NIO upgrades its central computing platform. What exactly is cabin-driving integration technology?

Today, NIO announced that the electronic and electrical architecture of its 2024 second-generation platform models will be upgraded from a domain control architecture to a central computing platform. The biggest change is the upgrade from the original separation of smart cabin and smart driving to the integration of cabin and driving.

Frankly speaking, this is not a very easy-to-perceive technology upgrade, but it is very important for the continued evolution of vehicle intelligence and performance release. In fact, the integration of central computing platforms and cabin driving has become a trend in the entire industry, and NIO is not the first car company to launch this technology.

The ultimate architecture for software-defined cars

The automotive electronic and electrical architecture carries various technological and intelligent functions on the car. With the increase in automobile intelligent functions, the traditional distributed architecture has accelerated its transition to domain control architecture in recent years. The number of ECUs on the car is greatly reduced, and control and management are easier.

As the computing power of automobile chips further increases, automobile companies have begun to consider how to further tap the computing power of automobiles and implement more cross-domain linkage functions to achieve the effect of 1+1>2. For example, the graphics capabilities of the smart driving chip are used for face recognition and 3D rendering, and the control capabilities of the chassis system are used to provide safety redundancy for the smart driving system, etc.

As a result, many suppliers and car companies have proposed the next generation of electronic and electrical architecture for cars - domain control fusion, or even a complete single-chip central computing architecture. Among them, the cockpit and assisted driving, which have the most intelligent functions, were the first to be integrated, which is the integration of cabin and driving. The so-called cabin-driving integration is to weld the originally independent cockpit chip and smart driving chip on the same board, so that the two can share part of the computing power. This method is called One Board.

In 2023, many car companies and suppliers have already launched cabin-driving integration and central computing architecture technologies: Tesla’s HW3.0 and HW4.0 integrate AMD cockpit chips and FSD chips, and   Xpeng G6 ( parameters | Inquiry ) The X-EEA electronic and electrical architecture version 3.5 adopted adopts a hybrid architecture of central computing + domain centralized; Jidu 01 also uses two domain controls, Qualcomm 8295 smart cabin chip and NVIDIA Orin X smart driving chip The vehicle’s cabin and driver are integrated. According to Jidu, the two domain controllers can share computing power, perception, and services. The smart cabin domain controller can support system-level security redundancy in the event of smart driving system failure, while the smart driving domain controller can support smart cabin 3D The ultimate AI interaction of human-machine co-driving map

Neusoft Ruichi has integrated a combination solution based on Horizon J5 + Xinchi X9 series chips on the high-performance all-in-one domain controller X-Box 4.0. The car brain released by Yikatong also integrates a Longying 1 smart cockpit chip from Xinqing Technology and a smart driving chip from Black Sesame Intelligence, which also achieves cabin-driving integration. The central computing platform released by NIO this time is also a typical cabin-driving integration solution: 1 Qualcomm Snapdragon 8295 smart cockpit chip + 4 NVIDIA Orin X smart driving chips.

Make car control easier

Just like domain control is simpler and more efficient than distributed architecture, the central computing platform brings the efficiency of the vehicle's electronic and electrical architecture to a higher level. At the same time, the length of the wiring harness is also greatly reduced, and the cost is also reduced.

According to NIO's description, the new central computing platform integrates more than 12,000 devices, overcoming technical problems such as PI/SI, EMC, and Thermal caused by high integration, and is 40% smaller than a separate cabin driving domain controller. , the weight is reduced by 20%.

At the same time, the central computing platform ADAM can eliminate the need for encoding and decoding between different domains in the car, eliminating the need for encoding and decoding chips, power supplies, heat dissipation, wiring harnesses, etc., and directly replaces Gigabit Ethernet through etched circuits on the circuit board. The cross-domain data bandwidth between the smart car and the cockpit has been significantly increased from Gigabit to 16Gbps, achieving a transmission rate increase of more than 10 times.

In addition to simplification and weight reduction, the biggest advantage of cabin-driving integration is the sharing of computing power, which allows car companies to fully tap computing power. NIO said that through system-level innovation, it can achieve a maximum of 256TOPS of computing power sharing for smart driving, smart cabin and vehicle control. For example, the identification of occupants in the car is a task of the smart cockpit chip, but the smart driving chip NVIDIA Orin X is better at AI computing and image processing. The computing power calling function of the central computing platform can use the Orin Efficiency, leaving the cockpit chip computing power for tasks it is better at.

In addition, cross-domain computing power sharing can more reasonably allocate various large computing power needs, such as end-side deployment of multi-mode large models, making the best use of resources, optimizing CPU/GPU/NPU usage efficiency, and giving full play to the super computing power of the central computing platform. force. NIO said that NOMI GPT, a large model innovative for car use, will be released soon and will be installed on a new generation of central computing platform.

Other brands also serve the same purpose. For example, Ekatong's integrated cabin-driving solution can also share the computing power of two chips, while reducing the vehicle wiring harness by 5%, R&D costs by 15%, and BOM costs by 20%.

All in One?

It is not difficult to see that the current integration of the central computing platform and cabin driving of car companies is just One Board, which enables the integration and interoperability of the smart cabin and smart driving chips. In their blueprint, the ultimate central computing platform should be All functions can be completed with one chip, and it can even include chassis, power and other domain control functions, that is, One Chip.

The relevant person in charge of Desay SV said: "With the development of cabin-driving fusion chips, the rapid evolution of SOA architecture technology, and the continuous growth of OEM demand for cost optimization, users' demand for intelligent functions continues to increase, and the superposition of multiple factors has promoted the single-chip The mass production application of cabin-driving integration is accelerating. It is expected that 2024 will be an important year of opportunity for the implementation of single-chip cabin-driving integration."

In other words, All in One, using one chip to control all vehicle functions may be the final form of automotive electronic and electrical architecture. Next-generation automotive chips like Nvidia and Qualcomm will have the ability to combine cabin and driving into one.

NVIDIA THOR has 2000 TOPS and 2000 TFLOPS of computing power. All of this computing power can be used for autonomous driving workflows, or it can be split and configured as part for cockpit AI and infotainment functions and part for assisted driving. . When computing in multiple domains, THOR can divide functions such as autonomous driving and in-vehicle infotainment into different task intervals, which can run simultaneously without interfering with each other. For example, the vehicle runs different systems such as Linux, QNX and Android at the same time on a THOR.

In addition, Qualcomm has also released the Snapdragon Ride Flex chip, with a maximum computing power of 2000TOPS. Ride Flex also supports cabin-driving integration, supporting computing in the entire car and Robotaxi-level autonomous driving capabilities.

However, letting one chip control the entire car may not be the best car control solution. Not long ago, I saw an interesting thought from a netizen: Humans and most animals on the earth are typical central calculations. The brain controls our thinking and actions, but not all the control behaviors that keep our bodies running are completed by the brain.

For example, the operation of subsystems such as the respiratory system, digestive system, and cardiovascular system is controlled by the brainstem, and the brain cannot perform strong intervention and control. In addition, the human body has many direct controls. For example, when we touch a hot cup of boiling water, our hands will not go through the brain to calculate the reaction and directly retract our hands.

Many emergency driving reactions are not directly controlled by our brains. For example, when an obstacle suddenly appears ahead, most people will not yield to the right of way as required by traffic regulations. Instead, they will follow the risk aversion mentality and swerve to avoid collision.

Therefore, some people believe that using a chip All in One for cars may not be the most ideal solution. For example, emergency solutions such as AEB emergency braking may be more suitable to use end control, which can get rid of the communication and decision-making delays with the AEB sensors and braking systems and the central platform, and brake faster.

Regarding this view, we interviewed a Weilai autonomous driving engineer. He said: "Car control is still too simple compared to the animal nervous system. If you consider delegating power to insignificant and small functions, it means that the control unit and computing power decentralization, this will return to the old path of distribution and rising costs. As long as the computing power and communication bandwidth have sufficient performance guarantees, single-chip central computing is definitely a better solution. Wait until the car truly realizes ONE CHIP before considering How can it be meaningful to decentralize power?”