The smart car chip war has started again.
This time, the protagonists are Mobileye and Ambarah, the former is currently the leader in the ADAS assisted driving market share, but in the high-end intelligent driving track, temporarily lagging behind NVIDIA; the latter in the past few years from edge products (including driving recorders, cabin interaction and commercial vehicle ADAS) to the high computing power market.
Today and in the future of the smart automotive industry, chips are becoming the top brand for car companies to market product technology for the C-end. New forces in China's automotive chips such as Qualcomm, NVIDIA and even Horizon are increasingly appearing on the big screen of the new car conference.
On the one hand, the revenue of chip manufacturers from the automotive industry continues to rise. According to the data, the automotive industry contributed nearly $1 billion to Qualcomm's revenue in fiscal 2021, and although its share of overall revenue is still low (less than 5%), the upward momentum has been clear.
"Chips will play a key role in the autonomous driving revolution in the automotive industry." NVIDIA CEO Jen-Hsun Huang predicted. At present, the company has won orders for high-end smart cars from Mercedes-Benz, Volvo, Weilai, Xiaopeng, Ideal, Zhiji and other brands.
One
On January 4, Mobileye unveiled its most advanced and powerful EyeQ Ultra system integration chip for autonomous driving to date at CES. Optimized for energy efficiency at 176 TOPS, the EyeQ Ultra is a streamlined chip for autonomous vehicles.
Built on the proven experience of the 7th Generation EyeQ chip architecture, this energy-efficient system integration chip delivers the performance and power consumption needed for all autonomous vehicles, especially all-electric autonomous vehicles. However, until the actual delivery and launch, it is necessary to fully realize the mass production of the vehicle specification level by 2025.
"Achieving an excellent ratio of performance and cost," has always been the product philosophy of the ADAS driver assistance market leader, and the EyeQ Ultra achieves an excellent balance of performance on different accelerators and general-purpose processors in the extremely high energy efficiency range.
The EyeQ Ultra's performance is equivalent to the sum of the performance of 10 EyeQ 5s, and with the help of a 5nm process, it can meet all the needs and application scenarios of L4 autonomous driving, while avoiding the additional energy consumption and cost of combining multiple system integration chips.
Like its predecessor, the EyeQ Ultra was co-designed with Mobileye software, enabling the chip to achieve extremely high energy efficiency ratios without sacrificing performance. This is also a new track for the current focus of NVIDIA, Qualcomm and other companies, starting from software to improve the efficiency of chips.
EyeQ Ultra features a set of designs with four proprietary accelerators, each designed for a specific task. These accelerators work with other CPUs, ISPs, and GPUs to form an energy-efficient solution. According to Mobileye's official data, the EyeQ Ultra has a hash rate of only 176 TOPS, but its energy efficiency ratio is much higher than other self-driving car solutions.
For example, the solution can simultaneously process input data from two sensing subsystems (including a pure camera subsystem and a subsystem that integrates radar and lidar) as well as the vehicle's central computing system, high-definition maps, and driving decision software, providing the performance required for consumer-grade autonomous vehicles at a reasonable cost.
In addition, in order to continue to "retain" the ADAS market share, two new EyeQ system integration chips for ADAS solutions are launched at the same time as EyeQ Ultra, EyeQ 6L and EyeQ 6H, based on 7nm process processes and featuring cost-effective pure camera solutions.
Among them, EyeQ 6L is the successor to EyeQ 4, and its package size is only 55% of ThatQ 4. This all-in-one windshield solution delivers higher deep learning hash rate (TOPS) at ultra-low power for energy-efficient entry-level and high-end (L2) ADAS. Samples of the chip began last year and mass production is expected in mid-2023.
The EyeQ 6H enables high-end ADAS and partial autopilot with a full-surround view camera configuration. In terms of force, it is equivalent to two EyeQ 5 system integration chips. But more importantly, the EyeQ 6H supports visualization and performs better at heavy AI workloads.
The centralized solution will provide all L2+ADAS capabilities, multi-camera processing capabilities including parking cameras, and the ability to support third-party applications such as visual parking and driver monitoring. Samples of the integrated chip for ADAS-compatible systems will begin this year and are expected to be in mass production by the end of 2024.
Considering that the new chinese forces such as Weilai, Ideal, and Xiaopeng have switched to the NVIDIA platform, Great Wall has also chosen qualcomm solutions on the next-generation platform, and Extreme Kr has become Mobileye's new position in the Chinese market.
On January 4, Mobileye and Extreme Krypton announced that they will collaborate to develop a new consumer-grade pure electric vehicle with L4 autonomous driving capabilities, equipped with six EyeQ 5s, equipped with a redundant perception system, RSS driving strategy and map crowdsourcing (REM) technology.
At the same time, Mobileye announced that it will strengthen its R&D capabilities in China, establish a local data center and expand Mobileye's China team to support the fast-growing business in China. According to the plan, after the launch of the Krypton 001, there will be a series of different models of the Krypton equipped with mobileye solutions.
Two
Unlike Mobileye, which continues to consolidate the coverage of the entire product line from L2 to L4, Ambarella (hereinafter referred to as "Aba") chose to target the ultra-high computing power track and released the latest AI domain controller chip CV3 series, with AI processing performance up to 500 eTOPS, which is 42 times higher than the previous generation of vehicle-grade SoC CV2 series.
Based on the fully scalable, energy-efficient CVflow architecture, the CV3 is equipped with up to 16 Arm Cortex-A78AE CPU cores, which is 30 times higher than the previous generation chip CV2. At the same time, CV3 integrates multiple sensors through a single chip for centralized AI perception processing, deep integration of multiple sensors, and path planning of autonomous vehicles.
Among them, we can pay attention to the advantages of Ambarco in visual ISP and millimeter-wave radar calculation units, for example, CV3 can run the 4D imaging radar AI algorithm of Auco (acquired by Ambarco at the end of last year), plus a new generation of ISPs, to achieve a higher level of environmental perception accuracy under complex environmental conditions.
The CV3 can support more than 20 cameras connected through MIPI VC at the same time, and can handle a full set of sensors with a single chip, adapt to typical L2+ sensor configurations, and provide a high-performance binocular stereo vision engine and dense optical flow engine to meet the needs of depth detection and motion perception.
As Ambarella's next-generation CVflow architecture, the new floating-point general-purpose vector processor (GVP) complements the Neural Vector Processor (NVP) to help the NVP engine share traditional machine vision and radar processing, as well as the Arm CPU to share floating-point intensive algorithms.
Similar to NVIDIA and Qualcomm's strategy, in addition to covering the entire autonomous driving domain (from ADAS assisted driving system to L4 automatic driving system), CV3 can also handle in-cabin perception at the same time, including driver and in-vehicle perception system.
The advantage of this model is to help car companies reduce software development costs.
Amba's approach is that automakers no longer need to develop different software stacks for their entry-level, mid-range and high-end models, because customers can use Amba's unified CVflow platform in all models, saving engineering costs while being able to react faster to market changes.
At the same time, the GPU unit under this architecture meets the rendering of 3D surround view (which is a key component of the multi-camera visual perception system), realizes isolation and security software updates for different domains through hardware security modules, is equipped with a high-speed INTERFACE for ultra-low latency communication, and can meet the software stack test verification similar to Tesla's shadow mode.
In the Chinese market, the MAXIPILOT 1.0 L2 intelligent cruise system based on the Ambarco CV2x series has been delivered to passenger car customers for large-scale mass production, and the upgraded L2+ function is expected to be mass-produced in 2022.
At the same time, Ambarco has mass-produced and delivered a variety of models of in-cabin visual interaction, driving recorder, panoramic surround view and other front-loading solutions landed, the cumulative market shipments of more than 300,000 units. With the start of offering CV3 samples earlier this year, Ambarco is one of the new players in the high-power SoC track and could bring white-hot competition for market prices.
According to the calculation of the Gaogong Intelligent Automobile Research Institute, by 2023, the first phase of the market pattern of high-computing chips that meet the large-scale mass production car market will initially take shape. Correspondingly, this year will be the golden time for chip manufacturers to compete for the pre-installation fixed point.
According to the judgment of Qualcomm CEO Cristiano Amon, the revenue potential of each smart car equipped with a chip will increase by more than ten times in the future, especially in the Chinese market, the rapid rise in the intelligent loading rate not only gives market participants enough distributable "cake", but also tests the speed of each company.
Three
The situation of you chasing after me has become the norm in the smart car chip track. Giving all other car companies the ability to PK with Tesla products and technologies is the goal of these chipmakers.
Also during this CES, Qualcomm released the new Snapdragon Ride Vision system, based on the 4nm SoC and equipped with a new software stack from Arriver (Qualcomm's acquisition process in progress), providing enhanced perception of chipsets with custom neural network architectures. The mass production delivery time for this platform is 2024.
Open, customizable systems are Qualcomm's clear strategy.
According to Qualcomm, Snapdragon Ride focuses on customers' broad autonomous driving needs, from scalable SoCs to integrated AD stacks, development platforms and tools to provide a comprehensive solution for L2-L3 level autonomous driving.
At the same time, combined with Qualcomm's inherent advantages in cockpit and networking, Snapdragon Digital Chassis has become a unique combination of Qualcomm's comprehensive entry into the intelligent car track.
The total order volume of more than $13 billion is the figure given by Qualcomm. This week, Volvo Cars became the latest signed customer of Qualcomm's digital cockpit, the latter's upcoming all-electric SUV and Polestarr3 SUV will be equipped with Qualcomm digital cockpit + Google Android operating system, and the new car is expected to be unveiled before the end of this year.
Up to now, Qualcomm has signed cooperation agreements with more than 40 OEMs such as BMW, GM, Hyundai, Jidu, Xiaopeng, Weilai, Volvo, Honda, Renault, etc., to integrate different Snapdragon digital cockpit platforms. According to data released by the Gaogong Intelligent Vehicle Research Institute, as of November this year, Qualcomm cockpit SoC has occupied more than 80% of the front-loading mass production share in China's high-end intelligent car market.
Among them, Desay SV is Qualcomm's latest Tier1 partner for the 4th generation Snapdragon cockpit platform, the former is currently one of the four suppliers in the Chinese market with more than 100,000 cockpit domain controller frontloads. According to the plan, the two parties will jointly build the 4th generation intelligent cockpit system of Desay SV based on the 4th generation Snapdragon cockpit platform.
Supported by the leading computing, AI, graphics and multimedia performance of the 4th Generation Snapdragon Cockpit Platform, Desay SV's 4th Generation Intelligent Cockpit System supports leading technologies such as multi-screen linkage, sound processing and AR, bringing users a rich immersive interactive experience.
Prior to this, Desay SV was the first Chinese supplier to mass-produce and deliver a Qualcomm Snapdragon 820A four-screen interactive digital cockpit based on the Ideal ONE model. At present, Desay SV's qualcomm 8155-based smart cockpit domain controller has received mass production orders from a number of car companies and is entering the delivery cycle.
Baidu's Jidu is the first car manufacturer in China to announce the use of the 4th generation Snapdragon cockpit platform, and the first mass production car of Jidu in 2023 will be officially unveiled, equipped with 5nm Qualcomm SA8295P, the 3D rendering performance of the GPU has a 3 times performance improvement compared with the current mainstream 8155 chip in the market, and its AI computing power reaches 30TOPS.
But for Qualcomm, autonomous driving is the "hard bone".
Just two months ago, Nvidia announced at its Fall 2021 GTC conference that Hyperion 8, a production-ready (reference design) platform that includes sensors, computing and software needed for autonomous vehicle development, will deliver to customers starting in 2024.
This end-to-end autonomous driving platform supports 12 cameras, 9 millimeter wave radar, 12 ultrasonic sensors and 1 lidar, which is almost the typical sensor configuration of the mainstream high-end intelligent driving mass production platform.
NVIDIA's goal is to gain enough market share on both high-end intelligent driving (for the private market) and autonomous driving (for the mobility and logistics markets) to provide customers with a sufficiently open end-to-end platform.
In addition, the company also targets the main software platform solution for the large-scale mass production car market, in which DRIVE Concierge is for in-vehicle artificial intelligence assistants and DRIVE Chauffeur is for autonomous driving, and cross-domain integration between the two.
In China, the three new forces of ideal, Weilai and Xiaopeng have all completed the switching of the intelligent driving core computing platform (all using the NVIDIA Orin solution), which is NVIDIA's stage victory in the intelligent electric vehicle track.
Judging from the current mass production plan, Chinese car companies will lead overseas car companies in 2-3 years to complete the replacement of high-end intelligent driving computing platforms. Luxury brands, including Mercedes-Benz and BMW, should change platforms around 2024-2025.
From the perspective of data estimates, similar to Qualcomm's absolute leading share in the field of intelligent cockpits, it is difficult to appear in the field of intelligent driving in the short term. Among them, Chinese companies such as Horizon, Xinchi Technology, and Black Sesame Intelligence are seizing the market.
But there is no doubt that whether it is a digital cockpit or intelligent driving, including high-end automatic driving, the competition of automotive chip companies has been determined, and small steps and fast running and big dry fast, than performance and fighting cost will be the main theme.