The explosive growth of artificial intelligence is based on powerful computing power, and the carrier that provides computing power is the chip. In recent years, the unicorn company that has received a lot of capital from domestic companies has a close relationship with the AI chip, and it also proves from the side. The importance and broad prospects of AI chips. However, as more and more start-up companies, Internet companies and traditional chip companies begin to enter the AI chip field, the risks inherent in them need to be taken seriously. AI chips will be integrated in China. What are the risks in overtaking the circuit industry? What are the risks involved? How can we seize this rare opportunity in the industry? 尴尬 产业 产业 产业 产业 产业 产业 产业 产业 产业 要 要 要 要 要 要 要 要The chip is the foundation of a new computing era. Controlling the core chip architecture from the source will gain the first-mover advantage, and it is of great significance to gain a new computing era dominance. It is also based on this consideration, Google, Microsoft. , Amazon, IBM and other global technology giants have invested heavily to accelerate the research and development of artificial intelligence core chips, the goal is to grab The strategic commanding heights in the new computing era control the dominance of the artificial intelligence era. China also attaches great importance to the artificial intelligence chip industry. The Ministry of Industry and Information Technology officially publishes the “Three-year Action Plan for the Development of a New Generation of Artificial Intelligence Industry (2018-2020)”. Emphasis on breakthroughs in smart sensors, neural network chips, and open source open platforms in several areas. According to the "2018-2023 Chinese Chip Industry Market Demand and Investment Planning Analysis Report" released by the Industry Research Institute, 2016, global artificial intelligence chips The market size reached 3.6 billion U.S. dollars, and it is expected to reach 11.1 billion U.S. dollars in 2021, with a compound annual growth rate of 25%. Combined with the size of China's artificial intelligence market, it is estimated that the market size of China's artificial intelligence chips will be approximately 1.5 billion U.S. dollars by 2016, by 2022 The market size will reach about 5 billion yuan, with rapid growth and huge development space. With the promising development prospects of the AI chip, more and more Internet companies and traditional chip companies have entered the AI chip field, and the risks involved need to be raised. People pay attention. At the recently held '2018 Semiconductor Market Annual Meeting', Tsinghua University When Wei Shaojun, director of the School of Microelectronics, said in a speech, “There are people in the investment community asking me whether I should invest in AI chips. My answer is, 'Investment can, but it must not be the last one'.' What are the challenges faced by Chinese companies investing in AI chips? From the perspective of the industry chain, artificial intelligence includes complex structures ranging from chip devices, computing devices, program platforms, to big data, functional layers, and applications. In such an industrial ecosystem, Some powerful Internet giants such as Google, Facebook, Amazon, Baidu and Apple, are involved in the vertical integration mode, they become the most powerful players. Intel, Qualcomm, Infinity and other international chip leaders are integrated chips , calculations and some software programs for development strategies, and strive to play the role of hardware infrastructure platform providers. Some powerful terminal brand companies, including Chinese smartphone manufacturers that have been rapidly developing in recent years, such as Huawei, Xiaomi, VIVO, OPPO, hope AI becomes a terminal equipment supplier. In this industry ecology, Chinese AI chip manufacturers are relatively positioned. It is difficult for them to fully cooperate with Nvidia. Intel will compete with vendors that provide artificial intelligence computing solutions for cloud devices. More development opportunities exist in the terminal market, namely for different applications, providing targeted chip or IP solutions. The problem is that the entire AI industry has so far been plagued by the lack of 'killer' applications. If we remove smart driving, smart cars, smart surveillance, face recognition, and voice recognition for a few application markets, we will find it difficult to find AI. The typical application scenario, especially the lack of application market that is closely related to people's working life, and non-AI support is not. In this way, an important issue will emerge - Is there an independent AI processor like a general-purpose CPU? If it exists, it What's the architecture? If not, then Chinese AI chip makers can hardly compete with international giants for the cloud market. The terminal market they are based on is fragmented, fragmented, and lacks the challenge of killer applications. Perhaps it is true that Wei Shaojun pointed out: 'The Chinese AI core with the primary goal of meeting specific applications The future is likely to exist only in the form of IP cores, eventually being integrated by various SoCs. 'So, today's part, and even most of the AI chip entrepreneurs will become the 'martyrs' in this technological transformation. New era seeks new ideas for IC design In the face of this challenge, architectural innovation has become an unavoidable issue for China's AI chips - developing a chip architecture that adapts to the market in light of the current status of the AI industry. Although such innovation will be very difficult In this regard, Wei Shaojun proposed the concept of 'software-defined chip'. From perception, transmission to processing, then to transmission and execution, this is the basic logic of the AI chip. Software is the core of the realization of intelligence, and the chip is the basis for supporting intelligence. Among them, software is the carrier for achieving intelligence. Technically, intelligent software is required to have its own learning capabilities, the ability to form knowledge and experience, the ability to continuously improve and optimize, the logic of logical reasoning, and the logic to make correct judgments. The chip will need to carry the required calculations, requiring high-performance computing capabilities, multi-task parallel computing capabilities, and extremely high energy efficiency. Flexible and efficient storage capabilities, real-time dynamic function conversion capabilities, etc. The linkage between the two will redefine the design concept of the AI chip, make the design of the AI chip more flexible, and also be more operable when facing different AI application requirements. . Synopsys China Chairman and Global Vice President Ge Qun also expressed the same view: 'Previously, the hardware and software boundaries in chip design are relatively clear. Generally, a layered design and optimization method are used. This helps to simplify the problem. However, Today's AI applications are no longer able to separate them. That is to say, the development of a good chip and hardware product must take into account the cooperation of software and system level, and even break the level boundaries, in-depth optimization. Nvidia The success of CUDA and GPU, Google's Tensorflow and TPU are all good examples. 'In fact, there is a close interaction between EDA tools and AI, which will play a crucial role in the architectural innovation of AI chips. Whether in the cloud Also on the terminal equipment, as the differentiated application requirements increasingly affect the chip design, for the cloud AI training, automatic driving, security monitoring The solution of various intelligent IoT devices needs to meet the special needs of AI applications from the aspects of application software, operating system, hardware architecture, IP, low power consumption, security, verification and testing, and more emphasis on the overall hardware and software efficiency of the entire system. , Security and Reliability. As the basic platform for chip design, EDA tools will be more AI applications as a key scenario to support chip designers to better complete the efficient implementation of various AI algorithms and applications on the chip. According to Gequn, Synopsys started using AI in early 1986. It uses artificial intelligence to support EDA, chip and system R&D. Currently, it is actively exploring innovations in design methodology and promoting the improvement of AI chip design capabilities. 2017 Synopsys With the establishment of a global artificial intelligence laboratory, we hope to work with the industry to explore new issues such as AI chip hardware and software collaborative optimization necessary to support artificial intelligence technology in a more open platform, and seek more effective methodologies, tools and solutions. Program. Based on the particularity of its application scenario, AI chips should have the following basic elements. Programmability to accommodate The evolution and application diversification; the architecture can be dynamically reconfigured, can adapt to different algorithms without significant reduction of efficiency; efficient architectural transformation capability; efficient computation and storage, low power consumption, low delay; low satisfaction Cost, small size requirements to enable access to home appliances and consumer electronics, mobile devices and the Internet of Things terminal market; application development is simple. The government and the company jointly promoted the 'artificial intelligence' in early March this year was written in the "Government Work Report", emphasizing To strengthen the application of new-generation AI R&D, promote the application and landing of AI in many fields such as medical care, pensions, education, culture, and sports. This shows that the Chinese government attaches great importance to the artificial intelligence chip industry. It also seized on the current AI industry. The key link in development. If China has gradually developed into one of the centers of the global AI industry, then the lack of application, data environment and shortage of talents will continue to be the major problems that restrict the development of China's AI chip industry. In terms of policy, the above Three levels of force will effectively promote the development of the AI chip industry. Academician Ni Guangnan of the Chinese Academy of Engineering stated: Intelligence is a very important development direction in the future. Because artificial intelligence is still created by us, we should better serve humanity, help human beings to get rid of complicated work, give repetitive labor to artificial intelligence, and humans should do more. More creative work. 'Ge Qun pointed out: 'This round of artificial intelligence bursts are symbiotic with big data. Without effective data, artificial intelligence, especially deep learning, is difficult to make progress. Nowadays, various industries and different There are various data islands in enterprises and institutions, and these data are all key factors in advancing artificial intelligence. If governments can promote the breaking of information islands, they will greatly promote the development of artificial intelligence industry. For the AI chip industry It also has a huge driving role. 'Whether it is chips, algorithms, data, applications, talents, or any short board will make artificial intelligence difficult to achieve rapid and balanced development. Domestic manufacturers should now maintain a certain degree of determination and the heart of artisans. Basic research must never fail. The government needs policy guidance and practical support from the national level. Only governments and enterprises Together, we can make China's artificial intelligence and semiconductor industry go further.
