An Energy-efficient and Dual-mode AI Accelerator for Wearable Lung Sound Monitoring

Hits:

First Author:Keyi Yang

Correspondence Author:C. Wang*

Co-author:B. Liu, Z. Huang, H. Ping, Z. Wen, Z. Shen, C. Huang and

Journal:IEEE Biomedical Circuits and Systems (BioCAS 2024) Conference

Included Journals:EI

Discipline:Engineering

First-Level Discipline:Electronic Science And Technology

Document Type:C

Abstract:Lung diseases seriously impact human health. Using deep neural networks (DNN) to pre-diagnose lung diseases based on lung sounds is a growing trend. However, most existing DNN hardware accelerators perform direct fine-grained multi-category classification for lung sounds, leading to low energy efficiency, as most lung sounds are normal and only a coarse-grained classification for abnormal detection is sufficient to pre-screening. This paper proposes an energy-efficient Artificial Intelligence (AI) accelerator based on a tightly two-stage hybrid neural network (TS-HNN) model for wearable lung-sound monitoring applications. First, a Two-Stage Data Buffering Scheme (TS-DBS) with a custom Local Memory Bank (LMB) is proposed to support the two-stage acceleration of the TS-HNN model. Second, by observing the channel correlation of two major convolution operators, a Reconfigurable Processing Element Array (RPEA) is designed for efficiently computing standard/pointwise convolution and depthwise convolution, enhancing the computation efficiency and energy efficiency. The FPGA experiment results show that the proposed AI accelerator can achieve an energy efficiency of 105.8 GOPs/W, and it only consumes 0.39 mJ/frame for normal lung sounds and 2.03 mJ/frame for abnormal lung sounds, which indicates energy efficiency improvement against the state-of-the-art design by 13.44 and 2.58, respectively.