What's missing in 2021? For the car, it must be a chip.
At CES 2022, the three major chip companies – Intel's Mobileye, Qualcomm and Nvidia – became leaders in locking in the brains of self-driving cars for the next decade.
Smart cars give AI chips only 2 to 3 years of opportunities, and there is not much time for domestic chips to break through.
The chip still has to hold the thigh of the car company
The automatic driving chip has been on the car, and in the comparison of the computing power of the new car not long ago, NVIDIA OrinX, Qualcomm Snapdragon Ride, and Huawei MDC have taken the lead and opened the mass production situation of AI chips.
Nowadays, with the continuous improvement of the perception accuracy, control accuracy and response speed requirements of automatic driving, the demand for large computing power main control chips will rise sharply, the investment of car companies in chips will become larger, and the relationship with technology companies will become closer and more delicate.
Danny Shapiro, Nvidia's vice president of automotive business, said: "Automakers who are no longer pioneers are finally realizing that if they don't change their thinking, they will be left behind forever." ”
At CES 2022, NVIDIA announced that it will provide several Chinese EV startups with the electronic brains of the cars of the future, while also partnering with other automakers, including Mercedes-Benz, Hyundai, Volvo and Audi.
Nvidia partners with start-up car companies
Nvidia is the darling of the current automotive autonomous driving industry, and the OrinX chip will appear on new models such as THEO ET7, Xiaopeng G7, and Zhiji L7. In addition, cars such as Jidu, WM, and Feifan will also choose to use OrinX to build intelligent electric vehicles. OrinX's single chip hash rate of 254TOPS and can be calculated 254 trillion times per second.
At the same time, Qualcomm has also demonstrated its ability to cooperate with car companies with quantity, and will carry out new cooperation with 37 car companies this year.
Qualcomm cooperates with car companies
But even as the automotive value chain appears to be tilting toward tech companies, chip companies have to agree to give automakers control over key parts of smart cars in order to win cooperation.
Qualcomm, for example, bought Veoneer last year for $4.5 billion to perfect all the software needed for its self-driving chips. But after announcing at this year's CES show that it won the first large-scale self-driving chip supply contract with GM, these software assets are not included, because GM will develop its own software.
An autonomous driving platform developed jointly by GM and Qualcomm
"Our software stacks are all developed in-house, so we won't use their parts," said Jason Ditman, principal engineer at GM's upcoming "Ultra Cruise" off-the-beaten-path product.
But for other automakers, Qualcomm needs to have all parts of its self-driving system ready, said Nakul Duggal, senior vice president and general manager of automotive at chip companies.
"Different automakers find themselves at different stages of preparation," he said, "and the key for automakers is that they have to be able to build relationships with the customers they're trying to acquire." ”
In 2021, Horizon's "Journey 3" chip can catch the express train of the ideal ONE, and its efficient and low-cost communication costs are indispensable. The project started to the mass production model in just about 7 months, shortening the feedback time and reducing the cost of communication, which will be a huge advantage in the cooperation of domestic chips in the future.
The realization of autonomous driving requires a lot of technology. These include computer vision algorithms that help cameras identify pedestrians, high-precision maps of the world's roads that are expanding, and "driving strategy" software for making millisecond-level decisions about a car's behavior when it comes to unexpected situations.
For chipmakers, this means they need to have every aspect of the technology ready, but at the same time be willing to let customers pick and choose.
Technology and data are sensitive
Chip computing power is the base of smart cars, and algorithms are the embodiment of real performance.
Whether it is traditional car manufacturers such as GM, Ford and Volvo, or new energy forces such as Weilai, Xiaopeng and Ideal, they are deepening their ties with partners in key technologies and emphasizing that they must grasp the key to core data.
Especially in terms of intelligent electrification, Tesla and Apple carry out full-stack self-research of chips and core technologies. "The same is true for the new car-making forces, they must control their own destiny and take the initiative to constantly iterate data with users." Horizon founder Yu Kai said.
In this case, the issue of technology attribution for automakers and tech companies becomes complicated.
Danny Shapiro says control of technology and data is a sensitive strip of land between automakers and tech companies. "Whether it is a generalization plan or a customized plan, whose hands should the data be in?"
In fact, Tesla had a close relationship with Mobileye before self-development, and Mobileye was the first to develop chips on autopilot. However, Mobileye has packaged its vision sensors, chips and autonomous driving software into an integrated solution for partners in the past, and this black-box solution has gradually deviated from the track of car companies and eventually been abandoned.
Li Xiang once said, "We originally used Mobileye's black box algorithm, and we couldn't touch the core algorithm, which is an area that suppliers are not open to." ”
Recently, Mobileye has changed its approach, starting to separate some of the features of its packaging scheme and allowing Ford to build its own technology on top of those features. This change is reflected in the cooperation with Ford.
"We give All the output to Ford, and they're going to run their own algorithms on top of our output," Mobileye CEO Amnon Shashua told Reuters.
It seems that in this temptation in this sensitive area, the supplier will have to make some concessions.
"Chip companies have very little choice and can only become more flexible because they also have to face important competitors from their peers." Phil Amsrud, senior principal analyst at IHS Markit, said.
When it comes to controlling simpler semiconductors such as engines, automakers used to rely on only three major suppliers – Infineon, Renesas and NXP. However, the market for chip companies that provide high-performance computing for automakers is relatively crowded, including companies such as Huawei and Amba Semiconductor, which have also entered the automotive industry, as well as upstarts such as Horizon and Black Sesame Intelligence, which are born for AI car chips.
Black Sesame Intelligence mentioned in the latest speech, "We have been building an open ecosystem, because the core business logic of the chip is scale, how to increase the amount, you need to really open the system, can support different customers to push different scenarios." ”
Huashan 2 A1000 Pro
This year, black sesame intelligent large computing chip will be really mass production on the car. Huashan No. 2 A1000 Pro hash maximum computing power of up to 106TOPS, which realizes the complete coupling of software and hardware on the basis of hardware, and the core components and software modules can be customized and replaced according to customer needs. "We have found that at this stage of development, automakers are increasingly taking on the role of technology in leading the way forward." Yang Yuxin, CMO of Black Sesame Intelligence, said.
Who will be "leftover"?
"We're in a phase where we have too many suppliers," Says Armsrud, "and if we look at the automotive industry from a traditional perspective, there will be a handful of suppliers left in the future." ”
Although the share of domestic chips in the traditional automotive chip market is less than 3%, and most of them are in the middle and low end. However, the intelligent era gives domestic chips the opportunity to challenge, but this window period does not allow local companies to have a little slack.
Yu Kai once said that the next three years is the most critical time window, if Chinese brands in the chip and operating system can not get the top two in the Chinese intelligent car market, basically out. By 2023, the overall situation is initially determined, and once the software ecology based on foreign chips is formed, it will be difficult to pull back.
In this short period of about three years, there have been intelligent driving products developed by Huawei, Horizon, Black Sesame Intelligence, and Xinchi Technology.
The first local chip company Horizon occupied the first-mover advantage, and last year it signed a contract with SAIC, BYD, Nezha and many other OEMs to prepare for the impact of the global market of smart chips.
Black sesame intelligent mass production is one step later, but the ability of large computing power and complete tool chain is the highlight. Guosen Securities pointed out that the demand for computing power for high-end autonomous driving has increased exponentially. L2 level of automatic driving requires > 10TOPS hashrate, L3 level requires 100~200TOPS hashing power, L4 level requires 400~600TOPS hashrate, L5 level requires at least > 3000TOPS hashrate.
Huawei's self-driving computing platform MDC810 has been installed on the BAIC Jihu, with a computing power of 400+ TOPS. Wang Jun said that the core of smart cars is computing power, and Huawei's advantage lies in its control over large-scale integrated software projects, and the accumulation of software talents and capabilities will play a role in the car.
Che Yun summary
In the face of new opportunities, China's new energy vehicle market has the ability to spawn giant-level local chip companies, and the window period of domestic chips is only a short period of two or three years.
In 1999, Xu Guanhua, a former minister of science and technology, proposed that the situation of "lack of core and little soul" in China's information industry was expected to be completely reversed in twenty years.
(Wang Ruihao's compilation also contributed to this article)