Source: Technology Daily
"Only looking at the design architecture, domestic AI chips are no worse than foreign ones, but this is not enough." Wang Yu, a professor in the Department of electronic engineering of Tsinghua University, pointed out at the science popularization Salon of "science cafe" held by the Institute of physics of the Chinese Academy of Sciences recently that this is equivalent to building on sand and the foundation is not solid.
The Science and Technology Daily reporter then interviewed Gong Guoliang, deputy director of the Brain Computing Research Center of the Institute of Semiconductors of the Chinese Academy of Sciences, who also expressed similar views.
Design architecture is remarkable
Wang Yu once participated in the establishment of artificial intelligence chip company Shenjian Technology in 2016. Now more domestic enterprises making artificial intelligence chips have emerged. His feeling is that he has done "not worse than foreign countries, and some are even better."
The "good" mentioned here mainly covers the design architecture level. The design architecture has become the strength of domestic artificial intelligence chips, there are certain reasons behind it.
"Traditional CPU and GPU chips use von Neumann-based computing architecture and computing modes based on instruction streams to run, while artificial intelligence chips mostly use brain-like or brain-like architectures, which can break through the computational bottleneck of the memory wall." Gong Guo Liang introduced that in scenarios where deep neural network algorithms are used such as speech recognition and image recognition, artificial intelligence chips have significant advantages over traditional chips.
Due to the above-mentioned differences from traditional chips, Gong Guoliang told Sci-Tech Daily that from the perspective of design architecture, domestic artificial intelligence chips can quickly follow up related technologies, which are almost at the same level as developed countries.
The recent news can give a glimpse of the fiery progress of domestic artificial intelligence chips.
On June 20, the Cambrian launched the second-generation cloud artificial intelligence chip "Siyuan 270"; on June 21, Huawei released the artificial intelligence mobile phone chip "Kirin 810"; on July 3, Baidu released the far-field voice interaction chip "Honghu".
Gong Guoliang introduced that most of the artificial intelligence chips developed by companies such as Huawei and the Cambrian are general purpose chips, which can be used for a variety of artificial intelligence application scenarios. In addition, many universities and research institutes are also developing artificial intelligence chips, which are mostly dedicated chips, which have strong performance in special application scenarios.
The industrial chain still has shortcomings
Although the development of domestic artificial intelligence chips is in the ascendant and commendable in terms of design architecture, experts point out that, overall, China has a weak foundation in the chip field and still has many shortcomings.
Wang Yu said that the latest equipment and processes for manufacturing chips in mainland China are one or two generations behind the international advanced level. Therefore, some artificial intelligence chips need to be sent overseas for manufacturing and packaging, and the integrity of the industrial chain is not good.
Gong Guoliang confirmed this in an interview. He told the Sci-Tech Daily reporter that the artificial intelligence chips used on mobile phones are typical examples. These chips often need to use the latest technology to reduce power consumption, improve integration and computing performance. They are high-end chips. The demand is very large, but mainland China is not yet With manufacturing and packaging conditions.
The weaknesses in the development of domestic artificial intelligence chips also include high-speed interfaces and dedicated integrated circuit IP cores. Gong Guoliang introduced that the latter is an example. They have relatively high computing performance, complex design, and are closely related to the manufacturing process. To achieve such a design module usually requires years of precipitation and accumulation.
"High-speed interfaces and dedicated integrated circuit IP cores are regarded as core technologies by the industry, and the use of them often requires foreign companies to authorize licenses, and companies that do not have these technologies have difficulty obtaining licenses in the short term." Gong Guoliang said.
To sum up in Wang Yu's words, generally speaking, domestic artificial intelligence chips have done a good job in the areas where intelligence is needed, but they have not done well in the areas where accumulation is needed.
"Overall, research in domestic related fields started late, and it really needs to accumulate and precipitate." Gong Guoliang believes that in the long run, China's artificial intelligence chip field should seize the opportunity to make up for weak links, and control key core technologies in its own hands. To avoid being controlled by others.