other · DR-NTU (Nanyang Technological University) · 2025
FPGA implementation of embedded AI
Wu, Haotian
Cited by 0Abstract and details only
Abstract
With the rapid development of AI (Artificial Intelligence) technology, its application and deployment on embedded platforms has become one of the key research focuses in the industry. YOLO (You Only Look Once) is the most frequently mentioned algorithm. However, traditional CPU (Central Processing Unit) unable to meet the real-time requirements. The power consumption of high-performance GPU (Graphics Processing Unit) is not acceptable for end devices. FPGAs (Field-Programmable Gate Arrays) achieve efficient computation and memory access, made them the mainstream choice for embedded deployment. Vitis HLS (High-Level Synthesis) enables the conversion from high-level to hardware languages, greatly simplifying the development process. This dissertation takes the lightweight but highly accurate YOLOv4-tiny model as the research object and realizes efficient acceleration on the Xilinx ZYNQ-7020 platform through overlapping architectures reusable PE (Processing Element) design, as well as various optimization strategies such as ping-pong buffering, operator fusion, data reuse and quantization mechanism with HLS primitive instructions. Through hardware and software co-design and integration, we achieved a frame rate of 15.58 FPS (Frames Per Second) at a clock frequency of 150 MHz while power consumption was only 2.514 W, providing real-time processing capability and low power consumption.
Cite this paper
Wu, H. (2025). FPGA implementation of embedded AI. DR-NTU (Nanyang Technological University).
Wu, Haotian. “FPGA implementation of embedded AI.” DR-NTU (Nanyang Technological University), 2025.
Wu, Haotian. 2025. “FPGA implementation of embedded AI.” DR-NTU (Nanyang Technological University).
Wu, H. (2025) ‘FPGA implementation of embedded AI’, DR-NTU (Nanyang Technological University).
@misc{wu2025fpga,
title = {{FPGA implementation of embedded AI}},
author = {Wu, Haotian},
journal = {DR-NTU (Nanyang Technological University)},
year = {2025}
}Made from the index record. Check names, capitalisation and page numbers against the paper. Need another style? AI for Research has over 10,000.