Intelligent Vehicle China Program: Chip Ecology Takes Shape | In Depth

New Era, New Rules: Intelligent Transformation in the Automotive Industry

Written by Zheng Wen and Zhou Changxian

Edited by Zhou Changxian

Under the banner of “intelligence”, the automotive industry is undergoing a savage transformation reminiscent of the African savanna.

On the plains of Africa, when animals start running towards the sun, the lion knows it can catch the antelope only if it is faster. Likewise, the antelope knows it has to outrun the lion to avoid being eaten.

Following this most basic of instincts, we embark on a journey. Like the evolution of life forms, the intelligence of cars has evolved from the spineless invertebrates with only distributed neurons to the highly intelligent primates with a huge brain capacity.

In this virgin land of “traditional automobiles,” the old world has already collapsed, and the new world has not yet arrived.

How can we create the new order of “intelligence”?

For most mass-produced cars, the electronic/electrical architecture uses a distributed functional architecture. With each new function added to new cars, a corresponding electronic control unit (ECU) is also added, and these ECUs communicate through the CAN/LIN bus.

A traditional car model has about 70 to 100 separate ECUs. As more and more functions were added, more and more ECUs were crammed in, causing the traditional distributed electronic/electrical architecture to buckle under the load.

The automotive electronic architecture began to seek change, moving from distributed to domain architecture, culminating in the process of central computing + domain control architecture, with the MCU slimming down to several highly powerful integrated chips.

The reduction of ECU burden involves the separation of dozens of ECU modules into software/hardware, which are then centralized into domain controllers. For example, in the era of EE2.0, XPeng Motors has reduced 60% of the number of traditional ECUs in the domain controller.

Consequently, the value of the supply chain is being recombined. The supply chain of the “old world” is long, with suppliers arranged hierarchically as Tier 1, Tier 2, Tier 3, etc. in orderly execution. But now, after a huge variable, the industrial model has been reconstructed, the value points have shifted, and many new modalities have emerged, with the industry chain greatly lengthened and intertwined.

With the advent of the “software-defined car” era, new windows are opening in the impregnable castle…

The Window of Opportunity is Brief but Precious

As a result of the decoupling of hardware and software, computing is becoming more centralized, while connectivity is becoming more dispersed.

The communication, computing, storage, and perception capabilities required for intelligent cars are all based on chips, and the software architecture must be developed based on the functionality of the chip. The software and chip must be tightly integrated, with the chip performing the critical task of bridging the gap between software and hardware.

Thus, high-performance chips have taken center stage, and there is increasing exploration and experimentation around them. Among them, the Chinese market is the hottest.Zhang Hongzhi, Vice President of Horizon, and Head of Business Development for the Intelligent Driving Product Line

Zhang Hongzhi, Vice President of Horizon and head of business development for the intelligent driving product line, told “AutocarMax” that the Chinese market is now the hottest market in the world for intelligentization. All chip companies in the world bring their latest chips to China and find Chinese brand car factories for cooperation.

In a sense, China has become the main battlefield for intelligentization development, like the Olympics or the World Cup. And we are in the center of this market. Zhang Hongzhi said that it is impossible not to be excited.

One possibility is that Chinese local suppliers like Horizon will have important development opportunities.

It is well known that chip technology follows Moore’s Law. On the premise of the same price, the number of components will double and the performance will double every 18-24 months. This terrifying evolution speed means that the chip industry chain can make an advanced technology into an outdated technology every two years. In other words, if you are slow, all previous investments will be wasted.

At the same time, the investment cost of chip design and development is very high. It can be seen that the competitors in this field are in a very difficult situation. So, for quite some time, the main competitors in this field have been international players.

The famous statement from John Chambers, the former CEO of Cisco, was “Fast fish eat slow fish”. He emphasized that the grasp of market opportunities in the Internet economy is different from that in the Industrial Revolution and requires both speed and precision. This analogy is even more appropriate in the chip industry, and it is also a reflection of the automotive intelligentization industry today.

The time window for the chip industry is very narrow. From PCs, to mobiles, to the automotive field, the entire computing platform and operating system were built in a narrow time window. Because, if the speed is slow, it is impossible to master the initiative and opportunity for innovation and value creation.

The biggest problem is that the time window is not directly given to the most advanced technology, but it selects the ecology that develops most rapidly. There will be thousands of applications developed on this, and these may be selected by millions or tens of millions of developers.

The speed of innovation iteration has become a key to winning. Building an ecosystem and being able to constantly iterate and innovate through ecosystem collaboration is a key step for us to win the “World Cup” final at our doorstep.

In such a narrow time window, in the position of the world’s intelligentization innovation center, can the ecosystem built by chips and operating systems be related to us? Horizon, a Chinese chip company is willing to try.”Professional careers for a generation of people lay the foundation for future development. To expedite the iteration process and seize the opportunity, we adopt an ecological approach, uniting industry participants to accelerate development” says Zhang Hongzhi.

As a deeply involved participant in the transformational process, Zhang Qian, the senior manager of product strategic planning at ZF, discovered that many development processes require Tier 1 collaboration, chip factories, and OEMs from the original “turnkey” model to the present state of deep collaborative binding.

This is a window of time that everyone is competing for, whether in chip terminals or the R&D of autonomous driving products. Participants in the industry chain should pursue efficiency. Once this period is missed, it’s lost for everyone.

“Going Solo” is Outdated

When describing the Chinese market, foreign research reports often use the term “fragmented.” Observing externally, China’s industry is scattered and lacks cooperation.

However, this is not entirely accurate. Horizon Robotics, for instance, has a clear positioning of itself as a Tier-2 company, offering chips and tools, software, and some reference algorithms. The company supports partners to achieve original equipment manufacturer status and creates value for customers.

Horizon Robotics’ founder and chief scientist Yu Kai has clearly stated that the core of the company’s business model is “All-Dimensional Altruism” and the core strategy is to create an ecological enterprise. Because altruism can benefit oneself, helping clients is the best way to succeed.

In the mind of Lu Peng, the Senior Director of Marketing and Product Planning at Horizon Robotics’ Intelligent Driving Product Line, it is always clear that hardware (on the SOC chip side) will continue to converge, while software will be diverse. This means that Horizon Robotics has a more pressing sense of competition than other industry participants, and its will to construct an ecosystem will also be stronger.

Only the most efficient platform can support the richest applications. There will be a large number of software ecosystems in the future, but in terms of high computational chips, there will be continued convergence and concentration. Therefore, Horizon Robotics must create a championship ecosystem.

In the past few years, Horizon Robotics’ pace of iteration has been rapid, not only maintaining the speed of releasing a new chip each year but also maintaining the efficiency of producing them on a large scale and in cars within one year of release, which is even more impressive.

“Within a year and a half, we will upgrade our computing architecture one generation.” Zhang Hongzhi further reveals.What’s the use? To improve performance without changing the hardware. At the launch event of Horizon Journey 5 in July last year, the chip’s frame rate data was announced to be 1283. Through optimization of software architecture and software stack, it reached 1531 a year later.

In the ecosystem, Horizon is clear about the only two things it can and must do to the extreme: one is to do a good job with the chip and improve its computational efficiency; the other is to do a good job with open-source tools.

Things are clear, but not simple. A variety of chips, from low-computing communication chips, power chips, and sensors, to medium-to-high computing MCU microcontrollers, and then to high-performance AI chips and main control chips, have a “wall” separating their industrial chains.

High-computing chips are like multi-taskers that can do anything – they have to handle images, graphics, audio, and even deep learning.

It usually consists of central processing units (CPUs), graphics processing units (GPUs), image processing units (ISPs), neural processing units (NPUs), digital signal processors (DSPs), memory (ROM/RAM), and other major units.

The difficulty of high-performance chips for intelligent connected cars is even greater. It has unique functions and performance requirements, and is more similar to consumer electronics chips in terms of process specifications, but also meets car-grade requirements, and has its own unique requirements for customization, specialization, and platformization.

Such a high-dimensional technical challenge is far from enough to evaluate with just one parameter, computing power. What truly tests strength is the comprehensive data processing efficiency of the entire system and the AI processing ability under the coordination of software and hardware.

At Horizon, two indicators are used to judge the quality of a chip: FPS/Watt (FPS divided by power consumption) and FPS/Dollar (FPS divided by money). The latter means how much money is needed to complete algorithm development on this chip.

Luo Shikoku does everything. Zhang Zhihong explained, “This is a very complex process. We need to constantly balance different heterogeneous calculations and their storage. We make it efficient enough by allocating and scheduling computing resources, optimizing the process, and optimizing storage in chip design.”

“The delivery of a chip is not just hardware, but a system.” Huang Ying, CEO of In-Charge Technology, told “100 People Forum/AutocarMax,” “Choosing an AI chip + toolchain vendor that can provide rich, complete, and mass-produced verified AI chips and tools can achieve twice the result with half the effort for our enterprise that focuses on middleware development.”

For a programmable chip, the quality of the toolchain directly affects the chip’s market.The efficient and open Horizon platform, which is built around the core of “chips + toolchains” and the mature development environment, is the key to building an ecosystem. It can help partners complete full-stack function development from hardware to software, from perception to regulation, and reach the stage of real-world vehicle testing in a short period of time.

According to a survey conducted by the Horizon team in the industry, if the software tools and software development efficiency are not sufficient, additional effort equivalent to 300 people working for two years may be required to implement high-level autonomous driving on a high-performance chip. This is just the cost of a car manufacturer to create a vehicle model.

YU Kai firmly believes that building a thriving software ecosystem is the true goal for which they should strive. They believe that the ultimate meaning of competition in the chip industry is also reflected in the software ecosystem, and whoever has better software will come out on top.

In order to be as open as possible, “no bottom line” is the policy of Horizon. They even overturned the four modes of cooperation between chip suppliers and customers in the industry:

  • The full-stack black-box mode is not open and is not adopted by Horizon;
  • Providing computing chips and toolchains for users to develop on top of them. This is NVIDIA’s conventional delivery method, which is adopted by Horizon;
  • TogetherOS mode, which opens the DSP underlying software to partners, speeds up their iteration speed, and is still a black box for NVIDIA, but is the main cooperation mode for Horizon currently;
  • Chip cooperation. Horizon can even provide support for its partners to create high-performance chips.

On July 29, 2021, Horizon finally made a breakthrough and launched the third-generation Horizon Journey 5 AI chip, becoming the only provider in the industry that can cover full-scene smart vehicle chip solutions from L2 to L4.

With its deep technical accumulation and mass production experience in the field of hardware and software, as well as richly verified AI tools development, Horizon Journey 5 is the only large-capacity AI chip in the world that can be mass-produced at a level of 100 TOPS (trillion operations per second).

In 2020, Horizon only had one customer, Changan. By 2021, its main factory partners had reached more than 20.

At the Horizon Journey 5 launch event, Horizon announced its intention to cooperate with a group of manufacturers, including SAIC Group, Great Wall Motors, Jiangling Motors, Changan Automobile, BYD, NIO, and Landtourer Automotive, on the mass production of the Horizon Journey 5 chip. Moreover, it also formed a joint venture with the Continental Group and reached a pre-research cooperation agreement based on Horizon Journey 5 with the Ideal Auto.

It is worth mentioning that more than half of the leaders of China’s independent brands have endorsed Horizon, highlighting its prominence.In terms of chip cooperation, the Ideal Automotive team’s happiness has increased significantly after switching from Mobileye to Horizon Robotics, to the point that founder and CEO Li Xiang exclaimed that Horizon Robotics is the most cooperative and professional team they have ever seen.

“They have a large number of teams working directly with us. When we are doing various perception tests and calibrations, whether it’s in the Changshu factory, on the test track, or on the controlled roads, they have professional engineers working with us. This is a huge difference in value compared to working with chip manufacturers in the past.”

In various stages of the intelligent automotive industry chain, the ecosystem based on chips is continuously demonstrating performance, and it is even more pleasing that Chinese companies represented by Horizon Robotics and its ecosystem partners are beginning to light up domestic solutions one after another.

The Growth of the “Chinese Solution” Ecosystem

By the summer of 2022, everything had fallen into place. Horizon Robotics’ hardware and software ecosystem partners have successively launched high-level autonomous driving products and solutions based on the Journey 5 platform, and have begun to fully meet the differentiated mass production needs of automakers for intelligent driving in various scenarios.

For example, Light Sail Intelligent Navigation has created a cost-effective high-level autonomous driving solution for high-speed and urban NOA; Jianzhi Robotics has launched a mass production solution for high-level autonomous driving and autonomous driving perception system; Juefei Technology has released an urban commuting fused positioning mass production solution; and Tianzhun Corporation has released a standardized reference solution for domestic domain controller mass production.

Yu Kai revealed that there are more than a hundred ecosystem partners on Horizon Robotics’ Tian Gong Kai Wu ®AI chip toolchain. With Horizon Robotics’ empowerment, they are flourishing and jointly creating China’s solutions for intelligent automotive.

However, there are still many difficult problems to solve, such as:

As chips continue to evolve, computing power continues to increase. Can the software architecture be designed to shield the changes in hardware, so that the impact of hardware changes on upper-layer software development is minimized?

Different car models may have different operating systems and connect to different clouds. Can the resulting changes not affect the parts that have already been developed, and minimize the impact on functionality?

The software iteration cycle is short. Can the architecture support continuous new requirements for OTA upgrades?

These are all questions that companies like Neusoft Reach need to answer when developing middleware. They are reminding the industry that a flexible, stable, and scalable middleware support platform is needed.

Although the widely used AUTOSAR architecture has standardized some functional modules, improved software reuse, and defined methodologies and systems, it can only solve some basic foundational issues and is limited in expanding future functionalities for large-scale intelligent cars, such as cross-core fusion and vehicle-to-cloud interaction.

Hu Jun, Deputy General Manager of Neusoft Reach's Autonomous Driving Business UnitActually, NeuSAR solution launched by Neusoft Reach aims to solve the above-mentioned problems in response to the AUTOSAR standard. NeuSAR provides standard basic software, middleware ASF and development tools, and provides a vehicle software development platform and corresponding tool chain, so that OEMs can quickly enter the application development phase.

However, with the advent of the software-defined automotive era, in the development process of a product, the traditional relatively closed Tier-1 and Tier-2 industry chains have been unable to meet the requirements, and instead require a more flattened, networked, and complex new cooperation model of “you have me and I have you”. This is precisely why Neusoft Reach and Horizon have been working together.

In the words of Hu Jun, Vice President of Neusoft Reach’s Autonomous Driving Business Unit, both parties have an attitude of high openness in their project cooperation with Horizon. The entire software deployment or implementation of the software stack is a state of “you have me and I have you”, a very three-dimensional and in-depth cooperation model.

Based on this cooperation model, Neusoft Reach and Horizon have conducted deep comprehensive cooperation and jointly created a series of products.

For example, based on JAC iEVS4, Neusoft Reach has built an automatic driving central computing platform that supports cross-domain integration of vehicle functions such as intelligent cockpit and intelligent driving, and has comprehensively opened up the development and mass production application chain of China’s autonomous driving industry’s self-developed chips, algorithms, software, and hardware.

In addition, based on Horizon’s Journey 5 series and Journey 3 series AI chips, the fourth-generation automatic driving domain controller X-Box launched by Neusoft Reach is a brand-new L2+ level domain controller standard product, which has L2+ level driving and parking functions and will achieve SOP next year.

Undoubtedly, after the scale effect is produced, the further acceleration of iteration and mass production speed will follow.

Take the cooperation between Qingzhou Zhixing and Horizon as an example. On May 18, the two sides reached a strategic cooperation. Within just three months, Qingzhou Zhixing launched a highly cost-effective high-level automatic driving solution.

It is expected that in the third quarter of 2022, the automatic driving sample car based on Journey 5 chip by both parties will carry out road tests. As of now, the Qingzhou Zhixing City NOA plan equipped with the Journey 5 chip has landed, and it is expected to open for trial rides in October 2022. By 2023, Qingzhou Zhixing’s high-level automatic driving solution based on Journey 5 chip will reach mass production level with the support of Horizon.”Horizon’s efficient and easy-to-use development toolchain has been the main reason for achieving significant landing results in such a short time, greatly reducing production time and saving a lot of manpower and material costs,” said Yu Qian, co-founder and CEO of QZ Intelligent Navigation.

During the process of developing a low-cost solution for high-level autonomous driving, the QZ Intelligent Navigation team found that the Journey 5 chip platform had excellent portability, especially in the perception algorithm for complex environments. For this feature, another partner of Horizon, Jianzhi Robotics team, had a deeper understanding. From newcomers to development, it only took 7 days, and the algorithm migration for the autonomous driving solution was completed in only two months.

It seems that the efficiency brought by the Horizon ecological strategy has already shown concretely in the marketization process and has affected some potential partners who suffer from “choice paralysis.”

Yang Jiaying, the product and project director of ZJTech, revealed, “Before we partnered with Horizon, we ourselves were very cautious about choosing high-performance chips and had also selected other domestic and foreign chip manufacturers. In comparison, Horizon’s toolchain maturity, software, and local support services are better.”

It is understood that ZJTech will develop navigation assistance and parking-assisted hybrid function products based on the journey 5 for the MAXIPILOT® high-end platform solution for passenger car companies.

At the same time, with new roles beginning to appear in the Horizon’s ecological system, the gap in the industry chain in subdivision fields is further filled.

On June 29, Horizon announced three official authorized hardware IDH (Independent Design House) partners: Shanghai Yingchi Technology Co., Ltd., Shanghai Jinmai Electronics Technology Co., Ltd., and Suzhou Tianhuai Technology Co., Ltd.

As Tier1.5, three partners will provide support for custom design needs for different Tier 1 based on the Journey 5, with rich software and hardware development experience and capabilities to accelerate the mass production of intelligent driving solutions for customers.

In just a month and a half, Tianzhun Technology has given the “reference answer” TADC-D52 domain controller, based on the dual Journey 5, which is a standard solution for domestic domain controller mass production, and has become the first high-integration, flexible, and open high-capacity domestic domain controller standard solution for mass production in the industry.

“In the new era of intelligent automobiles, we not only have technology brands like NIO, Xpeng, and Ideal, but more importantly, we need Chinese solutions. This is something we feel very necessary,” said Chen Liming, president of Horizon, moved by the achievements of ecological partners.

Conclusion:Although some fruits have been produced in various links of the industry chain, industry insiders generally believe that the era of intelligent automobiles has not yet truly begun.

Yu Kai frankly stated that intelligent driving is still in its infancy stage and has only achieved 10%. There is still 90% of work left to do, with a lot of things that can be done.

“From the perspective of competition patterns, we are still in the Spring and Autumn Period, with many thriving ecosystems.” Yang Jiaying predicted that “one day in the future, we may transition to the Warring States Period, where the main theme will be to unite and cooperate.”

“If there is an opportunity to promote the progress of the industry little by little in this complex industry with many years of accumulation, I think it is very remarkable.” Cao Bin said, “We are full of expectations and confidence.”

In this unprecedented era of opportunity, not only Yu Kai, Cao Bin, Yu Qian, and Yang Jiaying, but also more ecological participants will receive timely responses from history.

And the most important thing for these “Warring States warlords” right now is to cooperate, speed up innovation, meet user needs, and then form scale effects, taking the lead in the intelligent competition.

Therefore, the importance of the industry ecology is more prominent.

In James Carse’s book “Finite and Infinite Games,” he described two types of “games” in the world: finite games, whose purpose is to win; and infinite games, which aim to keep the game going forever. Finite games are played within boundaries, and infinite games play the boundaries, the purpose is to bring more people into the game itself, and to continue the game.

Yu Kai believes that this statement is very accurate. He believes that Horizon is building an ecology to allow more people to participate, to jointly expand the pie. In fact, it is an infinite game.

We firmly believe that under the participation and co-construction of ecological participants such as Horizon, the Chinese forces on the automobile industry chain will unite and play to their advantages, further promoting the full integration of software and related advanced technologies with automotive intelligent chips, and contributing to the high-speed development of China’s intelligent driving industry.

This article is a translation by ChatGPT of a Chinese report from 42HOW. If you have any questions about it, please email bd@42how.com.