In the automotive industry, domain controllers (DCUs) have quietly become popular.
Data shows that in the past year, the recruitment popularity of domain controller-related positions has even surpassed that of the three core components of electric vehicles (battery, motor, and electronic control system) and is approaching that of ADAS (Advanced Driver Assistance Systems). A search on recruitment platforms by Gasgoo revealed that automakers, auto parts companies, and even tech companies that have recently started manufacturing cars are all aggressively recruiting domain controller personnel, offering relatively high salaries.
In fact, many of the above-mentioned companies had already joined the wave of domain controller R&D before this, and many products have been successfully mass-produced and "installed" in vehicles. For example, XPeng Motors P7 has applied the Desay SV IPU03 autonomous driving domain controller, and other models will also be equipped with it in the future. Geely Xingyue L, which was launched in July this year, also adopted a domain controller and claimed to have full-domain FOTA capability. The penetration rate of domain controllers in new cars is rapidly increasing.
What exactly is a domain controller?
With extensive deployments and rapid penetration, what exactly is a domain controller? What amazing features does it possess?
Before explaining this issue, it is necessary to look at how the relevant systems or components operate before the domain controller is deployed.
The predecessor of the domain controller was the ECU (Electronic Control Unit), also known as the "vehicle computer" of a car. Its purpose is to control the driving status of the car and realize its various functions.
In the distributed E/E (electrical/electronic) architecture currently used in most traditional automobiles, ECUs are widely used in various low-level execution components such as the engine, transmission, and airbags, undertaking decision-making functions. Under this architecture, upgrading to intelligent functions requires increasing the number of ECUs.
As a result, the number of ECUs in cars has surged in recent years with the accelerated development of automotive intelligence. It is understood that from 1993 to 2010, the number of ECUs used in the Audi A8 model increased sharply from 5 to more than 100, and the number of ECUs installed in the Audi A8L also exceeded 100 in 2013.
With the acceleration of intelligent upgrades, especially the development of autonomous driving, simply increasing the number of ECUs is no longer a good strategy.
Since different ECUs come from different suppliers, automakers need to communicate and collaborate with these suppliers separately for both the development of car functions and later maintenance and upgrades. This process is cumbersome, which lengthens the car development cycle and increases the cost of human and material resources.
Furthermore, the increasing number of ECUs complicates power and data distribution wiring, making automated production increasingly difficult and requiring greater reliance on manual labor. Given rising labor costs, this approach is uneconomical. Additionally, the varying computing power of each ECU necessitates its own redundancy design, significantly increasing costs for automakers.
In addition, in most cases, the intelligent upgrade of a vehicle not only requires a significant increase in the computing power of a single ECU, but also requires efficient information and data exchange between various ECUs, and sufficient computing power redundancy to cope with various emergencies and ensure driving safety.
In a distributed architecture, the various ECUs are mostly connected via buses such as LIN/CAN, which have limited transmission speeds and cannot meet the needs of efficient information flow within intelligent vehicles.
From another perspective, this also makes it difficult for numerous ECUs to perform rapid collaborative upgrades, making continuous OTA (Over-The-Air) updates for the entire vehicle difficult to implement. It's important to understand that OTA has become a fundamental skill in the era of intelligent vehicles, aimed at reducing costs and improving user experience.
In short, the original technology was no longer sufficient, so domain controllers with new skills were brought in.
Domain controllers emerged from the centralization process of traditional distributed E/E architectures, essentially merging previously isolated ECUs into a centralized group control system. Through domain controllers, the number of ECUs in a vehicle can be reduced from over 100 to just a few DCUs, rapidly centralizing control functions and resolving the aforementioned issues of cost, safety, and upgrades.
Domain controllers can perform more functions with fewer components, which helps reduce costs. Aptiv found in a study of a vehicle manufacturer that using a domain controller could integrate nine ECUs and reduce the number of individual wires by hundreds, significantly lowering costs. Furthermore, the vehicle's weight was reduced by 8.5 kg, which helps reduce CO2 emissions and extends the driving range of electric vehicles.
Furthermore, compared to traditional distributed ECUs, domain controllers offer scalable computing power, more flexible over-the-air (OTA) updates for the entire vehicle, and a greater emphasis on software, enabling automakers to provide users with continuously iterative and upgraded functional experiences. In other words, automakers can upgrade vehicle functions solely through chip computing power and software algorithms without adding ECUs.
More importantly, the domain controller breaks the traditional bundled development model of perception + algorithm + ECU. Because the perception data processing of multiple sensors can achieve data fusion on the domain controller computing platform, this means that the vehicle can make safer decisions in a timely manner.
What are the different types of domain controllers?
Today, there are more and more domain controller products on the market, and the product categories are not entirely the same. Although there is no unified classification standard for domain controllers in the industry, currently, there are two main ways to classify domain controllers.
One method is to divide the area controllers by region, which can be further divided into front area controllers, left area controllers, right area controllers, etc. Due to the high concentration and difficulty, only a few companies such as Tesla and Aptiv currently use this classification method, and different companies also have different specific divisions of area controllers.
According to Open Source Securities data, the architecture of the Tesla Model 3 is equipped with a central computing module and three regional controllers, namely the front body control module, the left body control module, and the right body control module.
The left and right body control modules symmetrically divide some basic functions into areas, with each responsible for internal and external lights, door locks, windows, parking calipers, etc., within their respective areas. Compared to the left body controller, the right body control module also has two unique functions: thermal management and automatic parking assist system.
The front body control module is primarily responsible for power distribution to the various controllers in the vehicle. It can monitor the power consumption of each ECU in real time and promptly cut off power to ECUs that are static but consume high power. In addition, the front body control module also includes functions such as the headlights and windshield wipers.
Aptiv's SVA intelligent vehicle architecture consists of a central computing cluster and multiple regional controllers. This reportedly solves the problem of overly complex in-vehicle controllers, integrating all vehicle computations into the regional controllers. Simultaneously, it provides sufficient interfaces for future software updates and the addition of new features.
The company stated that the area controller is key in the advanced vehicle power and data distribution architecture, and the number of area controllers can be adjusted according to the vehicle's needs and complexity.
Compared to this classification method, the functional classification method is probably more familiar to everyone, and most car manufacturers or parts suppliers currently use this method. Currently, the main classifications include powertrain domain controllers, chassis domain controllers, body domain controllers, cockpit domain controllers, and autonomous driving domain controllers, with slight variations between different companies.
The powertrain domain controller, as the name suggests, mainly integrates powertrain-related functions and is primarily responsible for the optimization and control of the powertrain. Of course, powertrain domain controllers are not only found in traditional gasoline vehicles; they are also increasingly used in new energy vehicles as electric drive and control systems become more integrated.
The chassis domain controller is primarily responsible for specific vehicle driving control, requiring unified control of systems including power steering, vehicle stability control, electric brake booster, and airbag control. It is understood that in Huawei's CCA electronic and electrical architecture, the chassis domain controller comprehensively controls drive, braking, and steering.
The cockpit domain controller is responsible for the functions of the car's cockpit electronic system. It can integrate traditional in-vehicle information system (instrument) and in-vehicle entertainment system (IVI) functions, as well as driver monitoring system, 360-degree surround view system, AR HUD, dashcam and air conditioning controller functions.
The body domain controller is primarily responsible for the overall control of body functions. Given the current trend, and considering its relatively low safety level, the body domain controller is expected to be the first to integrate with the smart cockpit domain as automotive electronic and electrical architecture becomes more centralized.
Autonomous driving domain controllers possess capabilities for multi-sensor fusion, localization, path planning, decision control, wireless communication, and high-speed communication. They typically require external connections to multiple cameras, millimeter-wave radars, LiDARs, and other devices, performing functions including image recognition and data processing. Due to the massive computational demands, domain controllers generally require a high-performance processor capable of supporting different levels of computing power for autonomous driving. It is reported that Desay SV's IPU03 boasts a Xavier computing power of up to 30 trillion operations per second (30 TOPS), capable of processing massive amounts of data from vehicle radar, cameras, LiDARs, and ultrasonic systems in real time, running perception, localization, planning, and control algorithms.
What are the key components of a domain controller?
Regardless of the type, domain controllers are clearly under immense pressure, as they have taken on the task of controlling numerous intelligent functions, and their capabilities must keep up, which is no easy feat.
From the perspective of the domain controller's structure, achieving this requires the organic integration of multiple layers of hardware and software, including the main control chip, software operating system and middleware, and application algorithms. In particular, the main control chip, as the core component of the domain controller, will face higher requirements.
Taking intelligent cockpit domain controllers as an example, in the era of intelligent cockpits, the complexity of computation and processing has increased exponentially. Moreover, OEMs often need to pre-install high-performance hardware in the early stages, because only in this way can the utilization rate of the pre-installed hardware be gradually released later, and continuous iteration and updates be achieved through chip algorithms and software computing power.
Therefore, traditional functional chips are no longer suitable, and it is necessary to choose system-on-a-chip (SoC) chips that integrate multiple modules such as a central processing unit (CPU), AI processing unit, image processing unit (GPU), and deep learning acceleration unit (NPU). As OEMs increasingly favor hardware pre-embedded components in their intelligent vehicle competition, the use of a single high-performance SoC chip or multiple SoC chips has become the mainstream trend. Bosch's intelligent cockpit domain controller technology integrates multiple ECUs in the instrument and entertainment domains onto a single SoC (System-on-a-Chip).
It is understood that traditional smart cockpit chip manufacturers such as NXP, Texas Instruments, and Renesas, as well as consumer chip manufacturers such as Nvidia, Qualcomm, and Samsung, have also entered the market, making cockpit SoCs a focal point of competition among suppliers. Domestic chip manufacturers such as Horizon Robotics, SemiDrive Technology, and Huawei are also targeting the smart cockpit SoC chip market.
Equally important are the software operating system and middleware, which are primarily responsible for the rational allocation of hardware resources to ensure the orderly operation of various intelligent functions. As automotive electronic and electrical architecture evolves towards a domain architecture, the importance of the operating system and middleware within this domain architecture has significantly increased. Simultaneously, the proportion of system software controlling battery management, vehicle connectivity, and related services is gradually growing.
In terms of software operating systems, the system can be roughly divided into two types: real-time operating systems and non-real-time operating systems.
In terms of real-time operating systems, the key is rapid response. After receiving an input signal, the system can process it and provide feedback in a short time. The (latest) completion time of its processing tasks is known.
For this reason, real-time operating systems have high security and reliability, and are often used in the field of vehicle control, including traditional vehicle power, chassis, body and emerging autonomous driving.
Non-real-time operating systems, as opposed to real-time operating systems, are not about speed of response but rather about compatibility and development ecosystem. They are widely used in fields such as cockpit entertainment, such as Alibaba's AliOS and Google's Android Auto.
Another point worth mentioning is the application algorithm. Application algorithms are software programs developed based on the operating system and are a key area for differentiating car brands.
