning applied to document recognitionProceedings of the IEEE, 1998