The reporter saw on the official website of Tencent’s recruitment on July 14 that Tencent is already recruiting chip architects, chip verification engineers, chip design engineers and other positions. The position is set up under Tencent Technology Engineering Group (TEG).
Take the position of chip architect released on July 9 as an example. The job responsibilities of this position are: responsible for the demand analysis of AI chips and general processors or subdivisions, leading AI, processor chip architecture design, competition analysis and specification definition. Work requirements: Familiar with AI chips and general-purpose processor architecture and circuit design, first-line technical experts of the design company are preferred. The location of the above chip-related jobs can be Beijing, Shenzhen or Shanghai.
In the past, Tencent mainly deployed investment in the past, investing in Shanghai-based AI chip company Suiyuan Technology. Suiyuan Technology focuses on general artificial intelligence training and reasoning products. At the 2021 World Artificial Intelligence Conference, Suiyuan launched the company’s second-generation general artificial intelligence training chip “Yisi 2.0”.
In March this year, Shenzhen Baoan Bay Tencent Cloud Computing Co., Ltd. was registered and established. Its business scope includes integrated circuit design and research and development, which triggered the guess that Tencent would personally end up making cores.
In response to the above news, Tencent exclusively responded to the reporter saying that based on some business needs, they have tried to develop some chips in specific areas, such as AI acceleration and video codec, and non-universal chips.
According to Chinese chip industry insiders, AI chips are strongly related to application scenarios. AI acceleration and video codec may be applied to Tencent’s game business scenarios, and the more specialized the application scenarios, the narrower and narrower the design is relatively less difficult.
Baidu and Alibaba, among the BAT three giants of China, both entered the AI chip track before, and moved faster than Tencent.
In 2018, Baidu released the cloud-based full-function AI chip “Kunlun”, which included the training chip Kunlun 818-300 and the inference chip Kunlun 818-100. At that time, Baidu Chairman and CEO Robin Li introduced that “Kunlun” was China’s first cloud-based AI chip. Functional AI chip.
According to information released by Baidu, Baidu’s first-generation Kunlun chip has been mass-produced in early 2020, and more than 20,000 pieces have been deployed on a large scale. The second-generation Kunlun chip has been successfully taped out and will be mass-produced in the second half of 2021.
In June of this year, Baidu announced the establishment of an independent new company in its Kunlun chip business called Kunlun Core (Beijing) Technology Co., Ltd. Kunlun Core has completed independent financing in March, with a valuation of about 13 billion yuan.
Alibaba announced the establishment of Pingtouge Semiconductor Co., Ltd. at the Yunqi Conference in September 2018. Pingtouge was formed by the integration of the Zhongtian Microsystems Co., Ltd. previously acquired by Alibaba and the self-developed chip business of Dharma Institute. The main task is to engage in Self-research and development of chips and strategic layout.
In September 2019, Alibaba’s first self-developed AI inference chip Hanguang 800 was released. At that time, Alibaba said that Hanguang 800 has been mass-produced and only serves Alibaba internally, and does not intend to sell it independently to third-party companies for the time being.
Why get together with artificial intelligence chips to start businesses? Cambrian, an AI chip company listed on the Science and Technology Innovation Board in 2020, introduced in its annual report: As the current artificial intelligence technology represented by deep learning is widely used in daily life and traditional industries, the demand for computing power of underlying chips has been growing rapidly, and its growth rate has greatly exceeded the speed of Moore’s Law. Artificial intelligence computing often has the characteristics of large computing capacity, high concurrency, and frequent memory access, and the computing modes involved in different sub-fields (such as vision, speech and natural language processing) are highly diversified, posing huge challenges for the micro-architecture and instruction integration, manufacturing process and even supporting system software of chips.
Cambrian believes that, compared with traditional chips such as CPU and GPU, general-purpose smart chips can better match and support key computing operations in artificial intelligence algorithms, and have significant advantages in performance and power consumption. With the gradual release of the demand potential of the artificial intelligence market, general-purpose artificial intelligence chips will become mainstream products in this market in the future.
However, the initial investment required for chip development is extremely high. Taking the Cambrian as an example, the company has a net loss of 434 million yuan in 2020. Cambrian said that the main reason for the company’s continued losses was the further increase in R&D investment in accordance with the company’s strategic plan.
Another difficulty for AI chip entrepreneurship is to find the market. The AI chips of Baidu and Alibaba are currently mainly serving their own businesses, which provides a certain guarantee for the implementation of self-developed chips.
According to information released by Baidu, on April 23, Baidu created the “Baidu Smart Cloud-Kunlun Smart Ecological Center” in Hubei Yichang Dian Military Region, and plans to invest in the construction of an artificial intelligence computing center built by Baidu Kunlun chips. This is also the first industrialization of Baidu’s Kunlun chip ecology nationwide.
Alibaba Cloud introduced in July 2020 that Alibaba Cloud has deployed hundreds of cloud data centers in 22 regions around the world, including 5 super data centers. Alibaba’s self-developed AI chip Hanguang 800 is deployed on a large scale in super data center.