Domestic intelligent driving SoC chip suppliers are about to break through, in-depth analysis of the chip industry landscape

To join the " Smart Car Experts Autonomous Driving Industry Exchange Group", please add 18512119620 (same as WeChat), and note company-name-position to join the group

Automotive chips: Electrification and intelligence drive the increase in both volume and price of automotive semiconductors

According to their functions, automotive chips can be divided into four categories: main control chips, power chips, sensor chips and other chips (Figure 29). Among them, the main control chip includes computing chips and control chips, which are integrated circuits and are mainly used for information processing, computing analysis and decision-making; the power chip is a discrete device, which mainly converts electrical energy and controls the circuit; the sensor chip is mainly responsible for sensing the operating conditions of the car and converting the information into electrical signals.

As the core component of automotive electronic systems, automotive chips are widely used in all aspects of modern automobiles, responsible for controlling and managing various functions of the vehicle. From engine control to in-vehicle entertainment systems, from safety systems to autonomous driving technology, their importance is self-evident. At present, automotive chips have become the core of the development of the intelligent industry of new energy vehicles, and their scope of use covers five major sectors: body, instrument/infotainment system, chassis/safety, powertrain, ADAS and autonomous driving system (Figure 30).

According to the Black Sesame Smart IPO Prospectus, the global automotive chip market size is estimated to be approximately RMB 355 billion in 2023. With the continuous development of automotive electrification and intelligence, the demand for higher chip utilization and better performance continues to grow. With continued development and growing demand, it is expected that the global automotive chip market will exceed RMB 600 billion by 2030.

Compared with consumer-grade and industrial-grade chips, automotive-grade chips have more severe working environments, lower fault tolerance, longer service life requirements, and longer supply life cycles. They must be able to withstand the harsh and extreme temperatures, humidity, mechanical vibrations, shocks, and complex electrical and electromagnetic environments of vehicles in daily use. For example, automotive-grade chips need to adapt to a temperature range of -40°C to 150°C, and their general design life is 15 years or 200,000 kilometers.

The stringent specifications mentioned above jointly determine that automotive chips need to undergo a series of complex and rigorous testing and certification processes to ensure that they meet the relevant requirements of the automotive grade before they can be further put into mass production. So in a sense, automotive chips can also be defined as "chips with automotive-grade quality standards that can be used for automotive control", which is also the difficulty and technical threshold that distinguishes them from consumer-grade and industrial-grade chips. Chip automotive certification standards usually include the following three dimensions of control:

Quality management standard IATF 16949: Standard specification for automotive design, development and production quality management system. The content covers product safety, risk management and emergency plans, embedded software requirements, change and warranty management, and secondary supplier management. For automotive chip products, this management system needs to be followed from chip design to tape-out and then to large-scale production.

Reliability standard AEC-Q100: A universal reliability test standard for automotive-grade components, and an important reference guide for automotive industry parts suppliers. Automotive-grade chips must pass AEC-Q testing, and different semiconductor devices correspond to different test types, and different automotive parts also need to pass different levels of testing.

Since its initial release, AEC-Q100 has been revised several times, and the J version of the test certification standard document was released in August 2023. This is also the latest standard requirement currently used by chip companies to carry out AEC-Q100 test certification, which includes 7 major test items: accelerated environmental stress test, accelerated life test, package inspection test, wafer reliability verification, electrical characteristics verification, defect screening test and cavity package integrity test.

Functional safety standard ISO 26262: An international standard specifically developed for the functional safety of automotive electronic systems. This standard covers the functional safety requirements of the entire life cycle of chips, including project requirements planning, design, wafer manufacturing, and finally the entire process of packaging and testing. It aims to reduce the risk of chip failure during use to ensure that these safety-critical devices meet the requirements for use in automobiles.

Since automotive semiconductor products need to undergo a series of tests and certifications including the above three standards to enter the automotive supply chain, the certification cycle is long and there are high barriers. At the same time, considering product stability and test verification costs, automotive companies generally do not randomly update suppliers during the product life cycle of the same model.

Process

Compared with consumer electronics, automotive-grade chips mainly use mature processes due to higher demands for safety and stability. It should be noted that although most automotive chips are still based on mature processes, considering the increasingly rich functions of cars in the context of intelligence, the volume of information data is increasing day by day, which puts higher requirements on the computing performance of the main control chip; coupled with the continuous strengthening of the real demand of car companies for cost reduction and efficiency improvement, chip manufacturers are also constantly pursuing lower processes to achieve performance improvement and cost reduction.

Products such as MCU (Microcontroller Unit), CIS (CMOS Image Sensor), display driver IC, and MEMS (Micro Electro Mechanical Systems) sensors do not have high requirements for process technology and are generally based on mature processes (above 28nm).

The AI ​​chips, main control SoC (System on Chip), and GPU (Graphics Processing Unit) required for scenarios such as smart cockpits and smart driving are continuously pursuing advanced processes (below 28nm) to achieve breakthroughs in computing performance and cost.

Intelligent driving SoC chip: the central brain of intelligent driving cars

2.1 Smart driving SoC is the "central brain" of smart driving cars and is crucial to the performance of ADAS and ADS cars.

• Smart driving SoC is a SoC designed specifically for smart driving functions. It is usually integrated into a camera module or a smart driving domain controller as the "central brain" of a smart driving car. Smart driving functions generally involve three levels: perception, decision-making, and execution. Smart driving SoC is used in the decision-making layer, responsible for processing and fusing data from sensors in the perception layer, and then making driving decisions on behalf of human drivers.

• Intelligent driving SoC and solution industry chain: mainly includes semiconductor manufacturers, intelligent driving SoC and solution suppliers and terminal applications.

• ADAS SoC market space: The ADAS SoC market has expanded rapidly in recent years. According to Frost & Sullivan, in 2023, the global and Chinese ADAS SoC markets will reach RMB 27.5 billion and RMB 14.1 billion respectively; the global ADAS SoC market is expected to reach RMB 92.5 billion by 2028, with a compound annual growth rate of 27.5% from 2023 to 2028; the Chinese ADAS SoC market is expected to reach RMB 49.6 billion by 2028, with a compound annual growth rate of 28.6% from 2023 to 2028.

Due to its more advanced autonomous driving capabilities and complex functions, ADS SoCs are generally more valuable than SoCs for ADAS applications. According to Frost & Sullivan, the global ADS SoC market is expected to reach RMB 8.1 billion by 2026 and RMB 45.4 billion by 2030. China is expected to become the largest market in terms of ADS car sales. It is expected that the Chinese ADS SoC market will reach RMB 3.9 billion and RMB 25.7 billion by 2026 and 2030, respectively.

2.2 Intelligent Driving SoC Chips: With the upgrading of intelligence, the demand for intelligent driving chip computing power has increased

As the level of intelligence increases, the demand for computing power of intelligent driving chips increases.

• The intelligent driving SoC chip is the "central brain" of the intelligent driving car in the process of realizing high-level intelligent driving. It needs to uniformly analyze and process massive amounts of data in real time and perform complex logical operations, so its computing power requirements are very high. At present, the intelligent driving SoC usually needs to heterogeneously integrate CPU with general/special chips such as GPU, FPGA, ASIC, and SoC chips that integrate AI accelerators will be widely used to realize parallel computing of large amounts of data and complex logical functions. With the upgrading of intelligence, the computing power demand of intelligent driving chips continues to increase.

• Low-computing SoC chip: The computing power is usually 2.5-20 TOPS. At present, the L0-L2 level functions of intelligent driving have been mass-produced and applied in passenger cars. The price range of the car models is generally 100,000-150,000 yuan. In pursuit of high cost performance, they are mainly equipped with low-computing SoC chips. The product form is mainly a front-view integrated machine or a distributed driving or parking controller solution. Some models can provide high-speed NOA function.

• Medium-computing SoC chip: The computing power is usually between 20 and 80 TOPS. The price range of the models equipped with it is generally RMB 150,000 to 250,000. The supported product form is mainly a lightweight integrated parking domain controller solution; in terms of functions, it can usually realize functions such as high-speed NOA, city memory NOA and memory parking, and some models may provide city NOA function.

• High-computing SoC chips: The computing power is usually above 80 TOPS. With the maturity of L3 and above autonomous driving, the development of high-level intelligent driving functions requires new algorithms (Transformer+BEV+OCC) and more advanced vehicle EE architecture (central computing + regional control), which all require the support of SoC chips with higher computing power.

2.3 Intelligent Driving SoC Chips: Self-developed IP, tool chains, AI training platforms, etc. build differentiated advantages for chip manufacturers

In addition to stacking computing power, self-developed core IP, improving the development tool chain, and deploying AI training platforms are important means for major SoC manufacturers to build differentiated advantages.

• Self-developed core IP: In chip design, the configuration of heterogeneous IP is very important. Autonomous driving SoC chip manufacturers are constantly strengthening the research and development of core IP to maintain key competitiveness. For example, NVIDIA upgraded its existing GPU-based product line to a "three-core" strategy of "GPU+CPU+DPU", with layouts in both CPU and DPU; China's Black Sesame Intelligence launched two self-developed core IPs - NeurallQ ISP image signal processor and deep neural network algorithm platform DynamAI NN engine.

• Improve the development tool chain: The development tool chain of SoC chips is crucial. Only by forming a developer ecosystem can we build long-term sustainable competitiveness. For example, Horizon Robotics provides J5 with a complete set of edge computing platform algorithm implementation solutions, providing a wealth of development tools, examples, and model releases with a large number of built-in algorithm models to improve development efficiency and quickly deploy self-developed algorithm models on the Horizon Robotics computing platform.

• Layout of AI training platform: Autonomous driving data sets are crucial for training deep learning models and improving algorithm reliability. SoC manufacturers have launched self-developed AI training chips and supercomputing platforms. Tesla has launched the AI ​​training chip D1 and Dojo supercomputing platform, which will be used for training Tesla's autonomous driving neural network. In addition, training algorithm model products are becoming increasingly important, including 2D annotation, 3D point cloud annotation, 2D/3D fusion annotation, semantic segmentation, target tracking, etc., such as NVIDIA Drive Sim autonomous driving simulation platform, Horizon "Eddie" data closed-loop training platform, etc.

Automotive-grade chip industry landscape :

Domestic automotive chip manufacturers have achieved breakthroughs from 0 to 1 in most areas

China's automotive chip manufacturers have achieved a breakthrough from 0 to 1 in most areas and have begun to emerge in some segments, but the problems of weak industrial foundation, few product categories, and poor chip performance have not been completely solved, and the overall localization rate is less than 10%. On the sales side, China's automotive semiconductor market sales have increased year by year, and it has now become the largest single-country semiconductor market and the world's second largest consumer of automotive semiconductor products. IDC data shows that in 2023, China will account for 20.3% of the global automotive semiconductor market with a revenue scale of US$13.7 billion.

From the perspective of specific players, the concentration of each segment of the automotive chip market is generally high, and the global automotive chip giants are also showing a strong position in their respective segments. According to statistics and estimates by Semiconductor Intelligence, the top 5 manufacturers in the global automotive semiconductor market in 2023 will account for more than 50% of the market share. Among them, Infineon leads with a market share of 13.7% due to its leadership in the entire automotive chip market and power semiconductors; followed by NXP and STMicroelectronics (ST), with shares of 11.2% and 10.6% respectively; Texas Instruments (TI) and Renesas Electronics account for 8.9% and 7.0% of the market share respectively. More than 80% of the global supply of automotive chips is in the hands of these international IDMs (Integrated Device Manufacturers).

In addition, as the market demand for high-computing computing chips grows, Qualcomm, Nvidia and others are gradually dominating the computing chip market. Fuel vehicle sensor giants such as Bosch and Continental and emerging sensor leading manufacturers such as ON Semiconductor, OmniVision and Tyco Electronics are jointly dominating the automotive sensor chip market.

From the perspective of vertical industry segments, with the rapid development of new energy vehicles and intelligent connected vehicles, the degree of automotive electronics and intelligence is increasing, and the demand for automotive-grade chips is showing a rapid growth trend, making automobiles one of the important application areas that drive the growth of the semiconductor industry.

In terms of market size, the global automotive semiconductor market will be worth approximately US$67.4 billion in 2023. IDC predicts that the global automotive semiconductor market will exceed US$88 billion by 2027, and the attention and importance of semiconductor companies in the automotive industry chain will be further enhanced.

From the dimension of shipment quantity, IC Insights data shows that after global automotive chip shipments increased by 30% year-on-year to 52.4 billion units in 2021, the growth rate will still exceed 10% in 2022/23 (Figure 41); it is estimated that by 2030, global automotive chip demand will exceed 100 billion units.

Although the overall growth of the automotive market has stabilized, with the continuous advancement of automotive electrification and intelligence, the demand for various chips is increasing, including the demand for higher utilization and better performance, which has brought about a double increase in the amount of chips used per vehicle and the value of a single chip, driving a significant increase in the total value of chips for the entire vehicle, and bringing new growth opportunities to the automotive semiconductor industry. As the value of semiconductors per vehicle continues to grow, the attention and importance of semiconductor companies in the automotive industry chain will also increase further.

The electrification of automobiles increases the amount of chips used per vehicle

From the perspective of a single vehicle, the China Association of Automobile Manufacturers has calculated that the number of chips required for traditional fuel vehicles is 600-700; for new energy vehicles and vehicles with assisted driving functions, the demand will increase to 1,600 chips per vehicle; and for more advanced self-driving vehicles, the chip demand is expected to exceed 3,000 chips per vehicle. SIA statistics show that Hyundai Motor may have 8,000 or more semiconductor chips and more than 100

Electronic control unit. Combined with the natural growth of new energy vehicle sales, the total amount of automotive chips has broad room for growth. In addition, according to Deloitte statistics, by 2025, benefiting from the demand for electronic components in new energy vehicle battery management and electric powertrain, the BOM value of automotive electronic components will increase significantly (Figure 46).

The intelligentization of automobiles has increased the demand for computing power, driving up the value of chips

In terms of chip usage, taking autonomous driving, an important application in intelligence, as an example, the higher the level, the more sensors are required, and the sensor chip usage also increases accordingly. According to Deloitte statistics, L3 autonomous driving is equipped with an average of 8 sensor chips, while the number of sensor chips required for L5 autonomous driving is increased to 20. According to the calculation of iResearch, the average number of chips per smart electric vehicle will reach 2,072 in 2025, gradually widening the gap with traditional fuel vehicles.

In terms of chip value, the amount of information that a vehicle needs to process and store is also positively correlated with the maturity of autonomous driving technology, and the value of semiconductors is positively correlated with the processing power of the vehicle system. For example, the value of semiconductors for smart sensors at the L2 level is about $160-180 per vehicle; when upgraded to the L2+ level, it is $280-350 per vehicle; and when it reaches the L4/5 level, the value will further increase to more than $1,150-1,250 per vehicle.

Three major chip development trends:

Cross-domain integration, OEMs may deploy upstream in various ways, and there is great room for domestic substitution

4.1 Development Trend: Cross-domain integration with a single chip will become the main technology trend in the future

At present, the electronic and electrical architecture of mainstream car companies is mostly in the stage of moving from functional domain to cross-domain integration. At present, cross-domain integration in hardware is mainly concentrated at the controller level. We believe that as the core device supporting cross-domain integration, achieving cross-domain integration at the chip level through a single-chip solution will become the main technical trend in the future.

• Cross-domain computing: refers to using a single chip to achieve functions that would otherwise require multiple chips. Cross-domain integrated chip products can not only bring about further improvements in intelligent functions, but also effectively reduce costs and power consumption because a single chip can achieve capabilities that would otherwise require multiple chips.

• At present, chip manufacturers such as NVIDIA, Qualcomm, CoreDrive, and Black Sesame have already developed cross-domain fusion-related chip layouts and products. NVIDIA's Orin and Qualcomm's 8295 chips that support cross-domain fusion functions have been mass-produced, and NVIDIA's Thor and Qualcomm's Snapdragon Ride Flex chips with higher computing power are expected to be mass-produced in 24-25 years. Domestic CoreDrive and Black Sesame have also released their cross-domain fusion chip products. We expect that from 2024, cross-domain fusion chips and functions are expected to accelerate mass production.

4.2 Development trend: There is a large space for domestic substitution, and domestic suppliers are expected to stand out

We believe that the current localization rate of smart driving SoC chips is low, and there is a large space for domestic substitution. With the mass production of domestic smart driving SoC chips, domestic suppliers are expected to gradually stand out with their cost-effective advantages and localized advantages of good service and fast response.

• Horizon Robotics, Heizhima Intelligent and other domestic suppliers' mass-produced SoC chips are already competitive in terms of parameters compared with mass-produced products designed by Nvidia, a major international competitor.

• According to Horizon's prospectus, the company's products have certain competitive advantages over Israeli company A in terms of openness, system operating efficiency, service and responsiveness, and cost-effectiveness.

• Domestic suppliers have highly flexible business models and strong localization advantages. Domestic suppliers usually allow customers to choose any solution or any combination of components from the full stack of products from algorithms to software and development tools to processing hardware, which can meet the diverse and customized needs of customers. Local suppliers are usually able to provide additional services, including joint software development and consulting services and image tuning services for automotive OEMs and Tier 1 suppliers, with good services and quick response.

4.3 Development trend: OEMs may expand upstream in various ways, and the key is whether they can achieve economies of scale.

We believe that by comprehensively grasping the initiative in the industrial chain, reducing costs, and improving performance in multiple dimensions, OEMs may deploy upstream in a variety of ways, but whether they can achieve economies of scale is the key.

• Tesla started using its self-developed FSD chip in 2019; NIO, Xpeng, and Ideal have all chosen to cross-border develop their own intelligent driving SoC chips to achieve full-stack self-research from hardware to software, from chips to algorithms; many OEMs including Geely, BYD, Great Wall, SAIC, etc. have strategically deployed in the intelligent driving chip track by investing in chip companies.

• We believe that whether car companies choose to develop their own intelligent driving chips, as well as the depth of their participation in the chip track, is a comprehensive consideration of multiple dimensions, including taking the initiative in the industrial chain, reducing costs through in-house development, and having higher software and hardware coordination with self-developed algorithms to improve performance.

• Considering that the computing power and process requirements of intelligent driving SoC chips are continuously improving, development trends such as cabin-driver integration at the chip level have further increased the difficulty of R&D of intelligent driving SoC chips. The costs of IP licensing, tape-out, testing, personnel R&D, etc. will continue to increase. Therefore, whether the scale effect can be achieved to share the cost is the key to whether the OEM chooses to develop its own intelligent driving SoC chip.

Self-developed: Tesla develops its own AI chip D1 and builds its own supercomputing center Dojo

Tesla has developed its own AI chip D1 and built its own supercomputing center Dojo to support the large-scale artificial intelligence training needs of Grok 3, FSD, Optimus Prime robots, etc.

• In 2021, Tesla released the D1 chip at AI Day. The D1 has 50 billion transistors, powerful and efficient performance, can quickly handle various complex tasks, and is dedicated to machine learning. In May 2024, the D1 chip began production, using TSMC's 7nm process node.

• To achieve higher bandwidth and computing power, the Tesla AI team fused 25 D1 chips into one tile to operate it as a unified computer system. Each tile has 9 petaflops of computing power and 36 TB of bandwidth per second, and includes power supply, cooling and data transmission hardware. By using wafer-level interconnect technology InFO_SoW (Integrated Fan-Out, System-on-Wafer), 25 D1 chips on the same wafer can achieve high-performance connections and work like a single processor.

• Six tiles make up a rack, two racks make up a cabinet, and ten cabinets make up an ExaPOD. At AI Day 2022, Tesla said that Dojo will scale by deploying multiple ExaPODs. All of these together make up a supercomputer.

• In 2024, when Musk visited Tesla’s supercomputer cluster at the Texas Gigafactory (Cortex), he said, “This will be a system with approximately 100,000 H100/H200 GPUs and large-scale storage for video training of fully autonomous driving (FSD) and Optimus robots.” In addition to NVIDIA GPUs, the supercomputer cluster is also equipped with Tesla HW4, AI5, and Dojo systems.

Self-developed: Wei, Xiao and Li Auto cross-border self-developed intelligent driving SoC chips

NIO, Xpeng and Li Auto have all chosen to cross-border develop their own intelligent driving SoC chips to achieve full-stack self-research from hardware to software, from chips to algorithms.

➢ NIO: The first self-developed intelligent driving chip Shenji NX9031 has been successfully taped out. At NIO Day 2023, NIO released the first self-developed intelligent driving chip Shenji NX9031. Shenji NX9031 adopts 5nm automotive-grade process, has more than 50 billion transistors, supports 32-core CPU, and can be used with Tianshu global operating system. At NIO IN 2024, NIO announced that Shenji NX9031 was successfully taped out. According to Li Bin, a self-developed chip can achieve the performance of four NVIDIA Orin X chips. NIO's flagship sedan ET9 will first use this intelligent driving chip and SkyOS Tianshu system.

➢ Xiaopeng: Xiaopeng started to build a chip team in 2020. In 2022, Xiaopeng changed its chip design partner from Marvell, an American chip design company, to Socionext, which will be responsible for chip backend design. At the MONA M03 launch conference in August 2024, Xiaopeng officially announced the successful tape-out of Turing's self-developed AI smart driving chip.

➢ Ideal: Ideal has previously focused more on power semiconductors. Since 2023, it has strengthened its chip R&D team. The current team size is about 200 people, and it continues to promote chip architecture optimization and innovation, such as focusing on new technologies such as chiplet and RISC-V.

• We believe that car companies choose to develop their own intelligent driving chips, on the one hand for cost reduction considerations, and on the other hand because self-developed chips can also have a higher degree of software and hardware coordination with self-developed algorithms, thereby improving performance.

Investment: Many OEMs have strategically positioned themselves by investing in chip companies

Many OEMs including Geely, BYD, Great Wall, and SAIC have strategically deployed in the intelligent driving chip market by investing in chip companies.

• In 2018, Ecarx Technology and ARM China, subsidiaries of Geely Holding, jointly invested in Hubei CoreEngine Technology Co., Ltd., which focuses on the design, development and sales of advanced automotive electronic chips and is committed to becoming the world's leading provider of automotive electronic chip solutions. In June 2021, CoreEngine Technology's self-developed China's first 7nm automotive-grade SoC chip "Dragon Eagle No. 1" was officially released. In 2024, CoreEngine Technology's new generation of intelligent driving chip AD1000 was unveiled at Ecarx Technology Day. The computing power of a single AD1000 can support L2++ assisted driving capabilities, two chips with a total computing power of 512TOPS can meet the computing power requirements of L3 intelligent driving, and four chips with a total computing power of 1024TOPS can support L4 autonomous driving.

• Industrial capital with backgrounds in OEMs such as SAIC, Volkswagen (through its software company CARIAD), Chery, BYD, Dongfeng, FAW, GAC, and Great Wall have all invested in Horizon.

• Industrial capital with backgrounds in OEMs such as Xiaomi, Geely, SAIC, and Dongfeng has invested in Heizhima Intelligence.

A review of representative chip companies at home and abroad

NVIDIA can provide complete solutions from driving to cockpit, from software to hardware. Orin is currently NVIDIA's main intelligent driving SoC chip.

• NVIDIA’s NVIDIA DRIVE AGX™ platform includes all the hardware and software needed to develop autonomous driving features and immersive cockpit experiences. The platform is an open, modular platform running NVIDIA DRIVE™ OS, which, when used with supported sensors and accessories, enables manufacturers to build autonomous driving features and in-vehicle AI applications. These features and applications can be further enhanced through over-the-air updates (OTA).

• Since 2015, NVIDIA has entered the field of automotive SoC and in-vehicle computing platforms to provide basic computing capabilities for autonomous driving. Since then, NVIDIA has released an automotive-grade SoC chip almost every two years, and has continuously increased the level of computing power. In 2020, the Xavier chip had a computing power of 30 TOPS, and the Orin released in 2022 had a computing power of 254 TOPS. At the 2022 Fall GTC conference, a new autonomous driving chip Thor was released with a computing power of 2000TFLOPS@FP8 and 4000TOPS@INT8, replacing the previously released Altan with a computing power of 1000TOPS.

• Currently, NVIDIA's main intelligent driving SoC chip is Orin. The OEMs that have mass-produced it include SAIC Zhiji, Ideal, NIO, Xpeng, BYD, Volvo, Xiaomi, etc. In addition, customers also include Robotaxi and many other star companies such as Cruise, Zoox, Didi, Pony.ai, AutoX, etc.

• Nvidia’s next-generation intelligent driving SoC chip is Thor. Currently, Zeekr, Xiaopeng, Ideal, BYD, etc. have announced plans to install it.

Mobileye is one of the world's leaders in autonomous driving solutions, moving from basic assisted driving to advanced intelligent driving.

• As one of the global leaders in autonomous driving solutions, Mobileye has delivered 200 million chips worldwide, and more than 170 million cars are equipped with Mobileye technology. At the same time, as an important participant in China's advanced driver assistance system (ADAS) and intelligent driving market, in recent years, Mobileye has also reached long-term strategic cooperation with Geely Group, FAW Group, SAIC Group, Great Wall Motors, Dongfeng Group and other companies on a number of important models.

• From 2008 to 2021, Mobileye launched five chips: EyeQ1, EyeQ2, EyeQ3, EyeQ4 and EyeQ5. In 2023, Moblieye released EyeQ6H and EyeQ6HL. The computing power of EyeQ6H reaches 34 TOPS. By using 2-4 EyeQ6H at the same time, it can cover application scenarios such as Mobileye SuperVision, Mobileye Chauffeur (L3), and Mobileye Drive (Robotaxi).

In 2024, Mobileye announced the EyeQ7H chip at CES, with a computing power of 67Tops.

Mobileye released the EyeQ Kit, a software development kit (SDK) for the EyeQ system-on-chip, moving from a black box to openness.

• In the early days of smart driving, Mobileye's black box packaging design was ready to use after installation, which greatly saved car companies' usage costs. At that time, Mobileye became a supplier to many car companies in the ADAS field with its low-cost visual perception solution. However, with the development of smart driving technology, full-stack self-development has become a trend, and car companies' requirements for assisted driving have become more and more diverse, so Mobileye's black box solution has become a limitation.

• In 2022, Mobileye officially released the first software development kit (SDK) for the EyeQ system-on-chip - EyeQ Kit. Using EyeQKit, automakers can deploy differentiated algorithms and human-machine interface tools on the EyeQ platform based on the energy-efficient architecture of the EyeQ 6H processor.

• Through EyeQ hardware and software, automakers can access comprehensive Mobileye solutions, including computer vision, road information management (REM) autonomous vehicle mapping technology, and driving strategies based on the Responsibility Sensitive Safety Model (RSS). EyeQ Kit enables automakers to further tap into the power of Mobileye's system-on-chip, enhancing assisted driving capabilities while also helping automakers achieve personalized customization exclusive to their own brands.

5.3 Horizon

Horizon has a complete intelligent driving product matrix, which includes not only chips but also algorithms, Horizon Tiangong Kaiwu (tool chain), Horizon Tage (middleware) and Horizon Eddie (training platform) product lines.

• Algorithms: Comprehensive algorithm capabilities cover perception, environmental modeling, planning and control, and driving functions, and can meet the development requirements of advanced driver assistance and high-level autonomous driving solutions at all levels.

• Horizon Robotics: Horizon Robotics is Horizon Robotics' algorithm development toolchain, which includes a series of ready-to-use modules and reference algorithms. It has a user-friendly interface and rich auxiliary tools, enabling users to accurately and efficiently deploy algorithms and software on their hardware.

• Horizon Tage: Horizon Tage is a set of secure, simple and easy-to-use embedded middleware for advanced autonomous driving. It provides standardized automotive-grade services and tools to help speed up development, integration and verification, thereby significantly promoting and accelerating mass production.

• Horizon AIDI: Horizon AIDI is a software development platform designed to efficiently complete the automatic iteration and improvement of models. By providing various tools and application interfaces as well as simplified workflows, AIDI helps software developers optimize the entire software development process from deployment, training, verification, evaluation to iteration. (Source: Western Securities )

Due to limited space, to be continued ...