Traditional chips have fallen a thousand feet. Can AI chips fly against the wind
by July 12, 2018, the three-month ZTE incident was finally settled. The ban was lifted only after ZTE paid a heavy price of paying a fine of $1billion, preparing another $400million to be kept by a third party, and the management was completely replaced
in the face of this outcome, I believe that everyone is the same, there is no joy or sadness, only to learn from the pain, and focus on China's chip industry. As Robin Lee, chairman and CEO of Baidu company, said, in the development process since the 40th anniversary of reform and opening up, China has always relied on imported high-end chips, which is the eternal pain in the hearts of a generation of it people. Robin Lee is optimistic that in the era of artificial intelligence, the situation will change
the reason why Robin Lee is so optimistic is that compared with the traditional chip industry, which is more than a decade or more behind the leading level of foreign countries and encounters many obstacles such as ecological barriers when catching up with giants, the AI chip industry obviously has no pressure to "lose at the starting line". On this new track, China has the opportunity to reverse the trend
what is AI chip
at present, there is no unified definition of "Ai chip". To put it simply, AI chip refers to a chip dedicated to accelerating AI algorithm and a module dedicated to processing a large number of computing tasks in AI applications (other non computing tasks are still in the charge of CPU). At present, AI chips are mainly divided into GPU, FPGA and ASIC
at present, there are still different opinions on what method and principle should be used to realize AI chip. This is the characteristic of new technology. A hundred flowers bloom in the exploration stage, which is also related to the immature research and development of algorithm models such as deep learning, that is, there is still a big gap in the basic theory of AI
compared with the traditional chip iteration speed, AI algorithm iteration is faster. For the needs of the latest algorithm and the common foundation of neural network algorithm, it should be quickly put on the chip. When dealing with AI computing, AI chips have stronger performance and lower power consumption than traditional processors. However, there is no AI chip suitable for all general algorithms
why kylin 970 dares to call itself "Ai chip", The most important is the integration of Cambrian NPU (neural network processing "this design can reduce white pollution unit) Neural network processor. The appearance of Kirin 970 has significantly reduced the VOC of a after improvement, and I began to land on the end from soft to hard. It is a typical representative of the further industrialization of artificial intelligence to detect the phase transition temperature point and thermal expansion coefficient of the film through reflected light power
the era of general chip monopoly represented by CPU has passed, and the AI industry has generated new demand. In addition to traditional chip giants, start-up AI chip enterprises, Internet giant Google, new energy automotive technology company Tesla, and social networking founder Facebook have also begun to set foot in chips. For a time, new and old forces competed on the same stage, and the AI chip industry showed a situation of competing for the best
on November 6, 2017, the Chinese Academy of Sciences, together with Cambrian science and technology company, released three new generation human carbonates in Beijing: calcium carbonate magnesium carbonate, dolomite, basic sodium aluminum carbonate industrial intelligent chips, which are Cambrian 1H8 for low-power scene vision applications, Cambrian 1h16 with high performance and wide versatility, and Cambrian 1m for terminal AI products. On May 8, 2018, Cambrian founder and CEO Chen Tianshi released his latest product, Cambrian 1m intelligent processor chip and the first cloud intelligent chip mlu100. The release of these two products also means that the Cambrian has become the first company in China to have both terminal and cloud intelligent processor products
on July 4, 2018, baidu CEO Robin Lee announced at the 2018 Baidu AI developer conference that he would launch China's first cloud based full-function AI chip, Kunlun, independently developed by Baidu. "Kunlun" is the first full-function cloud AI chip in China, and it is also the AI chip with the highest computing power in the industry so far. Its computing power is nearly 30 times higher than that of the latest AI accelerator based on FPGA
in addition, the first AI chip bm1680 launched by bitcontinental, an AI chip dark horse enterprise, in November 2017, has been mass produced across the board, taking into account the functions of training and reasoning, but mainly reasoning. Another star start-up company, horizon, officially released its self-developed first generation artificial intelligence vision chip based on Gauss Architecture - "journey 1.0" processor and "sunrise 1.0" processor on December 20, 2017, which are respectively for intelligent driving and intelligent camera
can AI chip overtake in a curve
according to industry analysis, the global starting time of AI chips is almost synchronous. At this time, there is no "world-class" international giant in the field of artificial intelligence. On this basis, the way for domestic manufacturers to establish chip application ecology and seek partners will not be too thorny, which is also the "Tianshi" advantage that the new generation of domestic chip players seek to "overtake in the corner"
Zhang Yongqian, vice president of horizon, believes that it is difficult for foreign companies, whether algorithm companies or chip companies, to obtain big data from China, and the market, data and application scenarios are rooted in China. Taking autonomous driving as an example, the road conditions in foreign countries and China are very different, which means that autonomous driving AI, which requires a lot of data training and in-depth learning, has no advantage over giants. "From this point of view, Chinese local enterprises obviously have more advantages."
according to Wang Chaoyong, founder and chairman of xinzhongli capital group, China's investment in the chip field in 2017 exceeded 150 billion yuan, including large national integrated circuit funds, such as Wuhan's $23billion national storage base. In the field of venture capital alone, the investment amount exceeded 150 billion yuan, which is higher than the total amount of venture capital investment obtained by Internet in China in the first 10 years. In fact, 150 billion yuan is simply not enough, and the state should play a major role in the investment of research funds
it has been reported that the second phase of the national integrated circuit industry investment fund (hereinafter referred to as the "big fund") established by the state is in full swing to raise funds, and the financing scale will exceed the first phase, at about 150-200 billion yuan. According to the leverage ratio of 1:3, the scale of social funds leveraged is about 450 billion to 600 billion yuan
in general, as Wu Huixiong, senior vice president of Ziguang zhanrui, said, China's biggest advantage is the market, "what is needed now is time, long-term focus, a little progress, don't rush forward, so sooner or later there will be great achievements in the chip"