Analysis and overview of mainstream chip solutions Intelligent driving has become one of the preferred configurations for Chinese users when buying cars. The entire product layout from L0 to L2++ has become a key publicity point when car manufacturers sell cars. Intelligent driving is different from autonomous driving, whether it is Legal regulations and product definitions are clear and clear. Users’ own needs for smart driving products are also different. Finding new smart driving (assisted driving) functions is an urgent focus of car manufacturers. After high-speed/elevated assisted driving products have been launched, assisted driving on urban roads has been put on the agenda. Currently, the solutions we are focusing on are Mobileye’s EyeQ6/NVIDIA’s Orin and Thor/Horizon J5/Black Sesame’s A1000 . Qualcomm’s Snapdragon Ride FLEX 8540 chip-related demo has also been launched. The following five major solutions are divided into different parts of the world. In order to reduce costs, domestic chips can better reflect the cost performance, and the market will eventually hand over the report card.
01 Analysis of China’s Intelligent Driving Market Trends from 2020 to 2030
Currently, the classification of intelligent driving/autonomous driving by NHTSA (National Transportation Safety Administration) and SAE (Society of Automotive Engineers) is: L0~
L2++: Intelligent Driving/Assisted Driving L3 ~L5: Autonomous Driving From 2020 to 2030, the cost overview of smart and autonomous driving in the Chinese market is shown in the figure:
Image source: Stratedy Analytica Infineon
As can be seen from the above figure, the cost of L2 level intelligent driving is US$160, and the Chinese market in 2020 will be about 6 million vehicles. By 2030, the cost of L4/L5 autonomous driving will be US$970, and the Chinese market will be about 4 million vehicles. At the same time, it can be seen that lidar is gradually becoming standard equipment on L3 and above autonomous driving models. The intelligent driving configuration of L3 autonomous vehicles in 2025 will cost US$630, which is a reasonable price range. In 2025, the price difference between L2+ intelligent driving and L3 autonomous driving will be reduced to less than 100 US dollars. Among users with a price difference of 100 US dollars, if the law is relaxed, more people will choose L3 autonomous driving.
According to the current development of smart driving, until 2025, the Chinese market will still be dominated by L2 to L2+ smart driving. Laws, regulations and costs are important factors that restrict car manufacturers from developing autonomous driving above L3. In 2025, major car manufacturers will begin to develop L3 autonomous driving. 2023 and 2024 are the times for solution providers to focus on demonstrating L3 autonomous driving.
Lidar is the threshold for L3 level. It is still controversial whether to use lidar, but currently most OEMs have reserved lidar. At the current stage, fusion solutions have become mainstream, and cabin-driving fusion or cockpit fusion solutions are gradually being mass-produced, but the final central computing platform will still take some time.
It is expected that after 2025, L3 and above autonomous driving solutions will begin to be implemented, and central computing platforms will also begin to be implemented. In 2030, L4/L5 autonomous driving will gradually become mainstream, and the automotive industry will enter a new milestone, fully autonomous vehicles.
02Development Trend of Intelligent Driving Technology in China
Currently, the intelligent driving/autonomous driving level (SAE J3016) is divided into L0~L5. This division is more suitable for China's national conditions. L0~L5 stipulates the division of responsibilities between the driver and the vehicle, as shown in the following figure:
· L0~L2++ intelligent driving is still applied in LDW/FAW and other scenarios
· L3~L5 autonomous driving begins to allow drivers to leave the steering wheel, and the law still needs to be perfected
China's roads are very complex, and the probability of legal liberalization in the short term is very low. Currently, most OEMs are still focusing on intelligent driving in specific scenarios such as highways and elevated highways. In the short term, we also see new forces such as Xiaopeng starting intelligent driving on urban roads. Due to the involution of Chinese car manufacturers, the gap in technical strength between major car manufacturers will become smaller and smaller.
Changes in vehicle electronic architecture Domain controller>Convergence solution>Central computing platform (Bosch Automotive EE architecture)
When talking about the evolution of E/E architecture, Tesla cannot be avoided. The concept of functional domain division appeared in Model S in 2012. By 2017, Tesla further tried the central computing + regional controller framework on Model 3, and the prototype of central computing appeared.
Take SAIC Zero Beam Galaxy Full Stack 3.0 architecture as an example, which consists of 2 HPC high-performance computing units and 4 regional controllers. Two HPCs serve as the vehicle's computing center and are used to implement functions such as smart cockpit, smart driving, and smart driving redundancy backup. Four regional controllers are used to implement related functions in different areas.
GAC Aian's Xingling architecture is composed of three core computing units and four regional controllers, and uses high-speed vehicle Ethernet as the backbone, combined with 5G communication technology, to form a high-speed interconnection architecture inside and outside the vehicle. Among them, the three core computing units include the central computing unit, the automatic driving control unit, and the infotainment control unit. In this case, the central computing unit is equipped with NXP S32G399 gateway computing chip, the cockpit domain control can be equipped with Qualcomm 8155 or 8295 chip, and the smart driving domain control module is equipped with Huawei Ascend 610 high-performance chip.
In addition, there are Dongfeng Lantu ESSA architecture, Great Wall Motors (601633)-GEEP4.0 architecture, Xpeng X-EEA3.0, etc. The architecture designs of each company are not consistent and the names are also different, but the overall trend is still towards Central computing + regional control architecture moves forward.
Image source: Bosch
Regarding the central computing platform, it is expected that the probability of integrating smart driving/autonomous driving into smart cockpits is high in the future, because the safety requirements of smart driving are high and the safety level of integrated driving is lower. At present, China's domestic independent platforms have begun to develop their own central computing platforms, and the mainstream solution is central computing platform (Center Computer/HPC) + regional controller .
03China ’s mainstream smart driving chips and solutions
What problems does intelligent driving solve for users? In the final analysis, it is about partial scenes to the whole scene to finally achieve the purpose of driverless driving. Depending on whether the working conditions are used occasionally or full-time, as well as scenarios such as human-machine co-driving and driverless driving, we can solve: energy consumption issues, efficiency issues, comfort issues, and safety issues.
All implementations of intelligent driving/autonomous driving include four processes: perception > fusion > decision-making > control ; let’s take a look at the solution analysis of mainstream players.
1Intel solution Mobileye EyeQ5 -- 24TOPS/EyeQ6H --34TOPS
EyeQ5 is a 7-nanometer chip with a multi-threaded 8-core CPU, plus an innovative generation 18-core Mobileye vision processor. EyeQ5 is produced by TSMC, and its FinFET technology design at 10 nanometer nodes or below helps it greatly reduce energy consumption and greatly improve performance.
Image source: Mobileye
Both EyeQ6 and Ultra belong to the latest generation of chips, especially the Ultra chip calculation example can reach 176TOPS (L4 autonomous driving preparation).
Image source: Mobileye
Mobileye's EyeQ6 is divided into three versions : Light version , High version and Ultra version :
Image source: Mobileye
Recently, Ji Krypton Motors has announced that it will mass-produce Mobileye EyeQ5 High on a certain car in the first half of 2024, but there is currently no mass production model information for EyeQ6.
2NVIDIA solution Drive AGX platform/Orin chip and Thor chip
Image source: NVIDIA
NVIDIA DRIVE Orin™ SoC (system-on-chip) can provide 254 TOPS (tera operations per second) and is the central computer of smart vehicles. It is the ideal solution to power smart/autonomous driving functions, confidence views, digital clusters, and AI cockpits. With the scalable DRIVE Orin product family, developers can move from Level 2+ systems all the way to Level 5 fully autonomous vehicle systems by building, scaling and leveraging a single development investment across their entire fleet.
1 NVIDIA DRIVE Orin™ SoC (system-on-a-chip) delivers 254 TOPS (tera operations per second)
2 With DRIVE Thor, automakers can efficiently integrate multiple functions such as digital instrument clusters, infotainment, parking, and assisted driving on a single system-on-chip, thereby greatly improving development efficiency and the speed of software update iterations. 2000 TOPS and 2000 TFLOPS are all used for autonomous driving workflows.
NVIDIA DRIVE Thor is a next-generation centralized in-vehicle computing platform that runs advanced driver assistance applications and in-vehicle infotainment applications on a single safe and reliable system. The DRIVE Thor superchip leverages new CPU and GPU breakthroughs to deliver outstanding 2000 teraflops of performance while reducing overall system cost, and is scheduled to begin mass production in 2025.
DRIVE Thor can be configured in multiple modes, using all of its 2000 TOPS and 2000 TFLOPS for autonomous driving workflows, or it can be split and configured with one part for cockpit AI and infotainment functions and one part for autonomous driving. Assisted driving.
3 Horizon Program Journey Platform/J5 -- 128TOPS 2023 SOP
Horizon chip is currently the leader in domestic intelligent driving solutions. Li Auto has currently used this chip in its L7/L8/L9 mid-range models.
Image source: CSDN@KilnHuang
There are a wealth of interfaces available in the J5 chip:
J5 development board photos:
Image source: Horizon Developer Community
4 Qualcomm solution Snapdragon Ride platform/Ride Flex SoC:
SA8650 GEN2 for ADAS L2++ 2024 SOP
SA8775 GEN3 for center computer 2024 Q4 SOP
The Snapdragon Ride platform is built on a series of different Snapdragon automotive SoCs and accelerators, using scalable and modular high-performance heterogeneous multi-core CPUs, industry-leading GPUs, and energy-efficient AI engines and computer vision engines. Based on different combinations of SoCs and accelerators, the Snapdragon Ride platform can match the needs of each market segment of autonomous driving and provide corresponding computing power support according to different levels of autonomous driving systems.
Snapdragon Ride adopts a highly customizable SoC platform based on the 5nm process technology and has flexible scalability. It can not only help automakers and first-tier suppliers develop products with more energy efficiency and high heat dissipation performance, but also create customized products. It provides more choices for centralized solutions and points the direction for the evolution of next-generation automotive architecture.
Snapdragon Ride also has a rich software ecosystem that helps automakers bring greater safety and comfort to daily driving through software and applications optimized for complex use cases. Examples include autonomously guided human-like highway driving, as well as providing modular options for perception, localization, sensor fusion and behavior planning. These autonomous driving software stacks will help automakers and first-tier suppliers accelerate development and innovation, and enhance the autonomous driving experience from multiple aspects such as safety and Internet of Vehicles.
The Snapdragon Ride platform uses high-performance, low-power computing to bring open, customizable and complete advanced driver assistance systems and autonomous driving solutions to a wide range of models and categories, and will promote the future of autonomous driving.
5Huawei Solution MDC Platform/Ascend 910 -- 320TOPS
Ascend 910 is an AI processor with ultra-high computing power. Its maximum power consumption is 310W. As a highly integrated system-on-chip (SoC), in addition to the AI core based on DaVinci architecture, Ascend 910 also integrates It has multiple CPUs, DVPP and task scheduler (TaskScheduler), so it has self-management capabilities and can give full play to its advantages of high computing power.
Huawei's MDC intelligent driving computing platform is highly open source and can not only use Huawei's self-developed Hongmeng platform, but also support Adaptive AUTOSAR and ROS, and is equipped with supporting tool chains.
The MDC intelligent driving computing platform has also obtained the ISO26262 functional safety management certification issued by Rheinland, Germany, which proves the safety of the platform. At the same time, the MDC intelligent driving computing platform also has the characteristics of high energy efficiency and low latency.
The industry has estimated the computing power of L2-L5 autonomous driving. The computing power required for L2 level is <10tops, the computing power required for L3 level is 30-60tops, the computing power required for L4 level is >100TOPS, and the computing power required for L5 level is 500- 1000TOPS. The chip computing power of the Alpha S Huawei HI version equipped with Huawei's HI autonomous driving system reaches 400TOPS, far exceeding the computing power required for L4 autonomous driving. It is worth mentioning that according to Huawei’s original plan to reach L5 level autonomous driving in 2030, the high-end version of the MDC intelligent driving computing platform has a computing power of up to 800TOPS.
Huawei MDC Intelligent Driving first established a complete vehicle-cloud integrated solution for data collection, transmission, and storage, and collected massive data through a test fleet. The development of advanced driving assistance systems and autonomous vehicles requires the collection, transmission, and storage of massive data. Huawei Cloud Services provides data express, data injection, and data storage requirements to optimize costs and efficiency.
Data processing, algorithm development and model training
It provides high-performance massive data analysis and processing and data pre-annotation capabilities, reducing the time and cost of the data preparation phase for model training. It has optimized machine learning and deep learning frameworks, and its training and inference speeds are far ahead.
Autonomous driving simulation test
During the simulation phase, extremely strong GPU computing power is required. While consuming a large amount of computing resources, a large amount of temporary data will be generated, which also has extremely high requirements for storage bandwidth and latency. Huawei Cloud provides GPU accelerated cloud servers, allowing developers to build The simulation environment you need can be used flexibly as needed.
Autonomous driving development platform technical architecture
Develop autonomous vehicles to collect, transmit, store and manage massive amounts of data. Huawei Cloud provides massive, highly scalable storage and computing capabilities, as well as big data components such as Hadoop and Spark, and an AI one-stop development platform. It is pre-integrated with data preprocessing and semi-automatic annotation algorithms. Users can efficiently complete autonomous driving through the platform. On-demand model training and simulation testing help users reduce the complexity and cost of infrastructure deployment in the development environment.
Intelligent driving on highways and specific scenarios has become popular. Autonomous driving is coming from 2025. Major manufacturers are making technical reserves for 2025. Intelligent driving on urban roads is currently the biggest promotion point of major car manufacturers. Yixin Technology will provide Urban roads and autonomous driving in 2025 do your part!
*Note: The pictures in this article are all from the Internet. If there is any infringement, please contact us to delete them.