@inproceedings{e02adcdb66be44a7889513acba77120a,
title = "A Design of Lightweight Convolutional Neural Network Accelerator for IoT Devices",
abstract = "In this paper, we propose to design a Convolutional Neural Network (CNN) accelerator suitable for application in Internet of Things (IoT) devices. The CNN accelerator is trained on Modified National Institute of Standards and Technology (MNIST) images provided by TensorFlow and used as data. We simplify the structure of the accelerator by designing an optimized Multiply and Accumulate (MAC) that is common to all layers of the accelerator. We also quantized the values of the learned float 32-bit weights and biases to 8 bits. The design of the lightweight CNN accelerator with the proposed structure was implemented on Cadence's NC Verilog and Altera's Cyclone IV EP4CE115F29C7 to evaluate its functionality and performance. Despite the data loss due to the lightweight of the parameters used in the computation, the test results of the proposed CNN accelerator presented a high accuracy of about 95\%.",
keywords = "Accelerator, CNN, FPGA, MNIST, Quantization, Verilog-HDL",
author = "Song, \{Yeon Seob\} and Lee, \{Kang Yoon\}",
note = "Publisher Copyright: {\textcopyright} 2023 IEEE.; 14th International Conference on Ubiquitous and Future Networks, ICUFN 2023 ; Conference date: 04-07-2023 Through 07-07-2023",
year = "2023",
doi = "10.1109/ICUFN57995.2023.10200660",
language = "English",
series = "International Conference on Ubiquitous and Future Networks, ICUFN",
publisher = "IEEE Computer Society",
pages = "474--477",
booktitle = "ICUFN 2023 - 14th International Conference on Ubiquitous and Future Networks",
}