Huawei enters the intelligent driving market, and LiDAR and cloud platforms are "in full bloom"
Latest update time：2020-11-26
Reads：
To join the
"Smart Car Expert LiDAR Industry Exchange Group", please add WeChat 13816528214 and note that you want to join the group
Source: Soochow Securities Research Institute
If you need to obtain the PDF document, click the blue word " Smart Car Expert " above and reply "Huawei" to receive it
Intelligent connected vehicles are developing rapidly, and the market size will exceed 300 billion in 2025. With the rapid development of smart cars, the functions of smart cockpits and ADAS are constantly upgraded, and the number of sensors, chip computing power and single vehicle value are all increasing rapidly.
Huawei's extensive sensor deployment
Automotive ADAS sensors have their own advantages and disadvantages. Camera: The basic principle is that the lens forms an image, which can detect information around the driver, such as traffic lights and road signs, but the disadvantage is that it cannot detect the distance between obstacles and the car, and is greatly affected by weather and light.
Millimeter-wave radar: It detects through the reflection of millimeter waves, has a long detection distance, and is less affected by weather, but cannot detect pedestrians and trees, etc. Objects with low radio wave reflectivity.
LiDAR: It can detect the target position by emitting and receiving laser beams, draw high-precision 3D maps, and detect the distance between objects and vehicles, but it is expensive and greatly affected by weather. The current mainstream solution is to use multiple sensors to complement each other.
At the 2020 Beijing Auto Show, Huawei released an 8M front-view binocular camera, a super fisheye camera, a 77GHz millimeter-wave radar, supporting a variety of short-range, medium-range and long-range application scenarios, a 100-line equivalent LiDAR sensor, and a 4D imaging millimeter-wave radar. In terms of LiDAR, Huawei will mass-produce hybrid solid-state LiDAR by the end of 2021, which can achieve an equivalent of 100 lines. By around 2024, the next generation of Huawei's all-solid-state LiDAR will be mass-produced.
Huawei's MEMS LiDAR technology can effectively increase the detection distance and field of view of LiDAR. LiDAR can be divided into two types: mechanical rotation type and solid-state LiDAR. On July 2, 2020, the International Bureau of the World Intellectual Property Organization announced a Huawei LiDAR patent.
Huawei's product is a MEMS solid-state laser radar. Unlike traditional MEMS laser radars, which have one transmitting and receiving component, this radar uses multiple transmitting and receiving components. The patent drawing shows three ranging modules, each of which contains a laser transmitter 101a, a spectroscope 102a, and a receiver 103a. Although this design will increase the size of the radar, it can effectively increase the detection distance and field of view.
The automotive LiDAR industry is dominated by start-ups. Huawei's LiDAR competitors include Velodyne, Quanergy, Ibeo, and domestic companies such as Hesai Technology, RoboSense, and DJI.
Velodyne was involved in LiDAR business earlier, has accumulated certain technologies, and currently has the highest market share. Hesai Technology has strong technical strength, and its products are mainly aimed at medium- and high-speed driverless taxis. RoboSense not only provides radar products, but also provides corresponding algorithms, and its products have been used in low-speed logistics vehicles.
The solid-state LiDAR products that have passed automotive regulations and are in mass production mainly include Velodyne's Velarray and DJI's Tele-15 and Horizon. In comparison, Huawei's LiDAR has a larger horizontal and vertical field of view, a wider scanning range, and a lower vertical angular resolution, making the scanning more precise.
Huawei HI full-stack intelligent vehicle solution forms five major systems
Huawei has entered the field of smart cars based on its accumulated chips, operating systems, machine learning algorithms, cloud services and other basic technologies in the ICT field. On October 30, 2020, Huawei released the Huawei Smart Car Solution-HI brand.
HI full-stack smart car solutions include:
1) A computing and communication architecture that enables:
scalable hardware and sustainable OTA software upgrades. Huawei proposed a cross-domain integrated software stack (VehicleStack) on top of the computing and communication architecture (CCA) to jointly build a digital system, using microservices and microplug-ins, and based on the service concept to build a sustainable profit model for car companies.
2) Five major intelligent systems:
smart car cloud, smart networking, smart driving, smart cockpit, and smart electric.
3) A full set of intelligent components such as LiDAR.
HI technology helps the automotive industry achieve technological upgrades, quickly develop leading smart electric vehicles, and bring consumers the best travel experience.
Huawei enables the upgrade of automotive E/E architecture. As the automotive industry gradually replaces hardware-defined functions with software-defined functions, Huawei enables the transformation of distributed electronic + electrical architecture to computing + communication architecture.
The core of architecture upgrade is reflected in: hardware, software, and communication architecture upgrades.
1) Hardware architecture upgrade: From distributed to domain control/centralized, with higher computing power utilization, unified interaction, and coordinated vehicle functions.
2) Software architecture upgrade: The software architecture is layered and decoupled to promote software versatility and facilitate supplier management.
3) Communication architecture upgrade: LIN/CAN develops towards Ethernet to meet performance requirements such as high-speed transmission and low latency.
HUAWEI CLOUD - Intelligent Cloud Platform
An intelligent cloud platform is built based on the Ascend 910AI chip.
Intelligent car cloud services include:
autonomous driving cloud services (providing data services, training services, and simulation services), Internet of Vehicles cloud services (three-electric, intelligent driving, and intelligent cockpit data collection and storage), and high-precision map cloud services (building a dynamic map aggregation platform, not building maps by ourselves, but letting map suppliers present them on cloud services).
Huawei Autonomous Driving Cloud Service
The pain points of the autonomous driving cloud service industry include:
1) Massive data collection, but a small proportion of valid data, which requires high AI computing power;
2) Autonomous driving development involves multiple technology stacks and many isolated tools.
3) Virtual simulation requires a rich scene library and a high-performance simulation system.
4) There is a lack of evaluation standards and systems for listing, and a lack of a regulatory platform for commercial operations. Huawei has leveraged its years of experience in ICT technologies such as cloud computing, artificial intelligence, and Internet of Vehicles to build a unified full-stack cloud platform to help traditional automakers quickly launch autonomous driving and provide evaluation and regulatory services to evaluation agencies and government departments.
Huawei launched the autonomous driving cloud service, the Octopus system can provide data services, training services, and simulation services.
1) Data service: For massive amounts of raw data, data sets are automatically generated based on platform capabilities such as fusion annotation capabilities and multi-model parallelism.
2) Training service: The combination of AI chips and frameworks greatly improves training efficiency. In the training of a typical ResNet50 network, the Altas900 cluster has nearly doubled the training speed, while supporting mainstream deep learning frameworks in the industry, such as TensorFlow and PyTorch.
3) Simulation service: It can realize the efficient conversion of various data from vehicle-road terminals into simulation scenario libraries. In addition, it can realize simulation of cameras, lidar, millimeter-wave radar, vehicle dynamics, different weather and road conditions, etc.
In addition, the autonomous driving cloud service also includes evaluation services: it can analyze and evaluate modules such as takeover rate, traffic rules, perception, decision-making, planning, and control, serving vehicle evaluation agencies and improving evaluation efficiency.
Huawei Internet of Vehicles Cloud Service
Industry pain points of connected car cloud services include:
1) Massive data is heterogeneous, requiring high unified data processing capabilities;
2) Safety issues occur frequently and there is a lack of operational supervision on the safe operation of vehicles.
3) High requirements for platform security and reliability.
Huawei released the OceanConnect Internet of Vehicles platform to fully enable the digital transformation of automakers. Huawei released the OceanConnect Internet of Vehicles platform in Germany in June 2018, dedicated to enabling the intelligent networking of vehicles, the service-oriented transformation of automakers, and the intelligent evolution of transportation.
In the traditional model, consumers and car companies have little contact and lack stickiness. Huawei's Internet of Vehicles platform can collect and store status data and fault data of digital components such as smart driving and smart cockpits, forming a unified smart vehicle data resource pool, and then based on the powerful AI and big data capabilities of the cloud, it can monetize data assets and provide customers with more valuable automotive services, such as smart driving, fleet management, and preventive maintenance.
OceanConnect Huawei's Internet of Vehicles platform digitizes every car and every road.
1) Ecosystem enablement: Through the separation of data and business structures, we help automakers control digital assets, aggregate third-party content and application ecosystems, and build an ecosystem centered on automakers.
2) Connectivity enablement: Provide stable connectivity for cars, support billions of connections and millions of concurrent connections; meet the global operational needs of automotive businesses through globally accessible public cloud deployment capabilities.
3) Data enablement: By collecting and analyzing big data on vehicles, such as vehicle conditions and driving behaviors, digital portraits (Digital Twins) of people and vehicles are realized on the cloud. Through accurate analysis of vehicle owners’ driving behaviors and travel scenarios, intelligent content distribution and business recommendations are enabled
.
4) Evolutionary enablement: The Internet of Vehicles platform and V2X develop in coordination, from single-vehicle intelligence to vehicle-road collaborative intelligence, enabling future intelligent transportation and improving the overall safety and efficiency of social transportation.
Huawei released the three-electric cloud service:
integrating battery mechanism and data model to achieve battery safety warning and accurate life management. Based on its rich technical accumulation in the battery field, Huawei has launched the three-electric cloud service capabilities by combining cloud computing, AI, big data and other technologies. It can realize cloud visualization of vehicle status, battery fault warning, thermal runaway prevention and control, accurate evaluation of battery health status, accurate prediction of battery remaining life, and optimization of battery control strategy.
Huawei High-Precision Map Cloud Service
The pain points of the high-precision map cloud service industry are:
1) The qualification threshold requirements stipulated by surveying and mapping laws and regulations are high;
2) The security preservation requirements for massive map surveying and mapping data are high;
3) Data desensitization and map element extraction require high AI computing power and algorithms.
Huawei will build a national high-precision dynamic map aggregation platform, and enterprises with high-precision maps can present them on cloud services.
At the 2020 Beijing Auto Show, Huawei released a high-precision map cloud service, which is to build a national high-precision dynamic map aggregation platform. Through data cooperation with map vendors and partners, we can complement each other's strengths and provide customers with map data service capabilities with wider coverage, better quality, and more dynamic and vivid.
Huawei's high-precision map cloud service provides customers with five major service capabilities: storage and application compliance, autonomous driving application support, high-precision map distribution, dynamic map data distribution, and high-precision map data security. It serves four major scenarios: Internet of Vehicles location applications, intelligent connected industrial parks, autonomous driving simulation/operation, and autonomous driving services.
Intelligent network platform: 5G vehicle module + T-Box + Ethernet gateway
Huawei creates intelligent connected solutions to achieve high-speed connections inside and outside the vehicle.
1) Create an open terminal and cloud intelligent network solution to ensure that every vehicle is always online and has direct service;
2) The world’s first 2G/3G/4G/5G full-standard V2X open vehicle module, enabling partners to develop professional products;
3) OceanConnect Internet of Vehicles connection management cloud service, with global access, supports tens of millions of vehicles online at the same time;
4) Based on leading network technology, create an in-vehicle GE~10GE or higher Ethernet network.
Huawei's core products include: 5G+C-V2X vehicle communication module, T-Box, vehicle gateway, RSU, etc.
1)
Huawei
5G vehicle module MH5000: It not only enables vehicle terminals to have high-speed, low-latency 5G mobile communication capabilities, but also has C-V2X (Cellular Vehicle-to-Everything) communication capabilities for vehicle-road collaboration. Huawei's 5G communication module MH5000 highly integrates 5G and C-V2X technologies, uses the 5G baseband chip Balong5000, and has single-core multi-mode, high-speed, uplink and downlink decoupling, supports SA (5G independent networking) and NSA (5G non-independent networking) dual-mode networking, supports C-V2X and other features.
2) Huawei T-Box Platform: Huawei released the first generation of T-BOX platform as early as 2016, which can realize functions such as car networking, vehicle control, and data security. In 2019, Huawei released a new generation of T-box platform, which can greatly improve the response speed and operation speed of the smart cockpit. The response speed is increased by 50% and the operation speed is increased by 60% compared with the previous generation of products. It can realize vehicle anti-theft, network security protection, Bluetooth car keys, remote control, cloud service docking and other functions. At present, BYD has announced that its Han model will adopt Huawei's 5G communication module MH5000 with Balong 5000 as the core.
Device-intelligent driving system: chip hardware + OS + cloud service + sensor
From the perspective of the intelligent driving upgrade path, the current stage is in the L3 introduction stage.
In 2018, we entered the L2 level partial autonomous driving era, where the driver can take their feet off the wheel during driving, and the computing power requirement is less than 10TOPS. Representative functions include ACC with LKA, APA, etc. In 2020, we will gradually enter the L3 level conditional autonomous driving era, where the driver can free their hands and the computing power requirement is about 30-60TOPS. The driver does not have to monitor the system all the time, but must always be vigilant and intervene when necessary. Representative functions include TJP, RPK, etc.
By 2025, we will gradually enter the era of L4 highly automated driving. With the increase in performance of chips and algorithms, the automated driving function will be further upgraded, City Pilot, more advanced AP and other functions will emerge, and the E/E architecture will be further upgraded. By 2030, we will gradually enter the era of L5 fully automated driving, where the whole vehicle is completely controlled by the system, and the computing power requirement will even exceed 1000TOPS.
Huawei has created the MDC intelligent driving platform and promoted the development of intelligent driving through open cooperation.
1) Leverage the advantages of Huawei Cloud + AI to build a car-cloud collaborative intelligent driving platform, including: intelligent hardware platform (that is, chip platform, Huawei uses its self-developed Host CPU and AI chip, ISP chip, storage control chip to build the MDC domain controller) + intelligent driving OS + Octopus autonomous driving cloud service + ADAS software algorithm.
2) Establish certification standards and docking processes to create an open sensor ecosystem;
3) Support partners in developing intelligent driving algorithms and building differentiated applications, services and solutions that can flexibly adapt to intelligent driving scenarios.
4) Establish docking specifications and build an execution component ecosystem together with mainstream manufacturers.
5) Promote the implementation of industry standards and legislation for intelligent driving, build industry consensus, and jointly expand future industrial space.
Huawei is positioned in the incremental automotive market and has the ability to provide full-stack autonomous driving solutions. Its core products include: chip solutions + operating systems + ADAS algorithm software + cloud services. The company is positioned in the incremental automotive market and provides incremental components for automotive customers. Customers can choose to adopt Huawei's solutions according to their own needs.
In 2019, Huawei released the MDC300 and MDC600 platforms based on the Ascend 310 chip. On the eve of the 2020 Beijing Auto Show, Huawei released the new generation platforms MDC210 and MDC610, which provide 48 and 160TOPS computing power respectively to support L2+, L3~L4 autonomous driving.
End-Smart Cockpit System: Kirin Chip + Hongmeng OS + Application Ecosystem
Huawei has created the CDC intelligent cockpit platform, which collaborates in all scenarios and creates a new benchmark for experience.
1) Build the CDC smart cockpit platform to achieve seamless sharing of the entire industry chain including hardware, software and application ecology between smart cars and smart phones;
2) Build IVI modules based on Kirin chips for smartphones, leverage the scale effect of industry chain collaboration, and reduce hardware costs;
3) Based on Hongmeng OS, it shares Huawei's "1+8" ecosystem and realizes seamless interconnection across terminals;
4) Sharing smartphones enriches the APP ecosystem and improves the car-using experience. Open APIs enable cross-terminal partners to develop smart cockpit applications.
Huawei will build a smart cockpit platform and ecosystem based on Kirin chips and Hongmeng operating system. In 2019, it will launch Hicar car-connected interconnection solution to achieve a deep interconnection "1+1+N" model, that is, the interconnection of a mobile phone, a car machine and other smart terminals.
HiCar has more than 20 car manufacturers as partners in the ecosystem, more than 150 models of cooperation, and more than 30 applications. In 2021, the number of HiCar pre-installed models reached 5 million. In addition, on Hi Brand Day in 2020, Huawei launched the pre-installed product HMS for Car, the in-car projection HiCar, and the after-installed product in-car smart screen for C-end car users.
Harmony car OS is the first neutral and open OS developed for smart cockpits. HMS for Car and Hicar businesses are still part of the consumer business, with more than 400 people in the two business segments.
End-intelligent electric system: mPower+chip hardware+vehicle control OS+three-electric cloud services
Huawei builds VDC intelligent electric platform to enable car companies to innovate differentiated electric vehicle experience. Huawei focuses on electric vehicles and builds VDC intelligent electric platform to enable car companies to create differentiated user experience for different user preferences and provide customers with VDC hardware platform + vehicle control OS.
Introducing the network energy industry chain and technological advantages into smart electric vehicles, creating mPower multi-mode electric drive and efficient on-board charging products. At the Shanghai Auto Show in April 2019, Huawei debuted as an automotive Tier 1 for the first time and demonstrated a series of automotive digital solutions such as mPower smart electric, including on-board charging systems, motor controllers (MCUs), battery management systems (BMSs), three-in-one electric drive systems, all-in-one electric drive systems, and DC charging modules, aiming to provide automakers with power domain solutions for multi-mode electric drive, charging, and battery management systems.
MPower intelligent electric is Huawei's self-developed three-electric system, which mainly includes BMS battery management system, MCU motor control system, on-board charging system and off-board charging module, providing automakers with multi-form electric drive, charging and battery management system solutions.
In March 2020, Huawei's mPower smart electric product obtained the German Rhine TUV safety certification, indicating that the entire process system of mPower from research and development to production meets the ASIL D standard requirements.
Huawei provides HiCharger DC fast charging modules and on-board chargers (OBCs) for new energy vehicles. DC-DC is a DC conversion of the circuit, which converts the high voltage of the battery pack into a low voltage for use by on-board electronic devices. The on-board charger (OBC) of an electric vehicle refers to a charger fixedly installed on an electric vehicle. When charging, the vehicle battery is charged through the OBC to ensure system safety.
In April 2020, the new generation of HiCharger DC fast charging module was released, entering the charging pile field. The new generation of modules can reach 30kW (domestic version), and a 20kW version will be launched overseas. The two versions are compatible with the same size. Adhering to the design concept of "reliable, efficient, intelligent and low-noise", HUAWEI HiCharger DC fast charging module will effectively solve the pain points of the charging infrastructure industry.
The era of software-defined cars has arrived
The architecture of smart cars is composed of vehicle platform + peripheral hardware + chip platform + system software (operating system) + application algorithm software from bottom to top. With the great changes in the smart connected car industry, the concept of software-defined cars has become a consensus.
The distributed electrical and electronic (E/E) architecture used in traditional automobiles cannot meet the needs of current automobile development due to bottlenecks such as insufficient computing power, insufficient communication bandwidth, and inconvenience for software OTA online upgrades. The upgrade of E/E architecture has become the key to the development of smart cars.
E/E architecture upgrades include hardware, software, and communication architecture upgrades. Tesla has already achieved a central computing platform to control the entire vehicle, while traditional automakers/Tier 1 suppliers cannot do it all at once, so most of them are cross-domain integration solutions (i.e. 3 domains or 5 domains, etc.). The key variables to achieve software-defined cars are: chips + operating system + middleware + application algorithm software + data five core technologies. Whoever can grasp one of these links in the future may achieve an improvement in the status of the automotive industry chain.
Tesla's FSD chip is self-developed and used, leading the development of the industry, and belongs to the independent first level; NVIDIA, the global AI leader in the GPU field, and Mobileye, the automotive AI chip leader backed by Intel, belong to the first array; Huawei Technologies' strong self-built ecosystem belongs to the 1.5 array and is expected to quickly break through into the first array; domestic smart driving AI chip newcomer Horizon Robotics and others are in the second array.
Operating system:
Giants build basic platforms, and Tiers conduct secondary development to create differentiated products, which are the cornerstones of the software ecosystem. As domains gradually take shape, the amount of algorithm software and code that needs to be managed increases exponentially, and it is imperative to create an adaptive operating system.
Technology and Internet giants such as Huawei and Baidu are building a broad operating system basic software platform, aiming to create the "Google Android" of the automotive industry. Tiers are developing differentiated products based on the different secondary development of OEMs' sensors and autonomous driving algorithm solutions.
Software defines cars, and application layer functions are the touchstone. Application algorithm software engineering and integration are ADAS functions or cockpit applications, such as ACC adaptive cruise control, automatic parking and other functions. The algorithm system is mainly composed of three parts: perception fusion, decision planning, and control. Perception algorithm suppliers are relatively mature, and most of these players are sensor suppliers and technology startups.
Decision-making and planning algorithms mainly involve global path planning, behavioral decision-making, motion planning, etc., involving whole vehicle system solutions. Most of these players are car companies/technology Internet/L4 driving startups.
Control algorithms mainly involve the execution end, and most of these players are traditional chassis electronics and car companies. The software layout of car companies is from shallow to deep: software integration, decision planning, perception, and basic software (OS).
Huawei established the Intelligent Vehicle Solutions BU and officially entered the field of intelligent vehicles. Background review
Huawei has two main responsible organizations: ICT Infrastructure Business Management Committee and Consumer Business Management Committee. The Consumer Committee consists of two departments: Consumer BG and Consumer BG Regional Organization, responsible for the strategy and operation management of consumer business.
ICT consists of six departments: Carrier BG, Enterprise BG, Network Products and Solutions, Cloud & AI BG, ICT Regional Organization and Smart Car Solutions BU.
The Automotive BU is part of Huawei's ICT and is led by Huawei's Rotating Chairman Xu Zhijun. The Automotive BU is responsible for the company's end-to-end business in the field of smart cars, extending Huawei's ICT technology advantages to the smart car industry and providing incremental ICT components and solutions.
According to 36Kr, Huawei's Consumer BG is being integrated with the Intelligent Automotive Solutions BU, and the overall person in charge is Huawei's Consumer Business CEO Richard Yu.
The core members of Huawei's Automotive BU carry the genes of hard technology and automotive industry. President Wang Jun previously worked in Huawei's Japan Carrier Business Department and was the president of the FDD product line of Huawei's Wireless Network Business Department. Vice President Zheng Gang was a member of the Standing Committee of the Party Committee of BAIC Group, General Manager and Party Secretary of Beijing New Energy, and was awarded "China's Top Ten Chief Brand Officers".
Another vice president, He Liyang, was previously the head of Huawei's Western European Enterprise Business Department, president of Huawei's global solutions, and president of Huawei's business BG solutions. They all have deep project experience and proficient business capabilities, and together with Chief Technology Officer Cai Jianyong and Product Manager Li Zhenya, they form the leadership backbone of the Automotive BU. In addition to the executive team, the Automotive BU has selected a number of core technical backbones from the automotive ICT business field to form new businesses.
Huawei's profound ICT technology lays a solid foundation
Huawei has accumulated a solid technical foundation in the ICT field, including but not limited to chips, operating systems, machine learning algorithms, cloud technology, sensors, etc., which is fertile ground for cultivating Huawei's automotive business.
Comprehensive chip layout supports Huawei's powerful ecosystem
Huawei has a comprehensive chip layout, and five major types of chips are the foundation of Huawei's ecosystem. Huawei's HiSilicon Semiconductor was established in 2004 and has established a relatively complete chip product system.
HiSilicon chips are mainly divided into five categories in the general field: AI chip Ascend series, cloud computing processor Kunpeng chip, mobile phone SoC chip Kirin series, 5G base station chip Tiangang and 5G baseband chip Balong, and connection chip Lingxiao series.
In the automotive field, Ascend 310 and Ascend 910 are currently used for automotive autonomous driving reasoning and enterprise cloud training respectively, Kunpeng 920 is used as an intelligent driving CPU chip for general computing, Balong 5000 is a 5G communication chip, and Kirin 710A is a SoC for the cockpit domain.
New entrants such as Huawei and Qualcomm are jointly grabbing the market share of traditional automotive chip manufacturers. Cockpit chips are similar to consumer electronics applications, and the functional safety standards are higher than those in the consumer electronics field. Therefore, the computing performance of automotive cockpit chips is generally lower than that of mobile phones, but the reliability and stability are higher than those of mobile phones.
From the perspective of process technology, the chip process in the consumer electronics field has been popularized to 7nm, and some products have reached 5nm, while the chip process of traditional automotive chip manufacturers is still mainly 16nm\28nm, etc. The chip players in the traditional cockpit field are mainly NXP, Renesas, Infineon, TI, etc. Qualcomm, Huawei, etc., as leading companies in the field of mobile phone chips, are constantly seizing the market share of traditional automotive electronics with their advantages such as high chip computing power.
Ascend chip applications in AI computing
Ascend series smart chips provide computing power support for AI applications. According to algorithm classification, AI chips are divided into two parts: cloud training and edge/terminal inference chips. Inference chips are generally used in the edge field and use cloud-trained algorithm models for calculations.
Training chips are used in internal R&D/cloud computing for training algorithm models. Relatively speaking, training chips require higher computing power. From a technical perspective, AI chips are mainly divided into three categories: GPU, FPGA, and ASIC. The industry generally believes that GPU solutions are more versatile, support more algorithms, and have a superior ecosystem; ASIC solutions have better performance-to-power ratio and outstanding performance on a few algorithms; FPGA solutions are somewhere in between.
Huawei first launched the Ascend 310 inference chip in 2018, which can be used in edge computing and the automotive autonomous driving domain controller MDC platform. In addition, Huawei released the Ascend 910 training chip in 2019 for use in the cloud.
Huawei has built an automotive autonomous driving domain controller MDC platform based on the Ascend 310 chip. Ascend 310 is an efficient, flexible, and programmable AI processor. Based on a typical configuration, the performance reaches 16TOPS/INT8, 8 TFLOPS/FP16, and its power consumption is only 8W.
The energy efficiency ratio is higher than the current mainstream autonomous driving NVIDIA Xavier and Mobileye EyeQ4. In 2018, it launched the automotive autonomous driving MDC computing platform and high-level autonomous driving full-stack solutions, including the MDC 300 and MDC 600 platforms corresponding to L3 and L4 autonomous driving respectively.
MDC integrates Huawei's self-developed Host CPU chip, AI chip, ISP chip and SSD control chip, and through the underlying software and hardware integrated tuning, it leads the industry in time synchronization, accurate sensor data processing, multi-node real-time communication, minimized background noise, low power consumption management, fast and safe startup, etc. At present, Huawei has four intelligent driving domain computing platforms: MDC300, MDC600, MDC210, and MDC610.
The automotive AI inference chip landscape is clear and oligopolistic. The AI chip players in the ADAS field are mainly Tesla, Nvidia, Mobileye, Huawei, Horizon, etc. Tesla develops its own chips for its own use and has become a school of its own.
In terms of suppliers of AI chips, NVIDIA and Mobileye are in the absolute first tier. NVIDIA mainly focuses on L2+ and above high-level autonomous driving, and provides chips + basic software platforms (does not provide application software algorithms). Mobileye mainly focuses on the L0-L3 ADAS field, and provides an integrated solution of camera + chip + basic software + application algorithm.
Huawei is in the 1.5th tier because its products have not yet been installed in commercial models, and its mode is similar to Nvidia; Horizon and Mobileye have similar modes and are in the 2nd tier.
In the field of cloud AI chips, NVIDIA is the absolute market leader, and Huawei, Cambrian and others are catching up. In the field of cloud AI chips, NVIDIA is the absolute leader, accounting for 90% of the AI chip market share. This is mainly because NVIDIA has created a series of deep learning SDKs based on its GPU, including Cuda, cuDNN, TensorRT, etc., which lowers the threshold for developers to use GPUs for deep learning training and reasoning, speeds up computing, and it is difficult for other manufacturers to break through its application ecosystem in the short term.
Huawei released the Ascend 910 chip in 2019, which is manufactured using TSMC's 7nm EUV process, has up to 32 cores, and a thermal design power consumption of 350W. Its half-precision floating-point performance is as high as 256TFlops, the core area is 182.4 square millimeters, the computing density exceeds NVIDIA V100 and Google TPU v3, and the overall performance is as high as 512PFlops.
Kunpeng CPU chips are used in general computing
The latest Kunpeng 920 chip has achieved the strongest computing power for general computing, and its performance is better than similar chips from other manufacturers. Based on the ARMv8 instruction set, Kunpeng 920 is the industry's first 7nm data center ARM processor. It uses multiple means such as multi-issue, out-of-order execution, and optimized branch prediction, and is optimized for scenarios such as big data, distributed storage, databases, and cloud services, improving its performance.
Kunpeng 920 has 64 cores, integrated 8-channel DDR4, can provide multiple interfaces, main frequency up to 2.6GHz, total bandwidth 640Gbps. Kunpeng 920 is aimed at data centers, focusing on low power consumption and high performance, and its performance has reached the industry-leading level, especially in integer computing capabilities. The industry standard SPECintBenchmark score exceeds 930, which is 25% higher than the industry benchmark, and the energy efficiency is 30% better than the industry benchmark.
Balong and Tiangang chips are used in the communications field
Huawei's 5G communication chips include the Balong and Tiangang series chips. Balong 5000 is one of the few 5G baseband terminal chips that have been commercially available. Balong 5000 supports both NSA and SA networking modes and is compatible with 2G, 3G, 4G and 5G network standards.
Covering sub-6GHz and mmWave frequency bands, the peak download speeds can reach 4.6Gbps and 7.5Gbps respectively. BYD has announced that its Han model will adopt Huawei's 5G communication module MH5000 with Balong 5000 as the core.
The Tiangang chip is the world's first 5G base station chip, with outstanding performance in terms of integration, computing power, and spectrum bandwidth. The improvement in these three aspects has reduced the size of the base station by more than 50%, reduced the weight by 23%, and reduced the installation time by half compared to 4G. Therefore, Huawei's self-developed 5G base station can achieve multiple advantages such as small size, light weight, and strong performance, surpassing previous 4G base stations and achieving cost compression.
There are currently five 5G baseband chips released on the market, including Unisoc's Chunteng 510, Qualcomm's X50/X55, Huawei's Balong 5000, MediaTek's M70, and Samsung's Exynos Modem 5100. Only Balong 5000 and Qualcomm's X50 and X55 have been commercially used.
Hongmeng operating system, connecting infinite possibilities
Huawei Hongmeng is a distributed OS with a microkernel for all scenarios, which can achieve cross-platform collaboration. Hongmeng is the world's first distributed OS with a microkernel for all scenarios. Its original intention was to enhance the cross-platform capabilities of the operating system, including supporting all scenarios, across multiple devices and platforms, and coping with low latency and high security challenges.
Hongmeng OS has four major features:
distributed architecture, inherent fluency, kernel security and ecological sharing; it has a three-layer architecture: the first layer is the kernel, the second layer is the basic service, and the third layer is the program framework. In 2019, Hongmeng OS 1.0 was first used in smart screen products, and it is planned to be gradually used in more smart devices such as mobile phones, tablets, and cars from 2020.
The Hongmeng system has four major technical characteristics: distributed architecture, inherent fluency, kernel security, and ecological sharing.
1) Distributed architecture ensures system stability:
Hongmeng adopts a distributed architecture to develop cross-terminal distributed applications and ensure the stability of the system. Even if a part of the system fails, it can still continue to run.
2) Latency engine + high-performance IPC, higher communication efficiency: Hongmeng OS solves the problem of insufficient performance of existing systems and improves communication efficiency by using two major technologies: latency engine and high-performance IPC.
3) Microkernel + exokernel design, higher security: Hongmeng system adopts microkernel + exokernel design, in which the microkernel does not require root permissions, and the exokernel services are isolated from each other, thereby improving system security
.
4) Richer development environment and shared ecosystem:
Huawei provides an integrated development environment and
the Ark compiler that supports unified compilation of multiple languages, allowing application developers to significantly improve software development efficiency.
In August 2020, Huawei announced the Hongmeng cockpit operating system HOS, the intelligent driving operating system AOS, the intelligent vehicle control operating system VOS, and the cross-domain integrated software framework Vehicle Stack. The cross-domain integrated software stack (VehicleStack) can realize the interconnection and interoperability of the three operating systems. It is constructed based on the service concept and builds a sustainable profit model for car companies.
Huawei's autonomous driving operating system kernel (including virtualization mechanism) has obtained the industry's highest level of functional safety certification in the safety field (ISO 26262 ASIL-D), becoming the first operating system kernel in my country to obtain ASIL-D certification; at the same time, the kernel obtained the high-level information security certification in the security field (CC EAL 5+) in September 2019, marking that the system kernel has become the industry's first commercial OS kernel with dual high certification of Security & Safety.
The machine learning algorithm is powerful and can achieve fast/accurate/smart
Noah's Ark Lab and Smart Car Cloud Service Product Department are the core support teams for machine learning software algorithms. Huawei Smart Car Cloud Service Product Department and Noah's Ark Lab formed a joint technical team (Noah CV Lab & Octopus) to carry out research in business directions such as automated data identification, sensor fusion algorithms, SLAM/VIO algorithms, intelligent decision-making and reasoning, path planning and motion control, and intelligent transportation system simulation.
HUAWEI Octopus autonomous driving cloud service relies on the latest research results of the joint team and Noah's Ark Laboratory to optimize self-developed algorithms, and the accuracy of multiple algorithm models has reached the industry-leading level.
Huawei selected open source datasets for algorithm verification testing, and verified it on its own datasets to build billion-level data annotation capabilities. The quality of the algorithm is mainly judged by the test results of the dataset. The most important test tasks for autonomous driving include 3D object detection, 2D object detection, semantic segmentation, instance segmentation, scene flow prediction, optical flow prediction, depth estimation, etc. Among them, 3D object detection and 2D object detection are the most core benchmark task scenarios.
3D object detection datasets include Kitti, nuScenes, Lyft dataset, Waymo open dataset, appllo scape, H3D, etc. Among them, nuScenes and Waymo are the most important test sets. The COCO test set is used as the benchmark for 2D object detection.
Huawei
selects the industry's most authoritative open source dataset as the algorithm verification set for testing. By continuously optimizing the algorithm design, the data mining algorithm achieves SOTA performance on the dataset to enhance Huawei's competitiveness in the field of autonomous driving data iteration.
After the open source dataset obtains model verification, Huawei will also verify the closed-loop system of data mining on its own dataset and build high-quality data annotation capabilities at the billion level to meet the data scale requirements of mass production algorithms in commercial environments.
Huawei's machine learning software algorithms are strong. In July 2020, Huawei's Noah's Ark Lab and HUAWEI Octopus Autonomous Driving Cloud Service team Noah CV Lab & Octopus won the first place in the 3D detection track (mAP: 64%, NDS: 69%) in the 2nd autonomous driving dataset 2020 nuScenes Challenge. They were significantly ahead of the second place CenterPoint (UT Austin) by 3.1 mAP and 1.5 NDS percentage points, and exceeded the previous challenge champion model by 11.4 mAP and 5.7 NDS percentage points. As of July 2020, the Noah CV Lab team of Huawei's Noah's Ark Lab has firmly ranked first in COCO BBOX Detection (2D object detection) (the 2020 annual challenge has not yet started), 1 percentage point ahead of the second place.
Cloud services are accelerating
Huawei Cloud's business development has entered the fast lane, with rapid growth in revenue, number of paying users, and infrastructure scale.
Huawei's rapid development is closely related to the growth of its developers. In 2016, Huawei had only 25,000 developers in the cloud and computing fields, but now the number is close to 2 million.
Huawei plans to further expand its scale and launch the "Fertile Soil Plan 2.0" in 2019, planning to invest $1.5 billion in the development of cloud and program developers in the next five years. According to the Canalys report, Huawei accounted for 15.5% of the Chinese public cloud service market in Q2 2020, surpassing Tencent Cloud and Baidu Cloud to rank second, second only to Alibaba Cloud, with a year-on-year growth rate of 259.6%. At present, Huawei Cloud has launched more than 200 cloud services and more than 200 solutions, with an annual transaction volume of more than 1 billion yuan and more than 100,000 orders.
In China, Huawei Cloud has served multiple industries including government, Internet, automobile manufacturing, finance, and genetics, including more than 30 national ministries, more than 600 government and public utilities, 30 of the top 50 Internet companies, more than 20 large automobile companies, and 14 genetics companies.
Positioning as an incremental component supplier and fully cooperating with the automotive industry chain
Car company level
Huawei is positioned as a supplier of incremental automotive components and has developed strategic cooperation with major automakers. By May 2020, Huawei had established a 5G automotive ecosystem with 18 automakers, aiming to accelerate the commercialization of 5G in-vehicle technology in the automotive field. Huawei can provide a full-stack smart car solution, which can be based on the needs of automakers, including chip solutions + operating systems + ADAS algorithm software + cloud services.
The cooperation between car companies and Huawei can be divided into three categories:
Level 1: For car companies with weak software capabilities, Huawei, as Tier 1, provides a complete solution product (chip + operating system + algorithm software + sensor) for car companies to integrate;
Level 2: refers to car companies that have some software algorithms (such as fusion decision-making algorithms). Huawei provides basic chips + basic software (operating system) platform + sensors and perception algorithms, and car companies are responsible for fusion decision-making algorithms.
Level 3: refers to car companies with outstanding software algorithm capabilities (perception fusion + decision-making control algorithm). Huawei provides the basic chip + basic software (operating system) platform, and the car companies are responsible for the entire ADAS algorithm.
Component level
Huawei's smart car includes a "cloud-pipe-end" architecture. The cloud refers to the smart car cloud, the pipe refers to the smart network, and the end extends from the cockpit to smart driving, smart cockpit, and smart electric.
At this stage, Huawei products have not yet been mass-produced. In the next two years, a large number of companies will participate in Huawei's industrial chain and jointly contribute to the independent rise of China's automobile industry.
For full text
Click to follow the blue word "
Smart Car Expert
"
above
Reply "
Huawei
" to receive
2020 LiDAR Conference Event Preview:
(Click on the picture to view conference details)
Conference Background
Zhichexingjia and Yimao Information Technology will hold the 2020 (Second) LiDAR Advanced Technology Exhibition and Exchange Conference on December 15-16, and discuss with more than 30 industry experts on hot topics such as market trends, technological development, automotive-grade manufacturing, sensor fusion, and vehicle-road collaborative applications. They will analyze the most advanced LiDAR technology (including optical devices, light sources, receivers and processing units, etc.), and strive to create a comprehensive cooperation platform for information sharing, experience exchange, technical support and product display.
Click "Read original text" to register for the conference and reserve a booth (free for vehicle manufacturers only)
京公网安备 11010802033920号