Zosi Automobile developed and released the "2022 ADAS Domain Controller Key Components Trend Analysis Report".

Zosi Automotive Research Institute has conducted research and summary of the current mainstream high-computing power ADAS domain controller products and technical information in China, such as Huawei MDC and DJI ADAS domain control prototype.

This article will briefly analyze the key components such as CPU, MCU, storage, and interface of ADAS domain controllers.

CPU

In terms of CPU selection, each domain controller is pursuing large computing power chips in order to achieve L2+ high-end ADAS functions such as NOP/NGP, and the most widely used are NVIDIA Xavier and Orin. NVIDIA's Xavier and Orin have built-in stereo binocular hard core acceleration, which can directly output the disparity map (Disparity map) with a hard line, as well as an optical flow acceleration module, stereo binocular optical flow effect is much better than monocular, for stereoscopic binocular enterprises, the most core software asset is stereo matching algorithm, most of which are semi-global matching, but to really do well, it takes a long time to explore.

ADAS domain controller key component trend analysis

NVIDIA Orin chip basic parameters, image source: NVDA official website

The calculation of the depth map is mainly the work of the CPU. The depth map is followed by Freespace computing, which is mainly the responsibility of the GPU.

The MCU in the domain controller

DJI's ADAS domain control prototype adopts Texas Instruments MCU, image source: Zoth Automotive Research Institute

In general, MCUs for ADAS domain controllers are provided by Infineon or NXP, especially Infineon's TC297X/397X series, which has a high market share. The mainstream MCUs of Renesas, Infineon, and NXP all meet the ASIL-D standard, such as the MCU of the DJI ADAS domain controller engineering prototype, which uses Texas Instruments TMS570LC4357, which was launched in 2014 and is not ASIL certified, only AEC Q-100 certified.

In the second half of 2024, Infineon plans to start mass production of the new AURIX™ TC4xx using 28nm technology. The AURIX™ TC4xx family of microcontrollers is targeted at a wide range of data-handling applications such as advanced driver assistance systems, individual domain controllers, new energy and gateway systems.

Infineon TC4xx architecture diagram, image source: Infineon official website

The MCU is very important and is the most important part of ensuring that the domain controller achieves ASIL-C/D certification.

interface

ADAS domain controllers require rich interfaces (video interface, Ethernet interface, CAN interface, etc.) to connect various sensor devices, such as cameras, lidar, millimeter wave radar, ultrasonic radar, combined navigation, IMU, and V2X modules. The camera interface generally adopts GMSL, LVDS, FPDLink and other protocols, millimeter wave radar generally uses CAN/FD communication, and lidar needs to upload a large amount of data and uses Ethernet interface.

Some interfaces and Ethernet switches of DJI ADAS domain control prototype, image source: Zosi Automotive Research Institute

Above the DJI ADAS domain controller carrier board MCU are two Ethernet switches, namely Marvell's 88EA6321, on the left is a hard disk SATA interface, using Marvell's 88SE9171 chip to convert SATA to PCIe interface, most development boards do not have SATA interface, generally use USB interface. Next to the 88SE9171 is also a Taiwanese Winbond W25Q64JV, which is a NOR Flash with a capacity of 64Mb and is supposed to store a simple hard drive driver.

Marvell's first-generation automotive Ethernet switch 88EA6321 is a 7-port Ethernet Gigabit performance switch that fully complies with IEEE802.3 automotive standards, supports AVB (audio/video bridging function), and supports low-energy Ethernet to reduce power consumption. The 7-port Ethernet switch integrates two IEEE 10/100/1000BASE-T/TX/T ports, two RGMII/xMII ports (these two ports can be configured as one GMII) port, and one SGMII/SerDes port. The switch provides remote management capabilities to easily connect and configure the device. It is usually the bridge between the main processor and the MCU, namely the NVIDIA Xavier and Texas Instruments TMS570LC4357.

The 88EA6321 is an early product of Marvell, and Marvell's Ethernet switches have evolved into the third generation, but Marvell's more advanced products usually only cooperate with large manufacturers. The 88EA6321 targets less safety-critical applications such as body controllers, Infotainment controllers and gateways. The highest only supports 1G, although Tesla is also using this chip, even if the new products of traditional car manufacturers such as Volkswagen will not use such a low-bandwidth switch, the current most advanced design, up to 10G, that is, Gigabit Ethernet, generally supports 2.5G.

Some point cloud density high lidar, the peak rate may exceed 100M per second, 88EA6321 is not suitable for contact point cloud density high lidar (CAN is even more impossible, CAN up to 0.5M bandwidth), and the current mainstream millimeter wave radar are CAN or CAN-FD interface, a very small number of 4D millimeter wave radar optional Ethernet output, the default is generally CAN-FD.

storage

As in the field of consumer electronics, high-speed memory devices such as LPDDR DRAM, UFS, and eMMC have begun to be adopted on a large scale to meet the data transmission and storage requirements of the system and software-level algorithms of ADAS domain controllers. At present, the mainstream domain controller storage combination mainly adopts the form of "LPDDR+UFS", which is consistent with the mobile phone storage combination.

Micron automotive-grade LPDDR5, image source: Micron official website

Domain-controlled storage, Micron is at the forefront of the industry. In June 2021, Micron launched its first automotive UFS 3.1 memory device portfolio with cost/density advantages. Micron UFS 3.1 has twice the data read performance and 50% better sequential write performance than the previous generation UFS 2.1, meeting the growing demand for real-time local storage of sensor and camera data in ADAS systems and black box applications above Level 3. The ideal L9 ADAS domain controller is equipped with Micron's automotive-grade LPDDR5 DRAM memory and UFS 3.1 memory chip based on 3D TLC NAND technology. To date, Micron LPDDR5 is the industry's only memory product to achieve ASIL-D certification for automotive safety integrity.

With the improvement of the level of intelligent driving assistance of vehicles and the gradual application of high-speed/urban NGP, automatic parking AVP and other functions, it will further drive the DRAM capacity, bandwidth and product requirements of vehicle regulations.

In terms of capacity, according to Micron's data, the DRAM capacity requirement for L1/2 cars is about 8GB, while the L3 and L5 levels are increased to 16GB and 74GB, respectively.
In terms of bandwidth, the bandwidth of L2 DRAM is generally 25-50GB/s, the bandwidth can reach 200GB/s at L3, and the bandwidth will be increased to 1TB/s after L4.
In terms of products, the L2 level mainly adopts the basic DDR2/DDR3, and at this stage, L2 begins to upgrade to L3, and DRAM will gradually switch to DDR4/LPDDR4/LPDDR5/GDDR5.

At the UFS level, UFS is specifically defined by JEDEC as a high-performance memory alternative to e-MMC. It has become a major smartphone solution and continues to migrate to in-vehicle and other applications. UFS will eventually surpass e-MMC as the primary storage solution for automotive applications.

Deserialize

The representative model of the deserialization is Texas Instruments' DS90UB960. 360 panoramic fisheye lens surround view is usually completed by Infotainment, the effective distance of surround view is generally within 10 meters, it is impossible to do long-distance ADAS, generally only used for parking, ADAS domain controllers do not consider 360 surround view, moreover, a DS90UB960 is enough to correspond to 4 360 surround view image sensors.

DS90UB960 typical application diagram, image source: Texas Instruments

The figure above is a typical application diagram of Texas Instruments DS90UB960, that is, four YUV444 data with a frame rate of 30Hz of 2 million pixels, or four YUV420 data with a frame rate of 60Hz of 2 million pixels, the latter is more likely. The DS90UB954 is a simplified version of the DS90UB960, reduced from 4Lane to 2Lane. It is generally matched with the DS90UB953. It is speculated that Tesla's car driver status monitoring with this chip, because the LVDS output of the camera is not suitable for long-distance transmission, basically the camera must be equipped with a deserialization chip to convert parallel data into serial coaxial or STP transmission, so that the transmission distance is long and EMI electromagnetic interference is easier to pass the car regulations.

It is necessary to explain the data format of the camera, which is usually made of RAW RGB and YUV. There are three common levels of YUV, YUV444, YUV422 and YUV420. The formula for calculating bandwidth is pixels for RAW RGB× frame rate × bits×4, for example, a camera output 30Hz, 2 million pixels, then the bandwidth is 2 million x30x8x4, that is, 1.92Gbps, this bandwidth is too wide. YUV444 is a pixel × frame rate × bit ×3, or 1.44Gbps, YUV422 is a pixel × frame rate × bit ×2, or 0.96Gbps, and YUV420 is a pixel × frame rate × bit × 1.5, or 0.72Gbps. ADAS usually does not give much thought to color, and the YUV420 is sufficient. In addition to vehicles, YUV422 is generally used.

Table of Contents of the 2022 ADAS Domain Controller Critical Components Trend Report

Number of report pages: 109 pages

01. High computing power domain controller product solution

1.1 Huawei

1.1.1 Huawei Intelligent Driving Domain Controller Product Matrix

1.1.2 Features of Huawei Intelligent Driving Domain Controller

1.1.3 Internal architecture of Huawei's intelligent driving domain controller

1.1.4 Huawei MDC610 domain controller

1.1.5 Huawei MDC610 logical architecture

1.1.6 Huawei MDC610 low-power state

1.1.7 Huawei MDC610 appearance interface

1.1.8 Huawei MDC610 interface solution - CAN/Ethernet/Video Interface

1.1.9 Huawei MDC610 Typical Deployment Solution - Debugging/Mass Production Solution

1.1.10 Huawei MDC610 Thermal Specifications - Liquid/Air Cooled

1.1.11 Huawei MDC610 deployment location

1.1.12 Huawei MDC610 box size and installation space - liquid/air cooled

1.1.13 Huawei MDC610 Thermal Management Deployment Requirements

1.1.14 Huawei MDC610 adapts to the sensor

1.1.15 Huawei MDC300/F domain controller

1.1.16 Huawei MDC300/F external interface

1.1.17 Huawei MDC300/F internal structure

1.1.18 Huawei MDC300/F Technical Specifications - Interface/Machine/Computing Power and Power Supply

1.1.19 Huawei MDC300/F hardware design block diagram

1.2 Desay SV

1.2.1 Desay SV intelligent driving domain control product planning

1.2.2 Desay SV Autopilot Domain Controller IPU: Product Matrix and Comparison

1.2.3 Desay SV autonomous driving domain controller: software and hardware architecture and division of labor logic

1.2.4 Desay SV Autopilot Domain Controller IPU04: Internal Structure/Software Architecture

1.2.5 Desay SV Converged Central Computing Platform Aurora: Performance Features/Design Solutions

1.3 DJI Automotive

1.3.1 Domain controller appearance and interface

1.3.2 Domain control internal structure form - domain control motherboard/SOM socket/chip module

1.3.3 Introduction to domain-controlled motherboard devices

1.3.4 Domain-controlled Ethernet switches

1.3.5 Domain Control Deserialization

1.3.6 Domain-controlled debugging PCB

1.3.7 Block diagram of the domain control system

1.3.8 Domain Control Key Devices - Master Table/CPU/MCU/Deserialization

1.3.9 Control ECU appearance and interface

1.3.10 Control the internal structure of the ECU

1.3.11 Block diagram of the control ECU system

1.3.12 Control ECU key devices

1.4 Tesla

1.4.1 Tesla Autopilot function upgrade path: HW1.0→HW3.0

1.4.2 Tesla Domain Controller HW: Development Path

1.4.3 Tesla Domain Controller HW2.5 VS HW3.0

1.4.4 Tesla Domain Controller HW3.0: Internal Structure

1.5 Milli-Smart

1.5.1 Milli Intelligent Driving Domain Controller: Product Development Planning

1.5.2 Intelligent Driving Domain Control: "Little Magic Box 3.0" (IDC 3.0)

1.6 Fredek

1.6.1 Fritec Smart Driving Domain Control Layout

1.6.2 Fratech third-generation intelligent driving domain control ADC30 architecture diagram

1.6.3 Flextec's third-generation intelligent driving domain control ADC30 sensor configuration and function support

1.6.4 Fredeco's next-generation autonomous driving domain control ADC-X

1.7 Domain Controller Technology Benchmarking

1.7.1 Domain Controller Product Technology Benchmarking(1)

1.7.2 Domain Controller Product Technology Benchmarking(2)

1.7.3 Domain Controller Product Technology Benchmarking (3)

02. Comparison of key components of high-end ADAS domain controllers

2.1 Chip

2.1.1 Chip: NVIDIA

2.1.2 Chip: Horizon

2.1.3 Chip: Black Sesame

2.1.4 Chip: Qualcomm Snapdragon RIGE

2.1.5 ADAS domain controller chip manufacturer software stack solution

2.1.6 Comparison of Product Performance Parameters of ADAS Domain Control Chip Manufacturers (1)

2.1.7 Comparison of Product Performance Parameters of ADAS Domain Control Chip Manufacturers (2)

2.1.8 Comparison of Product Performance Parameters of ADAS Domain Control Chip Manufacturers (3)

2.1.9 Comparison of Product Performance Parameters of ADAS Domain Control Chip Manufacturers (4)

2.1.10 Comparison of Product Performance Parameters of ADAS Domain Control Chip Manufacturers (5)

2.2 MCU

2.2.1 Major ADAS domain control MCU vendors

2.2.2 MCUs: comparison of Infineon's TC series

2.2.3 MCU: Infineon TC2xx

2.2.4 MCU: Infineon TC3xx

2.2.5 MCU: Infineon TC4xx

2.2.6 Trend 1

2.2.7 Trend 2

2.2.8 Trend 3

2.3 Storage

2.3.1 Storage: Micron

2.3.2 Storage: Samsung

2.3.3 Storage: Hynix

2.3.4 Storage: Kioxia (former Toshiba Storage)

2.3.5 Storage: Beijing Sicheng (ISSI, wholly-owned subsidiary of Beijing Junzheng)

2.3.6 Storage Trend 1

2.3.7 Storage Trend 2

2.3.8 Storage Trend 3

2.4 I/O interface

2.4.1 LVDS/CAN/ETH

More Zoth reports

For report ordering and cooperation consultation, please send a private message.

Zoce's 2023 research report writing plan

Panorama of the Intelligent and Connected Vehicle Industry Chain (December 2022 Edition)

Self-owned brand OEMs autonomous driving	Automotive Vision (Domestic)	High-precision maps
Autonomous driving of joint venture brand OEMs	Automotive Vision (Overseas)	High-precision positioning
ADAS and autonomous driving Tier 1 - domestic	Surround Market Research (Local)	Automotive gateways
ADAS and autonomous driving Tier 1 - foreign	Surround Market Research (Joint Venture)	Data closed-loop research
Autopilot and cockpit domain controller	Infrared night vision	Automotive information security hardware
Multi-domain computing and regional controllers	Autonomous driving simulation (overseas)	Automotive security software
Passenger car chassis domain control	Autonomous driving simulation (domestic)	OEM information security
Domain controller rank analysis	Lidar - Domestic Chapter	Wireless communication module
E/E architecture	Lidar - Foreign Chapter	Automotive 5G convergence
L4 autonomous driving	Millimeter wave radar	800V high voltage platform
L2/L2+ autonomous driving	Automotive ultrasonic radar	fuel cell
Passenger car camera quarterly report	Radar disassembly	All-in-one battery
ADAS data annual report	Laser and millimeter wave radar rankings	Integrated die casting
Joint venture brand Internet of Vehicles	Autonomous driving of special vehicles	Automotive operating system
Self-owned brand Internet of Vehicles	Autonomous driving in mines	Drive-by-wire chassis
Autonomous driving heavy trucks	Unmanned shuttle bus	Skateboard chassis
Commercial vehicle ADAS	Unmanned delivery vehicles	Electronically controlled suspension
Intelligent cockpit for commercial vehicles	Unmanned retail vehicle research	Steering system
Networking of commercial vehicles	Autonomous driving of agricultural machinery	Brake-by-wire research
Intelligent chassis for commercial vehicles	Port autonomous driving	Charging and swapping infrastructure
Automotive intelligent cockpit	Modular reporting	Automotive motor controllers
Intelligent cockpit Tier1	V2X and road-to-road coordination	Hybrid report
Cockpit multi-screen and joint screen	Roadside IntelliSense	Automotive PCB research
Intelligent cockpit design	Roadside edge computing	IGBT and SiC research
Gauge and central display	Automotive eCall system	EV thermal management system
Smart mirrors	Automotive EDR research	Automotive power electronics
Dashcam	Smart car personalization	Electric drive and power domain
Car digital key	Automotive multimodal interaction	Automotive wiring harnesses
Automotive UWB research	In-car voice	Car audio
HUD industry research	TSP manufacturers and products	Car seats
Interactive	Autonomous driving regulations	Automotive lighting
In-vehicle DMS	Autonomous driving standards and certifications	Automotive magnesium alloy die-casting
OTA research	Intelligent network connection test base	Denso Shin-4 Chemical
Automotive cloud service research	PBVs and automotive robots	NIO, a new automaker
Automotive functional safety	Flying cars	A new force in car manufacturing - Xiaopeng
AUTOSAR research	Overnight Integrated Research	A new force in car manufacturing - ideal
Software-defined vehicles	Smart parking research	Autopilot chips
Software vendor	Car timeshare rental	Cockpit SOC
Passenger car T-Box	Shared mobility and autonomous driving	Automotive VCU research
Commercial vehicle T-Box	Digital transformation of automotive companies	Automotive MCU research
T-Box ranking analysis	Smart surfaces	Sensor chip
Xpeng G9 function disassembly	Model supplier research	On-board memory chip
Ideal L8/L9 function disassembly	NIO ET5/ET7 intelligent function disassembly	Automotive CIS Research
DJI forward-looking binocular disassembles with Tudatong LiDAR	Autonomous driving fusion algorithm

"Zoth Research Monthly Bulletin"

Included in Collection #佐思年报

223 pcs

PreviousAUTOSAR research: CP+AP integration, ecological construction, and localization will be the key direction

ADAS domain controller key component trend analysis

CPU