Accelerating Event-based Deep Neural Networks via Flexible Data Encoding

IF 0.8 4区工程技术 Q4 ENGINEERING, ELECTRICAL & ELECTRONIC

Ieice Electronics Express Pub Date : 2023-11-10 DOI:10.1587/elex.20.20230379

Yuanli Zhong, Yongqi Xu, Bosheng Liu, Yibing Tang, Jigang Wu

引用次数: 0

Abstract

Event-based deep neural networks (DNNs) have shown great promise in computer vision under difficult lighting conditions. However, existing hardware solutions cannot provide efficient event-based DNN accelerations owing to the characteristic of event streams, which are typically in low datarate and high-dynamic range. In this letter, we present a novel hardware design that can handle event-based DNNs according to the data characteristic of event streams. Furthermore, we provide a dataflow that enables flexible DNN data encodings (including both bitmask and compressed sparse row (CSR)) based on the event data characteristic for energy saving. Comprehensive evaluations based on four famous event-based benchmarks show that the proposed design can achieve higher performance and better energy efficiency compared with representative accelerator baselines.

查看原文本刊更多论文

通过灵活的数据编码加速基于事件的深度神经网络

基于事件的深度神经网络(dnn)在困难光照条件下的计算机视觉中显示出巨大的前景。然而，由于事件流的特性，现有的硬件解决方案无法提供有效的基于事件的DNN加速，事件流通常处于低数据量和高动态范围内。在这封信中，我们提出了一种新的硬件设计，可以根据事件流的数据特征处理基于事件的dnn。此外，我们提供了一个基于事件数据特征的数据流，该数据流支持灵活的DNN数据编码(包括位掩码和压缩稀疏行(CSR))，以节省能源。基于四个著名事件基准的综合评估表明，与代表性加速器基准相比，所提出的设计可以实现更高的性能和更好的能源效率。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Ieice Electronics Express 工程技术-工程：电子与电气

CiteScore

1.50

自引率

37.50%

发文量

119

审稿时长

1.1 months

期刊介绍： An aim of ELEX is rapid publication of original, peer-reviewed short papers that treat the field of modern electronics and electrical engineering. The boundaries of acceptable fields are not strictly delimited and they are flexibly varied to reflect trends of the fields. The scope of ELEX has mainly been focused on device and circuit technologies. Current appropriate topics include: - Integrated optoelectronics (lasers and optoelectronic devices, silicon photonics, planar lightwave circuits, polymer optical circuits, etc.) - Optical hardware (fiber optics, microwave photonics, optical interconnects, photonic signal processing, photonic integration and modules, optical sensing, etc.) - Electromagnetic theory - Microwave and millimeter-wave devices, circuits, and modules - THz devices, circuits and modules - Electron devices, circuits and modules (silicon, compound semiconductor, organic and novel materials) - Integrated circuits (memory, logic, analog, RF, sensor) - Power devices and circuits - Micro- or nano-electromechanical systems - Circuits and modules for storage - Superconducting electronics - Energy harvesting devices, circuits and modules - Circuits and modules for electronic displays - Circuits and modules for electronic instrumentation - Devices, circuits and modules for IoT and biomedical applications