As the underlying cornerstone of the intelligent driving system, autonomous driving chips are ushering in an industry outbreak period with the new wave of artificial intelligence and the acceleration of the trend of automotive intelligence. #AI##自动驾驶 #
AI chips replace the past CPUs, and combine high-performance MCUs to form an architecture of central computing plus regional controllers.
The central computing architecture can reduce the amount and type of chips in the whole vehicle by at least 10 times, and the functional integration and performance of a single chip can be increased by more than two orders of magnitude compared with the original, which greatly enhances the versatility of the chip and improves the chip reuse rate.
With the rapid growth of automotive intelligence demand, on-board large computing power chips are accelerating the iteration speed, and the proportion of chip costs in smart cars will also increase significantly.
AI chips are at the core of the intelligent vehicle industry chain in the process of electronic architecture moving towards central computing. #AI芯片 #
Intel CEO predicted in 2021 that by 2025/2030, chips will account for 12%/20%+ of high-end automotive material costs (4% in 2019), and the largest increase comes from assisted driving.
According to UBS disassembly, the cost of the Model 3 chip is about 1516 US dollars, accounting for about 6% of the total BOM cost, and the cost of each part of the chip is sorted as: electric drive>ADAS> entertainment system and power distribution > body control.
Pay attention to Leqing Think Tank and gain insight into the industrial pattern!
An overview of the autonomous driving AI chip industry
With the upgrade of autonomous driving from L1-L5, the automotive E/E architecture has changed from distributed to centralized, and computing power has begun to be concentrated, and the integrated design can reduce the redundancy requirements of computing power while greatly reducing the length of the vehicle wiring harness and effectively reducing costs.
When the domain is concentrated, the intelligent function upgrade will shift from increasing the number of sensors to increasing computing power, algorithm models and data training, so the computing power requirements for autonomous driving AI chips will become higher and higher, and the computing power demand will increase by more than 10 times with each level of improvement.
The requirements for chip computing power for autonomous driving are increasing exponentially:
The sales of large computing power chips are generally equipped with the debugged operating system, and reference design algorithms are provided to customers, and the sales form is close to the mobile phone chip mode.
When car companies choose a large computing power chip platform, it is not only to meet the needs of one car, but also to cover multiple different models at the same time, which requires the large computing power platform to have sufficient flexibility and scalability, and can provide supporting tool chains and software stacks at the software level.
The vehicle brand will decide which intelligent functions to achieve according to the positioning of the model, combined with the characteristics of each AI chip company (evaluation dimension: computing power, cost, power consumption, ease of development, isomorphism - compatibility of the chip platform with other systems) and the product roadmap of the chip company to determine which main control chip to choose, and build a domain controller based on the main control chip.
Requirements for the types of on-board chips and their computing power for the evolution of electronic and electrical architecture of intelligent vehicles:
Source: Bosch, Xinqing Technology
Autonomous driving chip market landscape
At present, the global pattern of automotive large computing power chips is clear, local companies are rising, new forces and independent head automobile brands are actively targeting NVIDIA, Qualcomm and other large computing power chips, AI chips are at the core of the intelligent car industry chain, AI chip product roadmap points to large computing power + cross-domain integration, and business models are becoming more open and flexible.
From the perspective of the chip solutions used in autonomous domain controllers, in November 2022, the top 5 chip market shares of autonomous domain controllers were Tesla FSD (43.2%), NVIDIA (19.4%), Xilinx (11.9%), Mobileye (7.6%), and Infineon (5.6%).
Tesla, NVIDIA, Qualcomm and other foreign companies lead:
From January to November 2022, the top 5 market shares of autonomous domain controller chips were Tesla FSD (40.9%), NVIDIA (13.3%), Xilinx (12.8%), Infineon (9.2%), and Horizon (8.5%).
In 2022, mainstream chips commonly used in autonomous driving domain controllers include Tesla FSD, Xilinx, Horizon J3, NVIDIA Orin, NVIDIA Xavier, mobileyeEyeQ5H, Infineon AURIX, etc.
Tesla insists on using self-developed FSD chips to consolidate its visual recognition advantages through ASIC solutions.
Qualcomm began to gradually move from cockpit chips to autonomous driving AI chips, and Great Wall became the first partner in China.
Competitive landscape of autonomous driving domain controller chips for passenger cars from January to November 2022:
Source: Gaogong Intelligent Vehicle
At present, the mainstream released model AI chips are mainly Intel Mobileye series, but starting from the next generation of models, the choice of NVIDIA chip manufacturers began to occupy the majority, including NIO, Xpeng, Lili, SAIC, WM and other car companies have chosen Orin chips as the next generation of AI chips.
NVIDIA launched Xavier in 2019, which can achieve L2+ or even L3 functions, with high maturity, and the domestic Desay SV is a Tier1 that is very deeply bound to NVIDIA.
NVIDIA released a new generation of chip DRIVE Thor, an automotive-grade system-on-chip (SoC) based on the latest CPUs and GPUs, providing 2,000 trillion floating-point operations per second performance, which will be mass-produced in 2024, and Geely's Zeekr is the first to announce that it will equip its models with Thor chips from 2025. #NVIDIA##TeslaModelY##华为 #
Domestic related layout manufacturers mainly include Huawei, Horizon, Black Sesame, Xinchi Technology, Cambrian, Xinqing Technology, etc.
Huawei completed the implementation of AI chips with the help of BAIC Alpha S, and subsequently formed cooperative relationships with Changan, GAC and salon brands.
In addition, the acceleration of autonomous driving chips on the car, domain control players and partners who have in-depth cooperation with chip manufacturers are also expected to benefit. #5月财经新势力 #
Autonomous driving AI chip loading tracking:
Source: Gasch Automobile, Soochow Securities, Xingxingcha
On the whole, the market pattern of large computing power chips has not yet converged, at present, NVIDIA and Qualcomm are at the forefront of change, Horizon mass production progress leads the domestic market, Huawei MDC or Nirvana returns, and domestic manufacturers such as Huixi Intelligent also have opportunities to break through under the architecture change. Large model developers and all links of the autonomous driving industry chain will enable AI chips to accelerate their evolution and usher in a new round of industrial opportunities.
Pay attention to Leqing Think Tank and gain insight into the industrial pattern!