An efficient loop tiling framework for convolutional neural network inference accelerators
Abstract Convolutional neural networks (CNNs) have been widely applied in the field of computer vision due to their inherent advantages in image feature extraction. However, it is difficult to implement CNNs directly on embedded platforms owing to excessive calculations of CNNs. Field Programmable G...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2022-01-01
|
| Series: | IET Circuits, Devices and Systems |
| Subjects: | |
| Online Access: | https://doi.org/10.1049/cds2.12091 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832546806397927424 |
|---|---|
| author | Hongmin Huang Xianghong Hu Xueming Li Xiaoming Xiong |
| author_facet | Hongmin Huang Xianghong Hu Xueming Li Xiaoming Xiong |
| author_sort | Hongmin Huang |
| collection | DOAJ |
| description | Abstract Convolutional neural networks (CNNs) have been widely applied in the field of computer vision due to their inherent advantages in image feature extraction. However, it is difficult to implement CNNs directly on embedded platforms owing to excessive calculations of CNNs. Field Programmable Gate Arrays have been popular in CNN accelerators because of their configurability and high energy efficiency. Given the highly parallel workloads of the CNN, a CNN accelerator with a 14 × 16 processing element array is designed in this study to accelerate the CNN inference. Besides, a loop tiling strategy for convolutional layers is proposed to efficiently transmit feature maps. Additionally, the roofline model is employed to explore the best tiling parameters for optimal performance. Finally, the accelerator written in Verilog‐HDL language is implemented on the Xilinx Zynq‐7045 evaluation platform. At an operating frequency of 200 MHz, the proposed accelerator can achieve a performance of 57.24 giga operations per second on You Only Look Once v2‐tiny and 78.39 GOPS on Visual Geometry Group‐16. The accelerator only consumes 224 DSPs, demonstrating a better performance compared with the previous works. |
| format | Article |
| id | doaj-art-5c266b81fdb642c6b995cbffb08579c8 |
| institution | Kabale University |
| issn | 1751-858X 1751-8598 |
| language | English |
| publishDate | 2022-01-01 |
| publisher | Wiley |
| record_format | Article |
| series | IET Circuits, Devices and Systems |
| spelling | doaj-art-5c266b81fdb642c6b995cbffb08579c82025-02-03T06:47:11ZengWileyIET Circuits, Devices and Systems1751-858X1751-85982022-01-0116111612310.1049/cds2.12091An efficient loop tiling framework for convolutional neural network inference acceleratorsHongmin Huang0Xianghong Hu1Xueming Li2Xiaoming Xiong3School of Automation Guangdong University of Technology Guangzhou ChinaSchool of Microelectronics Guangdong University of Technology Guangzhou ChinaSchool of Automation Guangdong University of Technology Guangzhou ChinaCompany of Chipeye Microelectronics Foshan Ltd. Foshan ChinaAbstract Convolutional neural networks (CNNs) have been widely applied in the field of computer vision due to their inherent advantages in image feature extraction. However, it is difficult to implement CNNs directly on embedded platforms owing to excessive calculations of CNNs. Field Programmable Gate Arrays have been popular in CNN accelerators because of their configurability and high energy efficiency. Given the highly parallel workloads of the CNN, a CNN accelerator with a 14 × 16 processing element array is designed in this study to accelerate the CNN inference. Besides, a loop tiling strategy for convolutional layers is proposed to efficiently transmit feature maps. Additionally, the roofline model is employed to explore the best tiling parameters for optimal performance. Finally, the accelerator written in Verilog‐HDL language is implemented on the Xilinx Zynq‐7045 evaluation platform. At an operating frequency of 200 MHz, the proposed accelerator can achieve a performance of 57.24 giga operations per second on You Only Look Once v2‐tiny and 78.39 GOPS on Visual Geometry Group‐16. The accelerator only consumes 224 DSPs, demonstrating a better performance compared with the previous works.https://doi.org/10.1049/cds2.12091field programmable gate arraysconvolutional neural netshardware acceleratorsinference mechanisms |
| spellingShingle | Hongmin Huang Xianghong Hu Xueming Li Xiaoming Xiong An efficient loop tiling framework for convolutional neural network inference accelerators IET Circuits, Devices and Systems field programmable gate arrays convolutional neural nets hardware accelerators inference mechanisms |
| title | An efficient loop tiling framework for convolutional neural network inference accelerators |
| title_full | An efficient loop tiling framework for convolutional neural network inference accelerators |
| title_fullStr | An efficient loop tiling framework for convolutional neural network inference accelerators |
| title_full_unstemmed | An efficient loop tiling framework for convolutional neural network inference accelerators |
| title_short | An efficient loop tiling framework for convolutional neural network inference accelerators |
| title_sort | efficient loop tiling framework for convolutional neural network inference accelerators |
| topic | field programmable gate arrays convolutional neural nets hardware accelerators inference mechanisms |
| url | https://doi.org/10.1049/cds2.12091 |
| work_keys_str_mv | AT hongminhuang anefficientlooptilingframeworkforconvolutionalneuralnetworkinferenceaccelerators AT xianghonghu anefficientlooptilingframeworkforconvolutionalneuralnetworkinferenceaccelerators AT xuemingli anefficientlooptilingframeworkforconvolutionalneuralnetworkinferenceaccelerators AT xiaomingxiong anefficientlooptilingframeworkforconvolutionalneuralnetworkinferenceaccelerators AT hongminhuang efficientlooptilingframeworkforconvolutionalneuralnetworkinferenceaccelerators AT xianghonghu efficientlooptilingframeworkforconvolutionalneuralnetworkinferenceaccelerators AT xuemingli efficientlooptilingframeworkforconvolutionalneuralnetworkinferenceaccelerators AT xiaomingxiong efficientlooptilingframeworkforconvolutionalneuralnetworkinferenceaccelerators |