This year will be the year when the high-power autonomous driving chips are concentrated on the car.
Author | Zhao Jian Liu Xiaoqian
The pace of adoption of autonomous driving is accelerating.
On January 12, the Ministry of Industry and Information Technology issued a document saying that in 2021, 3.521 million new energy vehicles will be sold, of which the new passenger car market equipped with a combination of auxiliary driving systems will account for 20%. Two years ago, the penetration rate of L2 assisted driving was only 3.3%.
The development of smart cars is inseparable from the brain of the car - the self-driving chip. Among the domestic autonomous driving chips, Black Sesame Intelligence, Xinchi Technology, Horizon, and Huawei are the main manufacturers. In addition, zero-run cars and NIO also have plans to develop their own chips.
This Wednesday (January 12), Black Sesame Intelligence and Xinchi Technology held their respective media communication meetings in Beijing at the same time to introduce the latest business progress.
Founded in 2016, Black Sesame Intelligent focuses on the track of autonomous driving chips, while the Xinchi Technology chip established in the late 2 years has a wider business scope, covering automatic driving, intelligent cockpit, central gateway, highly reliable MCUs, etc.
Autonomous driving chips have attracted much attention in the capital market. Black Sesame Intelligence has completed 6 rounds of financing, investment institutions include SAIC, NIO Capital, Xiaomi Yangtze River Industry Fund, etc., on January 12, it also announced that it had obtained strategic investment from Boyuan Capital under Bosch; Xinchi Technology completed a total of 4 rounds of financing, including Hechuang Capital, Sequoia Fund, Lenovo Venture Capital, Jingwei China, CDB Ronghua, etc.
For a long time, the self-driving chip market has been dominated by two companies, Mobileye and Xilinx. In 2019, Mobileye's market share was as high as 70%. In recent years, with the intelligent and electrification of automobiles, giants such as NVIDIA and Qualcomm in the era of consumer electronics have also expanded the field of autonomous driving chips.
In 2022, Nvidia's autopilot chip Orin will be listed with the delivery of NIO ET7 and Zhiji Automobile, and Qualcomm's Snapdragon Ride autonomous driving platform will also be launched in mass production of Great Wall Motors.
This year is also a year in which the domestic large-scale automatic driving chips are concentrated on the car. This time, domestic autonomous driving chip manufacturers have the opportunity to stand on the same starting line with international giants.
1. How much computing power is required for autonomous driving?
Over the past few years, computing power has been regarded as one of the simplest and crudest measures of an autonomous driving chip, presenting a state of "arms race."
In 2014, Tesla Model S was the first to apply Mobileye's first-generation EyeQ chip, with a computing power of only 0.256 TOPS (processor computing power unit, 1TOPS represents a trillion operations per second), which is almost the computing power of today's smart access control; in 2019, Tesla released the Model 3 equipped with a self-developed autopilot chip, and the computing power has been as high as 144TOPS In 2020, WEIO ET7 and Zhiji Automobile launched a chip equipped with Nvidia Orin chip, with a computing power of up to 1000 TOPS.
How much computing power does a self-driving car really need? Yang Yuxin, CMO of Black Sesame Intelligence, said that this problem can be viewed from two aspects. First, chip computing power will become a key point to attract users to buy, and now some car manufacturers say that consumers have asked the question of computing power in 4S stores.
Second, under the trend of software-defined cars, computing power embedding will become a common choice for car manufacturers, even if the computing power that is not used now, the future will upgrade the functions of the car through the form of OTA. In fact, the current mainstream models are L2 or L2+ class, and the maximum computing power of 100 TOPS is enough.
It is worth mentioning that the computing power of automatic driving that everyone often says is actually the computing power of the neural network acceleration unit - NPU, but to measure the performance of an automatic driving chip, in addition to the NPU, it also includes CPU, GPU, ISP, DSP, etc., which is a contest of comprehensive capabilities.
Like the CPU function of the computer, the CPU mainly processes the logical operations in the automatic driving scenario; the NPU belongs to the AI operation to realize the model detection of the neural network and other functions; the GPU is the graphics processing unit, which renders the camera's data image; the ISP is responsible for image signal processing, and the DSP is responsible for digital signal processing, processing large-scale data fusion and data operation.
At present, the largest computing power chip of Black Sesame Intelligence is Huashan No. 2 A1000Pro, hash rate 106TOPS, for L3 level automatic driving; the other two hash rates are slightly lower, namely Huashan No. 2 A1000L, hash rate 16TOPS, for L2 level automatic driving; Huashan No. 2 A1000, hash rate 58TOPS, for L2+ level automatic driving.
However, large computing power chips also mean higher costs, often more suitable for high-end and even flagship models. According to the data of Black Sesame, the large computing power chips with a computing power of more than 100 TOPS will be concentrated on the car from 2022 to 2023.
For example, many popular models such as the WEILAI ET7 and zhiji L7 equipped with NVIDIA Orin chips will be delivered this year.
Xu Chao, vice president of Xinchi Technology, mentioned: "There will be different models on a model, for example, the lowest version has nothing, the highest version is matched with everything, often these two models are the lowest sales, adding up to no more than 20%, the hottest model is 80%. ”
Compared with NVIDIA's large computing chip, the current V9 series of automatic driving chips of Xinchi Technology is not high, only 1 TOPS, but according to Dr. Tao Sheng, head of automatic driving of Xinchi Technology, L2.99 automatic driving can already be realized.
Tao Sheng said that L1 is the era that has passed, L2+ is the era that is happening, and L3, L4, L5 will be an era in the future. Xin Chi did not chase the big computing power from the beginning, but based on an achievable premise, it gave the most cost-effective solution. At present, V9T already has partners who use it to do the front-looking L2+ function.
Xin Chi judged that 2023 is an era in which L3 will enter mass production on a large scale, and made a product layout accordingly. According to the plan, Xinchi will launch an autonomous driving chip between 10 and 200TOPS this year - V9P/U, which has a higher computing power integration and can support L3 level automatic driving.
Unlike black sesame intelligent focus on autonomous driving chips, Xinchi is a product matrix play. In addition to the V9 series of autonomous driving chips, there are also smart cockpit chips X9 series, central gateway chips G9 series, and MCU car control chips.
2.Tier2 requires a flexible and open business model
With the improvement of the status of chips in the automotive supply chain, the division of roles in the traditional automotive supply chain is also quietly changing.
In the traditional automotive supply chain, the role of car companies, Tier1 (tier 1 suppliers), tier 2 (tier 2 suppliers) is a vertical linear relationship, car companies do not need to contact chip manufacturers as Tier2, some people quipped that "car companies people don't know where the door of the chip factory is open."
But three variables gradually broke the situation, the first of which was the lack of core in the car. Previously, AutoForecast Solutions released data that as of December 9, the global production reduction caused by lack of cores in 2021 has reached 10.272 million units, of which the Chinese automobile market will reduce production by 1.982 million units in 2021, accounting for 19.3% of the world.
In order to ensure production and supply, the chairman of many car companies has personally led the team to purchase chips.
The second change is that in the past, different car brands had their own separate supply chains, and now a car company has several sub-brands, and the main engine factory will merge the supply chain from the perspective of supply chain integration and supply chain security. Therefore, OEMs need to go deep into the chip manufacturers to communicate.
The lower level of change lies in the change of technology. Automotive electrical and electronic architecture is shifting from the traditional ECU to the current "domain controller" architecture, and will evolve to "central computing + area control" in the future. Under this trend of change, in-vehicle chips have gradually become the brains of smart cars.
In order to better define the demand, car companies have to contact chip manufacturers directly, and even cooperate in research and development. Xu Chao, vice president of Xinchi Technology, said that the traditional automobile development process does not require the participation of chip companies, most of which use the standard components provided by Tier1, and the electronic system differentiation of mainstream models is small; now the automobile companies have strengthened in-depth exchanges with chip companies, chip companies have entered the market more than 16 months in advance, and the demand for differentiation has increased.
These changes are also gradually reshaping the role of the industrial chain, and the traditional OEMs, Tier1, tier2 have begun to become OEMs to 1.5 level supplier systems. OEMs used to dock with Tier1, and now they are connected with chip suppliers, hardware suppliers, and integration test service providers.
In this model, how to more efficiently undertake customized needs from car manufacturers has become a problem that chip manufacturers must solve, which requires a more flexible and open business model.
Deng Kun, vice president of black sesame intelligent application engineering, explained: "When the chip is given to the main engine factory or Tier1, there is also a process of hands-on development. If the customer needs algorithm support, Black Sesame Intelligence provides a full-stack algorithm; if the partner uses its own algorithm, Black Sesame Intelligence can also support it, and cooperate through nested mode. ”
It is worth mentioning that although chip manufacturers will directly communicate with the main engine factory across Tier1, this is the cooperation of the "product layer" rather than the "commercial layer", and the direct orders of chip manufacturers still come from Tier1.
At present, domestic autonomous driving chip manufacturers have adopted a similar open cooperation model, compared to the traditional chip manufacturers can not customize the modification of the algorithm, to the car companies have brought great attraction, a representative counterexample is Mobileye.
Ideal Car's first model, the Ideal ONE, features Mobileye's EyeQ4 chip. However, because the algorithm of the Mobileye chip is fixed in the chip, it cannot be iterated at the visual perception layer, which is an "algorithm black box". Therefore, the 2021 ideal ONE replaces Mobileye's chip with the Horizon Journey 3 chip.
In the new era of smart cars and autonomous driving, mobileye, once the overlord of ADAS, seems to have fallen behind.
If you turn to the main models that will be listed in 2022 and beyond, almost all of them choose the autonomous driving chips of NVIDIA and Qualcomm, two technology giants. The only model that still uses mobileye chips for high-grade assisted driving is Geely's Krypton.
Of course, Mobileye lost the order of the head car company, but it is still a good choice in the face of waist car companies without algorithmic capabilities.
Every era has a company of each era, and China's autopilot chip manufacturers are in the best era.
END.