PUBLISHER: ResearchInChina | PRODUCT CODE: 2129076
PUBLISHER: ResearchInChina | PRODUCT CODE: 2129076
Cockpit domain controller research: L3 AIDV intelligent cockpit domain controllers are entering a boom period
Driven by multiple factors such as the continuous evolution of the automotive central integrated architecture and the rapid application of automotive AI foundation model technology, intelligent cockpit domain controllers tend to evolve from single-SoC cockpit domain controllers to AI cockpit domain controllers and cockpit-driving integration platforms. They essentially follow a clear path of "centralized computing power, functional integration, and decoupling of software and hardware." By form and time, the development of cockpit domain controllers can be summarized into four major stages:
L1: perception cockpit domain controllers: CPU: <100 k DMIPS, with moderate AI computing capability; DRAM interface: LPDDR3 / LPDDR4; memory bandwidth: <30 Gb/s. Single-chip support for fewer than 3 displays and fewer than 2 cameras. Connectivity and OTA, basic in-cabin perception capabilities (e.g., voice interaction), and cloud voice functions.
L2: partial cognition cockpit domain controllers: CPU: >100 k DMIPS, AI computing performance >10 TOPS; DRAM: LPDDR4 / LPDDR4X / LPDDR5; memory bandwidth: 30-70 Gb/s. Single-chip support for 3-5 or more displays with maximum 2K resolution; 2 or more in-cabin cameras. Support for edge foundation models with 1-20 billion parameters. Local voice command recognition plus cloud semantic understanding. Cognitive capabilities for both in-cabin and out-cabin scenarios are further enhanced.
L3: advanced cognition cockpit domain controllers: CPU: >200 k DMIPS; AI computing performance: >30 TOPS; GPU: >3 TFLOPS; DRAM: LPDDR5 / LPDDR5X; memory bandwidth: above 70 Gb/s. Cockpit-dedicated AI SoCs as well as advanced cockpit-driving integrated AI SoCs. Single-chip support for 6-8 displays with 2K / 4K resolution, heterogeneous display and heterogeneous touch. Support for 4 or more in-cabin cameras and edge foundation models with 50-150 billion parameters. Large model-empowered voice functions and all-scenario proactive perception.
L4: comprehensive cognition cockpit domain controllers: CPU: >500 k DMIPS; AI computing performance: >200 TOPS; DRAM: LPDDR6, GDDR7, HBM and other memory types; memory bandwidth: 300 Gb/s-1000 Gb/s. Single-chip support for 8-12 displays and 8 or more cameras. Support for edge foundation models with over 300 billion parameters.
L3 AIDV intelligent cockpits are seeing rapid growth. High-performance SoC-based cockpit domain controllers facilitate the automotive deployment of edge AI foundation models.
In 2025, 12.974 million passenger cars in China were equipped with intelligent cockpit domain controllers, with the market penetration rate increasing to 56.6%. From January to April 2026, the penetration rate of intelligent cockpit domain controllers further climbed to 63.2%. With the accelerated iteration of the EEA and the rapid advancement of vehicle intelligence toward AI, it is expected that by 2030, the penetration rate of intelligent cockpit domain controllers in passenger cars in China is expected to reach 98.3%, with the installations seen in 24.83 million cars.
In 2025, L3 cockpit domain controllers officially entered mass production, with installations in 221,000 vehicles. From January to April 2026, the installations rapidly increased to 226,000 vehicles, and the total number of vehicles fitted with such controllers for the year is expected to exceed one million. This also marks the official entry of the intelligent cockpit into the AIDV stage. As a key node connecting the past and the future, L3 cockpit domain controllers will become the core arena for industrial competition and technological innovation in the coming years.
Under the wave of AIDV technology, L3 cockpit domain solutions represented by SemiDriveX10, Qualcomm 8397, MediaTek MT8676, etc. have been deployed. With powerful edge AI computing power and breakthroughs in technical architectures, and with the operation of edge AI foundation models as the core, they have successfully supported efficient reasoning of physical world models and the real-time operation of multi-modal AI foundation models, enabling cockpits to truly possess the closed-loop intelligence capabilities of "perception-understanding-decision-execution", promoting the leapfrog evolution of L3 intelligent cockpits from "functional integration" to "AI native".
In 2026, SemiDrive's latest upgraded X10 is based on TSMC's 4nm automotive-grade process and is designed based on the ARMv9 architecture. The CPU performance reaches 250K DMIPS and the GPU performance reaches 3000 GFLOPS. The VPU supports 8K resolution codec and can drive 12 cameras and 8 displays at the same time, catering to mainstream high-end cockpits in all scenarios. At the same time, X10 also integrates 4 HiFi-5s Audio DSPs, which can support 16 microphones and 32 speaker arrays without the need for external audio chips, achieving multi-tone zone, immersive and ultimate sound experience.
In terms of software ecosystem, X10 adopts both QNX 8.0 Hypervisor and Android 16, taking into account functional safety and open ecosystem, and supports customers to conduct prototype development based on the same software baseline on the existing X9 platform, and then quickly and seamlessly migrate to X10, which can significantly shorten the model deployment and performance tuning cycle.
The core breakthrough of X10 lies not only in more parameters, but also in its new technical architecture specially designed for AI cockpits, which completely reconstructs the computing power distribution and system logic of traditional cockpits. Through a single-chip AI cockpit solution, SemiDrive X10 uses 80 TOPS NPU computing power and 154GB/s DDR bandwidth to support local deployment of 9B foundation models and flexibly adapts to mainstream edge foundation models such as Modelbest/Qwen.
In addition, through process and architecture optimization, X10 can achieve a 2-fold improvement in energy efficiency without water cooling. Compared with the traditional "cockpit chip + AI Box" solution, the content-per-car value can be lowered by RMB1,500~3,000, helping AI cockpits spread from high-end flagship cars to the mainstream consumer market, and promote popularization of AI technology.
Driven by the development of AIDVs and AI cockpit platforms, many mainstream suppliers such as Desay SV, AutoLink World, PATEO CONNECT+, Hangsheng Electronics, Megatronix, Visteon, and Bosch are also actively deploying L3 cockpit domain controllers to meet the surging market demand.
In August 2026, Freelander 8 was officially launched as the first production vehicle model under the independent new energy vehicle brand Freelander, a joint venture between Cherry and Jaguar Land Rover. It is also the world's first production vehicle model equipped with Qualcomm Snapdragon 8397. Besides the industry-leading SoC, its auxiliary chips and core materials feature "luxury configurations."
In terms of MCU, Freelander8 carries Infineon TC4D9XP. The chip's TriCore architecture has been upgraded from v1.6.2 to v1.8, with the main frequency rising from 300MHz to 500MHz. It supports up to 6 pairs of lock-step cores running at the same time. The computing power has approached that of low-end SoCs, making it one of the most advanced MCUs in the industry.
In terms of storage, the Qualcomm QAM8397P module is equipped with LPDDR5X memory with a capacity between 24-36GB and a speed of up to 8533Mbps. It can efficiently support the real-time operation of edge AI models and respond in milliseconds. In terms of flash memory, Freelander 8 adopts the current industry-leading UFS 4.0 standard from Kioxia. Compared with Micron SH023/SH021 (mostly UFS 3.1)commonly used by other companies, the single-channel bandwidth of UFS 4.0 can reach up to 23.2Gbps (about 2.9GB/s), the measured continuous read speed is about 4200MB/s, and the continuous write speed is between 2800MB/s and 4000MB/s. The overall read and write performance of UFS 4.0 is twice that of UFS 3.1, and its power efficiency is improved by approximately 45%. The UFS in Freelander8 is THGJFJT0T25BAB8, with a capacity of 128GB and a maximum speed of 4640MB/s.
Amid the development of AIDVs, mass production and adoption of cockpit-driving integration (including multi-chip or single-chip solutions) is accelerating.
The cockpit-driving integration central computing platform is one of the core forms in the evolution of cockpit domain controllers. The installation rate of cockpit-driving integration in passenger cars in China rose from 3.1% in 2023 to 7.3% in 2025. Driven by the evolution of EEAs and the trend of integration, the cockpit-driving integration central computing platform will continue to grow rapidly, with the penetration rate being expected to hit more than 30% by 2030.
As per installations, multi-chip cockpit-driving integration domain controllers are the current mainstream, accounting for 97.4% of the total installations from January to April 2026. In multi-chip cockpit-driving integration, there are various solutions, including dual-chip (1*intelligent driving SoC+1*cockpit SoC), three-chip (2*intelligent driving SoC+1*cockpit SoC or (1*intelligent driving SoC+2*cockpit SoC)), and four-chip (2*intelligent driving SoC+2*cockpit SoC) solutions.
Among them, the dual-chip cockpit-driving integration domain controller (1*intelligent driving SoC+1*cockpit SoC) solution prevails, including the combination of "1*NVIDIA Thor+1*Qualcomm Snapdragon 8 Gen 3", "1*intelligent driving chip (such as Qualcomm 8650, NVIDIA Orin, etc.) + 1*cockpit chip (Qualcomm 8295)", and "1*NIO Shenji+1*Qualcomm 8295".
At the same time, single-chip cockpit-driving integration platforms are rapidly expanding. At present, the market size is still small, but the growth is obvious. The penetration rate increased from 0.5% in 2025 to 2.0% during the period from January to April 2026. In 2026, single-chip cockpit-driving integration will witness accelerated mass production and application. Qualcomm 8775-based cockpit-driving integration is available in many main production vehicle models such as BAIC Arcfox, with installations growing rapidly. Meanwhile, Qualcomm 8797-based cockpit-driving integration lands in vehicles on a large scale with Leapmotor's high-end vehicle model - D19. New platforms such as SiEngine "Longying No.2", Horizon Robotics Starry, Sanechips Lanyue A1, and Renesas R-Car X5H have all announced that they will go into mass production in 2026-2027, which will further promote the quick development of the single-chip cockpit-driving integration market.
Driven by multiple factors such as central integration and cost reduction, single-chip cockpit-driving integration is expected to develop radically.
In October 2025, BAIC Arcfox aT5/S5 and other vehicle models equipped with the cockpit-driving integration platform based on Qualcomm 8775 were mass-produced and launched. This platform is jointly developed and supplied by AutoLink World and Zhuoyu Technology. AutoLink World is responsible for the development in the cockpit field. Through the embedded AI foundation models and the capabilities of cloud foundation models, it can provide a series of AI applications for vehicles. Zhuoyu Technology develops intelligent driving functions; relying on the cockpit-driving integration controller built on the Qualcomm Snapdragon SA8775P, it not only provides high-level intelligent driving capabilities covering all scenarios, but also supports the cockpit capabilities of AutoLink World.
The cockpit-driving integration domain controller is based on Qualcomm Snapdragon SA8775P and Renesas RH850U2A. It can access 4 screens, 12 cameras, 5 radars, and 12 ultrasonic radars, realize one-chip multiple-system solutions, cockpit-driving integration services and cockpit atomic services, create an intelligent interactive product that understands cars, roads, and users better by integrating domain controllers + peripheral equipment + perception algorithms + ecological services.
This domain controller not only delivers advantages such as powerful AI computing performance and ASIL-D functional safety features, but also supports ADAS functions including Lane Keeping Assist (LKA), Automatic Emergency Braking (AEB), and Automated Parking Assist (APA). It also enables L2+ intelligent driving functions such as highway NOA and HPA supported by ADAS software stack suppliers, as well as NCAP and regulatory functions such as the Camera Monitor System (CMS).