A Scalable CNN Accelerator with FPGA On-board Resource Optimization for Semantic Segmentation
Authors:
- 10 December 2021
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1158-1162
Abstract
In this paper, a scalable hardware accelerating system deployed on FPGA that can efficiently accelerate various types of convolution for semantic segmentation networks is designed. A hierarchical multiple line buffer structure is proposed to flexibly construct convolution dataflow and utilize FPGA on-chip resources. The accelerating system is validated on both ZYNQ series high performance SoC and XC7A100T board with a RISC-V core. The results show that the proposed system can be easily implemented and migrated to different FPGA platforms with high computation efficiency.Keywords
Funding Information
- Research and Development
This publication has 8 references indexed in Scilit:
- Wire-Aware Architecture and Dataflow for CNN Accelerators2019
- Sticker: A 0.41-62.1 TOPS/W 8Bit Neural Network Processor with Multi-Sparsity Compatible Convolution Arrays and Online Tuning Acceleration for Fully Connected Layers2018
- In-Datacenter Performance Analysis of a Tensor Processing Unit2017
- Deep Residual Learning for Image Recognition2016
- Design space exploration of FPGA-based Deep Convolutional Neural Networks2016
- Fully convolutional networks for semantic segmentation2015
- DianNaoACM SIGARCH Computer Architecture News, 2014
- Gradient-based learning applied to document recognitionProceedings of the IEEE, 1998