From April 10th to 12th, the 10th Anniversary Award Ceremony of Wu Wenjun Artificial Intelligence Science and Technology Award, known as the "Highest Award of China's Intelligent Science and Technology", and the 2020 Chinese Intelligent Industry Annual Conference were held in Beijing and Suzhou at the same time.
It is worth noting that this year's Wu Wenjun Artificial Intelligence Science and Technology Award for the first time in the fields of semiconductors, integrated circuits and software, microelectronics and other fields to set up artificial intelligence special chip award, awarded to the team and individuals who have made original and advanced achievements in the field of artificial intelligence chips, a total of 8 projects on the list, the main completion units include Beijing Horizon Robot Technology Research and Development Co., Ltd., Shenzhen Yuntian Lifei Technology Co., Ltd., Sichuan University, Suzhou Changguang Huaxin Optoelectronic Technology Co., Ltd., Southern University of Science and Technology, Shenzhen Tencent Computer System Co., Ltd., Staring Pat (Shenzhen) Technology Co., Ltd., etc.
After the meeting, Chen Ning, chairman and CEO of Shenzhen Yuntian Lifei Technology Co., Ltd., said in an interview with Nandu reporter that under the impetus of 5G and AI, two types of robots may have relatively large opportunities, one is industrial robots, and the other is in some special industries, service robots in special fields, such as solving the last kilometer of delivery robots, there will be large-scale industrialization breakthroughs in the next two years.
Ning Chen graduated from Georgia Institute of Technology with a degree in electrical engineering, and was a senior architect at Freescale Semiconductor and a chip technology director at ZTE. In August 2014, Chen Ning and his partners entered the intelligent security industry and officially founded Yuntian Lifei in Shenzhen, focusing on the research and development of visual artificial intelligence products.
Chen Ning, chairman and CEO of Shenzhen Yuntian Lifei Technology Co., Ltd
Nandu: Can you briefly introduce the award-winning "Video Edge Intelligent Neural Network Processor Chip"?
Chen Ning: The award-winning neural network processor chip is mainly based on some understanding of the deep learning algorithm framework.
Its architecture allows the processor to understand the algorithm, understand the scene, to carry and run these algorithms more efficiently, and to provide a higher PPA indicator, which is the performance power area. The performance indicators of these processors are oriented to the field of neural network computing, mainly in the inference calculation of the end and edge, especially in the inference of edge computing. Through the one-click deployment of a rich and flexible toolchain, the algorithms of various frameworks can be quickly deployed on such an edge computing platform to empower various specific scenario examples.
For example, in the two major areas of digital city and human life, video surveillance in digital cities can customize the algorithms inside these video cameras through flexible end-edge reasoning chip platforms. For example, for the analysis of massive video content, edge computing can provide a more cost-effective solution with lower latency.
Nandu: The national 14th Five-Year Plan also focuses on special chips, what is the current trend of special chips and general-purpose chips in research and development applications?
Chen Ning: General-purpose chips refer more to pan-scenarios like CPUs, mainly general-purpose chip platforms for control and scheduling. For example, the X86 architecture, the ARM architecture, including these control platforms in PCs and smartphones.
Dedicated chips, especially AI dedicated chips, our understanding is mainly for emerging deep learning algorithms, neural network computing, etc. Needs, it needs to provide high concurrency, high computing capacity of neural network processing capabilities in massive application scenarios, and this dedicated processing type is needed to provide such a powerful computing power platform. Dedicated chips have been a starting stage for the world in the past six or seven years, and in this process, Chinese companies still have more opportunities.
Nandu: How do you view the development prospects of intelligent robots?
Chen Ning: During the 14th Five-Year Plan period, intelligent robots will develop faster with the promotion of 5G and AI, unlike the past Internet era, which is the consumer Internet, and during the 14th Five-Year Plan period, it is more to promote the industrial and industrial Internet. In this process, I think there may be two types of robots that have greater opportunities, one is industrial robots, that is, intelligent manufacturing. The other type is service robots in some special industries and special fields, such as delivery robots that solve the last kilometer.
Whether it is an industrial park, an industrial park, or a delivery robot in a hotel or restaurant, I believe that these technologies have gradually matured, and what is needed is some breakthroughs in business models and operating models. Industrial robots and delivery robots may have another large-scale industrialization breakthrough in the next two years.
Intelligent robots themselves are a platform with strong demand for AI algorithms, especially vision and algorithms, so we are also pushing the edge computing platform of our neural network processor to empower industrial and distribution robots.
Nandu: What role does the breakthrough of upstream technology, especially at the chip level, play in the further development of intelligent robots?
Chen Ning: I think that the future of intelligent robots is not only the current industrial and distribution types, as well as home service robots, etc., the future of intelligent robots is actually an intelligent carrier of end-edge cloud collaboration.
That is to say, in the future, robots may have cloud operators to operate and manage like our smartphones, such as mobile phone operators such as Unicom and Mobile. However, at the level of robot ontology, we need to pay more attention to the intelligence of the ontology, the richness of the algorithm and the experience of the service.
This brings some new opportunities, and these opportunities are inseparable from the most core technical capabilities, that is, the computing power of the end-edge cloud and the service platform of the algorithm, especially the efficient training of the long-tail algorithm for various subdivision scenarios of the robot service, which will become a core capability.
Nandu: How do you view the advantages and shortcomings of the intelligent robot industry cluster in the Pearl River Delta?
Chen Ning: One of the core advantages of Guangdong's development of the robot industry is Guangdong's hardware production and manufacturing capabilities. Because of the information industry chain and manufacturing supply chain capabilities accumulated in the past two or three decades, Guangdong has become the world's electronic components information manufacturing capital, which is an advantage of Guangdong. Robots are also inseparable from the ability of such underlying production and manufacturing design.
But not only Guangdong, What China urgently needs to solve is the open source training platform and processor of the underlying core algorithm, that is, the ecological problem of dedicated AI chips, including the production and manufacturing of chips and other industrial chains. Then at the algorithm level of deep learning, the underlying framework for training these algorithms requires an open source platform, which is what we are currently missing and needs special attention.
Nandu: Yuntian Lifei's planning in chips?
Chen Ning: We still focus on the core technical capabilities of algorithm chipping, relying on this ability to build the ecology of our neural network processors, especially in the two major sectors of digital cities and human life, so that our application scenarios are becoming more and more abundant, and promoting the process of this rich application scenario, promoting an organic iteration from algorithms to chips to applications and data, so that the efficiency of our algorithms' training, application efficiency and research and development efficiency are gradually improved.
Written by: Nandu reporter Ma Ningning intern Xie Lulu