According to PulseStacks' report on March 18, the China Electric Vehicle One Hundred People's Forum (2024) was held at the Diaoyutai State Guesthouse in Beijing from March 15th to 17th. The forum, themed "Consolidating and Expanding the Advantage of New Energy Vehicle Development," extensively discussed topics including the new trends in global automotive industry development, green and low-carbon transformation paths for the automotive industry, new directions in intelligent connected vehicle development, new strategies for the integration and development of automotive cities, innovation in power battery industry, as well as the applications of big computing power, large models, big data, and the construction of charging infrastructure, and new changes in the automotive market and consumption.
On March 17th, during the Market and Consumption Forum, Horizon's founder and CEO, Yu Kai, stated that Horizon has already established deep cooperation with all mainstream Chinese automakers. "For example, we are now the largest supplier of intelligent driving chips for BYD, and also the largest and most important supplier of intelligent driving chips for NIO, including being a client of ours. Since last year, we have also begun cooperation with joint venture brands and established very deep strategic cooperation with Volkswagen Group."
He mentioned that this year is a big year for Horizon as they are launching the Journey 6 series chips. It's not just a single chip but a whole family series covering high-end, mid-range, including the integrated front-view machines. "Journey 6 is not just about enhancing some technical processing capabilities; more importantly, we are introducing a brand new generation of chip architecture called the Nash architecture. In advanced urban autonomous driving, we can support efficient computing. Through the Journey 6, which is composed of multiple chips, we can provide customers with a comprehensive solution for the first time, meeting various price points, scenarios, cost-effectiveness, and performance requirements."
At the same time, he shared his views on the development of operating systems. In the past, architectures and operating systems were software-centric, but now they are centered around intelligent computing and data flow processing. "An interesting point is that to excel in an operating system, it must have 'killer-level applications.' In the automotive operating system, 'killer-level' applications include two things: intelligent driving and comprehensive intelligent human-machine interaction brought by large models."
"I believe there is a great opportunity in the rare window of opportunity for the development of China's intelligent automobile industry to give birth to world-class Windows, world-class operating system companies, and world-class chip companies."
He believes that the truly competitive next-generation autonomous driving system should be evaluated through three dimensions: the first dimension is the standard scene pass-through rate; the second is commuting efficiency, and the third is behavioral personification.
"From the perspective of Horizon, from chips to operating systems, we will also focus on its 'killer-level' applications, which is advanced autonomous driving, to create benchmarks. What we call benchmarks is that Horizon will create a full-stack software solution and enable this software solution to be mass-produced and implemented in some key vehicle models. This actually allows Horizon's large number of Tier 1 partners to receive our favorable support, enabling them to cover more vehicle models during mass production, thus improving the overall delivery efficiency. So Horizon is still defined as a Tier 2 supplier, but our full-stack software capability can support partners to complete delivery."
Key Points:
Speech Transcript:
Respected guests, dear friends, good afternoon to everyone!
The speeches before me were truly splendid, reflecting the spirit of Chinese entrepreneurs. Each one is a king of battle, a king of the times, and the technology of autonomous driving is getting more impressive by the minute. As a supplier of intelligent driving computing solutions, many of the previous speakers are also my clients. So, from the bottom of my heart, I must say, only when everyone does well, can we truly prosper.
Firstly, I'd like to report to everyone on the commercial progress, product implementation, and overall market developments in the field of intelligent driving computing solutions that Horizon has achieved in the past year. Secondly, I want to return to the essence of a technician and share some thoughts on technology, especially on in-vehicle operating systems. Lastly, I'll share our insights on advanced autonomous driving, including urban autonomous driving, and discuss some technological trends.
Since 2020, Horizon has achieved the first commercial deployment of front-loading, which was on three models of Changan Automobile: UNI-T, UNI-K, UNI-V, and "KTV". Every year since then, we've been expanding our customer base. As of last year, over 30 major automotive companies have become our mass production clients.
Currently, essentially all mainstream Chinese automotive companies have engaged in deep cooperation with Horizon. For instance, we are now the largest supplier of intelligent driving chips for BYD, and also the largest and most significant supplier for the intelligent driving chips of NIO. Additionally, NIO is also one of our clients. Since last year, we've started collaborating with joint venture brands as well, and have established very profound strategic cooperation with Volkswagen Group.
We're fortunate to be closely aligned with the innovations in the Chinese automotive industry, where many models equipped with Horizon chips are gaining significant popularity among consumers. To date, over 110 production models have integrated Horizon chips, with many more in the pipeline for mass production and delivery.
According to the latest data, NVIDIA dominated the market for high-level autonomous driving (NOA) last year with a 48.9% market share, unquestionably securing the top position. Horizon, on the other hand, captured a 35.5% share, placing second in the market. Together, these two companies accounted for 85% of the market share, highlighting a clear duopoly at the top.
When considering the more traditional L2 advanced driver assistance systems (ADAS) market, Horizon initially held the third position in the first half of last year. However, with our efforts in the latter half of the year, our market share continued to grow. By the end of the year, we rose to second place in the market, capturing a 23.7% share. Horizon, Mobileye, and Renesas essentially dominated the largest share of this market.
This year marks a significant milestone for our product line as we introduce the Journey 6 series of chips. Unlike our previous Journey 2, Journey 3, and Journey 5 chips, which were standalone offerings, the Journey 6 is the first in a family series. It will cover both high-end and mid-range markets, including forward-looking integrated systems. The Journey 6 represents more than just a boost in technical processing capabilities; it signifies the introduction of a brand-new chip architecture known as the Nash architecture.
Named after the game theory pioneer John Nash (known for Nash equilibrium), the core concept of the Nash architecture lies in navigating complex scenarios and interactions with surrounding traffic participants in advanced autonomous driving scenes. So, how are we crafting this new chip architecture? In urban autonomous driving scenarios, we can support efficient computation. The results speak volumes: we're achieving several times to tenfold increases in computational efficiency compared to traditional convolutional neural networks, as well as enhanced support for advanced Transformer-based models.
Through the Journey 6, a multi-chip computational solution, we're able to offer customers a comprehensive solution for various price points, scenarios, cost-effectiveness, and performance needs.
Moving on, I'd like to share some thoughts on the operating system aspect. Chips and operating systems have historically been tightly integrated. Looking back at the evolution of computing platforms, such as the Wintel alliance in the era of personal computers, which tightly coupled Windows with Intel's x86 architecture. In the smartphone era, we saw the AA alliance between Android and ARM, representing software-hardware synergy.
In the realm of artificial intelligence in data centers, we're witnessing alliances like CUDA and GPU. Although this is typically driven by a single company, the collaboration between software, hardware, chips, and operating systems remains integral. So, naturally, we ask: What will the landscape of the next-generation computing platform, its software, hardware, and ecosystem, look like in the era of intelligent driving and smart cars?
Additionally, we must consider the evolution of the entire computing architecture, particularly how in-vehicle computers will evolve in the future. Today, we see intelligent driving and smart cabins as two separate domains controlled by independent chips. However, as we move towards an era of large-scale models and end-to-end Transformers, the goal is to unify autonomous driving and human-machine interaction under a single central computing domain.
This evolution entails a shift towards intelligent computing as the core, with the architecture and operating system supporting a diverse range of applications. The interaction between intelligent driving and smart cabins will become just one facet of these applications. This could lead to the emergence of entirely new computing architectures and operating systems, shifting from past paradigms centered around software to those centered around intelligent computing and data flow processing.
Reflecting on history, with each generation of computing platforms, we observe an interesting phenomenon: While users may not always accurately articulate their true computing needs, their demand for performance remains constant, evolving over time.
With the continuous improvement in computing performance of personal computers till today, it's clear that the computing power of PCs in the 90s couldn't meet our demands. That's why we kept iterating from 286 to 586. But now, unless there are specific reasons, we don't really think about replacing our computers. Similarly, the same scenario is happening in the iteration of personal mobile phones. What industry trend can we observe from this? It's that when performance doesn't meet user demands, a high degree of coordination between software and hardware is needed to gain a relative competitive advantage in performance.
When performance has surpassed user demands, it typically leads to the decoupling of software and hardware. For example, we've seen Microsoft's Windows starting to support ARM chips lately, and devices like Surface now support Qualcomm ARM chips. Similarly, we've seen Apple's Macintosh, which was previously combined with PowerPC, but recently Macintosh has also started supporting X86 chips. This shows that when performance has exceeded user demands, the combination of software and hardware is not as crucial as before.
When it comes to operating systems, it's essential to clarify some concepts here. The term "operating system" can actually have very different extensions and connotations. For instance, systems like QNX, AutoSAR in the automotive industry, or VxWorks all fall under the umbrella of operating systems. What we've observed is that when starting from the perspective of building an operating system itself, it generally doesn't lead to the creation of a dominant operating system in the ecosystem. However, approaching operating system development from the angle of vertical applications often yields significant success, as seen with Windows and Android.
Think about it: without Office, Windows wouldn't have been as successful. So, what is Office? Essentially, it's about white-collar productivity. For example, if your boss asks you to create a PowerPoint presentation, without Windows supporting Office, you wouldn't even consider it, right? Therefore, the demand for Office essentially drives the demand for Windows. Now, let's look at Android. Without the GMS suite, including Google Search, Gmail, and YouTube, Android wouldn't have such a strong appeal.
So, we observe an intriguing phenomenon: attempting to build a platform from the platform itself often fails, while starting from applications to build a platform often leads to success. Therefore, we need to move beyond just technical thinking and consider more business and ecosystem-oriented perspectives when conceptualizing major operating systems. A successful major operating system must enter a very large mainstream application scenario, then expand the platform and support a rich array of applications on top of it. So, one interesting point to note is that to excel as an operating system, it must have "killer" applications.
Now, what are the "killer" applications for automotive operating systems? Actually, we can already glean some clues from the insights shared by the previous speakers. In the realm of automotive operating systems, the "killer" applications are twofold: firstly, intelligent driving, and secondly, comprehensive intelligent human-machine interaction brought by big models. Therefore, the path ahead becomes clearer, and the goals become more defined. Moving forward, I believe there's a great opportunity in China's intelligent automotive industry to give birth to world-class companies in operating systems and chip manufacturing.
When it comes to "killer applications," the first "killer application" is undoubtedly autonomous driving. Now, let's analyze the trends in China over the past few years and in the future. For instance, the adoption and penetration rates of advanced driver assistance systems (ADAS) are increasing. Last year, they already exceeded 50%. Without a doubt, as technology continues to evolve, consumers will increasingly enjoy the safety and convenience brought by autonomous driving. Therefore, the penetration rate is inevitably moving towards 100%. At the same time, we can see that high-level autonomous driving, especially in the advanced autonomous driving market, has been continuously evolving in recent years.
In the price range of 200k to 300k, the adoption rate of advanced NOA equipment has already increased last year, so this industry is only going to get more bustling.
So, what's the good news then? The industry's increasing demand for high-performance chips is definitely good news for us. And for consumers, it means they'll be getting more value from the growing accessibility of technology. However, we also need to acknowledge that today's advanced autonomous driving systems, like NOA, still have many issues. Consumers aren't fully reaping the benefits due to high costs and less-than-stellar performance in various real-world scenarios.
When it comes to the value of technology for users, we can break it down into three levels: usability, user-friendliness, and user-love. Taking high-speed NOA (Navigate on Autopilot) as an example, currently, it may have achieved usability and is progressing towards user-friendliness. From our daily experiences, on highways and enclosed urban roads like overpasses and city loops, NOA can practically handle the entire journey without human intervention. Therefore, high-speed NOA can be considered usable, and it's approaching the stage of being user-friendly. However, whether it has reached the level of being loved by customers is debatable. As for urban NOA, it might still struggle even with basic usability.
Presently, urban NOA might require intervention up to ten times within a 100 km/h speed limit in busy city areas. Therefore, urban NOA is still in its early stages. We need to maintain confidence in its future while also holding an objective and realistic view of its current state. Only then can we continually improve the technology from being usable to user-friendly and eventually beloved by consumers.
Now, let's discuss the challenges in achieving user-friendliness where usability is lacking. Main factors include cost, speed of deployment, and intervention rates. However, achieving user-friendliness is more complex. For instance, during my commute to work, if I'm driving manually, I can easily change lanes and merge faster. But with autonomous driving, the system might be more conservative, resulting in lower efficiency, which could be frustrating or impatient for me. Therefore, achieving user-friendliness entails more nuanced demands.
Another aspect to consider is the development approach in autonomous driving technology. There are two main strategies: one aims to expand the scope of scenarios in which the technology can be used, albeit with lower overall intelligence. This approach is typical of past systems like Mobileye's advanced driver-assistance systems. The other strategy pursues full autonomy, resulting in a much higher level of automation but within a restricted domain, such as a defined geographical area, as seen in initiatives like Robotaxis. The ultimate goal now is to achieve advanced autonomous driving across all scenarios, with both high automation levels and a wide operational domain. This is a goal pursued by many Chinese automakers, including Horizon, who are supporting their partners in reaching this objective.
So, what defines a competitive next-generation autonomous driving system? We can evaluate it through three dimensions: standard scenario pass rates, commuting efficiency, and human-like behavior. The current technology, depicted by the inner blue triangle, emphasizes scenario pass rates, is less efficient in commuting, and lacks human-like behavior, often appearing rigid.
For users, this kind of experiential feeling isn't that great. So, how do we go about creating what we call a "graceful yet confident" personified system? Essentially, it's the goal that the next generation of autonomous driving systems aims to achieve. It can be said that currently, in terms of safety, we don't face significant issues with high-level autonomous driving. As He Xiaopeng just mentioned, the experimental data on safety actually shows a lower accident rate compared to human driving overall. So, what's the next step in embodying "graceful yet confident"?
From a horizon perspective, we're looking at everything from chips to operating systems, and even up to its "killer app," which is advanced autonomous driving. We'll be creating a blueprint for this. The concept of a blueprint here means that Horizon will develop a full-stack software solution, and enable this software solution to be mass-produced and implemented in some key vehicle models. By doing this, we're actually providing significant support to a large number of Tier 1 partners, allowing them to cover more vehicle models during mass production and replication phases. This, in turn, boosts overall delivery efficiency. Therefore, Horizon is still defined as a Tier 2 supplier, but our full-stack software capabilities support our partners in completing deliveries.
What's really backing us up behind the scenes is Horizon's world-class innovation and research capabilities in software algorithms. For instance, take the researcher from Horizon who made a huge impact last year with UniAD, which garnered significant attention across the industry and even won the Best Paper Award at a top-tier computer vision academic conference. Standing out among over 9,000 papers, it's safe to say it made a substantial splash in the industry, being the first end-to-end autonomous driving large model in the research community.
Recently, we introduced the Sparse4D algorithm to address the limitations of traditional dense BEV algorithms. With this sparse framework for autonomous driving perception computation, we have consistently ranked first on a series of public datasets. The innovative capabilities of our software algorithms support us in continually designing novel chip architectures, enabling our hardware to efficiently accommodate next-generation software algorithms. Additionally, our full-stack software algorithm research and development will facilitate our partners in developing their advanced software autonomous driving systems faster and more cost-effectively. In conclusion, Horizon continues its journey of collaboration and mutual success, leveraging our full-stack technology to create the most open innovation ecosystem. As always, true success lies in everyone's well-being. Thank you, everyone.
