The autonomous driving chip war

This year, the outbreak of the COVID-19 pandemic, the economic downturn, and the deterioration of the international political environment have filled the automotive industry with great uncertainty. Many consulting agencies predict that global auto sales will face a 10%-20% decline this year.

However, amidst the uncertainty, the automotive industry is very confident about the future direction. Several big news came out in the field of autonomous driving:

On June 23, Mercedes-Benz, which had just announced a peaceful breakup with BMW in the field of autonomous driving, announced a partnership with chip supplier Nvidia. It will use the latter's Orin chip to develop the next-generation in-vehicle computing system to provide computing power support for the L2-L3 autonomous driving functions that will be fully equipped in Mercedes-Benz mass-produced models in 2024, as well as automatic parking functions up to L4.

On June 25, Volvo Car Group announced that Volvo will reach a strategic partnership with Waymo, Google's autonomous driving company, to cooperate on L4 autonomous driving technology on a new electric vehicle platform and explore commercial scenarios such as autonomous driving online ride-hailing.

On June 26, Amazon officially acquired the American autonomous driving company Zoox, paying more than US$1.2 billion for the acquisition.

On June 27, Didi's self-driving online car-hailing demonstration operation was officially launched in Shanghai, and CCTV broadcast the entire process live. From this day on, Didi's self-driving test car in Jiading, Shanghai will be open to the public, and Didi has launched a "Future Travel" page in its APP for the public to apply for a test ride in a self-driving online car-hailing vehicle.

For a time, major companies almost started an arms race for autonomous driving. There is no doubt that all the companies involved realize that the cars of the future will be supercomputers running on wheels. The crucial role of high-performance computing chips in this arms race has become increasingly prominent.

1. Mercedes-Benz finds a new love, just because of it?

On June 23, four days after the suspension of its autonomous driving cooperation with BMW, Mercedes-Benz threw itself into the arms of chip supplier Nvidia, and the two parties reached a cooperation to develop a computing platform for Mercedes-Benz's autonomous driving models that will be mass-produced in 2024.

In the announcement a few days ago, the two parties also stated that "given the costs of establishing a shared technology platform and the current business and economic conditions, now is not an appropriate time to successfully implement cooperation." Being too expensive seems to be the key reason why the two parties decided to suspend their technical cooperation.

However, Mercedes-Benz's subsequent move to partner with Nvidia at the speed of light points to factors other than money. Generally speaking, cooperation between car companies will not affect the cooperation between car companies and suppliers, but the cooperation between Mercedes-Benz and BMW is different. Before reaching a cooperation with Mercedes-Benz, BMW had formed an autonomous driving alliance with Mobileye, the world's largest ADAS system supplier, to develop autonomous driving based on its EyeQ series chips.

The cooperation with BMW means that Mercedes-Benz will use Mobileye's chips to build key autonomous driving computing units. This may be the most important disagreement between the two parties. Sam Abuelsamid, chief analyst at foreign consulting firm Guidehouse, said, "I suspect that the two automakers can't reach a consensus on the platform to be used. Now, Orin looks like a more powerful solution compared to Intel/Mobileye's products."

Judging from the public information, Sam's analysis is not without reason. Mobileye's next-generation autonomous driving chip EyeQ 5 has a computing power of 24TOPS (24 trillion operations per second), while Nvidia's Orin, released at the end of last year, has a computing power of up to 200TOPS. In addition, Mobileye has always been strong in its cooperation with car companies in the past (although it promised that EyeQ 5 would be more open), and the functional modules it provided were often "black boxes" to the OEMs; while Nvidia's Drive AGX software platform for autonomous driving took an open path from the beginning, which can support car manufacturers to independently develop algorithms on their computing platforms.

In fact, before this, when Mercedes-Benz was exploring the development of self-driving online ride-hailing vehicles , it chose NVIDIA's Drive PEGASUS car computer because of the high requirements of the technology on chip computing power. The official announcement on June 23 means that Mercedes-Benz has fully turned to NVIDIA in the selection of chips in the era of self-driving, and will expand the cooperation between the two parties to Mercedes-Benz's mass-produced models.

Waymo, which has reached a strategic cooperation with Volvo on autonomous driving, relies on Google's technical strength in the field of AI and uses its own TPU. Although the computing power of the TPU used by Waymo for vehicles has not been announced, according to Waymo officials, after using TPU, the performance of its autonomous driving system has increased by 15 times.

The status of chips in autonomous driving can be described as a "hidden champion". You can't see its existence from the appearance of the vehicle, but it is definitely the number one contributor to the smooth operation of an autonomous vehicle.

2. The autonomous driving competition is also a chip competition

Whether it's Mercedes-Benz abandoning BMW and joining hands with Nvidia, or Volvo and Waymo's strategic alliance, or Didi's self-driving ride-hailing service, the big news that happened last week shows that both auto companies and technology companies have placed autonomous driving in a crucial position: in the short term, autonomous driving functions are an important part of automotive product strength; in the long run, after L4 autonomous driving is put into large-scale application, it may completely change the business model of the automotive industry.

The basis for all these changes is a small chip. In order to gain a competitive advantage in autonomous driving capabilities, companies participating in this competition either conduct independent research and development or form alliances, just to find a high-performance autonomous driving chip. There is a very typical example in the industry: Tesla.

As the leader of smart electric vehicles , Tesla has cooperated with the two mainstream autonomous driving chip manufacturers in the current market. However, due to the strength and closedness of Mobileye, Nvidia's inability to reduce power consumption and high development costs, the cooperation failed to last long. In order to give full play to the advantages of integrated software and hardware in autonomous driving, Tesla took the lead in independently developing the FSD autonomous driving computing platform among car companies, with a computing power of 144TOPS. FSD's computing power support for autonomous driving mainly comes from two AI chips, and its single chip computing power is about 72TOPS.

So far, Tesla's FSD still holds the record for the computing power of mass-produced cars' autonomous driving. Tesla believes that FSD is sufficient to support its upcoming Full Self-Driving function.

There is no doubt that the competition for autonomous driving is also a competition for chips. The entire automotive industry's emphasis on autonomous driving and even its overall shift will create a huge demand for autonomous driving chips. If a company occupies a considerable share of the autonomous driving chip market, then the corresponding market value may be worth hundreds of billions of dollars.

At present, attracted by the huge market, four new or old forces have emerged in the field of autonomous driving chips:

The first category is established ADAS chip/autonomous driving chip suppliers such as Mobileye.

This type of enterprise is the one that has participated in market competition since the automotive industry began to develop advanced driver assistance systems (ADAS). The competitive strategy of these enterprises for autonomous driving is to continuously upgrade their existing products through the technology and customer resources accumulated in the ADAS market to achieve a smooth transition to autonomous driving. A typical example is Mobileye's continuous iteration of the EyeQ series chips.

In addition to Mobileye, veteran automotive semiconductor suppliers such as Renesas, NXP, Texas Instruments, and Denso all have their own autonomous driving chip plans.

The second category is semiconductor giants that cross fields and see opportunities in autonomous driving chips.

For example, Nvidia mentioned above, its main business previously was GPUs for consumer electronics and data centers, but after Nvidia realized the demand for high-performance chips for autonomous driving, it quickly entered this market. It has currently launched three generations of products, Drive PX, Drive AGX Xavier, and Drive Orin, and has received orders from many car companies.

Qualcomm, whose main business is communications and which dominates baseband chips and mobile phone SoCs, launched the Snapdragon Ride autonomous driving computing platform at this year's CES after its attempt to acquire NXP to enter the autonomous driving competition failed. According to Qualcomm's official information, this computing platform built on Qualcomm chips has a maximum computing power of 700TOPS and can support L4--L5 autonomous driving.

Before Qualcomm, Huawei, whose main business is also communications and consumer electronics, has already released the autonomous driving computing platform MDC 600. This computing platform is integrated with 8 Ascend 310 AI chips, with a maximum computing power of 352TOPS.

The third category is autonomous driving chip startups born out of new opportunities.