AI chips are also known as AI accelerators or compute cards, which are modules dedicated to processing a large number of computing tasks in AI applications (other non-computing tasks are still handled by the CPU).
Artificial intelligence chip industry competition pattern Analysis of the development trend of artificial intelligence chip industry
The industry's definition of AI chips still lacks a rigorous and accepted set of standards. The broader view is that chips for artificial intelligence applications can be called AI chips. Due to the diversity of requirements, it is difficult to have a single design and approach that can be well adapted to every situation. As a result, a variety of novel designs and approaches have emerged specifically for AI applications, covering all levels from semiconductor materials, devices, circuits, and architectures.
Artificial intelligence chips have two outstanding features: first, the algorithm and the chip are highly compatible, improving computing power for different needs of terminals and clouds; The second is intelligent chips specifically for subdivided application scenarios, such as voice recognition chips, image recognition chips, video surveillance chips, etc. The current artificial intelligence is in the early stages of industrialization, and all countries are on the same starting line. The Chinese government has paid great attention to artificial intelligence from top to bottom, and has completed a series of top-level design at the policy level. The management of data sovereignty and the localization of application scenarios with a large amount of data will further help the rise of local chip companies in China. As foreign-funded enterprises rooted in China, they should also actively participate in the tide of China's artificial intelligence development, carry out win-win cooperation with local companies in technology, market and talent, and jointly help the development and growth of the Chinese industry intelligence.
According to the "2022-2027 Chinese Industry Development Analysis and Investment Risk Prediction Report" released by the Puhua Industrial Research Institute of the Chinese Research Institute, it is shown that:
Section 1 Analysis of the overall situation of the global artificial intelligence chip market
First, the development process of the global artificial intelligence chip industry
Since the development of the global semiconductor industry, there have been three transfers.
The first stage: transfer from the United States to Japan. Japan began to comprehensively learn American semiconductor technology from assembly, and the era of electrical appliances changed to the era of PCs;
The second stage: transfer from the United States and Japan to Taiwan and South Korea. Japan's chip industry declined due to pressure from the United States and the weak local economy, and the PC era continued to prosper;
The third stage: transfer from the United States, Taiwan, and South Korea to China. China's chip industry has a huge downstream market, abundant labor, and is backed by national industrial policies and active social capital. The PC era has changed to the era of Internet of Everything and artificial intelligence, stimulating the demand for AI chips.
Second, the global artificial intelligence chip market size
Chart: Global AI chip industry market size from 2019 to 2021 (unit: billion US dollars)
Data source: Puhua Industrial Research Institute
The global AI chip market size in 2021 is about $26 billion. With the maturity of artificial intelligence technology and the continuous improvement of digital infrastructure, the commercial application of artificial intelligence will accelerate and promote the rapid growth of the AI chip market.
Third, the regional distribution of the global artificial intelligence chip market
Europe and the United States are still the leading regions for the development of the global artificial intelligence chip industry
To a certain extent, the innovation of urban AI technology also represents the development level and development potential of the city's AI chips. The city is a comprehensive carrier for the integration and application of AI technology innovation, and it is also a centralized experience place for human beings and AI technology to produce comprehensive perception. In the past few years, major cities around the world have played a differentiated role in the development of AI technology, built their own ecosystems, and achieved initial results in enabling industrial applications and boosting regional economic development, setting off human thinking, cognition and action for a new round of industrial revolution.
As AI applications have landed at the city level, cities have gradually become the main battlefield for AI innovation and integration applications. While the key success factors of AI technologies vary around the world, they generally build ecosystems that facilitate the integration of technology and cities.
Fourth, the global artificial intelligence chip industry competition pattern
In recent years, various forces are working on AI chips, including traditional chip design, IT vendors, technology companies, the Internet and start-ups, etc., and products cover CPU, GPU, FPGA, ASIC, etc.
According to the list released by relevant market research institutions, the top ten are still European, American, Korean and Japanese companies, and domestic chip companies such as Huawei HiSilicon, MediaTek, Imagination, Cambrian, Horizon Robotics and other companies have entered the list, of which Huawei HiSilicon ranks 12th, Cambrian ranks 23rd, and Horizon Robot ranks 24th.
Chart: Global AI chip industry ranking
Data source: Puhua Industrial Research Institute
5. Development trend of global artificial intelligence chip industry
(1) More efficient large convolution deconstruction/reuse
On the basis of standard SIMD, CNNs can further reduce data communication on the bus due to their special multiplexing mechanism. The concept of multiplexing is particularly important in very large neural networks. How to reasonably decompose and map these super-large convolutions to effective hardware has become a direction worth studying.
(2) Lower Inference compute/memory bit width
Perhaps one of the biggest evolutions of AI chips is the rapid reduction of neural network parameters/compute bit widths — from 32-bit floating-point to 16-bit floating/fixed-point, 8-bit fixed-point, or even 4-bit fixed-point. In the field of theoretical calculation, 2 bits or even 1 bit wide parameter has gradually entered the field of practice.
(3) More diverse memory custom design
When the computing component is no longer the design bottleneck of the neural network accelerator, how to reduce the access delay of the memory will become the next research direction. In general, the closer to the computation, the faster the memory speed, the higher the cost per byte, and the more limited the capacity, so new storage structures will emerge.
(4) Sparser large-scale vector implementation
Although neural networks are large, there are actually many cases with zero as input, and sparse computing can effectively reduce useless energy efficiency. The team from Harvard University proposed an optimized five-stage pipeline junction for this problem, with a trigger signal output at the last stage. After the activation layer, the necessity of the next calculation is predetermined, and if it is found that this is a sparse node, the SKIP signal is triggered to avoid the power consumption of the multiplication operation to achieve the purpose of reducing useless power consumption.
(5) Integration of computing and storage
The main point of computing and storage integration (process-in-memory) technology is to use new non-volatile memory (such as ReRAM) devices, add neural network computing functions to the memory array, thereby eliminating the data migration operation, that is, realizing the neural network processing of computing and storage integration, and can be significantly improved in terms of power performance.
In recent years, AI technology has made breakthroughs. As an important physical foundation of AI technology, AI chips have great industrial value and strategic position. However, from the perspective of the general trend, it is still in the initial stage of the development of AI chips, and there is huge room for innovation in both scientific research and industrial applications. Now not only international giants such as NVIDIA and Google have successively launched new products, but also domestic Baidu, Ali, etc. have laid out this field, and AI chip startups such as Cambrian have also been born. In the case that the traditional chip field such as CPU and GPU is far from the world, China's AI chips are expected to achieve corner overtaking.
At present, traditional chips can no longer meet the requirements of artificial intelligence for chip performance and computing power, how to build efficient artificial intelligence chips, and effectively combine chip technology and artificial intelligence technology has become a hot topic.
Based on a large number of meticulous market research, this report is mainly based on a large number of materials and data published and provided by the National Bureau of Statistics, the Ministry of Commerce, the National Development and Reform Commission, the National Economic Information Center, the Development Research Center of the State Council, the Ministry of Industry and Information Technology, the China Industry Research Network, and a variety of relevant newspapers and magazines across the country and overseas, and objectively and multi-angle analysis and research on the Chinese intelligent chip market.
On the basis of summarizing the development process of Chinese intelligent chips, the report combines various factors in the new era to give a detailed and prudent prediction and demonstration of the development trend of Chinese intelligent chips. The report has detailed information and rich charts, both in-depth analysis and intuitive comparison, which provides insight into opportunities for artificial intelligence chip companies in the fierce market competition, and can accurately and timely adjust business strategies for their own environment.
For more industry data details, you can click to view the "2022-2027 Chinese Industry Development Analysis and Investment Risk Prediction Report" by Puhua Industrial Research Institute of China Research Institute.
Follow the headline "Central Research Institute Puhua Research Institute" and reply to "Free Report" to get an excerpt from a free report, pick it up quickly!