Written by: Roomy
Edited by: Chaoxian Zhou
“Thirty years in the east of the river, thirty years in the west of the river.”
This old Chinese saying accurately reflects the rise of Japan’s semiconductor industry and the fall of the American semiconductor industry. At its peak, Japan’s semiconductor industry had a market share of 50.3%, while the American semiconductor industry was in decline. Intel once gave up resistance and withdrew from the memory industry to switch to processor production due to the pressure of Japanese competition. Silicon Valley elites quickly went to Japan for inspections and found that the Japanese were too fierce in research and development, and the quality of the storage chip technology far exceeded that of Silicon Valley.
The lesson here is clear – “technology” is the Sky Piercer that splits Silicon Valley.
The rest of the story is well-known. Today, Japan’s global semiconductor market share is only 10%, and the Ministry of Economy, Trade and Industry predicts that by 2030, the production share will drop to 0.
The reasons for Japan’s decline are varied, but “technology” is the most important two words. As the Bell Lab invented semiconductor point-contact transistors, the Silicon Valley civilization era began. Meanwhile, Japan’s semiconductor industry was transitioning to the professional wafer manufacturing industry, but was extremely slow in doing so.
The old story has come to an end, but a new story has emerged with new protagonists – technology companies in China and the United States.
Over the past thirty years, from computers to smartphones, huge successful technology companies have been born, such as Microsoft, Intel, Qualcomm, and Apple… The stories are exciting, but Chinese technology companies lack a sense of presence.
Today, without a doubt, the third-generation smart terminal is smart cars. What will happen as the era changes?
The United States is seeking answers and Tesla, the leader, has provided the answer. Japan is also seeking answers and the Ministry of Economy, Trade and Industry has drafted a ten-year semiconductor revitalization strategy. However, everyone knows that the answer to this battle will ultimately take place in China.
Within this vast land, cars are replacing mobile phones and becoming the largest segmented market of the semiconductor industry. Can Chinese technology companies make breakthroughs in the fields of smart chips and operating systems, and once again perform a new thirty-year story?
Japan’s thirty-year loss in the semiconductor industry is enough to serve as a warning to all technology companies. Horizon founder and Chief Scientist, Kai Yu, has a good saying – “If you want a future, you must have today – Seeing Jinzhou needs to be done, and so does Tashan.”
He believes that in this smart electric vehicle revolution, ultimate thinking is critical. “Sometimes the next two or three steps are not easy to see, but if you extend the time scale, the future is still quite clear.”
As a global leader in edge artificial intelligence chips, Horizon has never been limited to the chip products themselves since its birth.
After 7 years of foundation building, from Journey 2, Journey 3 to Journey 5, Horizon has produced three car-grade chips with a shipment volume breaking 1 million pieces and won more than 40 fixed-point installations by car models. So far, Horizon has given an answer to the “breakthrough from 0 to 1”. But this is just what Yu Kai called “Jinzhou”.
He believes that Horizon needs to take a higher perspective to answer the question of “Can we make a difference?”
As we can see from the story of Silicon Valley in the US, the innovation ecosystem is the key to winning the future. In the era of innovation, speed is extremely important and opening collaboration requires opening up the boundaries in the middle.
This is the trip to Tashan of Horizon to create an ecology of “open and win-win from 1 to N.”
“Openness and collaboration are always the most important”
“To do something that no one else can do in the market.”
People familiar with Yu Kai have a unanimous evaluation of him as “easy-going,” but his style of doing things is bold and decisive. This is his personality and also a characteristic of Horizon.
In the movie “The Battle of Jangjin Lake,” Wu Qianli said to his younger brother: “If an egg is picked open from the outside, it is destined to be eaten. If you can peck it open from the inside, it is likely to be an eagle.”
Nowadays, the competition in the field of chips can be summarized in two sentences: China is becoming the “arena” of the world’s top automotive intelligent chips; China has performed well in the terminal product brand, but chips and OS still require breakthroughs.
This is the breakthrough point of the industry. Yu Kai hopes that Chinese chip companies can all break through from inside and become eagles.
“In the era of intelligent cars, intelligent chips and operating systems are the most important core technologies and the digital foundation of the entire innovation ecosystem.” Yu Kai defines the importance of chips and OS systems in this way.
Nowadays, whether it is Nvidia, Mobileye, or Qualcomm, the most advanced intelligent driving chips are not first mass-produced by Japanese car companies or German car companies, but by independent brands.
“China has become a source of innovation, and we are playing at home.” Yu Kai is clear about the advantages, disadvantages, and breakthrough points. However, he believes that there is no “passing on the bend” at home.
In his view, hardware has always been the strength of American companies. Intel and Qualcomm have established overall advantages in the processor and chip fields that other companies have not yet surpassed.
In the early days of entrepreneurship, Yu Kai believed that it was essential to think about where the base was. “We must face the objective gap between Chinese companies and international giants in hardware. However, outside the range of international giants, Chinese companies still have opportunities.”
Despite this, Horizon has set long-term goals: to become the world champion.
Yu Kai combined two groups of people, one mainly engaged in artificial intelligence algorithms, including deep neural network algorithms and other technical aspects, and the other mainly engaged in chips and software and hardware systems, embedding intelligence into everything in a low-cost way.
Horizon is like a powerful eagle, and it zealously pecks open the technological cracks of the automotive chip industry.
From 2019 to 2021, Horizon maintained a rhythm of releasing a generation of high-performance autonomous driving chips every year, from Journey 2 to Journey 3 to Journey 5. The computing power continues to increase, and each generation of chips was quickly mass-produced.
Currently, Horizon is the only Chinese company that has achieved mass production of automotive-level intelligent chips. The Journey series of chips has shipped more than one million pieces and signed more than 45 front-loading production projects with more than 17 car companies. Horizon has successfully completed multiple mass production project deliveries.
Regarding the construction of the ecological system, Horizon always takes a dual-track approach. It not only provides chips and optional mass-production-grade algorithm assistance to partners to accelerate the production and landing of intelligent driving applications, but also provides development tool support to partners for efficient development. It can also open mass-production-grade algorithms and provide chip OEM support.
“In the era of innovation, openness and collaboration are always the most important.” This is Yu Kai’s persistence and also Horizon’s mission.
To build an ecosystem, the middle boundaries need to be opened up
Making chips is only the first step, making a system and an ecosystem is the complete chain.
As for which point is the most challenging, Yu Kai said, “Choosing the path.”
As the saying goes: Don’t be exquisitely wrong; be roughly correct. Horizon deeply understands that even if the path is correct, if the foundation is not strong, the house built on top of it will not be stable. Therefore, Horizon has spent seven years researching and developing, landing, and has built a solid foundation.
Especially with the launch of the Journey 5 chip, its high computing power exceeds 100T, mainly benchmarked against Nvidia’s Orin high-level autonomous driving chip. This is a landmark product. Horizon, which previously placed its foothold “outside the range”, has stepped onto the stage to compete with world giants such as Nvidia and Qualcomm.
What is the competition among these giants like?In the chip product industry, stability and advanced technology are the two most important factors. The chips that support advanced level autonomous driving in the market mainly include NVIDIA’s Orin-X, Qualcomm’s SA9000A/B, and Huawei’s Ascend 610. In terms of speed and performance, Horizon Robotics is in the top tier.
However, most chip suppliers do not want to stop at the chip itself but gradually extend to the software layer to build an industrial ecology.
“Chips play the most upstream role in the entire information industry ecology, so they are naturally participants in the ecology,” said Yu Kai.
He summarized various cooperation models between chip companies and host factories.
The Mobileye mode, also known as the “Black Box faction.” After Mobileye develops the chip architecture, chips, operating system, and intelligent driving software and hardware systems, it delivers everything to the automobile manufacturer. In other words, this is equivalent to a black box, and the automobile factory has no participation in the development process of the autonomous driving technology.
This cooperative method is clearly not enough. Because the development of the operating system and autonomous driving software system requires time, and coupled with the time required for delivery to the entire vehicle development, the overall innovation cycle will take 6 to 7 years. This will seriously slow down the innovation speed of the vehicle manufacturer.
Mobileye’s dilemma is how to balance the conflict between meeting the personalized needs and economies of scale of different car companies. Therefore, Mobileye is not favored by car companies with self-research dreams, and the most typical representative is Tesla.
NVIDIA, known as the “computing power faction,” maintains an overwhelming share in the independent GPU market. NVIDIA uses “software and hardware decoupling” as its banner, develops GPU architecture into chips, and then wraps it with the operating system CUDA to allow car companies to develop autonomous driving software and hardware systems.
Compared to Mobileye, NVIDIA’s advantage is not only higher computing power but also the openness brought by “software and hardware decoupling.” Since most car companies are unable to have self-research capabilities like Tesla, deep algorithm cultivation has become a “compulsory course” for NVIDIA.The third mode is called TogetherOS mode. After the development of BPU and SoC, the underlying software in the middle is put together with the entire vehicle through an open-source cooperative mode. This includes not only the entire autonomous driving application software, but also deep into the underlying operating system, which can more efficiently call various resources under the operating system.
The fourth mode is the BPU authorization mode. The highly coordinated development of the entire vehicle system from chips to operating systems and then to autonomous driving software and hardware will greatly improve iteration speed. It is equally competitive with Tesla.
This mode is called the “Differentiation Empowerment Group.” What is the starting point of this mode?
Horizon Robotics is very clear: algorithms are only logical things, and software, hardware, and even custom-designed special chips are also needed. Therefore, Horizon Robotics iterates on artificial intelligence chips with a technical path of one generation per year – Gauss architecture, Bernoulli architecture, and Bayesian architecture.
From a shipping volume of over one million pieces to Well Journey 5 chips, which are known for being efficient, open, and safe, it can be seen that Horizon Robotics’ current product positioning can balance both computing power and algorithms and is more open.
The collaboration between Horizon Robotics and Ideal Automobiles confirms this. Within seven to eight months from project initiation to mass production, Ideal Automobiles created the industry’s strongest AEB, independently developed the NOA function in the shortest time, and realized the world’s first high-level auxiliary driving plan based on an 8-megapixel camera.
Behind this record, Yu Kai believes that the most important thing is openness and collaboration. “Promoting technological innovation in the entire industry should not keep the baby to oneself.” Last July, Horizon Robotics and Ideal Automobiles also reached a cooperation agreement based on Well Journey 5.
Yu Kai’s goal is clear: find a balance point between Horizon Robotics and the entire vehicle factory, and make Chinese smart chips break through as soon as possible in the arena.
Based on the above practice and thinking, Horizon Robotics has proposed a new mode that supports the development of self-developed chips by some automakers – the BPU authorization mode – providing software toolkits, chip reference designs, and technical support to help automakers develop their own chips.
“In the era of innovation, speed is extremely important, and open collaboration means opening up the middle boundary.” Yu Kai sees this very clearly and, in cooperation with the entire vehicle factory, has made the “middle boundary” more systematic.
“Willing to be that eagle”
Yu Kai once predicted that the competition for car-grade chips will end by 2023.Thus, the window of opportunity is limited. Horizon’s strategy must be clear – simply supporting car companies in making cars is not enough. Horizon must open its platform and support automakers in developing their own chips. By doing so, it also solves the issue of who holds the “soul” of the intelligent electric era.
Although new car makers such as NIO and XPeng are building fully self-developed self-driving capabilities, it is difficult for traditional car companies to self-develop chips. The establishment of software departments by Volkswagen and other car companies with massive investments provides a glimpse of this possibility.
Based on this, Horizon has decided to “open up, and open up again.” The plan is to open up the underlying technology platform in multiple dimensions to different ecological partners, accelerating the development of smart automotive applications and the production of intelligent cars. “Horizon does not produce hardware for mass production, does not bundle software, and does not provide closed solutions” – this is the creed it follows.
Starting last year, Horizon announced the creation of an open-source real-time operating system, TogetherOS.余凯 has clearly stated that “the operating system should always be open source and open to everyone from the very beginning.” On the Horizon 5 chip journey, Linux, QNX, Android, AliOS, HOS and other mainstream operating systems are supported.
However, the future business model is centered around the chip + algorithm + toolchain + development platform, where competition will inevitably be concentrated on ecological environments. That is to say, to win this battle, the strongest forces must be gathered to form a special force and attack team.
At the Internet Information Office Symposium, the General Secretary also made it clear that “there is a significant difference between us and international advanced levels and core technologies. A prominent reason for this is that backbone enterprises have not formed a synergy like Microsoft, Intel, Google, and Apple.” The General Secretary specifically mentioned the ATI alliance in the United States, where Microsoft’s Windows operating system is paired with Intel’s chip.
余凯 believes that this is a very detailed observation and that historically, there is no separation between chips and OS. For example, Microsoft or Android may have been at the forefront of technology, but in the end, they were actually ahead in terms of the application ecology. With so many users and developers worldwide, there is a digital foundation on which to innovate and develop various application software.
Throughout the history of the entire information industry, the most successful operating systems with the broadest ecology tend to choose a chip architecture for tight alliances. For example, the “AA” alliance during the era of smartphones, which was the alliance between ARM and Android; the AI era, where NVIDIA’s own GPU architecture is paired with its own operating system, CUDA.This is the ecological system that needs to be improved in the automotive chip industry, and Yu Kai is willing to take Horizon Robotics to do this. “We should use extremely realistic methods to strive for extremely idealistic results,” Yu Kai said.
In fact, NVIDIA’s operating system CUDA is coupled with NVIDIA’s own GPU architecture, and has formed a preliminary ecological alliance for smart electric vehicles. This is a dangerous signal for Chinese chip companies.
Therefore, Horizon Robotics decided to run on two tracks: on the one hand, it will develop Journey, the strongest autonomous driving chip in China, and on the other hand, it will not only open-source its OS, but also provide some car manufacturers with BPU IP authorization, and strive to build the “Wintel” (Windows+Intel) and “AA” of the intelligent car era.
Horizon Robotics provides BPU’s IP authorization, supports some host manufacturers to develop their own chips, and of course also offers Horizon Robotics’ own Journey chip, because not all host manufacturers should develop their own chips. Yu Kai believes that “the benefit of BPU IP authorization is to enhance the differentiation competitiveness of host manufacturers and speed up the pace of research and innovation.”
The window of opportunity for competition is precious, but open platforms are also a double-edged sword. However, Horizon Robotics, which is “always on the survival line, 10 years, 20 years, and 30 years are all on the survival line,” is willing to be that eagle.
Correct direction, then talk about strength
In a previous interview, a media outlet commented on Yu Kai, “You seem to have entered a complex world from a simple world.”
Yu Kai agrees and disagrees.
He agrees that as an innovative enterprise, Horizon Robotics considers how to think uniquely rather than follow the crowd. “In the post-Moore’s Law era, how do we innovate? When others are talking about how many T’s of computing power they have, we should think about unique and innovative differentiated strategies. This is the new Moore’s Law for the AI era that we have proposed.”
But he disagrees that Horizon Robotics’ own positioning has not changed. Chen Liming, the president of Horizon Robotics, said, “Horizon Robotics has always positioned itself as a Tier 2 enabler in the industry, hoping to use the Journey chip as the cornerstone to build the overall ecosystem of the intelligent automotive industry.”
Horizon Robotics’ positioning principle remains unchanged, standing on the edge of the spotlight, not in the center.
Yu Kai once said that he takes both a top-down and a bottom-up approach to move forward, constantly dynamically adjusting the balance. He knows well that the root cause of the loss of the Japanese semiconductor industry lies in not being able to dynamically adjust the balance according to changes.
Adjustments come with risks, first of all, the strategy needs to be correct.”In the era of artificial intelligence, we propose a new standard that emphasizes speed over computing power,” said Yu Kai. “Using only a fraction of the power of competitors to achieve more powerful computing.”
In 2016, Horizon Robotics summarized the new Moore’s Law with the formula: Real AI Performance = Peak Computing Power × Effective Utilization of Computing Resources × AI Algorithm Efficiency.
Coincidentally, in 2020, a paper by MIT scholars published in Science proposed the view of improving computing performance in the post-Moore’s Law era, which coincides with the new Moore’s Law for AI computing proposed by Yu Kai in 2016.
The direction is correct, but the strength must also be considered.
For example, the Journey 5 chip designed based on the new Moore’s Law only uses half of the chip area and half of the computing resources compared to NVIDIA’s flagship Orin-X chip. But, when using advanced neural networks for computing, the Journey 5 chip can still achieve high computing performance, especially in more advanced neural networks.
The standard for a good chip is not just computing power, but also software running efficiency. Chips are simply a means to an end, and it is ensuring that the software runs faster on top of the chip that matters. History has repeatedly proven this point – only chips + operating systems can guarantee computing efficiency.
Therefore, in the design of Journey 5, Horizon Robotics specifically optimized for latency in both hardware and software. This enables Journey 5 to process ultra-high images captured through an 8-megapixel camera, and output fully-preserved results in just 60 milliseconds, while most products on the market can only achieve around 150 milliseconds of latency.
In addition, the competitive advantage of the Journey 5 chip is also demonstrated through the launch of the open-source TogetherOS operating system without any commercial purposes. Horizon Robotics collaborates with partners to build a car OS base with rich software and hardware interfaces to facilitate the development of various products.
“We believe that the operating system should not be a profit-making tool, but rather a public technological resource,” said Yu Kai.
Under new competition, new technology players and leaders will inevitably emerge. The path to global expansion is clear, and Yu Kai’s benchmark object has shifted from Mobileye to NVIDIA, not only because of their comparable chip strength, but also because of their similar thinking paths.
NVIDIA’s GPU + CUDA and Horizon Robotics’ Journey + TogetherOS are becoming strong competitors for each other.
“We are the most knowledgeable in algorithm software for chip hardware companies, and the most knowledgeable in chip hardware for algorithm software companies. This is one of our core technological advantages,” said Yu Kai confidently.
Conclusion:Overall, there are three open and win-win business models for Horizon:
-
Providing BPU and SoC level journey chips and operating systems (OS) to help automakers complete the development of their autonomous driving software and hardware systems;
-
Providing BPU and SoC level chips for car manufacturers to develop their own autonomous driving software and hardware systems using the operating system;
-
Providing BPU IP to support automakers in developing their own SoC and implementing operating systems and autonomous driving software and hardware systems for vehicle development.
Under different open modes, the OEMs have different levels of innovation synergy, initiative and leadership. This is a real commercial demand and true commercial wisdom. Because having friends is to have a future.
Today, Horizon is accelerating its chip development speed, significantly reducing R&D investment, promoting the leapfrog development of the automotive chip industry, and striving to achieve an open ecosystem.
We believe that the thinking and practice of Horizon is not only a process of innovation and upgrading of its own business model, but also a portrayal of Chinese companies’ daring breakthroughs in the field of chips and operating systems in the era of intelligent vehicles, which has significant implications for the development of the 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.