Produced by | Huxiu Auto Group
Author | Wang Xiaoyu
html On September 22, Huxiu learned that the autonomous driving computing chip company Heisema Intelligent announced that it has completed two rounds of strategic and C round financing in the near future. After the investment, the strategic and C rounds of financing are worth nearly US$2 billion (approximately RMB 12.9 billion), and Hei Sesame Intelligent officially entered the ranks of "unicorns".strategic round is invested by Xiaomi Yangtze Industry Fund and Fusai Automobile and other leading domestic industry companies; the C round of financing is led by Xiaomi Yangtze Industry Fund, followed by Wentai Battle Investment, Wu Yuefeng Capital, Tianji Capital, Yuanhe Puhua, Lenovo Venture Capital , Linxin Capital, China Automobile Chip Industry Innovation Strategic Alliance, etc.
This is also the first investment in the upstream core chip link of upstream of the car after Xiaomi announced its car manufacturing.
This round of financing will provide financial support for the company's next generation high-performance large computing power autonomous driving platform, the company's business development, and the talent team. In addition, the C+ round of financing is currently progressing smoothly.
, founded in 2016, is undoubtedly a lucky person. Against the backdrop of "domestic substitution", it has entered the "unicorn" camp in the chip field, and was also selected by Xiaomi, which has a car-making halo. Everything seems so smooth, but there must be a lot of efforts and attempts behind it. Domestic chips are rising.
1. Xiaomi builds cars, chips start with
If it was a fuel car more than ten years ago, Xiaomi might invest in engine and gearbox manufacturers. However, overseas companies such as Japan Aisin, , Germany ZF , the United States Allison, and Canada Magna all occupied nearly 90% of the automatic transmission market share until 2017.
is now rebuilding pure electric vehicles. Xiaomi is no longer concerned with the "old three items", no longer mechanical parameters such as horsepower and torque. It is the range and computing power determined by power batteries and high computing power computing chips.
Starting from June this year, Xiaomi launched the "buy, buy, buy" model, investing in many companies in the fields of power batteries and autonomous driving. In the previous few attempts, Xiaomi invested in advanced assisted driving system algorithm solutions. This time, investing in Heisema Intelligence is a layout in the development of automotive-grade autonomous driving computing chips and platforms that are deeply rooted in the underlying layout.
Black Sesame Intelligent Technology, named after Black Sesame + Sesame Open Door, is an automotive-grade autonomous driving computing chip and platform research and development enterprise. It was established in 2016. The biggest feature of its business is both "soft and hard".
Because compared with other autonomous driving chip companies on the market, Heizhima Intelligent not only provides high computing power computing chips, but also provides complete autonomous driving and vehicle-road collaboration solutions.
You should know that today's smart electric vehicles are undergoing the development of distributed architecture towards domain control and centralized architecture. Traditional distributed hardware architectures face the needs of multi-dimensional perception and the needs of massive unstructured data processing. Generally, each new application function is added, corresponding perception, decision-making, and execution are added.
, a new car-making force represented by , Tesla , adopts a centralized architecture in the automotive electronic and electrical architecture, that is, using a "computer to control the entire vehicle". Under the minimalist interior, there is a Tesla self-developed FSD chip, which supports the FSD system to achieve the efficiency of 2,500 searches every 1.5 milliseconds, predict various possible situations, and find the safest, most comfortable and fastest autonomous driving paths.
Faced with the ever-evolving autonomous driving technology, Tesla's approach has never changed - chips that require higher computing power. But Tesla's FSD chips have also taken many detours, from the earliest purchase of chips from Mobileye EyeQ3 and Nvidia DRIVE PX2 to the ultimate choice of self-developed development path.
Mobileye EyeQ series chips have been in a "monopoly position" in the ADAS industry for a long time. Although it provides car companies with integrated chip + algorithm software solutions, EyeQ chips based on the "black box mode" make car companies unable to perform more functional iterations. The closed model has always been criticized by the industry. NIO and Ideal Auto are both in the next generation of products, and choose to abandon the Mobileye solution.
So, domestic car companies are actually re-taking Tesla's old path, but the rapid rise of domestic chip manufacturers can accelerate the transformation from "outside procurement" to "internal supply". The most typical example is that as China's first independent brand, , FAW Group, , chose Heisema Intelligent in cooperation with autonomous driving platforms.
From the end of 2019 to the end of 2020, in just one year, FAW Group and its subsidiaries signed three cooperation contracts with Heizhima Intelligent. The two parties have carried out all-round cooperation in the fields of autonomous driving chips, visual perception algorithms and data, and even intelligent driving platforms.
This strategic financing was jointly established by FAW Group, Fuao Automobile and Desai Xiwei to participate in the investment. Heizhima Intelligent cooperates with FAW to support L3 driving and L4 parking automatic driving functions, and will be applied to Hongqi flagship SUV models.
In addition, several popular brands in the current new energy vehicle market, SAIC and Dongfeng Yuexiang are both customers of Heisha Smart.
According to the calculations of the Dongwu Securities Research Institute, the AI chip market size will reach US$9.2 billion by 2025, with a CAGR of 45.0%, and will reach US$18.1 billion by 2030, with a 10-year compound growth rate of 28.8%. It is foreseeable that with the explosion of the smart car market, the demand for AI chips will be pushed to a new high.
2. computing power is both the strength
and the CEO of NIO. Li Bin. once mentioned a point of view: "Horizontal plus computing power is the new standard for defining high-end smart electric vehicles."
It is true that today may still be a long way from full autonomous driving, but the strength reserves for computing power are imminent. The competition for computing power is a bit like the competition between engine power and torque of fuel vehicles in previous years - you can not use it, but you can't do it.
The original distributed architecture of traditional cars can generally realize low-level assisted driving. Since there is relatively little sensor information to be processed, the use of MCU chips can meet the computing requirements.
With the arrival of high-level intelligent driving, it needs to process a larger amount of unstructured data such as pictures and videos. Relying on traditional MCU chips alone cannot meet the exponential growth computing needs. At this time, the AI chip installation can be quickly, accurately and accurately calculated.
For example, the amount of data generated by L3 level autonomous driving is 2.3GB/s, and the computing power requirement is above 129TOPS; the data volume of L4 level autonomous driving is 8GB/s, and the computing power requirement is above 448TOPS. If you consider the functionally secure redundant backup of , the computing power requirement may be doubled.
When car companies have a benchmark high-level autonomous driving capability, chip computing power becomes the primary indicator.
NIO's new flagship model ET7 is equipped with 4 Nvidia Orin chips, which is said to have a computing power of 1016TOPS. But in fact, there are only two for autonomous driving calculations and decision-making, one for redundancy and the other for training neural network models. The actual computing power used during autonomous driving is 762TOPS.
Zhiji Auto currently uses Nvidia Xavier chip, with a computing power of between 30-60TOPS, and supports a sensor layout solution that supports camera + radar sensing. Next, Zhiji will also upgrade the chip to multiple Nvidia Orin X chips, and it is publicly disclosed that its computing power is between 500-1000+TOPS.
Huawei 's pioneering work in the autonomous driving industry, BAIC ARCFOX Alpha S Huawei HI version is equipped with Huawei's customized computing platform MDC810, with a computing power of 400+TOPS. Another representative of the domestic core of
, Hei Sesame Intelligent released Huashan No. 2 A1000 Pro in April 2021. Hei Sesame Intelligent released Huashan No. 2 A1000 Pro. In July of the same year, the successful filming of the film was achieved, which means that large-scale production can begin. The chip
adopts a 16nm process, with a computing power of 106TOPS in INT8, and a computing power of 196TOPS in INT4, with a typical power consumption of 25W, which also means that the overall energy efficiency is as high as 8TOPS/W than that of . In terms of functional applications,
can support high-level autonomous driving including automatic parking and urban road to highway scenarios. In addition, this chip also supports multiple chip cascades to form a more powerful computing power platform.
Looking at the current industry, the computing power competition has begun.The computing power of Mobileye EyeQ5 is 24TOPS, Nvidia Xavier is 30TOPS, Nvidia Orin's high computing power version Orin X is 200TOPS, Huawei MDC is 48-160TOPS, and Tesla FSD is 144TOPS. Judging from the numbers alone, Heizhima Intelligent's chip computing power level is indeed not inferior to that of foreign-funded enterprises.
"The chip computing power continues to hit new highs, which is largely due to the car companies' thinking about the new business model."
Hei Sesame Intelligent Technology CEO Shan Jizhang once said in an interview: "Car companies have now produced a new model, that is, after hardware pre-embedding in the car, they can make money through software upgrades. Car companies do not know how much computing power is needed in the future, but they can first pre-embed the computing power and then upgrade its functions."
For example, NIO ET7 and Zhiji L7 are both sold for more than 400,000 yuan, which is equivalent to the C-class car level of traditional luxury car companies. For new car manufacturers whose brand background and manufacturing foundation do not yet constitute barriers, they can make breakthroughs based on the technological attributes and service ecology of their products. But for smart cars, what is more important is that they can be upgraded and evolved. The research and development of
autonomous driving technology will never be achieved overnight. On the basis of hardware, we need to pay more attention to the development ecosystem, supporting software and tool chain construction.
In order to cooperate with high computing chips, Heizhima Intelligent has opened an artificial intelligence development platform called "Shan Hai". The open platform has more than 50 AI reference model library conversion use cases, which can effectively help car company customers lower the threshold for algorithm development. In other words, Heizhima Intelligent can provide an overall solution of chip + algorithm according to the needs of car companies, and car companies can also write their own algorithms on the chip.
In addition, Heizhima Intelligent also cooperates with sensor manufacturers and algorithm manufacturers including Paulon, Sotorian, and Newe Technology to create a standardized FAD reference platform that can be deployed quickly and help improve the system development efficiency of car manufacturers and autonomous driving enterprise customers.
Finally, for vehicle-road collaboration scenario applications, Heizhima Intelligent also opened the vehicle-road collaborative roadside computing platform: FAD Edge. This platform can sink the computing in the cloud to the edge layer, and complete most of the calculations at the edge computing node to meet the ultra-low latency requirements of vehicle-road collaboration.
was written at the end
In the PC era and the mobile phone era, when applications and software are both hardware first on the eve of large-scale development, because the iteration cycle of hardware is long and the iteration cycle of software is short. Therefore, the premise of rapid iteration and expansion of functional performance of software is to first prepare the performance and computing power of hardware first.
The stage that the automotive industry is now entering is precisely: the automotive industry is beginning to become electronic, or even consumer electronic trend.
As Yang Yuxin, the CMO of Heisha Intelligent, said, the first step for car companies is to leave enough redundancy in hardware and computing power to leave enough space for software algorithms and innovation. This is why everyone now requires computing power first, and the current requirements for computing power in the market are faster than iterations imagined.
In other words, the leap in computing power is to help us open up the pattern.
This content is the author's independent view and does not represent the position of Huxiu. Reproduction is not allowed without permission. Please contact the authorization matters. [email protected]
. Those who are changing and wanting to change the world are all in Huxiu APP
