Shenjian technology, a domestic AI chip unicorn, was announced to be acquired by Xilinx, the founder of FPGA. The specific transaction amount is unknown. This has caused great repercussions in the industry. Xilinx said that Xilinx’s strategy to evolve from FPGA devices to adaptive computing acceleration platform providers is to accelerate the deployment of FPGA acceleration technology from cloud to end applications. The neural network pruning technology optimized by Shenjian technology can be applied to Xilinx FPGA devices to achieve breakthrough performance and the best energy efficiency in the industry, This is a major step for the new CEO of Xilinx to improve the R & D capability of machine learning and promote strategic development after the press conference of ACAP in March.
You know, since Actel and Altera were successively acquired by MICROSEMI and Intel, Xilinx has occupied the leading position in the FPGA industry. Shenjian adopts Xilinx high-performance FPGA platform and integrates original in-depth learning technology. It has accumulated profound AI technical and industrial advantages and is popular in the market.
Since 2017, Xilinx has become the main investor of Shenjian technology together with other well-known investment institutions around the world. The acquisition seems to be a happy one. Xilinx said that FPGA is the most suitable product for innovation and entrepreneurship. The in-depth learning scheme of Shenjian technology has always been developed on the Xilinx platform. Over the past two years, Xilinx has not only witnessed but also accompanied the growth of the company, including last year’s investment and introducing end customers to help them grow. Now the acquisition can be said to be a natural thing, and the combination of the two sides is certainly more effective than joint efforts.
What kind of chemical reaction will occur after the combination of the two sides? How will the layout change in the future? Xilinx’s FPGA is mainly used in traditional fields such as communication, industry, aviation and national defense. However, with the emergence of emerging technologies such as AI, cloud computing, 5g and autopilot, the application scope of FPGA is expanding. AI machine learning technology based on FPGA can cover almost all mainstream applications, including 5g, automobile, industry, medical treatment, consumer electronics, data center, etc.
Similar to CPU and GPU, FPGA chip can provide computing power for the training and reasoning of deep learning algorithm. Especially in reasoning applications, FPGA has advantages in power consumption and performance, and has become one of the core technologies of AI applications. From the market point of view, FPGA has now entered the list of AI mainstream processors. From end to edge to cloud, FPGA will develop rapidly iteratively with its advantageous power consumption performance ratio and adaptive ability.
In AI cloud computing, in terms of data center acceleration alone, FPGA will have flexibility, lower energy consumption and higher ROI. Therefore, most data centers will adopt the mixed mode of FPGA + CPU + GPU. On the AI edge side, FPGA has the advantages of flexibility, low power consumption and multiplexing. Using only a single chip can bring security, confidentiality and multi-sensor fusion technology. In the process of continuous improvement and optimization of the algorithm, the flexibility advantage of FPGA will always be welcomed.
At the cloud computing level, Xilinx’s partners include Alibaba, Baidu, Tencent and Huawei in addition to Amazon AWS, the leader of foreign cloud services. On the AI side, Xilinx FPGA chip has accelerated its penetration in autopilot, embedded vision and industrial Internet of things.
Especially at the level of automatic driving, Xilinx and Shenjian technology have increased their horsepower. Xilinx is the second largest ADAS semiconductor supplier. Between fiscal year 2013 and fiscal year 2017, the average annual growth rate of revenue in the field of vehicle use of Xilinx banner has reached 60%, and it is cooperating with several large automobile manufacturers to carry out automatic driving plan. On June 26, Shenjian technology announced the launch of dphiauto, an embedded AI computing platform based on FPGA, and officially joined the autopilot track. It is reported that it has signed cooperation agreements with well-known automobile host manufacturers and primary suppliers in Japan, North America, Europe and China.
In the AI field, Xilinx and Shenjian will work together to give full play to the advantages of technology, channel, platform and landing, and promote the deployment of machine learning in FPGA from cloud to end application field. Will an amazing AI opponent surface? Will NVIDIA or Intel, its benchmark, have similar actions? In any case, a series of actions after Xilinx’s new CEO took office show that Xilinx wants to take the next big step!
Shenjian technology, Cambrian and horizon, as three representative players of domestic AI chips, are the top companies in this round of AI chip entrepreneurship wave, and their every move is eye-catching. Now, Shenjian technology has made a perfect turn, which is also known to other start-ups. AI chips are always faced with problems such as difficult landing of streamers and difficult finding of application scenarios. Now, the acquisition of Shenjian technology, one of the three representative players, may indicate that the AI chip market is about to enter a new wave of integration.
And the AI chip war will enter a new watershed. When the capital bubble is coming, many of them are on the verge of collapse. The future AI chip competition will be more real. The ultimate chip competition is not performance, but service. “NVIDIA’s products are not much better than others simply from the perspective of cost performance. But why in the server field, customers are willing to endure five or six times more expensive than others and use NVIDIA’s products? This is because in GPU and vertical industries, NVIDIA can also provide services in vertical industries at the same time.” Yao song’s words are simple and profound.
Therefore, the future competition of players gathered on the AI chip track is not just performance, how to combine the chip with the application scenario, how to build an ecosystem, and how to provide services from chip to board to software application layer. The road of competition in the future is still long.