According to IHS data, the global automotive semiconductor market is projected to reach $68.2 billion by 2025, including approximately $17 billion for analog ICs, $11 billion for discrete devices, $10.1 billion for logic ICs, $8.7 billion for memory ICs, $8.5 billion for microcontrollers, $6.6 billion for optical semiconductors, and $6.3 billion for sensors and actuators.
Based on the specific application areas of semiconductors in intelligent vehicles, automotive semiconductors can be divided into computing chips, memory chips, sensor and actuator chips, and communication chips related to intelligence, as well as energy supply chips related to electrification. Meanwhile, with the increasing complexity of processed events, there is also the possibility of integrating several different types of chips together to form a system-on-a-chip (SoC). Typically, an SoC chip includes one or more processors, memory, analog circuit modules, mixed-signal modules, and on-chip programmable logic, thereby effectively reducing the development cost of electronic/information system products, shortening the development cycle, and improving product competitiveness.
Computing and Control Chips: These chips, primarily microcontrollers and logic ICs, are mainly used for computation, analysis, and decision-making. Similar to the human brain, they can be divided into main control chips and auxiliary chips. Main control chips include MCUs (microprocessors), CPUs (central processing units), FPGAs (field-programmable gate arrays), and ASICs (application-specific integrated circuits), while auxiliary chips include GPUs (graphics processing units) responsible for image processing and AI chips (intelligent computing chips).
Memory chips: mainly used for data storage, specifically including DRAM (Dynamic Memory), SRAM (Static RAM), FLASH (Flash Memory), etc.
Sensor chips: Primarily used to detect and sense external signals, physical conditions (such as light, heat, and humidity) or chemical compositions (such as smoke), and convert the detected information into electrical signals or other desired forms to transmit to other devices. Specifically, they include CIS (CMOS image sensors), MEMS, current sensors, magnetic sensors, gyroscopes, VCSEL chips, and SPAD chips (used in LiDAR).
Communication chips: mainly used for sending, receiving and transmitting communication signals, including baseband chips, radio frequency chips, channel chips, power line carrier communication chips, satellite navigation chips, etc.
Energy supply chips: These are mainly used to ensure and regulate energy transmission, and are primarily discrete devices. Specifically, they include power management chips (AC/DC, LED driver chips, etc.), transistors (IGBT, MOSFET, etc.), diodes, thyristors, etc.
From a chip type perspective, traditional CPUs used for central computing can no longer meet the computing power requirements of intelligent vehicles, giving rise to system-on-a-chip (SoC) solutions that integrate AI accelerators. In the era of distributed architecture, the ECU was the core of the vehicle's functional system, with its main control chip being a CPU, used only for logic control (AND, NOT, addition, or subtraction). As the E/E architecture accelerates its upgrade from distributed to domain controller/central computing, the domain controller (DCU) is replacing the ECU as the standard configuration for intelligent vehicles. During this upgrade process, the computing power and functionality of the CPU alone can no longer meet the needs of vehicle intelligence. SoC solutions, which heterogeneously integrate CPUs with general-purpose/dedicated chips such as GPUs, FPGAs, and ASICs, have been pushed to the forefront, becoming the main arena for the computing power arms race among major AI chip manufacturers.
In a System-on-a-Chip (SoC), each processor chip performs its specific function. The CPU is responsible for logic operations and task scheduling; the GPU, as a general-purpose accelerator, can handle neural network calculations and machine learning tasks such as CNNs, and will undertake the main computing work for a considerable period of time; the FPGA, as a hardware accelerator, has the advantage of programmability and performs excellently in sequential machine learning such as RNNs/LSTMs/reinforcement learning, playing a prominent role in some mature algorithm areas; and the ASIC can achieve optimal performance and power consumption, and as a fully customized solution, it will highlight its value in autonomous driving algorithms.
From the perspective of application scenarios, computing chips can be divided into smart cockpit chips, autonomous driving chips, and vehicle body control chips.
(1) Smart cockpit chip
Chip structure: The main solution is a heterogeneous fusion of SoC with "CPU + functional modules". Taking the Qualcomm 820A series of smart cockpit main control computing chips as an example: The Qualcomm 820A chip adopts a 14-nanometer process. In terms of overall performance, it can achieve a startup time of less than 3 seconds for the hypervisor and QNX system, a startup time of less than 18 seconds for the Android system, and a startup time of less than 3 seconds for the reversing camera. After further disassembly, it can be divided into four major modules: (1) CPU, which adopts a 64-bit quad-core processor (Qualcomm® Kryo™ CPU) with a main frequency of up to 2.1GHz, used for scheduling and management of all hardware resources; (2) GPU, which adopts Qualcomm Adreno530 GPU, which can support multiple 4K ultra-high-definition touch screen displays and realize one chip for multiple screens; (3) DSP, which adopts Qualcomm® Hexagon™ 680 DSP, which can support the simultaneous input of 8 camera sensors without increasing the CPU load; (4) LTE modem module, which ensures that the vehicle obtains continuous mobile connectivity during driving. In addition, the chip can be equipped with Qualcomm's deep learning software development kit (SDK) – Qualcomm Snapdragon Neural Processing Engine (NPE), which enables the integration of advanced driver assistance systems based on machine learning.
Competitive Landscape: Renesas, Nvidia, Qualcomm, Intel, and Samsung have distinguished themselves in the mid-to-high-end cockpit chip market thanks to their superior chip performance and supply chain. Qualcomm, Samsung, and Nvidia, in particular, have significantly reduced the R&D costs of next-generation architectures (the R&D costs for 7nm and 5nm processes are exorbitant) due to their massive shipment volumes and technological reserves in the mobile phone and consumer electronics sectors, allowing them to secure a leading position in the smart cockpit chip market. Currently, Qualcomm has achieved near-monopoly in emerging flagship models in China, with its cockpit product iteration speed almost simultaneously with mobile phone product updates (Samsung and MediaTek's cockpit chips lag behind mobile phone chips by at least one generation). According to Qualcomm data, its automotive chip order backlog exceeded $8 billion in 2021, with monthly shipments of main control chips reaching millions of units. Among domestic manufacturers, Huawei and Horizon Robotics have quickly gained popularity with their Kirin 990A and Journey 2 chips, respectively. Similar to Qualcomm, Huawei boasts strong R&D capabilities, a robust HarmonyOS ecosystem for the Internet of Things, and iterative development capabilities comparable to Qualcomm. The Jihu Alfa S is the first vehicle to feature the Kirin 990A, a single chip capable of simultaneously powering a 12.3-inch LCD instrument cluster, a 20.3-inch 4K touchscreen, and an 8-inch HUD, achieving an overall computing power of 3.5 TOPS (Qualcomm's latest cockpit chip, the SA8155P, achieves 3 TOPS). Horizon Robotics, with its open development platform and complete toolchain, has also attracted OEMs, with its Journey 2 cockpit chip already selected for the Changan UNI-T model.
(2) Autonomous driving chip
Chip structure: The main SoC heterogeneous solution is "CPU+GPU+NPU". Taking NVIDIA's autonomous driving main control computing chip Xavier series as an example, the SoC chip mainly includes three modules: control unit, computing unit and AI acceleration unit: (1) Control unit (CPU): 8-core Carmel CPU based on ARM architecture; (2) Computing unit (GPU): Based on NVIDIA Volta architecture, the single-precision floating-point performance can reach 1.3TFLOPS at 20W power, the Tensor core performance is 20TOPS, and when the power is increased to 30W, the computing power can reach 30TOPS, with strong performance and programmability; (3) ASIC (AI acceleration unit): Includes two ASIC chips, Deep Learning Accelerator (DLA) and Programmable Vision Accelerator (PVA), which are designed to improve CPU performance (perf/watt).
Competitive Landscape: Based on supply methods, solutions can be divided into two main camps: integrated hardware and software solutions and open solutions. Intel (Mobileye) and Huawei are representative domestic and international providers of integrated hardware and software solutions for autonomous driving, offering a complete package of solutions that bundle sensors, chips, and algorithms. The advantage of this approach is that it helps OEMs with insufficient self-development capabilities to quickly implement mass production in vehicles. Mobileye's chip series shipped 54 million units by the end of 2019, holding approximately 70% of the global ADAS market share (2019). Tesla adopted Mobileye's EyeQ3 as the main control chip for autonomous driving in its early Autopilot HW1.0 system.
In addition, Huawei also announced the provision of a full-stack solution. The Huawei MDC computing platform adopts a "unified hardware architecture, a single software platform, and a series of products," and will be first mass-produced on the Jihu Alfa S. NVIDIA and Horizon Robotics are representative domestic and international open-source autonomous driving solution providers. Both possess completely open ecosystems and user-friendly toolchains, allowing OEMs to purchase services at any level, including chips and algorithms. Currently, NVIDIA's autonomous driving solutions have been adopted by numerous emerging manufacturers and domestic brands, including XPeng, Li Auto, NIO, and SAIC Motor. Horizon Robotics is also continuously advancing the deployment of autonomous driving, having announced that it will replace Mobileye as the new supplier of the main autonomous driving control chip in the Li ONE, which will be equipped with two Journey 3 autonomous driving chips. We believe that in the early stages of the intelligent vehicle industry, some OEMs will choose integrated hardware and software solutions, taking into account factors such as cost, development cycle, and system stability. As the industry matures, leading OEMs, having developed considerable algorithm capabilities, will tend to choose more open computing platforms and combine their own scenario-based algorithms with a complete development toolchain to meet differentiated needs.
(3) Body control chip
Chip Structure: Body control chips have relatively low computing power requirements, typically using 8-bit or 32-bit MCU chips. The essence of the body control domain is to integrate functions such as keyless start system (PEPS), anti-pinch ripple system, and air conditioning control system on the basis of the traditional body control module (BCM). Therefore, the main chips are still automotive-grade MCUs. Based on the different data throughput of the chips, automotive-grade MCUs can be mainly divided into three types: 8-bit, 16-bit, and 32-bit. Among them, 8-bit MCUs operate at frequencies between 16-50MHz, have the advantages of simplicity, durability, and low cost, and are mainly used in body control areas such as windows, doors, and wipers; 32-bit MCUs have the highest operating frequency, better processing power and execution efficiency, and are more widely used, mainly in the powertrain and cockpit domains.
Meanwhile, due to the continuous improvement in the performance of 8-bit MCUs, which now meet the application requirements of low-end 16-bit MCUs, coupled with the gradual reduction in the cost of 32-bit MCUs, the market share of 16-bit MCUs is gradually shrinking. According to IHS data, the global automotive-grade MCU market is projected to reach $7.35 billion by 2025, with 32-bit MCUs accounting for 76.6%.
Competitive Landscape: Foreign manufacturers hold a high degree of monopoly, while domestic manufacturers are rapidly rising amid the industry's chip shortage. According to IHS data, foreign manufacturers, leveraging their first-mover advantage, have largely monopolized the global automotive-grade MCU market, including NXP (14%), Infineon (11%), and Renesas Electronics (10%). Since the end of 2020, the automotive industry's chip shortage has intensified, with imported MCU inventory tight and prices high. Against this backdrop, the domestic automotive-grade MCU market is accelerating import substitution. Currently, established domestic automotive-grade MCU suppliers include BYD Electronics, Jiefa Technology, and ChipON Microelectronics.
