A low-cost Hardware Architecture of Convolutional Neural Network for ECG Classification

Convolutional neural network (CNN) is one of the most important structure of the machine learning and have been widely used in medical diagnosis due to its effectiveness. The main aim of this study is to build a 1-D CNN model for electrocardiogram (ECG) classification and design a low-cost hardware architecture of this neural network, and this paper propose a novel architecture using efficient convolution PE array that can efficiently reuse weight data and pixel data to reduce resource utilization. The proposed CNN acceleration scheme and architecture are demonstrated by implementing end-to-end the proposed 1-D CNN for inference on FPGA Zynq ZC706, the overall throughput achieves 26.6GOPS consuming only 2.51k LUTs and 0.79W at 200MHz, and the hardware efficiency is 10.59 GOP/s/kLUT.