In-depth: The battle for autonomous driving among chip giants

Since the beginning of the year, a wave of new car models have come to the fore, and their most eye-catching features are all concentrated in the field of autonomous driving.

NIO ET7 announced that it will be equipped with LiDAR, and the autonomous driving domain controller integrates four NVIDIA Orin SoCs, with a theoretical computing power of 1016TOPS, the highest among all the models released so far. Zhiji Auto, a joint venture between SAIC and Alibaba, also announced that it will be equipped with Xavier or Orin X chips, with a maximum computing power of more than 1000TOPS.

NIO ET7 autonomous driving computer configuration

The new car Mocha launched by Great Wall WEY brand will be equipped with mass-produced solid-state laser radar, and will also use Qualcomm's Snapdragon Ride computing platform in 2022. Mobileye, an autonomous driving company under Intel, announced that it will be the first to equip Geely's models with EyeQ5 chips this year...

It is obvious that as the brain of the autonomous driving system, autonomous driving chips have obviously become a "battlefield for car companies", and the competition is becoming increasingly fierce.

Since the rise of ADAS technology, or L1~L2 autonomous driving technology, the autonomous driving chip market has long been controlled by two players, Mobileye and Xilinx. The former's annual shipments have reached nearly 20 million pieces, while the latter's has exceeded 7 million pieces.

However, with the development of technologies such as sensors and AI, the hardware architecture of autonomous driving systems has shifted to a domain controller architecture (data from all sensors are aggregated and processed and calculated in a unified manner), and the computing power requirements for computing chips are also getting higher and higher.

Nvidia, the GPU king, and Qualcomm, the mobile phone chip king, have successively launched high-computing-power autonomous driving SoCs, and began to snatch market share from Mobileye and Xilinx from Tesla, NIO, and Great Wall.

On the other hand, Intel and AMD, PC chip giants based on X86 architecture, acquired two veteran players, Mobileye and Xilinx, at sky-high prices of US$15.3 billion and US$35 billion respectively, hoping to combine their own high-computing chips with the products and experience of veteran players to seize the autonomous driving and smart car markets.

It is in this context that Nvidia, Intel and Qualcomm have begun to cooperate with different car companies to compete in the autonomous driving chip market.

A war for autonomous driving chips is underway among the four major consumer electronics chip giants: Intel, Nvidia, Qualcomm, and AMD.

1. The battle for autonomous driving chips begins as the four giants take action

At the "China Electric Vehicle 100 Forum 2021" high-level forum held last week, Miao Wei, deputy director of the Economic Committee of the National Committee of the Chinese People's Political Consultative Conference, said in his speech that in 2020, the penetration rate of L2 autonomous driving in my country has reached 15%.

Miao Wei, deputy director of the Economic Committee of the CPPCC National Committee

That is to say, in 2020, nearly 3 million vehicles equipped with L2 autonomous driving entered the market, which is more than twice the size of the new energy vehicle market, and this market is still expanding rapidly.

Major automakers are stepping up efforts to occupy such a huge market, and the competition for L2 autonomous driving is becoming increasingly fierce.

At the same time, the battle for L2 autonomous driving chips is also beginning.

Previously, the dominant players in L2 autonomous driving chips were Mobileye and Xilinx. However, starting from 2020, this phenomenon is gradually being rewritten. In 2021, the status of established manufacturers such as Mobileye and Xilinx in L2 autonomous driving chips may gradually be threatened.

1. Nvidia quickly attracts new forces in car manufacturing

At the beginning of 2021, two new car manufacturers held new car launch conferences. NIO released its fourth model and its first sedan ET7, and Zhiji Auto, a joint venture between SAIC and Alibaba, officially released two mass-produced models. These three models will be launched in the next two years.

Coincidentally, the autonomous driving solutions used in both cars come from NVIDIA.

The Adam autonomous driving computer of NIO ET7 uses four NVIDIA Orin SoCs, and the theoretical computing power of a single chip can reach 254TOPS. In actual operation, two Orin SoCs are involved in the calculation, which can process 8GB of data generated by sensors per second. The third Orin SoC is used as a redundant backup at any time. The fourth Orin SoC performs localized machine learning to continuously improve the vehicle's autonomous driving performance, while providing a personalized driving experience based on the driver's personal preferences.

Zhiji Auto has two solutions for autonomous driving. One is a visual perception solution that uses NVIDIA Xavier SoC with a computing power of 30 to 60TOPS. The other is a redundant solution that is compatible with the lidar hardware and software architecture, with a theoretical computing power of 500 to 1000+TOPS. The 500TOPS computing power version is equipped with two Orin SoCs.

In addition to NIO and Zhiji Auto, domestic new car-making forces Li Auto and Xpeng Motors have also reached cooperation with NVIDIA. NVIDIA has fully taken over the top domestic new car-making forces.

Ideal Auto will adopt Nvidia's solutions in its next-generation models to be launched next year, while Xiaopeng P7 has already adopted Nvidia's Xavier autonomous driving chip.

It is worth noting that NIO's three current models on sale are ES8, ES6 and EC6, and Ideal's current model on sale is Ideal ONE, both of which use Mobileye's autonomous driving solutions. Within a year, the two brands have successively reached cooperation with NVIDIA.

2. Mobileye is unwilling to give up and EyeQ5 chip will be installed in Geely this year

Mobileye is not reconciled to being "abandoned" by NIO and Ideal.

Earlier this month, Chedongxi interviewed Erez Dagan, vice president of Intel and executive vice president of products and strategy at Mobileye, during which he talked about the relationship between Mobileye and NIO.

Chinese media interviewed Erez Dagan online

Erez said that in 2020, Mobileye provided more than 50,000 systems to NIO, including three-eye camera systems and ADAS solutions.

However, as the leader in the ADAS chip industry, Mobileye prefers to cooperate with large manufacturers.

“We had to make a difficult decision about who to choose as the market leader for our solutions in China,” Erez said. “As mentioned at CES this year, our choice was to partner with the larger and stronger Geely Auto Group.”

At the same time, he said that with the addition of more advanced functions, we need to choose an OEM manufacturer that is stronger and more capable than NIO.

Erez said in an interview: "We still have a good relationship with NIO. We still have a contract with NIO for the supply of fleets for Mobility as a Service (MaaS) operations and an extension of the contract for the three-eye solution."

Last September, Mobileye and Geely jointly announced that the Mobileye SuperVision autonomous driving system will be officially mass-produced this fall, and the first model will be a Geely model.

Among them, SuperVision contains two EyeQ5 chips and 11 perception cameras, which can realize end-to-end full-stack L2 autonomous driving capabilities, high-precision maps and parking functions.

3. Qualcomm's autonomous driving has achieved initial results and will be officially mass-produced in 2022

Compared with smart cockpits, Qualcomm's actions in the field of autonomous driving can be described as "low-key." At the beginning of last year, Qualcomm launched the Snapdragon Ride autonomous driving solution, which can provide different SoCs for different autonomous driving systems such as L1/L2, L2+/L3, and L4.

Just on December 30 last year, Qualcomm and Great Wall Motors reached a cooperation in the field of autonomous driving. Great Wall Motors will adopt Qualcomm Snapdragon Ride platform and mass-produce it in Great Wall Motors' "Coffee Smart" models. It is expected to be installed in Great Wall's high-end mass-produced models in 2022.

Test vehicle equipped with Qualcomm Snapdragon Ride platform autonomous driving solution

Qualcomm Snapdragon Ride has a variety of hardware configurations. For L1/L2 autonomous driving, it is equipped with a single Qualcomm Snapdragon ADAS application processor; for L2+/L3 autonomous driving, a single SoC with a computing power of 30 TOPS can be used. If it is L4/L5 autonomous driving, multiple SoCs can be used, with a computing power of more than 700 TOPS and a power consumption of 130 watts.

4. AMD spends $35 billion to acquire Xilinx

Last October, AMD announced that it would acquire Xilinx, the world's largest FPGA manufacturer and the originator of FPGA, in an all-stock transaction with a total value of US$35 billion. This transaction amount was more than one-third of AMD's market value at the time (approximately US$94 billion).

In addition to Xilinx's FPGA (field programmable gate array) architecture, AMD's automotive business is also one of the key businesses of the merged company.

These four recent "wars" are actually just a microcosm of the chip giants in the field of L2 autonomous driving. In 2021, with the increasing popularity of L2 autonomous driving and the continuous evolution of automobiles from distributed computing to central computing, the "war" of autonomous driving chips among the four giants will continue to escalate.

2. Autonomous driving computing architecture transformation: Chip giants rush into the market

When ADAS technology, or L1~L2 autonomous driving technology, first emerged, its basic functions included ACC adaptive cruise control, AEB active braking, and LKA lane keeping systems.

To achieve these functions, the vehicle needs to sense other external vehicles, pedestrians and other traffic participants, as well as lane lines.