Bats cross the border to play chips.

Not long ago, Tencent was exposed to start recruiting various jobs related to chip R&D and design. Recently, Tencent responded by saying that it is trying chip R&D in specific areas based on some business needs, such as AI acceleration, video codec, etc., rather than general-purpose chips.

In fact, it is not unusual for an Internet giant like Tencent to develop its own chips. In the downstream and terminal of the chip industry, a large number of vertical head enterprises whose business involves AI hardware products have brought chip self-research into the framework of future development strategy.

Cross-border core research has become a trend.

In addition to vertical enterprises and platforms in the chip industry chain, there is an obvious cross-border trend in chip research in recent years. Of course, self-research here mainly refers to R&D and design, excluding manufacturing, packaging and testing, which can only be done by the manufacturer.

Judging from the player base of cross-border self-developed chips, there are three main categories at present.

The first category is Internet companies such as Tencent, Ali and Baidu. For example, Ali's first self-developed AI reasoning chip "Heguang 800" was publicly released as early as 20 19, and Baidu's first-generation Kunlun chip was also mass-produced early last year.

The second category is pan-electronic and digital consumer goods enterprises, such as Xiaomi, Huami, Gree and Xpeng Motors. Xiaomi released the first self-developed mobile phone image related chip this year. Gree also said some time ago that the self-developed chip has passed the mass production verification. As a wearable manufacturer, Huami released the first self-developed chip "Huangshan 1" as early as 20 18, and Xpeng Motors also said earlier that the self-developed chip project has been carried out simultaneously in China and the United States.

The third category is AI algorithm companies such as Yi Tu Science and Technology and NavInfo. In 20 19, Yi Tu released the first self-developed cloud AI chip "Xun", and in February this year, one of the goals of NavInfo's new release was to develop its own intelligent networked automobile chip.

It can be found that cross-border self-developed chip companies all have one thing in common: they all have hardware business involving AI technology. For giants like BAT, cloud computing and automobile-related businesses mostly need self-developed chips, while for intelligent hardware companies like Xiaomi, most products need self-developed chips. As shown in the figure, these algorithms and software companies are more interested in developing chips as third-party suppliers.

The threshold for self-research is not high.

In the mobile phone industry, the self-research of advanced process chips is beyond imagination. At present, only Huawei and MediaTek have chip design capabilities in China, so we can say that they have certain self-research capabilities regardless of manufacturing.

However, the objective reason why the players' self-developed chips mentioned above get together is that these fields are different from mobile phones, and the difficulty of chip design and manufacturing is much smaller, that is, the threshold is lower.

The requirements for chip manufacturing vary greatly in different fields. In the field of mobile phones, the most advanced technologies of chip manufacturers such as Snapdragon, MediaTek and Kirin are generally within 7nm, while many technologies of mainstream products of automobile chips are at 14nm and 28nm. In addition, 40nm chips are also widely used in storage, communication and other fields.

The more advanced the technology, the more difficult it is to design and manufacture. However, the related chips of automobiles, cloud servers and non-mobile consumer electronics generally have low technological requirements and are less difficult to design and manufacture.

In addition, in the industrial chain, the production capacity of advanced process chips such as 7nm and 5nm is also smaller, and only a few head manufacturers have mastered the relevant production equipment and technology, while the threshold of chip technology and equipment for 28nm and 40nm processes is lower, so there are more manufacturers with production capacity and greater production capacity.

Generally speaking, the chips of automobile and cloud server related products are easier to design and produce because the process requirements are not as high as those of mobile phones. Based on this premise, self-developed chips are actually a matter of time and money for giants, and these are just their advantages, so self-developed chips are easy to get started.

The Commercial Motive of Self-study

Although the objective environment is friendly to self-developed chips in some fields, the real reason to promote these head players to independently develop chips is commercial considerations.

First, the self-developed chip link is completely controllable, which reduces the risk of being "stuck". Huawei's ban sounded the alarm. For a giant like BAT, similar sudden obstacles may still occur in the field of highly competitive advanced technologies and products, and the design and research and development of self-developed chips can be controlled by themselves, thus eliminating the possible "black swan" incident.

Second, the chip is self-developed, which may be more efficient and lower in cost. Chip design and manufacturing will eventually serve a complete product together with software and hardware. In the field of mobile phones, whether mobile phone manufacturers can use the latest chips depends on the progress of chip manufacturers such as Qualcomm, MediaTek and Samsung. Moreover, after the manufacturer gets the chip, it is necessary to adapt and optimize the chip and mobile phone according to the specific products and needs.

The advantage of self-research is that you can design a set of chips with corresponding functions according to your own needs, without multi-party docking and running-in. The efficiency of business and project promotion is much faster, and the cost of final mass production application may be much lower than that of finding a third party.

Third, self-developed chips are an important source of core competitiveness. In industries such as automobiles and home appliances that accelerate the development of electronization and digitalization, the quality of chips greatly affects the final experience of products. For example, in the automotive field, the autopilot chip is closely linked with the autopilot capability, the calculation algorithm is stronger, and the autopilot experience is often better.

If the enterprise's algorithm ability is very strong, then the self-developed chip can integrate these abilities, which can be used in its own products, and can also serve other players as a third-party chip product supplier to enhance its liquidity and industry discourse power.

There are also pitfalls in chip self-research

Judging from many successful cases of self-developed chips at present, this business model is likely to become a reality in more and more fields, especially in industries such as automobiles, home appliances and servers, which are in the stage of rapid development of intelligence.

However, it should be noted that the low difficulty and low threshold of self-developed chips are mainly the objective environmental advantages that head enterprises can enjoy. There are also many failures in domestic self-developed chip models, often because the research and development cycle is too long and the technical threshold is too high. Therefore, small and medium-sized enterprises, or enterprises that do not have sufficient cash flow for a long time, need to develop their own chips, and the risks may continue to increase, or even enter a dead end.

Of course, head enterprises with capital and certain technology are not completely risk-free. One of the most obvious but easily overlooked traps that chip self-research mode may bring is low market tolerance. Generally speaking, the final return of self-developed chips may not cover the initial cost, and the scope and scale of final application are very small.

For example, a manufacturer spent two years developing chips for its own car products. Although the efficiency is higher and the cost is lower, the final car sales are not ideal, and the advantages of the self-developed chip are not fully reflected in the commercial value. This is a failure.

At present, the main purpose of many self-developed chip players is to use their own products. Although the self-developed mode can control the cost, whether the cost saved by the self-developed chip can cover the research and development cost is also a problem that needs to be well pressed.

In addition, the time cost and capital cost of self-developed chips are very high, especially in the iterative stage of chip upgrade. The cost of self-developed chips may double because of technical barriers. If the long-term profit of the business cannot cover this, then this model is a chronic disease.

In short, under the background that chip development takes time, effort and money, no player can underestimate the risk of self-developed mode.

Liu Kuang's official account number, ID: ID: liuquan110+00.