As a talents program specially launched by the Ministry of Industry and Information Technology's talent exchange center for the integrated circuit industry, the 'Core Power' talent program is like a big tree. Domestic and international IC experts, institutions and investment and financing institutions and other resources are attached. On this lush branch, a good industrial culture and ecological environment has been constructed. This seminar is closely followed by hot famous people and attracts universities, enterprises, research institutes, investment and financing institutions from artificial intelligence chips and related fields. The relevant practitioners, researchers and more than 200 people came to participate. From universities and colleges in countries such as China, the United States, Britain, and other countries, the technology geek in the business world understood the cutting-edge technologies and hot topics of current artificial intelligence.
Wang Zhen, director of the core dynamic talent program IC wisdom valley project office, welcomed the arrival of distinguished guests and detailed the functions of the talent exchange center of the Ministry of Industry and Information Technology and the development plan of the core dynamic talent program. He stated that the core dynamic talent plan is based on the Talents are at the forefront, through the integration of high-quality intellectual resources at home and abroad, the establishment of an exchange platform for the sharing of resources such as parks, enterprises, experts, and talents, so as to build a dynamic, rich-value integrated circuit industrial and humanistic environment. This year's online and offline activities In total, there are more than 170 games. To further enhance the development advantages of artificial intelligence in China and speed up the construction of innovative countries, the center organized and organized the artificial intelligence seminar. We hope everyone can use this platform to share resources and learn and exchange. The center is willing to join you. Work together to build a platform for exchanges and cooperation among domestic and foreign companies, researchers, researchers, and gather colleagues in the industry to contribute to the vigorous development of China's integrated circuit industry.
Professor Wang Zhongfeng from the School of Electronic Science and Engineering at Nanjing University introduced the VLSI signal processing in his speech at "VLSI Optimization for DSP and Deep Learning". He explained the VLSI optimization of turbo decoding from two aspects: high-speed design and low-power design. The VLSI optimization of deep learning and deep neural networks concludes that in the design of modern integrated circuits, optimization of VLSI for signal processing systems can greatly increase power or speed; DNNs are the basis for deep learning; in practical applications , The effective implementation of DNNs is very much needed.
Prof. Wayne Luk, a member of Imperial College London, and a member of the British Royal Academy of Engineering, Wayne Luk, “Development of Reconfigurable Systems: From Research to Practice,” describes how artificial intelligence technology can increase application efficiency, custom computing, and reconfigurable technology, through Microsoft and Amazon illustrated the hardware acceleration cloud server and pointed out that the three directions from research to practice are artificial intelligence function requirements to customizable hardware and software computing architectures, computing architectures to customizable scientific platforms: Hardware and software automation design and optimization tools , Research platform to ecosystem: Accelerate AI application development to reduce costs.
Wu Nanjian, a professor at the Graduate School of the Chinese Academy of Sciences, introduced heterogeneous integration from the aspects of image sensors, memory, processors, neural networks, etc., in his speech on the research and development trend of artificial intelligence vision system chips, and analyzed data bandwidth, working power consumption, and so on.
Zheng Lirong, Dean of School of Information Science and Engineering at Fudan University gave a speech titled "Internet of Things Edge Computing and On-Chip Intelligent System."
He Lei, professor at the University of California, Los Angeles, shared the deep learning and AI chips and AI Chips strategies of large companies, including CPUs, GPUs, FPGAs, ASICs, and brain-like chips (NPUs) in the speech "AI Chip Architecture and Trends." Illustrates the main processor in the AI industry and illustrates three major application areas of AI Chip: training, cloud reasoning, marginal reasoning, and the prospect of unified AI and future AI chips.
In the afternoon, the seminar continued. Wang Qi, CEO of Nanjing Kaiding Information Technology Co., Ltd. pointed out in the speech “The Mutual Promotion and Development of Artificial Intelligence and EDA” that artificial intelligence is an opportunity for rebirth of EDA, and EDA and artificial intelligence can promote each other. The semiconductor industry trends and challenges are explained in terms of semiconductor IC process, IP design, chip design and manufacturing, and large-scale system design. It also illustrates the application of artificial intelligence in place and route, and the application of machine learning in analog design. Visual processing algorithm evolution.
In his speech “Hardware / Software co-design for Efficient Deep Learning Inference”, Chen Zhongmin, vice president of S&T chip research and development, explained that the development of the AI era was driven by algorithms, data, and computational power, and analyzed the development logic of dedicated AI computing hardware. The requirements of the application scenario for dedicated AI computing hardware illustrate why AI hardware requires high energy efficiency, how to improve computational efficiency, and reduce power consumption. The deep AI inference solution is shared from hardware and software co-design.
Wang Shijin, vice president of the HKUST Academy of Science and Technology, pointed out in his speech “The latest development of artificial intelligence technology and introduction of deep learning platform for artificial intelligence” that the trend of artificial intelligence is the key technology hardwareization, the integration of technical methods, and the openness of the algorithm framework. AI service specialization, introduced the successful experience of industrial artificial intelligence to date from deep learning, big data, and computing power. Illustrated the three phases of artificial intelligence from computing intelligence, perceived intelligence, and cognitive intelligence. Knowledge intelligence in the translation of spoken language, knowledge reasoning, breakthroughs in common sense reasoning and artificial intelligence innovation applications. Wang Shijin also pointed out that the pain points of deep learning of enterprises, including data security and the use of low-cost open source learning platform can not be compatible; simply use open source framework to build simple , Large-scale deployment performance, platform stability is difficult to guarantee; Enterprises themselves are more difficult to build a learning platform.
Shi Yuyu, an associate professor of computer science and engineering at the University of Notre Dame in the United States, gave an example analyzing the general application of artificial intelligence in the speech of Chip.ai vs. Net.ai: Scaling for Edge Inference of Deep Neural Networks. Artificial intelligence hardware, artificial intelligence chip The challenges, elaborated on performance gaps, energy efficiency gaps, etc., proposed remedies to bridge the gaps and cited examples of successful quantification. At the same time several questions were raised: Why are the most quantified, even for binary networks? Can a bit-reduced-one network still operate to some extent? To maintain the same accuracy, what is the overhead caused by weight quantization? How do you determine the optimal number of bits to obtain the best compression result?
At the round table forum, the guests discussed fiercely about the "opportunities and challenges facing China's artificial intelligence chips."
'The first three doctoral students I took, I just went to the hardware company just graduated. Last year I met three of them at Google.' Shi Yuyu, an associate professor of computer science and engineering at the University of Notre Dame, said this topic could not help but Make complaints.
In recent years, the Internet industry has been popular, and artificial intelligence algorithms represented by deep learning have become hot spots. Many people think that software dominates the development of artificial intelligence technology. 'A lot of students regardless of what they learn, they all want to be software companies. It's because of the high salary. It's about getting started. ' Shi Yuyu said.
But experts said that behind the rapid development of artificial intelligence, big data, and cloud computing, chips are still needed as support.
Shi Yu-Yu believes that artificial intelligence chips will continue to be hot, 'many AI chips are for some specific scenes, such as unmanned driving. With the increasing number of AI applications, special AI chips are needed. The more it has, the more stringent its performance requirements will be, the more it will form an ecosystem of applications, and the demand for chips will continue to grow.'
“When we submit a product to go out and the customer experience is fundamental, then in order to improve the customer experience, algorithms and chips are indispensable.” Chen Zhongmin, vice president of chip research and development at Shenjian Technology, believes that the mutual promotion of algorithms and chips will be a long-term phenomenon.
In the era dominated by CPU, China's chip industry lags behind for too long. With the rise of artificial intelligence technology, we have started to synchronize with developed countries in the field of AI chips.
In this case, some Chinese AI chips claim to be 'global leaders'. However, experts believe that performance is not the most important one. What really should be concerned is whether these chips are closely integrated with applications.
'For a single computing core, (Chinese and foreign) are doing almost the same. 'Chen Zhongmin said, but the AI chip's instruction set is custom, when the user goes to use it must learn to use this hardware vendor's tools, so Must pay attention to the development and promotion of the tool chain. 'It's like building a car, but it doesn't teach people how to open it. It will greatly affect whether or not your chip pays.'
Industry insiders believe that Nvidia's GPUs are becoming mainstream because the ecology is doing well and development is convenient. From the perspective of ecosystems such as friendliness, ease of development, and migration algorithms, there is still a gap between us and developed countries.
Shi Yu-yu pointed out that China still lags behind in certain areas of the chip. For example, the core technology of high-speed ADC/DAC components is still in the United States, and exports are banned. In the future, it may become a bottleneck in the industry. Moreover, the domestic chip talent is wild. It has also appeared: 'With so many semiconductor graduates, where are the people who make the chips come from?' said Wang Qi, CEO of Nanjing Kaiding Electronic Technology Co., Ltd., The national semiconductor industry still lacks 400,000 engineers. 'There is a need for the market in the short term. Do chips, but (manpower) restrictions make us unable to meet the needs of the market.