The hottest search for China's AI chip industry ma

  • Detail

Explore the opportunities of China's AI chip industry from the "top four AI chips"

recently, the topic of AI has been heating up

Ma Huateng and Dong Mingzhu also repeatedly mentioned the development of artificial intelligence and chip industry. Ma Huateng also mentioned Tencent's future investment in AI chips

Deng Zhonghan, an academician of the Chinese Academy of engineering and the founder and chief scientist of Zhongxing microelectronics group, said that "AI is an important driving force for a new round of scientific and technological revolution and industrial change. AI, like important fields such as aerospace and military industry, must realize the autonomous control of the underlying core chips in order to ensure national security, public security and information security."

in fact, China's chip industry has lagged behind the West for a long time, while the artificial intelligence track is in the ascendant, and the outside world generally believes that it is a good opportunity to overtake on curves

AI chips are in the ascendant

in recent years, artificial intelligence has attracted much attention. In July 2016, artificial intelligence was included in the 13th five year plan for national scientific and technological innovation for the first time; In July 2017, the new generation of artificial intelligence development plan was issued. The relevant report also mentioned: "promote the accelerated development of emerging industries, deepen the R & D and application of big data, artificial intelligence, cultivate new generation of information technology, high-end equipment, biomedicine, new energy vehicles, new materials and other emerging industrial clusters, and expand the digital economy." It can be seen that AI will inject new momentum into future economic development

relevant people said that the legislation on artificial intelligence will be included in the research project, and an in-depth investigation and demonstration will be carried out around relevant legal issues, so as to provide a strong legal guarantee for the innovative development of artificial intelligence

AI applications and products known to the public include all kinds of intelligent speakers, intelligent cars, unmanned aerial vehicles, etc., but the most basic reason why AI can realize intellectualization is its chip and a large amount of data collected. The chip provides powerful computing power, so that machines can use a large amount of data to "deeply learn" to obtain a certain law, and finally realize the intellectualization of applications according to the law

it can be seen that AI chips occupy an important position in the whole AI industry chain. AI chips can be divided into general purpose chips (GPU), semi customized chips (FPGA), fully customized chips (ASIC) and brain like chips

according to the 2018 artificial intelligence chip research report published by aminer team of Tsinghua University, the mainstream architecture of AI chips has experienced a series of changes. From the most common CPU at the beginning, to the later with the development of high-definition video, VR, AR games and other industries, GPU (chip of graphics card) products have made a rapid breakthrough. At the same time, people found that the parallel computing characteristics of GPU just meet the needs of AI algorithms and big data parallel computing, so they began to try to use GPU for AI computing

after entering 2015, GPU's low performance power consumption ratio has limited its application. The industry has begun to develop special chips for artificial intelligence, that is, fully customized ASIC chips, which further improve the computing efficiency and energy consumption ratio through better hardware and chip architecture

industry analyst wangshuyi said that for most AI chip companies, it is ultimately necessary to survive in the market and find suitable application scenarios to land, which has become the inevitable choice for the development of most AI companies. Therefore, the development of AI industry has gradually developed from technology driven to scene and application driven, and fully customized chips are becoming more and more popular

Tang Weiwei, director of product strategy of bit Continental, once analyzed at the AI new product launch in October 2018 that behind the AI chip architecture change is the sharp increase in the increasing demand for AI computing power from all walks of life. The popularity and development of the Internet have produced a large amount of video, voice, image, text and other data. These data have encountered many problems in the process of transmission, storage and retrieval. The industry has an urgent need to make better use of these data to meet its own business needs

"Ai chip top four" molding

Wang Shuyi said that at present, AI chips can be divided into three echelons, and the first echelon has broken out of the encirclement and become the company of choice for the actual industry standard configuration; The second tier is still competing and has the opportunity to survive in some specific scenarios. For example, many companies at home and abroad do not rule out the opportunity for some companies to reach the general market and become giants; The third tier, companies that cannot find application scenarios and do not have enough money for the winter, will gradually lose their vitality in this cold winter

it is reported that at present, in the field of domestic AI chips, there are four companies with different characteristics called "four outstanding AI chips", which are Huawei Hisilicon, bitcontinent, Cambrian and horizon

bitland, a manufacturer established for more than five years, has promoted the process of blockchain chips from 55nm to 7Nm all the way, accounting for more than 70% of the global market share. Its prospectus shows that its revenue exceeded US $2.5 billion in 2017 and the first half of 2018, becoming the second largest IC design manufacturer in China after Hisilicon, with a valuation of US $15billion

blockchain chip is only the well-known side of bitland, and its relatively low-key AI chip focusing on the security field. Bitland set up a chip R & D team in 2015 and decided on the strategy of developing tensor processors for deep learning. After more than a year of research and development, it achieved original innovation, re innovation and integrated innovation in 2017 and 2018, respectively. It launched two generations of cloud AI chips, sold chip based deep learning boards and high-performance servers, and focused on security scenarios

in the application scenario of security, bitland has developed a suanfeng high-density intelligent server SA3 for video structure. This product can process about 90 channels of 1080p video data. It has been used for video intelligent analysis of public choice electronic universal experimental machine models, functional security, transportation and other scenes, technical facilities and computing services of video AI cloud, as well as video intelligent analysis of parks, radio and television and other scenes. A few days ago, bitland has reached strategic cooperation with Fuzhou municipal government departments, China Mobile Hangzhou Research Institute, Unicom Research Institute, etc., and its AI chip based computing cloud and servers, edge computing boxes and other products continue to be commercially available

in the Cambrian period, many things have also been done in the past year. For example, its cloud smart chip mlu100 and board card were officially launched, and a new round of financing was completed with an estimated value of $2.5 billion. Another example is to participate in the Symposium on curve display industry of private enterprises, and the Cambrian 1H blessing of Huawei Kirin 980

previously, Cambrian, together with eight partners and duanyun, appeared at the 2018 security Expo, which also indicates the determination to enter the security industry. Cambrian united with Xingchen semiconductor/sigmastar (the former mstarsmartcamera Division) and Lu Shenshi (a domestic cutting-edge company in the field of 3D vision) launched a system solution integrating Cambrian terminal intelligent processor IP products to show practical applications with chips + scenes

horizon recently made headlines again because of its $600million financing, and its valuation also rose to $3billion. The leader of this round of financing is SK Hynix. According to official news, breakthroughs will be made in the automotive processor architecture and the third generation processor architecture in the next year

the difference between horizon and the other three chip companies is that most of the founding team came from the algorithm industry, so the company has always announced that it will take the path of "software and hardware integration", focusing on both algorithm chips. In the past few years, horizon has launched two chips, journey and sunrise, as well as matrix, an autonomous driving computing platform. Other products have not appeared in public information

ai vision chip "rising sun 1.0" released at the end of 2017 marks that they have officially entered the field of smart city and smart retail. Horizon believes that their opportunity lies in extending their chip and algorithm capabilities in the field of autonomous driving, including chip based real-time processing of massive data, as well as large-scale intensive training and model optimization capabilities, to the video security market

the last one is Huawei Hisilicon, which is valued at $35billion. In addition to a series of Kirin chips, Hisilicon has also been working in the field of smart cities for a long time. On October 10, 2018, at the 2018 Huawei full connect conference, Hisilicon announced all in AI and released a full Scenario Oriented technical solution, which is a full stack solution including chips, chip enablers, training and reasoning frameworks, and application enablers

specifically, based on Hisilicon's "Da Vinci" architecture, Hisilicon released the server chip shengteng 910 and the terminal chip shengteng 310. Both new products will not be sold separately, and will eventually be sold in the mode of AI accelerator card, accelerator module, server and all-in-one machine

as a manufacturer with strong capital and scientific research strength, Huawei's ambition is to establish its own set of AI ecosystem from the bottom chip to the upper application, not just a simple chip manufacturer

the above four AI chip companies have different strategic characteristics, including those that focus on specific scenarios and those that need to do AI ecology. Wang Shuyi said that the two types of companies are not contradictory. On the one hand, they must be commercialized and combined with specific application scenarios. Only when they reach a certain scale, fully customized ASIC can have economic benefits, so it is also possible that general chips can be used in special scenarios

there are about two kinds of ecological enterprises, in addition to "ppt" companies. One is the head companies that have overcome the siege, such as NVIDIA, Intel and Xilinx abroad; The other is the one who will fight out of the siege in the future

in the future, how the "Ai chip top four" can overtake in corners under the competition of foreign giants who take the lead still needs the test of time and market

Copyright © 2011 JIN SHI