Proposed Edge Computing Neural Accelerator

2024pp. 1–8

Citations Over TimeTop 21% of 2024 papers

Vikram Singh, Rajesh Verma, Deepak Kumar Saini, Krishan Berwal, Sanotsh Kr Vishwakarna

Abstract

Edge computing is revolutionizing data processing, especially for applications like surveillance and traffic control systems. The data being processed closer to the source (e.g., cameras or sensors), with this edge computing reduces communication latency, bandwidth consumption, and power usage, addressing limitations of centralized cloud systems. The integration of machine learning (ML) in edge environments enhances surveillance systems, enabling real-time analytics, such as object detection and activity recognition. Convolutional Neural Networks (CNNs), widely used in video analytics, facilitate instant object recognition at the edge. However, the computational demands of deep neural networks (DNNs) require hardware accelerators like DNN accelerators for efficient deployment. These accelerators, implemented using technologies like Field-Programmable Gate Arrays (FPGAs) and Application-Specific Integrated Circuits (ASICs), improve latency and energy efficiency. Key benefits include parallelism, optimized memory management, and energy-efficient processing, essential for resource-constrained environments like surveillance drones. This paper explores the architecture of deep learning accelerators, such as the Edge Computing Neural Accelerator (ECNA), focusing on their role in real-time surveillance systems. By leveraging binary-weight convolutional layers and systolic array designs, ECNA significantly enhances processing speed and energy efficiency, enabling real-time video analytics while consuming minimal power.

Citations Over TimeTop 21% of 2024 papers

Abstract

Related Papers