At the recently concluded NVIDIA GPU Technology Conference (GTC 2024), NVIDIA co-founder and CEO Jensen Huang announced that he would expand his cooperation with BYD, and BYD will use NVIDIA DRIVE Thor chips in the future to improve the level of autonomous driving and other digital functions of its models;
For a time, Nvidia DRIVE Thor was in the limelight. According to official information, DRIVE Thor is capable of multi-domain computing, which is equivalent to having cross-domain integration capabilities. Going back, at CES 2024, cross-domain integration has become one of the important highlights in the field of automotive intelligence. For example, Bosch showcased a new cockpit based on Qualcomm's Snapdragon Ride Flex platform, which can also support intelligent cockpit and ADAS functions through a single SoC. For another example, the C1200 family of black sesame intelligent cross-domain computing chips "Wudang" series was unveiled, and JOYNEXT Zhixing, a subsidiary of Joyson Electronics, demonstrated the central computing unit product (nCCU) based on this chip.
According to statistics, in 2023, the penetration rate of Continental's automotive intelligent cockpit will rapidly increase to 61.5%, and the penetration rate of L2 and above autonomous driving systems will increase to 40.8%. It seems that it was yesterday for a single ECU to be integrated to form intra-domain integration, and now, the integration of cabin and driver, cross-domain integration is opening a new page for the development of automotive intelligence.
Cross-domain integration touches the "heart" of car companies
"The cross-domain integration of cabin and driver is a clear trend, which is inseparable from the improvement it can bring to the whole vehicle design, and the significant advantage is the reduction of costs. Wang Tao, director of the user experience module of Desay SV Technology Center, told China Automotive News, "Under the cross-domain integration, two SoCs belonging to two domains are concentrated into one to achieve different functions of the two domains; ”
Tang Wenbin, deputy general manager of Huayang General Motors, a wholly-owned subsidiary of Huayang Group, agrees with this view. "Some parts can be shared through cross-domain integration, which not only reduces the number of parts, but also improves the utilization rate of parts, and can also avoid subsequent repeated development and design work, so as to achieve cost reduction from the perspective of vehicle design and manufacturing. "In addition, cross-domain integration can help OEMs achieve a holistic design approach and optimize vehicle performance." The further compactness of the relevant structural design, coupled with the reduction of wiring harnesses and interface parts, will help to improve the stability and reliability of the vehicle. "Cost reduction must be the advantage of cabin and driver integration that can most attract car companies and supply chain enterprises. Qian Qian, vice president of the Central Research Institute of Hangsheng Group Technology Center, told reporters, "Cross-domain integration can solve the problem of efficient utilization of computing power through integrated design, and different departments can also collaborate on the process to improve development efficiency." Reflected in the vehicle products, it is the improvement of product strength, the data transmission between systems will be faster, the system efficiency and response speed will be higher, the adaptability of the system in complex scenarios will be improved, the user's human-machine co-driving experience will be improved, and the platform extensibility of the model can also be improved with the help of cockpit and driver integration. ”
However, the "germination" of cross-domain integration of cabin and driver integration is just beginning, and the actions of car companies are naturally fast and slow. Wang Tao told reporters: "At present, there are three different attitudes in the market towards the cross-domain integration trend of cabin and driver integration. The first is 'embracing', some auto brands firmly take the lead in carrying out cross-domain integration from low-end models based on cost and other factors, and then gradually promote it to mid-to-high-end models; the second is 'wait-and-see', because it involves adjusting the electronic and electrical architecture of the whole vehicle, which may bring many new problems such as functional safety, information security, data security, hardware redundancy, etc., so some auto brands are in a wait-and-see attitude; Positioning is sufficient to meet consumer needs, so there is no strong demand for cross-domain integration. ”
Multiple factors drive the implementation of technology
"While analyzing the benefits of cross-domain integration to car companies, we need to pay attention to the needs, experiences and expectations of end users. They are the underlying logic that promotes the integration of cabin and driver as a new technology. Tang Wenbin emphasized, "Consumers' requirements for intelligent driving and intelligent interaction of automobiles are constantly increasing. Technological trends such as big data and large models can empower the improvement of the intelligent level of automobiles, and cross-domain integration is the 'base' that plays a supporting role. At the same time, consumers are increasingly viewing the car as a third place, and the integration of cabin and driver can further improve the user experience, bring them more use scenarios, and improve the smoothness of interaction. The driving force from user expectations will drive more technological innovation and industrial upgrading. ”
"From the perspective of functional design, the cabin and driver integration can meet the needs of users for a multi-modal interactive experience. In the current environment of fierce competition in the terminal market, this is an aspect that all car brands are concerned about. In addition, when the computing power reaches a certain level and further challenges high-complexity tasks, perception, planning, decision-making, and control will definitely become centralized. Of course, there is Moore's Law as proof, which is enough to show the inevitability of technological realization. Wang Tao told reporters. "We should also look at this development trend in the context of the evolution of the electronic and electrical architecture of the whole vehicle. Qian Qian believes that "since the development of automotive intelligence, the functions have been continuously enriched, the amount of ECU assembly is also increasing, and the distributed architecture is limited by the efficiency of information interaction and cost efficiency, which cannot meet the new needs of automotive intelligence." "The traditional distributed architecture cannot support the implementation of the currently imagined functional scenarios, which is a very critical factor. Lu Xingwang, Product Director of JOYNEXT China, pointed out, "For example, the cockpit needs to be connected to other domains in the car, and the traditional electronic architecture cannot support the realization of related applications. ”
"In order to achieve the further upgrade of the cockpit and intelligent driving, it is necessary for us to continue to open up the 'two pulses of appointment and supervision'. Massive amounts of data are transmitted to each other, which means that the bandwidth and efficiency of data transmission need to be effectively improved. In addition, autonomous driving and intelligent cockpits have independent and common requirements for sensors and body data, and the integrated design of cabin and driver is conducive to achieving the above requirements. Qian Qian said.
Different convergence paths have their own considerations
Regardless of whether action is taken or not, the industry has basically formed a consensus on the trend of cabin and driver integration, and the difference is nothing more than the speed or the sooner or later it is realized. However, on the specific technical route, there are different understandings in the industry.
"Regarding the path and paradigm of convergence, there are two typical ones. First, it starts with the intelligent cockpit, increasing the complexity of computing power and software, and integrating the cockpit with intelligent driving. Second, starting from the car body, the functional units that undertake in-vehicle communication or interoperability, such as the network connection unit and the gateway system, are integrated, because these parts have similar requirements for safety, stability and other performance. The traditional body and chassis have their own uniqueness, and these domains will not be further integrated until at a later stage. Lu Xingwang introduced. "The convergence from the intelligent cockpit domain to the intelligent driving domain should be more acceptable to the industry. In Tang Wenbin's view, "although the algorithm update and iteration of the intelligent driving domain is very fast, the computing power is better, and there is a large amount of data training, its functions are relatively mature, stable and the goal is unified." In contrast, the intelligent cockpit domain has very significant customized development characteristics. Different brands have different specific needs, but they all focus on paying attention to the user's ride and human-computer interaction experience. Therefore, for vehicle companies, the integration of intelligent cockpit domain to intelligent driving domain is easier to enhance product competitiveness. "The direction of convergence is determined by multiple factors, and although the paths are different, the goals are the same. The user's experience and scenario will determine the scope of convergence, and computing power and security will determine the direction of convergence. Wang Tao told reporters, "Taking typical intelligent cockpit chips as an example, these products are better than GPUs and can support display functions well, but they are relatively lacking in NPU (neural network processing unit) capabilities, so they cannot integrate high-level autonomous driving functions for the time being." On the contrary, the chip NPU in the autonomous driving domain is powerful, and the support for the display is not ideal. Therefore, it is still necessary to return to the user's perspective and consider which experiences and scenarios are what they need most, and these functions should be integrated into the cockpit domain first. For example, in the cockpit domain, in addition to the integration of head-up display (HUD) and electronic rearview mirror (CMS), automatic assisted parking (APA) has been further integrated. Of course, there are also some variables, for example, from the perspective of AI computing power, it seems more logical for the cockpit domain to be integrated into the autonomous driving domain, but some chip companies have launched cockpit chips that support AI computing, and in a longer time period, the deployment of AI computing is likely to be divided into vehicle end and cloud. This just shows that there are many possibilities for the path and direction of integration, which need to be decided according to the specific scenario and experience. ”
"The difference between the two convergence routes in terms of goal achievement, efficiency and difficulty depends on the technology accumulation and strength of different companies, as well as the technological collaboration and development of the entire industry, and it is difficult to simply judge which path will be universally adopted. Qian Qian told reporters, "Compared with the One-Box or One-Board cockpit integration solution, which simply integrates the two technical architectures in one 'box', the real cockpit integration is the One-Chip solution, and the core is the unification of the technical architecture." Therefore, the fundamental integration from cockpit to intelligent driving or intelligent driving to cockpit integration is actually who can give priority to the unification of the integrated architecture between the two different technical architectures, which also includes the unification of the knowledge system. Only when the underlying technology is unified can true cross-domain integration be realized. ”
Convergence is a comprehensive capability challenge
Although the integration of cabin and pilot is gradually approaching, as the first attempt of cross-domain integration, there are still many technical difficulties to be overcome. "Frankly speaking, the integration of cabin and driver is almost all difficult. We need to explore new software and hardware architectures and communication architectures, which will test the R&D capabilities of car companies and Tier 1 component suppliers. In addition to the technical difficulties, the management challenges should not be underestimated. Due to the need to connect the cockpit and intelligent driving functions, the R&D team must have strong horizontal integration and coordination capabilities. Wang Tao said.
"Compared with a single domain controller product, the R&D of domain controllers with cross-domain convergence has changed a lot. Communication protocols, real-time data, security, and data interaction are the difficulties. "At the same time, cross-domain integrated domain control products also need to deal with the access mechanisms of different domains, and more importantly, consider the stability and reliability of the system as a whole." "The main difficulty in the integration of intelligent cockpit and intelligent driving is the difference in technical architecture. Intelligent cockpit technology focuses on human-computer interaction, infotainment, etc., which mainly requires CPU and GPU to provide general computing power and graphics computing power respectively, while intelligent driving covers the integration and application of technologies in the fields of sensors and data processing, involving complex real-time decision-making and control, and mainly requires GPU and NPU to support neural network-based intelligent driving algorithms and process scheduling. Although both include artificial intelligence applications, the cockpit is mainly the application of generative large language models, while intelligent driving involves decision-making and regulatory algorithms, and its technical architecture is significantly different. Qian Qian told reporters, "From the perspective of architecture design, the challenges are reflected in the integration and collaboration of different functional modules, the engineering challenges of complex architectures, the allocation of hardware resources and software security isolation, the data transmission and collaboration between different systems, the differences in cross-domain security and performance requirements, the collaboration of development teams in different fields, and the gap of compound technical talents." Tang Wenbin mentioned: "Cross-domain integration has also brought about changes in software. The importance and complexity of software are getting higher and higher, as the hardware selection may gradually become unified, the implementation and optimization of functions will become more and more dependent on software, and the requirements for real-time and functional safety of intelligent cockpit and intelligent driving are different. "In the early days of cross-domain convergence, software development was indeed more complex than ever. Qian Qian believes that "different chips, upper-level operating systems and application ecosystems are often different, so software development needs to choose different development tools and methods for the functions and characteristics of different chip products." At the same time, different functional safety levels also require more safety issues to be considered in the development process to ensure the stability and reliability of automotive systems, which increases the difficulty of software development. In addition, the cross-domain convergence architecture requires the integration of algorithms so that the software can respond quickly in the context of rapid changes, and the requirements for real-time performance are getting higher and higher. "It should be pointed out that although the early development is difficult and the cost is large, with the maturity of the cross-domain convergence technology architecture and the full development of software, more and more functional components and middleware will be reused, and the later development process will be greatly simplified.
"The design of domain controllers also needs to take into account the increasing functional requirements and rapidly evolving technological innovations. With the continuous improvement of the intelligent level of automobiles, the cross-domain integrated control platform must also be able to adapt to new functional requirements and technological changes. Qian Qian believes that "car companies need to strengthen technical cooperation with enterprises in different links in the upstream of the industrial chain, and carry out all-round collaboration in the whole process from demand design, architecture research and development to data sharing in the later stage." Traditional Tier 1 parts suppliers need to vertically integrate internal enterprise resources and external industrial status, build full-stack R&D capabilities for software and hardware, and transform from parts suppliers to system solution providers. ”
The "price war" does not hinder the integrated promotion of cabin driving
At a time when cross-domain integration is accelerating, the wave of highly "involution" in the domestic automobile market is also affecting the application of technology in its own way.
"Since the beginning of 2024, the vehicle market has been flooded with price reductions, involving almost all car companies, all power types and most price ranges. Lu Xingwang said, "In this context, some car companies are working around the selling point of being more cost-effective, and promoting the integration of more functions in the cockpit, such as the integration of interior and exterior vision, which reflects a more obvious trend in entry-level model products." There are also some car companies that strive to make 'differentiation', and even begin to accelerate the further implementation of the central computing unit. "The safety, reliability and stability of the product need to be further improved, a large amount of data collection and training are still needed to optimize and iterate, and due to the integration of multiple technical fields, the R&D cost is bound to be high, and the cost control is an unavoidable challenge when the 'price war' of the auto market is in full swing." Tang Wenbin said, "The subsequent scale commercialization time depends on a number of factors, including technology maturity, laws and regulations, market competition situation, etc. "The initial cost of new technology development is relatively high, and it will be a serious challenge to control costs while maintaining product performance. Especially in the current situation of serious 'involution' in the domestic auto market, the initial application of cross-domain integration new technologies may be affected by cost pressures. Qian Qian expressed the same concern. "The emergence or even continuation of the 'price war' may not affect the marketization process of cross-domain integration. On the contrary, I personally think it might serve as a reverse booster. Wang Tao believes that "the underlying logic of the 'price war' is cost competition, and cross-domain integration happens to bring significant cost savings, which is also a market factor that promotes the accelerated application of car companies." From this point of view, I am optimistic that the cabin and driver integration will be widely promoted from entry-level and middle-level car products. "In my opinion, the commercialization of cockpit fusion has arrived. It is expected that with the continuous maturity of technology, the gradual reduction of costs and the further release of market potential, more car companies will gradually realize mass production and application in 2024~2025, and their market scale will gradually expand. Qian Qian told reporters, "Perhaps, not all models will achieve the integration of cabin and driver." Due to cost constraints, low-end models may not be equipped with cockpit systems in the short term. However, the cross-domain fusion products of a single chip will gradually mature with the technology to reflect the cost advantage, and then achieve more commercial applications. Although there is a slight difference in expectations, in an interview with reporters, industry insiders without exception made a judgment that the mainland automotive industry in 2024 has ushered in the "first year" of cabin and driver integration.
Text: Ma Xin Editor: Pang Guoxia Layout: Zhao Fangting
Add a chicken leg to the editor!