【活动报道】LPN Cruciform 学术沙龙——芯片革命:可编程芯片FPGA–人工智能与大数据时代的加速引擎

【Event Report】LPN Cruciform Academic Seminar – Chip Revolution: Programmable Chip FPGA–Acceleration Engine in the Age of Artificial Intelligence and Big Data

 

题目:芯片革命:可编程芯片FPGA–人工智能与大数据时代的加速引擎

时间:2019年04月25日下午6:00

 

地点 : B304-Lecture Theatre 2, UCL Cruciform Building, Gower Street, University College London, WC1E 6AE

 

本次LPN Cruciform 学术沙龙,我们邀请到了英特尔英国研发中心的资深工程师石侃博士主讲,介绍了可编程芯片—FPGA(field-programmable gate array),及其特点和应用,FPGA在AI(artificial intelligence)时代的独特优势及FPGA的新型应用实例。

 

在人工智能与大数据日益高速发展的现代,可以进行快速、大容量数据计算的FPGA芯片受到广泛的关注。首先,FPGA作为一种特殊的高性能芯片,由现场可编程门阵列构成,理论上是一种“万能芯片”,可以实现“任何”功能。与CPU(central processing unit)相比,FPGA根据算法定义架构,并行处理数据流,具有快速、高并行、高吞吐量的特点,石博士分别用飞机和高铁生动形象的比喻了CPU和FPGA的特点。其次,FPGA作为一种高性能的AI芯片,石博士分别在性能、灵活性、同构性、成本和功耗五个方面与CPU、GPU(Graphics Processing Unit)、ASIC(application-specific integrated circuit)进行了对比,其中,CPU具有性能一般的不足,GPU具有高功耗和高成本的劣势,ASIC具有低灵活性和高研发成本的缺点,唯有FPGA具有综合的比较优势,是最为理想的AI芯片,在人工智能和大数据方面具有广泛的应用。最后,石博士介绍了FPGA由寡头垄断,如Intel Altera,Xilinx, 等公司,主要应用于通信、网络、信号处理、航天军工、高性能计算等,主要的用户有华为,爱立信,诺基亚,中兴等。他重点介绍了微软在大数据时代的转型,微软的Catapult项目和Brainwave项目,并介绍了FPGA在数据中心、AI、云计算、数据分析、网络安全、视频译码、金融加速、基因组学等方面的新应用。

本次沙龙内容丰富精彩,讲解生动并配合很多视频介绍,参会人员与嘉宾进行了热烈的跨学科交流和探讨,反响热烈。

本次沙龙由伦敦大学学院李会良博士和国王学院王国庆博士组织。

 

 

【主讲嘉宾介绍】

石侃,博士毕业于伦敦帝国理工大学电子工程系,现任英特尔公司资深芯片研发工程师,微信公众号“老石谈芯”主理人。在工业界,石侃一直深耕基于可编程芯片的数据中心网络加速、网络功能虚拟化、高速有线网络通信等领域的研发和创新工作,有丰富从业经验。在学术界,石侃曾经针对可编程逻辑芯片、近似计算、计算机数学等方向在学界顶级会议和期刊上发表过多篇研究论文,并两次获得欧盟HiPEAC论文奖。


【Event Report】LPN Cruciform Academic Seminar – Chip Revolution: Programmable Chip FPGA–Acceleration Engine in the Age of Artificial Intelligence and Big Data

In this LPN Cruciform Academic seminar, we invited Dr. Kan Shi, a senior engineer from Intel UK R&D Center, to give a lecture, and introduced the programmable chip-FPGA (field-programmable gate array), its characteristics and applications, FPGA in AI (artificial intelligence) The unique advantages of the times and new application examples of FPGAs.

In the modern era of the increasingly rapid development of artificial intelligence and big data, FPGA chips that can perform fast and large-capacity data calculations have received extensive attention. First of all, as a special high-performance chip, FPGA is composed of field programmable gate arrays. In theory, it is a “universal chip” that can realize “any” function. Compared with CPU (central processing unit), FPGA defines the architecture according to the algorithm and processes data streams in parallel. It has the characteristics of high speed, high parallelism and high throughput. Dr. Shi used an airplane and a high-speed rail to vividly compare the characteristics of CPU and FPGA. . Secondly, FPGA, as a high-performance AI chip, is comparable to CPU, GPU (Graphics Processing Unit), and ASIC (application-specific integrated circuit) in five aspects: performance, flexibility, isomorphism, cost, and power consumption. A comparison is made. Among them, the CPU has the disadvantage of general performance, the GPU has the disadvantages of high power consumption and high cost, and the ASIC has the disadvantages of low flexibility and high R&D cost. Only the FPGA has comprehensive comparative advantages and is the most ideal AI. Chips have a wide range of applications in artificial intelligence and big data. Finally, Dr. Shi introduced that FPGAs are monopolized by oligopolists, such as Intel Altera, Xilinx, and other companies. They are mainly used in communications, networking, signal processing, aerospace and military industry, high-performance computing, etc. The main users are Huawei, Ericsson, Nokia, ZTE, etc. . He focused on Microsoft’s transformation in the era of big data, Microsoft’s Catapult project and Brainwave project, and introduced FPGA in data center, AI, cloud computing, data analysis, network security, video decoding, financial acceleration, genomics, etc. new application.

The content of the seminar was rich and exciting, with vivid explanations and many video introductions. Participants and guests had a lively interdisciplinary exchange and discussion, and the response was enthusiastic.

The seminar was organized by Dr. Huiliang Li from UCL and Dr. Guoqing Wang from King’s College London.

【Introduction to the guest speakers】

Kan Shi, graduated from the Department of Electronic Engineering of Imperial College London with a Ph.D., is currently a senior chip R&D engineer at Intel Corporation, and the manager of the WeChat public account “Laoshi Tanxin”. In the industrial world, Shi Kan has been deeply engaged in R&D and innovation in the fields of data center network acceleration, network function virtualization, and high-speed wired network communication based on programmable chips, and has extensive experience in the industry. In academia, Shi Kan has published many research papers in top academic conferences and journals on programmable logic chips, approximate computing, computer mathematics, etc., and won the EU HiPEAC Paper Award twice.


【主办方介绍】

 

伦敦博士联盟(London PhD Network,简称LPN)成立于2015年9月,是以伦敦为轴心的各知名高校和研究机构的博士生,博士后,导师和教授组成的青年高科技学术团体,一直受到中国驻英使馆科技处的大力支持和关怀。伦敦博士联盟以促进跨学科交流和青年学者的职业发展为宗旨,精心组织一系列学术交流研讨会,定期邀请学者和专家进行讲座,内容涵盖生物医学,医疗健康、人工智能、能源交通、大数据、机器人技术等多个国家重点投资发展领域,获得广泛好评,至今已有超过800名活跃会员。同时伦敦博士联盟也为热衷产业化和商业化的精英人士提供咨询平台,积极推动伦敦地区青年学者的科研交流和合作。