The Media under the Outing of 100 People focuses on the Evolution of the Automotive Travel Industry Chain
Author: Roomy
Yu Kai, the founder of Horizon, cannot afford to be complacent.
“Horizon must remain focused.” This is a sentence full of awe that Yu Kai, the founder of Horizon, said when he launched Horizon Journey 5.
Even if Horizon has completed two generations of production, Journey 2 and Journey 3, of “vehicle-grade” chips with a production volume of over 400,000 and over 40 fixed-point production projects by only six years, and even if the first two generations of production chips have laid a good foundation, Journey 5 has also been highly praised by many potential customers.
“China has become the birthplace of global smart car innovation, just like the World Cup final from the beginning.” Yu Kai said that the competition for chips is extremely fierce, and Horizon must always remain focused, even with a sense of crisis.
Those who pay attention to Horizon may know that the company has always maintained a “pessimistic” attitude: pessimistic about the process and optimistic about the endgame. The endgame refers to the fact that by 2030, L3 and above autonomous driving will become standard, and every car will be equipped with AI chips. Thinking from the endgame, Musk’s ideas are similar to those of Horizon, and the difference between Tesla and the others is to achieve autonomous driving with low-cost sensors and high computing power.
To get a chip on board, it must pass the “vehicle-grade” certification, which is a hard fact. The requirements for vehicle-grade chips are zero defect rates, and millions of shipping volumes are required to prove it. Therefore, the players in the automotive chip market are basically super giants, such as Intel, which has decades of history.
As a start-up chip company, it is very difficult to break through from 0 to 1 without a production history or historical data on quality defect rates. From Horizon, Cambricon, to Black Sesame, all chip start-ups have to start with “how to persuade car companies to carry their own chips” when they leap out of Horizon.
Previously, Dan Jizhang, the founder and CEO of Hei Zhi Ma Intelligence, and Zhao Lidong, the founder and CEO of AI chip company Suiyuan Technology, both stated that “breaking through chip technology is the hardest thing from 0 to 1.” Horizon took about five or six years to break through from 0 to 1.
As competition gradually progresses towards high-level autonomous driving, domestic chip companies need to use stronger chip strength to engage in head-to-head battles with old-school chip giants such as Nvidia and Qualcomm, with almost no room for leverage.No matter whether it’s Horizon Journey 5 or Black Sesame A1000 Pro, it is not an easy task for these Chinese chip companies to make a broader breakthrough in the field of autonomous driving. However, the rise of representative chip companies is crucial to the intelligent development of China’s auto industry.
Not an Easy Battle
It is well-known that Horizon Journey 5 is facing a tough battle.
As Horizon Journey 5 is launched, Horizon has become the only whole vehicle intelligent chip solution provider covering from L2 to L4 in the industry. At the press conference, Yu Kai also announced a series of intended cooperation partners, including many newly-established automakers and traditional auto companies.
Horizon Journey 5 is the third-generation vehicle-level product following the first production of the vehicle-grade AI chip, Journey 2, and Journey 3. It has both large computing power and high performance, with a single chip AI computing power of 128 TOPS, supporting 16-channel camera perception computing, and millisecond-efficient collaboration. It can meet the requirements of multi-sensor fusion, prediction, planning and control for autonomous driving.
In terms of the core architecture, this is also the core technology of Horizon. The CPU part of Journey 5 adopts the 8-core ARM Cortex A55, and the AI computing unit adopts the third-generation Bayesian architecture of the dual-core BPU with independent intellectual property rights. This architecture adopts large-scale heterogeneous process computing, and selects the best computing mode according to the scenario configuration, greatly reducing the power consumption and delay of the computation while improving the computing efficiency.
Since the launch of Journey 2 in 2019, Horizon’s philosophy for automobile intelligence technology is to achieve overall efficiency of technology development and application iteration based on a whole-vehicle intelligent computing platform that features high performance, large computing power, safety, reliability, openness, and ease of use.
It is worth knowing that as the level of autonomous driving increases, the chip computing power needs to increase exponentially. However, there are only a few globally that can achieve mass production of computing platforms with large computing power.
Shan Jizhang, the founder of Black Sesame, said in an interview at the Shanghai Auto Show this year: “Considering the supply chain security issue, each automaker will have several supplier choices. In addition, almost all chip suppliers are involved in the rapid iteration of technology. In fact, there are very few chip suppliers in the market, especially for high-level chips.”
The large computing power vehicle-grade intelligent driving chips that have been launched including Tesla FSD, NVIDIA Orin, and Mobileye Eye Q4, among others. Qualcomm and Huawei are also investing in autonomous driving onboard chips. NVIDIA will launch a 1000T computing power chip in 2024… Developing powerful computing chips specifically for autonomous driving is becoming a new battlefield for chip giants to compete.Although computing power is not the only criterion for judging performance, the balance between power consumption and computing power is key. However, measuring performance in terms of computing power has become an industry consensus, with software optimization being the ultimate manifestation of computing power.
“Horizon 5’s performance will surpass Tesla’s FSD chip,” said Yu Kai. It is reported that Tesla’s FSD dual-core chip has a single-chip computing power of 72 TOPS. However, looking at the computing power data, Horizon 5 is lower than Nvidia’s next-generation chip Orin, which has 254 TOPS.
However, in measuring the FPS of AI performance parameters, Horizon 5 maintains its advantage. Horizon 5 calculates at 1283 frames per second, becoming the most powerful computing performance among the currently revealed chips. In addition, Horizon 5 also has a minimum delay of 60 milliseconds and a minimum power consumption of 30 watts.
“Horizon 5 is a key node in the development of the next stage for Horizon, targeted at L4 advanced autonomous driving chips.”
According to a technical engineer from an automaker, Horizon 5 is not only a breakthrough for Horizon, but also, in the context of the global chip industry and intelligent trajectory, domestic chip companies represented by Horizon have acquired a ticket for high-end autonomous driving chips.
Advanced autonomous driving is bound to be the most fiercely contested field in future technological competition. Due to the focus and decisive position of the Chinese market, including Nvidia and Qualcomm, the world’s top chip companies, their latest autonomous driving chip production launch has focused on this area. For Horizon, the competition may become even more intense in 2022.
The demand for autonomous driving chips is growing rapidly, and high-computing and high-efficiency autonomous driving chips are particularly scarce. The huge demand in the market has rapidly propelled startups such as Black Sesame and Horizon.
“The competition for intelligent driving chips has never been a regional league from the beginning, but a world-class finals. Horizon’s goal is to become the world champion in the chip field,” said Yu Kai. For an automotive chip company, the most important thing is to find customers and achieve mass production.
At the launch event, Horizon reached a cooperation agreement with numerous automakers including SAIC Group, Great Wall Motors, Jianghuai Automobile, Changan Automobile, BYD, NIO, and Link Tour. Horizon also reached a pre-research cooperation agreement with Ideanomics based on Horizon 5, accelerating the popularization of high-level autonomous driving functions.
At the beginning, everything looks good, but can Horizon 5 gain the advantage to win this race and leverage the huge market in the name of domestic AI chips? Overall, it’s not easy.# Translation
The CEO of Horizon Robotics, Li Dongzhao, once said a straightforward quote: “If the first one makes money, the second one can make money too, but it would be very difficult for the third one. Basically, there is no market”.
The Horizon Robotics team agreed with this view during the interview.
“Nowadays, after a carmaker uses a chip supplier and has a complete supply chain, even if the second chip company’s chips are almost as good or even better, it is difficult to break into the market. It’s not that there are no opportunities, it’s just hard.”
Of course, Waymo 5 won’t let opportunities slip away.
Yu Kai said: “Electricity and calculation power will be the two pillars supporting smart automobiles. In the future of China’s smart automobile industry, we hope that Ningde Times will support the electricity side, and Horizon Robotics will be on the computational side.” In other words, Horizon Robotics aims to become the No.1 chip manufacturer in China, just like how Ningde Times became the No.1 battery manufacturer.
“Once we have made the breakthrough from 0 to 1, we will feel like we are on top of the world, and many car manufacturers will approach us.” Currently, in the field of autonomous driving, Yu Kai thinks that Horizon Robotics has only two opponents, Mobileye from Intel and Nvidia.
Horizon Robotics obtained the confidence it needs after a single Wafer of Waymo 5 was successful, with all tests passing within 15 hours.
As a manufacturer of vehicle-grade chips, Heizhima Intelligence’s strength is also worth mentioning. Danji Zhang indicated that Heizhima is more advanced in its layout in this field, and “only Nvidia and Heizhima provide high-specification chips worldwide.”
He believes that “the trend for the development of smart cars is clear, and companies that joined the industry around the same time have all had a certain amount of accumulation. The next step will be to see who can achieve mass production faster.”
In April of this year, Heizhima Intelligence released the vehicle-grade autonomous driving chip A1000 Pro. It is its second high-performance and high-calculation power autonomous driving computation chip for vehicle certification after A1000. Equipped with a high-performance GPU, it can support high-definition 360-degree 3D panoramic image rendering and covers L3/L4 advanced automatic driving functions.
Heizhima’s autonomous driving chip is targeting Tesla. According to official information, successful Wafer testing for A1000 Pro allowed Heizhima Intelligence to be one of two high-caliber autonomous driving chip manufacturers that met the ISO26262 vehicle certification function safety standard; their plan is to achieve mass production and go public by the end of 2022.
Heizhima Intelligence and Horizon Robotics could be referred to as two dark horses in the domestic chip manufacturing field. Both established at approximately the same time around five to six years ago and both have launched high-caliber, vehicle-grade chips; A1000 Pro and Wafer 5 will both enter mass production by 2022.In terms of computing power, Horizon and Black Sesame hold different views but they are not contradictory. Yu Kai believes that “blindly putting emphasis on computing power does not truly generate user value. The true value lies in the actual performance of software running on the chip.” However, Shan Jizhang believes that “high computing power is based on considerations of functional redundancy and safety is crucial for autonomous driving.”
From a market competition point of view, the two dark horses in the chip field will exhibit strong competitive relations in the process of vying for the favor of car manufacturers. As mentioned earlier, car manufacturers will consider the completeness of the chain of supply and will not easily change chip suppliers. The question will be who can bring their product to market first.
Currently, Horizon’s Journey 2 and Journey 3 have delivered over 400,000 automotive-grade chips to customers and delivered over 40 mass production models. Horizon’s VP and general manager for intelligent driving products Yu Yinan revealed that out of these 400,000 chips, half of them are used for in-car intelligent interaction and the other half for autonomous driving.
In terms of progress, Black Sesame is slightly slower than Horizon. The first chip has already entered into mass production cooperation with SAIC and the second A1000 chip is in the process of mass production. In the second half of this year, over 100,000 mass production units will be implemented on commercial vehicles, and production for passenger vehicles will begin next year.
Currently, Horizon’s mass production models using the Journey chip have been recognized by the market, while Black Sesame needs to double their efforts and surpass the trust threshold of car manufacturers.
In addition, Cambricon has revealed that they will release an autonomous driving chip named “Xingge”, which has over 200Tops of computing power and uses a 7nm process technology. In computing power, it is comparable to Nvidia’s Orin and surpasses Mobileye’s EyeQ6, which may bring some pressure to Journey 5 and Black Sesame.
From the perspective of the rise of domestic chips, the success of Horizon, Black Sesame, and Cambricon is critical for breaking through the blockades of chip giants such as Nvidia and Qualcomm. “Chinese companies must have breakthroughs in the field of core chips. Without breakthroughs, the future will be very bleak.” Yu Kai maintains his usual pessimism towards this development process.
“The Chinese automobile industry urgently needs independently researched and developed high-performance, high-reliability, and high-safety core chips. Only in this way can we guarantee the future development of the Chinese automobile industry.” Chunxin Technology CEO Qiu Yujing also expresses a similar viewpoint.In the field of automotive industry, grasping more initiative in chip design has become a must. “NVIDIA is not the rule maker, what it has is the inertia of everyone around the use of CUDA,” said Zhao Lidong, CEO of Horizon Robotics, who has been calling for “fighting at the top end and doing the most high-end chips for startups!”
Compared with the mainstream car-level chips in the current market, domestic chips have achieved a good balance between computing power and power consumption, with the strength to compete with foreign chip manufacturers.
The advantage of domestic chips lies in their openness. As Yu Kai said, “Horizon Robotics does not produce mass hardware, do not bundle software, and do not provide closed solutions.” Horizon Robotics is positioned as Tier-2, and plans to open the underlying technology platform in multiple dimensions to different ecological partners, empowering independent R&D, accelerating the development of intelligent applications for automobiles and the process of mass production of intelligent vehicles.
“We dare to expose the platform for customers to develop on, rather than satisfying a particularly segmented, self-implemented function, because Horizon Robotics has been oriented towards deep learning, a new generation of AI technology since its inception, which is relatively unconstrained and can be more open,” said Yu Kai.
Currently, autonomous driving chips face some common problems, such as hardware computing power accumulation and lack of an open ecological network. Zhao Lidong also believes that AI chip start-ups need to build an ecosystem in the next step. Building an ecosystem is far more difficult and complex than making chips, and requires the participation of many partners and customers.
Closed or open? Horizon Robotics did not hesitate too much. “I don’t think there is any advantage in being closed,” Yu Kai has always insisted.
Therefore, the Horizon Journey 5 chip has three important advantages: efficiency, openness, and safety. In addition to the chip itself, Horizon Robotics has also released the Together OS automotive operating system. With reference algorithms, the development of car companies is not starting from scratch but from a relatively mature stage, resulting in a faster development speed.
There are always competitions, but it is difficult to win. The journey of domestic chip companies has just begun.
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.