The development of AI chips cannot be separated from the maturity of artificial intelligence technology. Since its inception in 1956, artificial intelligence has experienced three major waves. In the 21st century, due to the improvement of computer performance and the generation of massive data, as well as breakthroughs in machine learning and CNN technology (Convolutional Nerual Networks, Convolutional Neural Networks), algorithms, computing power and data have all met the commercialization requirements of artificial intelligence Artificial intelligence has ushered in a stage of rapid development.
In fact, the rapid development of the artificial intelligence industry is inseparable from the current physical foundation, the chip. It can be said that "without chips and AI", whether it is possible to develop chips with ultra-high computing power and meet market demand has become an important factor in the sustainable development of artificial intelligence.
In recent years, the AI chip industry has developed rapidly, and many companies have made arrangements. But from the three stages of the chip's inception, development, and maturity, artificial intelligence chips are still in their infancy.
Giant emerging, met industry bottlenecks
According to different application scenarios, AI chip design can be divided into three parts: cloud training, cloud inference, and terminal inference. The cloud training chip is mainly based on Nvidia's GPU. The new competitors are Google's TPU and Xilinx and Intel, which are deeply involved in FPGA. In terms of cloud inference, representative companies include AMD, Google, Nvidia, Baidu, and Cambrian.
In terms of terminal inference, due to the gradual explosion of demand for application scenarios such as mobile terminals and autonomous driving, layout companies include traditional chip giants and start-ups, such as Qualcomm, Huawei Hisilicon, Horizon, Cambrian, and Bitmain.
It is not difficult to find that in the market structure, although the traditional chip giants currently occupy the dominant position of the AI chip market. However, the difficulty of landing AI chips is a common problem that plagues giants and emerging players.
Source: Tsinghua University Future Chip Innovation Center
Huang Chang, co-founder and vice president of Horizon, told Eotech that the difficulty in landing AI chips is first because everyone has encountered a common bottleneck in technology, which is also the so-called "von Neumann bottleneck."
One of the keys to improving the performance of AI chips is to support efficient data access. In the traditional von Neumann architecture, data is extracted from the memory outside the processing unit and written back to the memory after processing. The AI chip itself is based on the von Neumann architecture, and simple functions are perfectly fine.
However, due to the speed difference between the computing component and the storage component, when the computing capacity reaches a certain level, the speed of accessing the memory cannot keep up with the speed of the data consumed by the computing component, and the increase of the computing component cannot be fully utilized. Iman bottlenecks, or "memory wall" issues, have long been a problem for computer architectures.
At present, a common method is to use a hierarchical storage technology such as a cache to mitigate the difference in speed between operations and storage. However, the amount of data that needs to be stored and processed in AI chips is much larger than previously common applications. This makes von Neumann's bottleneck problem more and more serious in the application of AI. "It is no exaggeration to say that most of the hardware architecture innovations proposed for AI are fighting this problem." Huang Chang added.
However, it is also due to the technical difficulties of artificial intelligence chips that both giants and emerging talents are on the same starting line, which provides domestic enterprises with a good surpassing "track". This also avoids the traditional giants taking advantage of their existing advantages to quickly shake off their opponents.
Ask the world how to overtake the curve
On June 20, 19, Cambrian launched the second-generation cloud "Siyuan 270"; on June 21, Huawei released artificial intelligence mobile phone chip "Kirin 810"; on July 3, Baidu released artificial intelligence chip far-field voice Interaction chip "Hongye"; On October 29, Horizon released the AIoT edge computing artificial intelligence chip "Sunrise II".
It can be found that the layout of domestic companies in the field of AI chips has begun to take shape, and there is a battle. But if we want to win the world, we still need to improve some shortcomings.
In response to the development of domestic AI chips, Ni Guangnan, an academician of the Chinese Academy of Engineering, has repeatedly stated that the threshold for chip design is very high, and only a few companies can afford the high-end chip R & D costs, which also limits innovation in the chip field. China can learn from the successful experience of open source software, lower the threshold for innovation, improve the autonomy of enterprises, and develop domestic open source chips.
"Open source software is becoming the mainstream of the current software industry, and the chip industry can also adopt the open source model." Ni Guangnan emphasized that in terms of chip development, the new RISC-V instruction set is a new type that can reduce the IP cost of processor chips. mode. Enterprises are free to use RISC-V for CPU design, development, and add their own instruction set for expansion. RISC-V has a very good effect on the optimization of the current AI chip architecture and cost control.
Regarding the AI chip architecture, in fact, there have been many remarkable cases in Chinese enterprises, such as the Da Vinci architecture of Huawei, the Cambricon-X architecture of the Cambrian, the CAISA architecture of Jinyun Technology, and the Bernoulli architecture of Horizon. .
Compared with the artificial intelligence chip architecture, China should pay more attention to the integrity of the artificial intelligence chip industry chain.
The latest equipment and technology for manufacturing chips in China are many generations behind the international advanced level, so some artificial intelligence chips need to be sent abroad for manufacturing and packaging. This will cause insufficient chip production and high prices. As a result, many downstream products using its modules cannot be mass-produced, creating a vicious circle and not conducive to the development of the industry.
As a pioneer in the field of domestic edge AI chips, Canaan Technology has mastered the 16nm process technology as early as 2016. The reason why the current AI chip process technology is still 28nm is mainly due to price and shipment restrictions.
CCID Consulting's "White Paper on the Development of China's Artificial Intelligence Chip Industry" shows that the size of China's artificial intelligence chip market has maintained rapid growth. In the cloud sector, the global cloud market accounted for 17.0% in 2018; it is expected to reach 22.15 billion yuan in 2021 and a CAGR of 51.23%. In the terminal field, it will reach 8.41 billion yuan in 2021, with a CAGR of 59.3%.
Faced with such a broad market, I hope that domestic companies can concentrate on breaking through the bottleneck and reaching the world.
From [Moore News]