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

Tier 1 companies such as Bosch have installed millimeter-wave radars on vehicles, using them to sense the speed and distance of the vehicle in front to implement ACC functions. The data sensed by the radar is generally processed and calculated on Xilinx chips to implement vehicle control functions.

At the same time, Mobileye, which excels in computer vision technology, chooses to use cameras to perceive external vehicles and lane lines. The images captured by the cameras will be calculated on its self-developed EyeQ series chips and the results will be output, thereby realizing AEB automatic emergency braking, lane keeping and other functions.

After ADAS technologies based on millimeter-wave radar and smart cameras have been developed for a period of time, Tier 1 and car companies have discovered that the perception capabilities of radar and cameras can be combined to achieve more advanced functions - such as the ICC integrated intelligent cruise system, which is now commonly referred to as L2 autonomous driving technology.

When working, the radar and camera will monitor the information of other road participants in front of the vehicle, such as vehicles, pedestrians, bicycles, etc. At the same time, the camera is responsible for sensing the lane line information. With the position and speed of other vehicles and the lane line data, the vehicle can automatically drive in a single lane and automatically accelerate and decelerate according to the distance and speed of the vehicle in front.

During this process, both the camera and the radar will sense the road conditions ahead, and the perception results need to be summarized and compared before making driving decisions. Therefore, the manufacturer will select one sensor (radar or camera) as the main controller, receive data from another sensor, and perform unified calculations here.

Since Mobileye has extensive experience in the field of visual perception, most car companies will choose to use the visual perception solution provided by Mobileye (camera + EyeQ chip with perception algorithm written in it).

Millimeter-wave radars use products from companies such as Bosch and Continental, while the chips used for calculations inside the radar are still products of Xilinx.

Precisely because L1~L2 autonomous driving generally uses such hardware architecture, this market has always been dominated by Mobileye and Xilinx.

Let’s look at two sets of data.

In 2020, Mobileye's EyeQ chip shipments reached a new high of 19.3 million units, and it has reached cooperation with 28 automakers.

The same is true for Xilinx. From 2016 to 2020, the average sales volume of automotive chips was 19.3 million sets. In 2020, more than 7.5 million sets of devices were used in ADAS.

It can be said that most vehicles equipped with ADAS systems use chips from these two players.

But changes occurred around 2015, when Nvidia launched the Drive PX series of autonomous driving SoCs, hoping to provide computing power for the car's autonomous driving system.

The most important of Drive PX's first customers is Tesla.

At that time, Tesla had realized that autonomous driving technology would be the core function of smart cars. Coupled with the rise of deep learning technology, Tesla hoped to use deep learning and other AI technologies to create the world's most powerful autonomous driving system.

Unlike traditional L1~L2 which mostly use rule-based algorithms, because of the use of deep learning technology, Tesla needs to change the hardware architecture of the system, from radar and camera calculations (and integration) to the role of autonomous driving domain controller.

Tesla developed an autonomous driving domain controller based on Nvidia's Drive PX chip. In the early days, it still chose Mobileye's perception system, but also installed its own cameras to collect road data.

When working, Mobileye's visual perception system and millimeter-wave radar will output the perception results to Tesla's domain controller, which will fuse the perception results, perform calculations, and make driving decisions.

In this step, Tesla is equivalent to handing over the perception part to suppliers and making the final decision algorithm by itself. In fact, large automakers such as BMW, GM, and Ford are also currently doing this.

Later, when Tesla had enough data, it began to abandon the perception solutions of suppliers such as Mobileye and purchased cameras and radars on its own. All the raw information was processed inside the domain controller, and the perception results were obtained by itself, and then fusion and decision-making were carried out on its own.

Tesla's approach was the most advanced autonomous driving system architecture in mass-produced cars at the time, and is also the domain controller architecture that is now rapidly gaining popularity.

The emergence of this architecture has completely changed the rules of the game.

On the one hand, driven by Tesla, more and more car companies are choosing to develop their own perception technology, which no longer requires Xilinx computing chips to be installed in camera or radar systems. The raw data collected by radar and camera can be directly processed and calculated in the domain controller.

On the other hand, under this architecture, all sensor data must be transmitted back to the domain controller for processing. At the same time, the types and numbers of sensors are constantly increasing, which puts higher demands on the computing power of the domain controller.

In other words, it places higher demands on the most critical computing chip in the domain controller.

It was precisely because of this trend of technological change that NVIDIA chose to use the SoC chips popular in the consumer electronics field to build autonomous driving chips in 2015, and together with Tesla, promoted the transformation of the L2 autonomous driving system.

Of course, due to Tesla's rapid development, in 2019 it decided to leave Nvidia behind and directly make its own autonomous driving chips to provide higher computing power, but it still proves that this change has become a trend.

The transformation of autonomous driving computing architecture towards centralized computing architecture has naturally given new opportunities to consumer electronics chip giants such as Intel, AMD, and Qualcomm.

After all, the most powerful chips on the planet are products of consumer electronics chip giants.

Nvidia was the first to respond, starting with the Drive PX product in 2015, and has been steadily advancing by launching new autonomous driving chips every year.

In 2016, Intel chose to acquire Mobileye at a sky-high price of US$15.3 billion, buying itself a ticket to enter the autonomous driving and even smart car industries.

Qualcomm, a giant in mobile SoC, first launched the 602A automotive SoC in 2014 to provide computing power for the cockpit. In 2016, it went a step further and launched the 820A SoC, which can also provide computing power for ADAS systems, hoping to enter the autonomous driving market.

At the same time, Qualcomm also chose the same path as Intel, hoping to gain vehicle installation experience and customer resources by acquiring NXP, the world's largest automotive chip supplier, for more than US$40 billion, but was ultimately rejected by regulators.

Later, Nvidia and AMD announced the acquisition of Arm and Xilinx respectively last year. The acquisition prices were also unprecedentedly high, reaching US$40 billion and US$35 billion respectively.

At this point, the four major chip giants in the consumer electronics field have all entered the fields of autonomous driving and smart cars, and a new round of battle for automotive chip computing power has begun.

3. Sky-high acquisitions continue to take place. Autonomous driving has become a new battlefield for giants

In addition to product layout, chip giants have never stopped investing in mergers and acquisitions, with the total investment and mergers and acquisitions exceeding US$90 billion in the past few years. Several chip giants have already made big bets in the fields of data centers, AI, industry, and automobiles.

Investment and M&A of the four chip giants in recent years

1. Intel buys Mobileye and Moovit for $16.2 billion

In 2017, Intel spent $15.3 billion to acquire Mobileye, marking the beginning of the chip war.

In fact, Intel also relies on Mobileye to gain a foothold in the L2 autonomous driving market.

In 2017, Intel reached a cooperation with Waymo, Google's self-driving car company at the time (later spun off as a subsidiary of Alphabet). The self-driving systems used in Waymo's tests all adopted Intel processors for calculations and decision-making.

However, Intel has very few resources in the field of L2 autonomous driving. After acquiring Mobileye, more than 25 automobile brands, more than 300 models, and more than 60 million vehicles in the world are equipped with Mobileye's L2 autonomous driving technology, allowing Intel to gain a firm foothold in the field of L2 autonomous driving.

Intel's entry into the L2 autonomous driving field has undoubtedly been very successful.

Earlier this month, Mobileye released a self-developed LiDAR system integrated chip. It uses silicon photonics technology to integrate lasers and relies on FMCW (frequency modulated continuous wave) to calculate the distance, speed, and direction of objects. It will be officially launched in cars in 2025.

Mobileye develops its own lidar chip

This self-developed LiDAR chip will be manufactured using Intel's technology. Intel has advantages in silicon photonics processing, wafer fabs, and IP, and can integrate active and passive laser components on the chip.

In addition to lidar, Mobileye has brought its self-developed autonomous driving technology to market through mobility as a service (MaaS), vehicle as a service (VaaS), and ride as a service (RaaS).

In May last year, Intel spent US$900 million (approximately RMB 5.83 billion) to acquire Israeli travel service company Moovit, which provides MaaS services.