Intelligence on Silicon: From Deep-Neural-Network Accelerators to Brain Mimicking AI-SoCs

📋 论文概要

本文综述了从深度神经网络加速器到类脑人工智能SoC的发展路径，旨在解决云、边缘、移动三大应用场景下AI技术的高效处理需求。文章分析了当前AI硬件面临的挑战，并展望了未来智能芯片的演进方向。

💡 主要创新点

• 提出从专用DNN加速器向类脑AI SoC的演进路线，覆盖云、边缘、移动三个层次
• 系统性地总结了AI技术在信息硬件、软件和网络中的广泛应用，并指出专用加速器在服务器端的关键作用

🏷 关键词

📄 原文摘要

lifestyle of all society. AI technology is widely used in most information hardware, software, and networking, underlying all consumer technology: smartphones, home appliances, and the Web. Figure 1.2.1 shows 3 main application areas for AI technology: Cloud, Edge, and Mobile. The first is Cloud AI where intelligent applications are processed by server machines at data centers with results sent to edge controllers (such as speakers) and through to mobile devices (such as smartphones). As AI applications become more popular, servers need to incorporate dedicated AI accelerators to speed up the ever-increasing demand for AI processing with low energy consumption [1, 2]. In addition to AI processing in servers, recently many edge devices such as CCTV or AI speakers (as depicted in Figure 1.2.1) are required to process AI algorithms

👥 作者与机构

Hoi-Jun Yoo, KAIST, Daejeon, Korea, 1. Introduction

While, currently, Artificial-Intelligence technology is affecting all industrial paradigms, it is also impacting