What is the extent of Horizon's ambition to be both a "solo player" and a "pace-setter"?

Outing 100, a media company under the Outing Group, focuses on the evolution of the automobile transportation industry.

Founder Yu Kai always speaks his mind in his signature style, which comes not only from his straightforward personality, but also from the confidence and courage brought on by Horizon Robotics’ rapid growth.

“We were the first AI chip start-up company in the world and launched the first vehicle-grade AI chip in China…” Yu Kai said. 

Actually, Horizon Robotics had a hard time in the past few years compared with older companies with more funding, talent, and experience. “We wanted to surpass the competition in new markets and new arenas,” Yu said. This year’s Shanghai Auto Show was one of those arenas.

Located in the passenger car pavilion, surrounded by potential partners of the future, Horizon Robotics signed a cooperation agreement with Ideal Auto. Despite the overwhelming impact of the wave of electrification and the entrance of technology giants with abundant resources, Horizon Robotics still holds its own.

Yu Kai believes this is an opportunity for technology companies like Horizon Robotics. “This is the moment when consumers and automakers are rushing forward,” he said. “Only when there is real change, there can be opportunities. There are no opportunities in stagnant water for a thousand years.” He appreciates the entry of these tech companies and believes it’s a positive thing.

“As a matter of fact, if Tesla hadn’t taken off, it would have been difficult for us to survive this year.”

Of course, with the acceleration of intelligent manufacturing by automakers, demand for chips is increasing. Since Horizon Robotics has seized many “firsts,” it hopes to take the lead in this new arena.

“We have been working in this market for a long time, and we have a strong competitive edge. When an opportunity comes, we seize it.”

Yu Kai doesn’t view this as ambition, but as a responsibility. In the culture of Horizon Robotics, there are eight words: “Endure loneliness and achieve customer success.” Technological breakthroughs in chip development do not happen overnight, and Yu Kai hopes that everyone at Horizon Robotics has the patience to stick with it and make their technology solid.”Making chips is a long-term endeavor that requires patience. Even though some may perceive me as low-key, I do not care about how others view me. What’s more important is how I view the technology that Horizon is creating. By being humble and avoiding arrogance, we can truly pursue long-term success.”

Thanks to this spirit, Horizon’s business in the intelligent driving race is thriving. As a company that specializes in both algorithms and chips, their cooperation with IDEAL has been an effortless partnership.

Reportedly, IDEAL will soon be implementing Horizon’s Journey chip in their future models to achieve L2+ level assisted driving functions.

The Journey 3, a new generation of high-efficiency intelligent chips for vehicles, was announced by Horizon last September. Within half a year, the company has revealed an ADAS-centered Horizon Matrix® Mono assisted driving solution featuring Journey 3.

This program has undergone a comprehensive upgrade based on the vision perception product in the mass production phase of Journey 2. It now supports a 120-degree horizontal view field and up to 8 million ultra-high-pixel front cameras, ensuring more extensive and higher-precision visual perception capabilities.

Journey 3 will become the world’s first AI chip with a front 8 million-pixel camera for automated driving, leading the way into the era of 8 million pixel vision.

The perception product of Journey 3’s front view incorporates high-precision visual mapping and positioning algorithms that can achieve lane-level positioning accuracy without the need for a high-precision positioning box.

The improvement in auxiliary visual perception is reinforced by features such as ramp entrance and exit detection, construction avoidance, correspondence with lane and traffic elements (traffic lights, signs, and road arrows), and increasing the adaptability of ADAS system scenes.

“Collaboration between automakers and chip manufacturers is indeed the current trend,” insiders explain. Prior to partnering with IDEAL, Horizon was a provider of chips for SAIC passenger vehicles’ autonomous driving systems.

Currently, both Horizon Matrix® for autonomous driving and Horizon Halo™ for in-vehicle intelligent interaction are in mass production and have been verified. Last year, Horizon’s chip shipment exceeded 100,000, mainly for intelligent driving and intelligent interaction.

This year’s figures will not be disclosed, but Mr. Yu Kai left us with a very confident sentence: “Horizon’s dream is to be the world’s top-tier chip company in the field of intelligent driving and artificial intelligence within the next decade.”Compared with last year, the shipment volume has increased significantly. We can assure our partners that chip supply is sufficient.

With years of accumulated expertise in software and hardware technology, as well as rich experience in front-end manufacturing, Horizon has released a complete vehicle intelligent solution for all scenarios, integrating interior and exterior intelligence to make autonomous driving more reliable and improve human-machine interaction.

“As the level of autonomous driving increases, the ultimate goal in the future is to have a central computer that forms an AI center in the vehicle, integrating all car AI computing resources with sufficient computing power to complete all AI calculations from environmental perception to human-machine interaction,” said Yu Kai.

“Vehicle computers will definitely embed the growth of all software for the next three years,” Yu Kai asserted.

According to Horizon’s calculations, several TOPS of computing power are sufficient for L2 autonomous driving, while dozens of TOPS are needed for L3 and hundreds for L4. Only at L5 level would thousands of TOPS of computing power be required.

“We never pursue 1,000T like NVIDIA,” Yu Kai has always emphasized. He believes that computing power does not truly reflect the real performance of an autonomous driving computing platform. Comparing the high and low of computing power is misleading.

At present, Horizon’s chips can achieve the same effect with several hundred TOPS of computing power as NVIDIA’s 1,000 TOPS. By fully integrating software algorithms and chip architectures and controlling power consumption, Horizon has remained at a relatively good level.

This year, Horizon will provide a chip called “Journey 5,” with computing power of 96T, and another chip called “Journey 5Pro” with computing power of 128T, which is expected to be mass-produced and installed in vehicles in the second half of 2022.

The planned Journey 6 chip has a planned computing power of 400TOPS and will adopt the 7nm process technology for the car. The engineering samples are expected to be launched in 2023, and mass production is expected to be achieved in 2024.

Based on the high computing power and high functional safety level of the Journey 5, matched with high-performance perception algorithms and comprehensive development verification tools, Horizon’s Horizon Matrix® FSD full autonomous driving solution supports 12 camera schemes and optional lidar, achieving point-to-point unmanned autonomous driving, urban parking and automatic summoning, and complete coverage autonomous driving in all scenarios.

In addition, Horizon plans to further expand the open chip “ecosystem” to promote the prosperity of intelligent automobiles in the future.”The underlying chip is not only inseparable from the ecosystem, but also the cornerstone of the ecosystem, because intelligent cars are essentially defined by software, and software needs to be supported by underlying chips.

Just as seen in the PC and mobile phone era, smart cars will also move forward along the path of IT, and the only way to win is to accumulate software and ecosystem richness supported by chips.

Horizon has established deep cooperation with major automakers such as Audi, BYD, Changan Automobile, Great Wall Motor, Dongfeng Motor, Guangzhou Automobile Group, Hongqi, Jiangqiaojituan, Ideal Automotive, Chery Automobile, SAIC Group, and Tier1 such as Desay SV, Eastsoft Rui Chi, Continental Group, Freetech, FOG, Huayang, and quickly built an open and win-win intelligent car chip ecosystem.

“Is there only one winner in the end?”

In response to this question, Yu Kai gave the following answer.

“Autonomous driving is a function, not an application. Therefore, its standard can only be one. Therefore, ultimately autonomous driving will not flourish.”

In his opinion, the future of autonomous driving chips is definitely convergent, there may be very few players, and even only the first and second places, and the first place may have a huge competitive advantage over the second place.

Horizon is working hard to become one of the “few players”.

The following is an excerpt from an interview with Yu Kai in the media:

Media: Do you think that the rapid entry of technology companies has disrupted the original release process and plan?

Yu Kai: I think it’s great that these technology companies have joined in. In the past, the development of automotive technology was led by some more traditional first-tier suppliers. Now it has become an industry driven by technology companies, which actually promotes faster development of the industry. Otherwise, the previous automotive industry seemed too conservative, and young people were unwilling to come in and work. Now young people have rushed in, which is great.

Now the whole industry, I think, is in a state of change. It’s not about overtaking on bends, it’s actually about changing lanes and overtaking. So I think it’s a good thing, and I hope everyone runs faster and better.

Media: Do you think how much computing power is enough for level 4?

Yu Kai: For level 4, a few hundred TOPS is enough. We calculated it three or four years ago, which was the earliest in the industry. L2 requires a few T, L3 requires tens of T, L4 requires hundreds of T, and L5 requires thousands of T. Now when we do L3, the hardware usually goes up to L4, because the hardware needs to be pre-embedded and the software upgraded.

The onboard computer must embed the growth of the entire software in the next three years, so generally speaking, the current experience for L3 is to buy L4 computing power, and the experience for L4 is to install L5 computing power.

Media: Is there only one winner in the end?”## 余凯:

When everyone has the same experience of a thing, there will be fewer ultimate winners. I feel that the future of the automatic driving chip will not be diversified, but will be converged to just a few players, possibly even only the first or second place. Moreover, the first place may have a huge advantage over the second place.
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.