Recently, ByteDance's official recruitment website has opened a number of semiconductor-related positions, mainly located in Beijing and Shanghai. According to the news, Byte is self-developing cloud AI chips and Arm server chips, so it is recruiting talents. In response to this, the byte official replied: it is setting up relevant teams to do some exploration in the field of AI chips.
ByteDance will also enter the semiconductor industry?
On March 16, Reuters reported that ByteDance is self-developing cloud AI chips and Arm server chips, but the plan is still in its early stages.
The news started from the 12 newly opened positions on the company's official recruitment website, mainly located in Beijing and Shanghai, all related to semiconductors.
ByteDance gave an official reply: it is setting up relevant teams and doing some exploration in the field of AI chips.
China is the world's largest purchaser of semiconductors, but its domestic chip production capacity lags behind the United States, Japan and South Korea. In this regard, some netizens gave a thumbs up to the layout of Bytedance:
For a long time, Chinese technology giants have relied on foreign manufacturers such as Qualcomm and Nvidia. If China stepped up to design its own chips, under the background of today's chip shortage, Internet companies can reduce their dependence on third-party supplies and reduce costs.
Recruitment positions, focus more on chip design
On the official website, these 12 positions include chip application (ARM hardware and software optimization), chip application (ARM hardware, Beijing/Shanghai), chip CAD engineer, chip synthesis/STA engineer, chip DFT engineer, chip back-end design engineer, chip CAD etc.
It can be seen that these positions focus on chip design, and a job description requires candidates with experience in chip design servers and electronic design automation (EDA) tools (software used to design advanced chips).
In the era of artificial intelligence, algorithms are king.
The most common type of artificial intelligence chip is the graphics processing unit (GPU), which is used to train artificial intelligence algorithms, while the field programmable gate array (FPGA) and application-specific integrated circuit (ASIC) are used to apply the trained artificial intelligence algorithms Data input in the real world.
As we all know, TikTok has attracted about 100 million users in the United States, and its popularity has been comparable to YouTube. The magic of its fame lies in its "content recommendation algorithm."
The algorithm used by TikTok is independent of the domestic Douyin. This algorithm is mainly used in TikTok's "For You" interface to recommend the next video by analyzing user behavior. Therefore, the content and user information used by the algorithm are more difficult to handle.
Combining byte services, whether it is Toutiao, Douyin or TikTok in the 5G era, its platform must carry more content, which requires server chips such as data centers and edge computing.
Self-developed cloud AI chip and Arm server chip
Since Google released TPU in 2016, it has become a trend for international Internet giants to enter the chip field.
Cloud AI chips and Arm server chips are the main ways for many Internet companies to enter the chip market.
First, let's talk about cloud AI chips.
When it comes to the global cloud AI chip market, NVIDIA still has an advantage against the sky, but due to the different ecology built by Internet companies, their requirements for chip performance are also different.
In this case, self-developed chips have become a development path.
However, compared with AI cloud chips that are difficult to develop and have clear goals, GPUs based on AI acceleration are currently still the mainstream in cloud applications. In February of this year, Byte invested in the local GPU chip design rookie Moore thread and deployed it to the GPU field.
Second, the Arm server chip.
Due to the increasing market demand for server chips, the Arm server chip market has attracted the participation of many manufacturers, and Internet companies are one of them.
Joining the server chip competition can not only reduce their dependence on third-party supplies, on the other hand, reducing costs is also extremely attractive. For example, Amazon has developed the Arm server chip Graviton.
Therefore, choosing the Arm architecture to increase self-research is conducive to further adapting to the market.
However, in addition to the Arm architecture, the open source architecture RISC-V is also gradually emerging. According to reports, in January this year, Byte also invested in the research and development of RISC-V.
Chinese Internet giants lay out the chip field
However, like most technology companies involved in chips, even if they successfully design their own AI chips, they still need to outsource their manufacturing work to independent foundries, such as TSMC and SMIC.
At present, the crisis of chip shortage is spreading around the world. This situation began in the second half of last year and has worsened in recent months. The production capacity of TSMC and other foundries are stretched because they are giving priority to responding to the needs of automakers and consumer electronics companies.
According to data from the research company ICInsights, of the 143 billion US dollars of chips sold in China in 2020, only 22.7 billion US dollars will be produced in China, while the China-based company will only produce 8.3 billion US dollars.
Beginning last year, the US sanctions on hardware manufacturer Huawei Technologies Co., Ltd. prevented the Chinese telecommunications giant from purchasing components including semiconductors, thereby weakening its smartphone business.
In China, the traditional three-person Internet giants known as BAT-Baidu, Alibaba and Tencent, have all developed internal chipsets.
The head of China's Ministry of Industry and Information Technology (MIIT) said this month that it will vigorously support its semiconductor industry.
Chinese search engine giant Baidu has completed a round of financing for its Kunlun AI chip unit, which is valued at approximately US$2 billion. Baidu is also considering turning the department into an independent company to commercialize its chip design capabilities.
Alibaba launched its first AI inference chipset in 2019, named Hanguang 800. Earlier this year, the first core processor IP Xuantie 910 was also released, which can be used for 5G networks, AI and autonomous driving.
Chinese AI unicorn SenseTime also began to develop local AI chips in 2019.
Semiconductors are like the pulse of this era. The existence of factories represented by TSMC allows other companies to participate in the chip industry in a more convenient way.
I don’t know if Byte and other Chinese Internet will also "beat" the pulse of the semiconductor industry.