Smart cars continue to spread and conquer the traditional automobile market, and driving behavior accelerates AI. According to the National Development and Reform Commission, China's smart car penetration rate will reach 82% in 2025, and the number will reach 28 million. It will reach 95% in 2030, or about 38 million vehicles.
Most of the new cars produced this year and next year have been equipped with more advanced driver assistance functions, and behind the support of intelligence is a fierce computing power chase.
This year, the milestone event of the smart car was born: the super 100TOPS hashrate chip was put on the car. Weilai ET7, Zhiji L7 and many other smart models are equipped with bai topS level chip Nvidia Orin X.
While Chinese car companies are taking the lead in global intelligence, domestic chip companies are also competing for the international arena of underlying technology. Recently, Horizon announced that a number of fixed-point car companies of its TOPS chip "Journey 5", including domestic new energy sales no.1 car company BYD and new car-making forces self-Youjia Car, will start mass production in 2023. In terms of computing power, Journey 5 is the only chip in China that breaks the hundred T and can be mass-produced against Nvidia Orin.
Wang Chuanfu, Chairman and President of BYD Co., Ltd. (left), and Yu Kai, Founder and Chief Scientist of Horizon (right)
With the official announcement of the journey 5, the curtain of mass production of domestic TOPS-level large-scale computing chips has officially opened, opening the pattern of "two strong competitions" with NVIDIA.
The challenge of "computing at scale"
Since last year, "big computing power on the car" has become a high-frequency vocabulary in the industry, but "just smell its sound and do not see its things", this year Nvidia Orin chip was officially released, the slogan became a reality, and car companies once raised the large computing power to an important position.
With the development of autonomous driving to a higher level, application scenarios have become more complex from high-speed to urban areas. At the same time, the speed and precision of perception, prediction and regulation are required. On the one hand, this has higher requirements for software algorithm capabilities, while on the other hand, this means to perform larger-scale calculations, while also burying sufficient computing power for subsequent software iterations. Before the emergence of top 100 chips, the computing power of a single chip on the market was difficult to meet the needs of high-level automatic driving.
Li Bo, head of Lotus's intelligent driving business, took smart phones as an example and talked about the importance of chip computing power reservation: the current mobile phone can meet the current functional needs, but it is difficult to meet the functions after a certain period of time. It is foreseeable that intelligent driving obviously has five years, ten or even longer of development, considering that the future intelligent driving car needs to continue to upgrade the latest features, and it is necessary to leave enough computing power space.
High-end intelligent driving has released a huge demand signal for large-scale computing chips, but the mass production of large-scale car rules chips still needs to tackle many key technologies. For example, in terms of chip technology, there is a need for advanced packaging technology, core IP technology, and breakthroughs in high-computing chip architecture; in terms of vehicle certification, a series of test licenses such as functional safety process, functional safety product certification, vehicle specification reliability certification, and ASPICE software certification are required.
In addition, the mass production process of large computing power chips takes a long time and is also a major difficulty. It takes 18 to 24 months from the processor design tape-out, after which it takes 12 to 18 months to develop the vehicle specification certification system program, and then to the model introduction test and verification at least 12 months to 24 months, and finally mass production and deployment, the whole process lasts three to five years.
Time and technology make it more difficult for large computing chips to get on the car, but keep the cloud open to see the moon, and this year's large computing chips for high-level automatic driving ushered in mass production.
In March this year, NVIDIA's DRIVE Orin chip was officially launched, and the Top 100 CHIP landed. Coincidentally, on April 21, Horizon and BYD's official announcement broke the tranquility of NVIDIA's solitary position. Its Journey 5 chip will be mass-produced on BYD models next year, which is also a TOPS-class chip. On April 28, Horizon once again announced a fixed-point cooperation with Ziyoujia Automobile on Journey 5, which will be equipped with many models of Journey 5 chips and start mass production and delivery next year.
Journey 5 is the third generation of Horizon vehicle specification products, a single chip AI computing power of up to 128 TOPS, and has obtained the globally recognized automotive functional safety standard - T V ISO 26262 ASIL-B functional safety certification, is the first domestic automotive grade AI chip based on the ISO 26262 functional safety process development. As early as last year, Horizon Journey 5 has successfully streamed the film, and was officially released in July, announcing on the spot that it has obtained the first mass production cooperation intention of a number of car companies.
The development cycle of smart driving chips is generally long, but high-energy players can shorten the time limit for chips to get on the car. Horizon said that this is because it has a strong software algorithm gene, which can run the understanding of terminal applications through the entire chip design and mass production cycle, which not only improves the performance of the chip itself, but also accelerates the efficiency of mass production verification.
In the past, Horizon released a chip every year, and the following year it could achieve mass production loading. Today, Horizon "Journey 5" is about to be mass-produced. At the same time, the "Journey 6" designed for the automatic driving architecture of vehicles above L4 is already on the road, and this rhythm responds to the phrase "the world's martial arts, only fast is not broken".
Although the development time of Horizon established in 2015 is not as good as that of players such as NVIDIA and Qualcomm, on the new track of large-scale automatic driving chips, Horizon can develop AI chips that benchmark international top players in a few years with the prediction and accumulation of forward-looking technologies, and impact the pattern of the automotive chip industry dominated by foreign companies.
Why Journey 5?
The beginning is the idea, the end is mass production. If there is only one inspection standard for the intelligence of automobiles, it is mass production.
The earliest realization of vehicle-grade AI chip front-loading mass production in China is Horizon. In 2020, Horizon Journey 2 opened the first year of mass production, and last year, the Deployment of The Journey 3 chip to the 2021 Ideal ONE set off a wave of climax, while reaching cooperation with many car companies and Tier1 and continuing to become popular.
When Zhengcheng 5 announced the mass production time, car companies also ate reassuring pills, and domestic large computing power chips finally had the ability to land.
In fact, not long before BYD officially announced the adoption of Journey 5, it also established cooperation with Nvidia, and ATD will be equipped with DRIVE Orin chips and DRIVE Hyperion platforms on some models in early 2023. Not only BYD, but also many OEMs have both Horizon and NVIDIA in their chip whitelists. Car companies with huge pressure on transformation costs are willing to vote on the horizon with real money and silver. Behind this, what kind of strength does horizon have?
Taking the flagship products orin X and Journey 5 that have recently been mass-produced by both sides, the single computing power is above 100TOPS. The hash rate of The Orin X is higher than that of Journey 5, but in terms of measured energy efficiency, Journey 5 is better.
The horizon indicates that this is due to the forward-looking anticipation and painstaking practice of the new Moore's Law.
The classic "Moore's Law" is that the number of transistors that can fit on an integrated circuit doubles approximately every 18 months, that is, the computational performance doubles every 18 months. Now, whether it is a mobile phone or a computer, performance is gradually becoming saturated, and in the case of slowing performance growth, the new Moore's Law is born: that is, the real AI performance is the actual processing speed (FPS) than the power consumption per watt, or the actual processing speed (FPS) is proportional to the cost per unit.
Through the self-developed Bayesian BPU computing architecture and the integration of the latest neural network algorithms, the 128TOPS computing power chip can complete the task with only 30 watts of computing power consumption, which is necessary for the cost control of electric vehicles.
On the typical Backbone 1080P model, the AI performance of the Journey 5 is comparable to that of Orin-X, but the energy efficiency is 330% of that of TheIn-X; under the preferred high-efficiency model, the performance of the Hardware and Software Combination 5 exceeds that of Orin-X, and the energy efficiency is as high as 870% of that of Orin-X.
Horizon has proposed the "post-Moore's Law" since 2016, relying on the understanding of software algorithms to continuously invest and improve in the efficiency of AI algorithms. In the future, even if the hardware has been determined, the processing efficiency can be continuously improved through compiler optimization and optimization of software algorithms.
High-computing chips have opened the era of mass production, and NVIDIA and Horizon have taken a step ahead. However, for how to turn what you have into what everyone can have, horizon strength is definitely not lost to NVIDIA. Yu Yinan, vice president of Horizon and general manager of autonomous driving products, said, "Chip + underlying algorithm empowerment is the key to the rapid landing of intelligent driving, we help customers get on the horse, and finally let the horse run faster and faster." ”
Tier 2's mission: to be the top playmate of car companies
Chip, the "Everest" of contemporary hard technology, needs to be looked up to and conquered.
Under the new industrial structure changes, the relationship between automobile OEMs and supply chain enterprises has become more and more delicate, and the debate of "soul theory" has always been a cloud on the head of car companies. For car companies that are eager to master the definition of products independently, the supplier of full-stack delivery is not a good supplier, but the chip manufacturer who only sells hardware is also not.
There is no doubt that what car companies need most is chip suppliers who can provide hardware foundations and open platforms. Therefore, in addition to technical strength, products that can obtain mass production points must also have the open ability of platforms and tool chains.
Unlike Mobileye's closure, both NVIDIA and Horizon employ an open delivery model. Most of the cooperation between NVIDIA and domestic car companies is mainly to provide chip hardware, and it is also equipped with software stack tools. Horizon has positioned itself in Tier 2 from the beginning, not only to car companies, but also to Tier1 to provide the underlying chips and algorithms, but also to provide an open toolchain to help partners build their own application software capabilities, while Horizon also built an open source real-time operating system Together OS, and tried to open up BPU IP licensing to some automakers. In these ways, when it comes to openness, Horizon is better than NVIDIA.
At the same time, compared with Nvidia, Horizon has joined forces with a number of car companies and Tier 1 to achieve mass production of application software algorithms, covering the fields of automatic driving and intelligent cockpits. It is reported that Horizon's urban autonomous driving plan based on Journey 5 will also be announced within the year. This means that Horizon's tool chain has been fully mass-proven and is more efficient and easy to use.
There is also a clear phenomenon this year, whether it is NVIDIA or Mobileye, Qualcomm or Horizon, the first production models of its most advanced intelligent driving chips are All Chinese independent brands, and the center of the intelligent automotive industry is moving to the Chinese market.
Domestic intelligent electric vehicle companies have risen together to drive the growth of new enterprises in the industrial chain, and in turn, local suppliers can also increase service efficiency for car companies. I remember when Horizon and Ideal Car were working together, Horizon engineers said that every day was overtime and mingling with Ideal employees. I have to admit that in the service system, in the face of domestic car companies, Horizon has an innate advantage.
First, the cost of communication has declined, horizon is usually stationed team and car companies to polish products together, compared to suppliers on the other side of the ocean can only rely on a few emails, a few engineers to solve the problem on the spot, face-to-face communication can directly understand each other's needs and ideas.
Second, the corporate culture is common. Whether it is a development concept or a sense of service, local partners often have more commonalities and it is easier to run into. Horizon employees once said, "Everyone regards the customer's business as their own business, and thinks that if the customer's business is successful, we will succeed." I remember that when working with the ideal, a project manager postponed the wedding date for 6 months in order to keep the mass production time of the ideal project, and the ideal team was also very moved, and it was more like a team for a long time. ”
Based on the Tier2 concept, excellent R & D strength and local service advantages, Horizon Journey series chips have shipped millions of pieces and obtained more than 50 front-loading mass production projects. On the road to mass production of domestic TOPS large computing chip opened by Journey 5, in addition to BYD and Ziyoujia, more fixed-point cooperation will be announced within the year.
The significance of these cooperation is not only that the large computing chip can get on the car, but also means that for the smart car track, China has finally got rid of the weak situation in the pc and smart phone era in terms of underlying technology, and has the strength to compete with the top players in the world, which is the indispensable digital cornerstone for winning the race.