On October 16, 2018, Xilinx, the global FPGA chip giant, held the Xilinx Developer Conference in Beijing. The Global president and CEO of Xilinx, Victor Peng, delivered a keynote speech entitled "A platform for flexible and intelligent world." As the Certificate Member of Xilinx in the field of smart car electronics in China, Autocruis was invited to participate in and the company co-founder and CEO Dr. Guan Yanfeng and CTO Dr. Jing Shen attended the conference together. At the conference, Dr. Jing Shen gave a keynote speech on "Design and Development of FPGA-Based Computer Vision Systems" . During the period, Victor Peng and Dr. Guan Yanfeng conducted in-depth exchanges on topics such as “AI Development Trends and FPGA Application Trends” and took a group photo.
President and CEO of Xilinx Victor Peng delivered a keynote speech
Xilinx CEO Victor Peng first introduced the company's three strategies of “data center priority, accelerating mainstream market growth and driving flexible computing”. He said that based on FPGAs, SoCs and MPSoCs, Xilinx introduced the industry's first Adaptive Compute Acceleration Platform (ACAP) VersalTM. ACAP was developed by over a thousand engineers of the company for 4 years , which invested more than $1 billion. It integrates a scalar processing engine, adaptive hardware engine and intelligent engine, as well as cutting-edge memory and interface technologies to provide powerful, adaptive heterogeneous computing acceleration for a wide range of applications during the rapid development of AI and dense edge computing. At the same time, ACAP has a range of tools, software, libraries, IP, middleware and frameworks that conform to industry-standard design processes.
Dr. Guan Yanfeng and Victor Peng have an in-depth exchange and take a group photo
Based on TSMC's 7 nm FinFET process technology, the Versal portfolio is the first platform to combine software programmability with hardware acceleration and flexibility in a specific area. It includes six product portfolios: Prime, Premium, HBM, AI Core, AI Edge, and AI RF. They use groundbreaking AI engine technology to meet the new demands of low-latency AI inference for a variety of applications, and support applications such as wireless and radar.
Victor Peng said that Xilinx has been in the automotive market for more than a decade. The Zynq-7000 series devices are widely used in Tier1, such as BOSCH. As of 2018, there are already 29 brands and 110 models using Xilinx chips and solutions, and the cumulative chip sales exceeded 40 million.In the booming field of smart car electronics, in addition to the Zynq series, Xilinx has accumulated a rich product portfolio including MPSoC series and AI Edge series, which are widely used in smart cars such as Car networking, ADAS, and autonomous driving. In July of this year, Xilinx entered into a strategic partnership with Daimler to develop AI -based in-vehicle systems and applications for Mercedes-Benz driverless cars.
The future of the smart car market is an extremely large market, and Xilinx will invest more in the smart car field to provide more powerful and flexible platforms and tools for developers and innovators in the hardware and software industry,” concludes Victor Peng. “We are very much looking forward to the excellent smart car high-tech companies like Autocruis and automakers going hand in hand to create a world of flexibility and intelligence in the future !”
随后，关艳峰博士首先对Xilinx发布业内首款自适应计算加速平台Versal和新一代加速器卡Alveo表示祝贺，感谢Xilinx一直以来对自行科技的支持和帮助。
他表示，随着全社会出行安全、出行效率以及舒适性需求的提升，汽车智能化正逐渐成为汽车产业创新升级的核心驱动力，智能化问题已经成为世界各国争夺未来汽车行业全球话语权和标准制定权的焦点问题，各国纷纷制定和出台相关法规标准。在中国，自行科技也在积极参与，与此同时致力于提供可大规模量产的，前装车规级的汽车智能化产品和服务。
智能汽车需要实时可靠地感知车、人、道路以及基础设施所传递的各类信息和信号，为ADAS和自动驾驶提供决策和执行支持，根据具体功能和ADAS阶段的不同，涉及的计算量从几TOPS到几百TOPS不等，而且OEM不断升级变化的需求也对供应商和开发者能够迅速迭代提出了新的要求。自行科技结合多年在FPGA平台开发优化CNN技术的经验，构建出完善的设计、开发、验证和量产测试工具链，形成了成熟的训练平台、仿真平台以及产品开发平台。其中，就深度神经网络而言，自行科技为了适配各类神经网络的快速优化和产品开发，独立开发了基于FPGA平台的神经网络状态机，大大加速产品开发进程，更新迭代一个新网络的周期可以在1个月完成。
在交流中，关艳峰表示，Level 4+ 级无人驾驶是汽车智能化的重要诉求，但无人驾驶之路非常漫长，这需要汽车、半导体和通信等传统产业的持续发展以及深度融合，更需要新型传感器、新型处理器和AI等新产业发展以及成熟。自行科技将在视觉感知、多传感器融合感知、驾驶员状态感知、车辆主动安全以及车载AI软硬件平台等方面贡献自己的力量。
公司目前基于CNN网络技术开发出业内最具性价比的FPGA加速方案。以普遍应用的VGG-16为例，如果不经过深度优化即使在MPSoC-ZU9中进行前向推理计算时，帧率只能达到19.3fps，但是自行科技通过权值压缩、并行化加速设计和算法优化后，可以在zynq 7020中实现超过30fps的帧率，同时性能不损失，这得益于自行科技在混合型精度的神经网络与FPGA深度结合方面的努力和积累。
关艳峰还强调，自行科技非常看好FPGA在ADAS和自动驾驶领域的未来发展和应用。公司从创立伊始，就专注在智能汽车电子领域，开发出基于FPGA的前向ADAS系统、驾驶员监控（DMS）系统、智能全景影像系统，多传感器融合等产品，逐渐形成了车内车外联合感知决策，远距感知与近距离感知联合协同的技术和产品路径，是少数掌握ADAS与自动驾驶核心算法并实现产品量产的企业。凭借产品高性能、高可靠性、低功耗以及在产品安全性上的突出优势,自行科技获得了国内众多传统车厂(乘用车与商用车)、新能源和互联网造车公司和Tier1汽车零部件供应商的高度认可。
最后，Victor Peng与关艳峰博士就下一阶段的合作进行了深入沟通，并达成共识，双方将继续深化合作关系，加强沟通交流，携手同行，共创共赢！