https://scholars.lib.ntu.edu.tw/handle/123456789/607332
Title: | A Multiplier-Less Convolutional Neural Network Inference Accelerator for Intelligent Edge Devices | Authors: | Hsieh M.-H Liu Y.-T Chiueh T.-D. TZI-DAR CHIUEH |
Keywords: | Convolutional neural networks (CNNs);FPGA;inference acceleration;multiplier-less;Acceleration;Codes (symbols);Computer graphics;Convolution;Electric power utilization;Energy efficiency;Network coding;Neural networks;Neurons;Program processors;Quantization (signal);Convolutional neural network;Encodings;Graphic processing unit;Graphics processing;Inference acceleration;Multiplierless;Processing units;Field programmable gate arrays (FPGA) | Issue Date: | 2021 | Journal Volume: | 11 | Journal Issue: | 4 | Start page/Pages: | 739-750 | Source: | IEEE Journal on Emerging and Selected Topics in Circuits and Systems | Abstract: | As the demand for neural network operations on edge devices increases, energy-efficient neural network inference solutions become necessary. To this end, this paper proposes a compact 4-bit number format (SD4) for neural network weights. In addition to significantly reducing the amount of neural network data transmission, SD4 also reduces the neural network convolution operation from multiplication and addition (MAC) to only addition. MNIST and CIFAR-10 CNNs with SD4 weights achieve results similar to their FP32-trained counterparts. The difference between the top-1 accuracy of 4-bit ResNet CNN for ImageNet and the baseline FP32 CNN is less than 0.5%. In the hardware design, we have implemented a multiplier-less convolution acceleration circuit. Compared with the 8-bit weight circuit, the power consumption and area of a 4-bit 3times 3 convolution circuit are reduced by nearly 50%. This work also proposes a systematic CNN deployment solution consisting of software CNN training and hardware acceleration. The proposed FPGA-based accelerator for VGG7 image classification achieves a peak throughput of 345.6 GOPS when running at a 100-MHz clock rate. The proposed convolution accelerator's power consumption and energy efficiency are 1.19W and 289. 5 GOPS/W, respectively. Compared to the CPU implementation of VGG7-128 inference, the multiplier-less acceleration circuit is 4.8 times faster and achieves 384 times higher energy efficiency. ? 2011 IEEE. |
URI: | https://www.scopus.com/inward/record.uri?eid=2-s2.0-85118637908&doi=10.1109%2fJETCAS.2021.3116044&partnerID=40&md5=e236c839c3dfdf42880fa626eca1c0ff https://scholars.lib.ntu.edu.tw/handle/123456789/607332 |
ISSN: | 21563357 | DOI: | 10.1109/JETCAS.2021.3116044 |
Appears in Collections: | 電機工程學系 |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.