Recently, Shenjian Technology was acquired by the American FPGA leader Xilinx. According to industry sources, the acquisition amount may be around 300 million US dollars. After the news came out, many netizens appeared to be outraged, and netizens said that China AI chips Troika: The Cambrian, the horizon, and Shen Jian Technology were thus abolished by the United States, and the government did not act.
Perhaps artificial intelligence is at the forefront, and the importance of artificial intelligence chips is actually overestimated by netizens. Shenjian Technology is one of the few in China, less than a lot, and whether it is for Shenjian Technology, For Xilinx, the completion of this acquisition is a win-win situation for both companies.
Shenjian Technology's acquisition by Xilinx is a win-win situation for the two companies
There are some similarities between Shenjian Technology and Cambrian. The founders are all technical singers from the Chinese Academy of Sciences. Their research results have been repeatedly captured abroad, and they belong to 'learning and doing business.' However, there are differences between the two. It was the Cambrian who chose to take the path of a dedicated chip, and Shen Jian Technology chose Xilinx-based FPGA to be its own artificial intelligence chip.
Specifically, Cambrian took the designed circuit diagram to TSMC to stream, while Shenjian Technology used its optimized neural network pruning technology on Xilinx FPGAs, and then used these FPGAs to run. Artificial intelligence application.
In other words, it is that Shenjian Technology has a strong dependence on Xilinx FPGAs. On the other hand, Xilinx also needs Shenjian Technology to help it develop the artificial intelligence application market. Therefore, the two companies are 'closed In a win-win situation, everyone can't do without it. In this case, the acquisition of Shenjian Technology by Xilinx is normal from a commercial point of view.
At present, China's AI chip companies have sprung up, and more than 40 have completed various financings. Most of them are often announcing how their chips hang on Nvidia, but the specific application is hard to see, and the ability to hematopoietic Unstable. Since Xilinx is willing to invest in Shenjian Technology to continue to burn money in order to make FPGAs better explore the artificial intelligence market, it is also very good. After all, Shenjian Technology is not much for China, one less. A lot of conditions will not threaten national security, nor will it affect the industry's monopoly by foreign capital. In this case, it is still commercial to commercial, after all, this little thing, the government does not need to interfere.
The technical route of Shenjian Technology makes it difficult to become an industry giant
As mentioned in the previous paragraph, Shenjian Technology is one of the few, and one less, state in China. The reason is that Shenjian Technology chose Xilinx-based FPGA as the AI chip. This technical route makes Shenjian Technology is hard to become an industry giant.
In recent years, with the rise of artificial intelligence, global integrated circuit design companies have launched their own programs, Intel pushes many core CPUs, NVIDIA pushes GPUs, Xilinx and Altera push FPGAs.
The advantage of FPGA is the performance-to-power ratio. For example, the product of Xilinx's biggest competitor, Altera, is an example. Earlier, Altera's FPGA for artificial intelligence has a performance-to-power ratio of up to 50GFlops/ W, if the artificial intelligence hardware selects the FPGA, it not only has obvious advantages in performance and power consumption ratio for Intel many core chips, but also dominates the GPGPU of NVIDIA. The higher performance-to-power ratio means operation and maintenance. Can save electricity costs.
If the performance-to-power ratio is the advantage of FPGA, then the price/performance ratio is the disadvantage of FPGA. FPGA can flexibly change its internal logic through programming, and it also means a lot of waste of computing resources. If you develop a special one and write The same chip into the FPGA circuit, the theoretical cost of production is only a fraction of the FPGA.
Another problem with FPGA is that the frequency is relatively low. This problem is brought out by the mother's womb. There is no medicine to save. With the advancement of competitors, continuing to do FPGA-based smart chips may not even have the advantage of performance-to-power ratio. maintain.
FPGA is a very special chip. It has better versatility than CPU before writing software. It has ASIC-like performance after writing software. This makes FPGAs have an advantage in publicity. When comparing functions, Compared with the high versatility when not writing software, when comparing performance, the outstanding specificity after writing the software overwhelm the opponent. However, the specificity and versatility of the actual application cannot be expressed at the same time, just like the innocence of childhood. Mature maternity can be manifested in different periods of the same person, but cannot be expressed at the same time.
Although FPGAs have achieved some success in some emerging applications, with the rise of emerging fields, their market size and importance continue to increase. Due to the lack of cost performance and low frequency, FPGAs will eventually be replaced by dedicated chips. .
In other words, SZ Technology's FPGA solution is only suitable for some emerging markets, or some market segments with small market capacity. For those large-scale mainstream markets, FPGA is difficult to compete with competitors.
It is for this reason that Shenjian Technology has limited meaning for China. However, for Xilinx, Shenjian Technology has a good meaning, because it is related to Xilinx's pioneering artificial intelligence market. Even if the market performance is normal in the future, Keeping the sign of artificial intelligence is also excellent for raising the stock price.
Chinese artificial intelligence chips are not too few, but too many
At present, a large number of domestic startups are catching up with artificial intelligence, and even Internet giants such as Ali and Baidu have come to the fore. These artificial intelligence chip companies have won the favor of capital and obtained high valuations. A thriving scene.
Although it seems that China's artificial intelligence chip industry is flourishing, in terms of specific applications, domestic artificial intelligence chips are rarely accessible to consumers. In addition to the Cambrian licensing of IP to Huawei, it has achieved tens of millions of applications. In addition, other domestic artificial intelligence chip companies in the market promotion and application performance is very general.
There are too many artificial intelligence chip companies. One problem is repeated construction and decentralization. The first is to spread the investment of artificial intelligence chips in the whole society. Secondly, the power to build the industrial ecological chain is dispersed. Finally, the scale is promoted. The potential of the application is weakened, and it is easy to become a bit of food for everyone, but no one can eat enough.
Tieliu believes that domestic artificial intelligence chip companies, what they really want to do, are not how to hang NVIDIA at the PPT of the conference, but to combine with specific applications, use the chips and sell them. The domestic lack is not the design. Companies that use artificial intelligence chips lack the enterprises that use these domestic artificial intelligence chips. Only when everyone uses domestic AI chips, will they have the opportunity to erode Nvidia's market share and not be broken by foreign giants.