Autonomous Driving Chips: The Late Cambrian and Its Competitors

The biggest change of Cambrian Technology in 2021 is the entry into the autopilot chip track. It is easy to understand that the Cambricon makes autonomous driving chips: autonomous driving is the biggest application scenario of AI in the future, and the global auto market lacks cores. As an AI chip design company, there is no reason not to enter the market. But is it too late for the Cambrian to set off ON this popular track with many competitors? Some competitors say that on the autonomous driving chip track, if mass production is not available in 2020, they will be out. How does Cambrian respond to this?

At the Cambrian financial report briefing held on August 23, Chen Tianshi, the founder and CEO of Cambrian Technology, answered the questions from the reporter of “China Electronics News” through the investor interactive platform. Regarding the above questions, Chen Tianshi gave his answer : “At present, L2-level in-vehicle smart chips have been applied in some models of some car manufacturers, but they cannot support high-level intelligent driving. The intelligent driving technology is far from mature, and the basic hardware required for high-level intelligent driving is still in the early stage of development. The Cambrian is at the right time for the layout of in-vehicle smart chips.”

At present, in terms of the core indicators of autonomous driving chips, the computing power of a single SoC of NVIDIA Atlan reaches 1000TOPS (samples are expected to be provided to developers in 2023); the single computing power of the Journey 5 chip released by Horizon last month is 128TOPS (it is expected that in the second year of next year) Quarterly mass production of autonomous driving domain controller hardware based on the Journey 5 chip), Black Sesame announced at the Shanghai Auto Show that the maximum computing power of a single chip of Huashan No. 2 can reach 196TOPS. The Cambrian announced in July this year that the computing power of a single chip for autonomous driving is 200TOPS. It is still under development and is expected to come out in two years. Can the Cambrian match other rivals?

In the face of the fiercely competitive market structure, at the second extraordinary general meeting of shareholders in 2021 held on August 19, Chen Tianshi said: “Small companies like us have no historical burden, and what we have to do is to Take the main product to the extreme without compromising on other markets or other products.”

Running into the autopilot chip track

The Cambrian is a low-key “running into” the autopilot chip track, and the business carrier that carries the car chip is the Cambrian Xingge.

On January 5, 2021, Cambricon Xingge (Nanjing) Technology Co., Ltd. was established with a registered capital of 30 million yuan. The company is 100% owned by Cambricon, but the “automotive chip” is not specified in the relevant business. According to Tianyancha, the business scope of Cambrian Xingge (Nanjing) Technology Co., Ltd. is mainly artificial intelligence basic software development, artificial intelligence application software development, artificial intelligence basic resources and technology platforms, artificial intelligence industry application system integration services, artificial intelligence theory and artificial intelligence. Algorithmic software development, etc.

In February 2021, industry insiders revealed that Wang Ping, McKinsey’s global managing partner and head of the automotive industry in Greater China, joined Cambrian Xingge, which confirmed that Cambrian is about to enter the field of intelligent driving.

On May 3, at the Cambrian 2020 annual performance briefing, Chen Tianshi said in response to whether he had plans to enter the automotive industry: “The company has been evaluating business and product strategies in this application field, and Cambrian is the leader in AI chips. Enterprises are one of the few manufacturers that have the ability and potential to build a unified intelligent ecosystem for cloud-side-end vehicles, so we will not be absent from such an important application scenario.”

From July 8th to 10th, the 2021 World Artificial Intelligence Conference (WAIC) was held in Shanghai. Cambrian participated in the exhibition. A huge painting of “car” hung in the center of the top of its booth. Synergistic advantages to meet the growing computing power demand of autonomous driving”, announcing that the Cambrian officially entered the automotive chip.

Also at 2021WAIC, Chen Tianshi disclosed for the first time the key data of the Xingge in-vehicle smart chip under development: over 200TOPS AI performance, 7nm process, automotive-grade, independent safety island, mature software tool chain, and support customer algorithms through an open software platform Continuously update and iterate to efficiently support the needs of high-level intelligent driving. Chen Tianshi said: “Intelligent driving chips have a very close relationship with the chips we have made in the past, so it is a very natural extension to expand the cloud and edge chip categories to automobiles.”

Regarding the entry of the Cambrian into the automotive chip market, some people in the industry are not optimistic, because the entry threshold for automotive-grade chips is very high, and the leading companies have basically been established, and at the same time, they have expressed doubts about the technical capabilities of the Cambrian. Yu Kai, founder and CEO of Autopilot Chip Horizon, once said at the end of 2020: “Everyone has seen the hot scene of new energy vehicles this year. If you want to enter the ‘final’ circle, you must have reached mass production of chips in 2020, otherwise It’s already out. Because the pace of development will be very tight, the ‘game’ will be over by 2023.”

Is it too late for the Cambrian to enter the game? Chen Tianshitou responded to reporters at the financial report meeting: Intelligent driving is far from mature technology as an emerging field, and the basic hardware required for high-level intelligent driving is at an early stage of development. The layout is timely.

At the same time, some car companies have given positive signals of approval for the future development of Cambrian car chips. On July 16, Cambricon disclosed that Cambricon Xingge Technology has introduced a number of strategic investors, among which Weiran, Shang Qi and Wending have attracted much attention. Because Weiran (Jiangsu) Investment Co., Ltd., with Li Bin as the chairman, is an affiliate of Weilai Automobile; SAIC Group is the major shareholder of Shanghai Shangqi Investment Management Partnership (Limited Partnership); Ningbo Meishan Bonded Port Area Aspiration Investment Co., Ltd. It is a wholly-owned subsidiary of CATL. Companies in the upstream and downstream automotive industry chains such as NIO, SAIC, and CATL have become strategic investors in Cambrian Xingge.

It should be said that autonomous driving technology is indeed still in the early stage of development, and it is far from mature, and it is not too late to enter the game. Judging from the investment and layout of AI chip startups by traditional car companies, they still adopt the strategy of placing eggs on multiple blue chips, indicating that the chip market is not finalized. SAIC is both an investor in Horizon and an investor in Black Sesame, and now it has become a strategic investor in Cambrian. On August 9, SAIC Motor stated on the investor interaction platform that in terms of chips, the company has invested in more than a dozen domestic head chip companies, including Horizon and Black Sesame.

Can the performance be maximized without the baggage?

It is understood that Xingge Technology’s in-vehicle smart chips will integrate self-developed automotive intelligent processors, automotive CPU cores (mostly authorized by ARM companies) and other automotive computing cores, and various automotive peripheral interfaces. It is used to process the perception data collected by various sensors such as the vision of smart cars, and plan and control intelligent driving according to the processing results of the perception data.

At present, there is a lot of competition in the automotive AI chip market. There are not only startups such as Horizon, Black Sesame, Xinchi Technology, Leapmotor, etc., but also foreign giants such as NVIDIA and Intel, and another Chinese giant, Huawei, which includes chips. Including full-stack solutions to join the market.

The autopilot chip competition is mainly in several key dimensions, one of which is facing “tickets” and “ship tickets”. Different from other consumer Electronic products, automotive chips have stricter requirements for safety, and have a considerable certification process. It is not easy to get a ticket. This may be one of Yu Kai’s mention of whether he can enter the “final circle”. premise. Not to mention established companies such as Nvidia and Intel Mobileye (because these companies are already in the game, and their shipments are not of an order of magnitude. For example, in 2020, Intel Mobileye’s EyeQ chip shipments are 19.3 million pieces, which is the same as 28 cars. It is the same AI chip startup company. Black Sesame and Horizon’s automotive-grade chips have come out. So far, Horizon has launched three chips, two of which achieved pre-chip mass production last year. The latest data shows that Horizon’s vehicle chip shipments have exceeded 400,000.

The second is the chip computing power TOPS. The autonomous driving industry generally believes that the chip computing power required for L2-level autonomous driving is less than 10TOPS, the L3-level requires 30-60TOPS, the L4-level requires more than 100TOPS, and the L5-level requires more than 1000TOPS. Because of this, the chip computing power TOPS has become the core weight of each chip competition.

Horizon released the Journey 5 chip last month, announcing that it has 128TOPS (INT8) computing power. Horizon expects to achieve mass production of autonomous driving domain controller hardware based on the Journey 5 chip in the second quarter of next year. Realized the mass production SOP of the SuperDrive autonomous driving solution based on the Journey 5 chip.

Black Sesame Intelligence released the Huashan No. 2 A1000 Pro autonomous driving chip at the auto show. The INT4 computing power of a single chip can reach 196TOPS. It adopts a 16nm process, and the CPU adopts a 16-core Arm v8 core. The typical power consumption is only 25W, and it supports 16 channels. High-definition camera input, support ASIL-B level functional safety, built-in ASIL-D level safety island. This chip will provide engineering samples in Q3 this year, and provide a development platform in Q4, which is not far from being carried into mass production vehicles.

At the Shanghai Artificial Intelligence Conference, Chen Tianshi gave the parameters of the autonomous driving chip that Cambrian is designing. The computing power exceeds 200TOPS. The intelligent driving chip inherits the integrated, unified and mature software tool chain of Cambrian, and adopts the 7nm process. It has an independent safety island and meets vehicle-level standards, and is positioned as a “high-level autonomous driving chip”. It should be said that, judging from the chip parameters given by the Cambrian, there may be a fight with the entrepreneurial brothers in the “upper bunk”. And Chen Tianshi said that the Cambrian has no historical burden, and what we need to do is to make the main products to the extreme.

At present, compared with international giants, there is still a gap between Chinese start-up companies. In April this year, NVIDIA’s Huang Renxun announced in his kitchen speech at 2021GTC that NVIDIA’s latest smart car and autonomous vehicle chipset, DRIVE Atlan, has a computing power of 1000TOPS per chip and will be applied to L4 and L5 automatic drive. The chip will begin sampling to automakers and developers as early as 2023, and it may be available in models that go on sale around 2025.

The battle of autonomous driving chips is getting faster and faster. At present, all companies are giving “target values”. There are still certain variables in this competition. However, from design to tape-out to mass production, each link has a process and time that must be completed, and it is not easy to run away.

The third is a battle of hard and soft platforms. The computing power of a single chip, TOPS, is a key indicator, but it is not the only one. Autonomous driving is a complex system that requires collaboration between the vehicle, the road, the cloud and the side. Therefore, in addition to the core, there are software and hardware collaborations, platforms and tool chains in its competition.

When investors mentioned the differences with other companies including Horizon, Chen Tianshi said that Cambrian is one of the few companies in the world that has comprehensively and systematically mastered the core technology of smart chip and its basic system software development and productization. It can provide a series of intelligent chip products and platform-based basic system software with cloud-edge-device integration, software and hardware collaboration, training and inference integration, and a unified ecosystem.

In fact, Yu Kai, who was born as an AI algorithm, also adopts the core-making idea of ​​”integration of software and hardware”. At the same time as the release of Horizon Journey 5, Horizon also released a high-performance vehicle intelligent computing platform based on Journey 5, an open source secure real-time operating system TogetherOS, and Horizon, which integrates full-scenario autonomous driving and in-vehicle linkage experience. Matrix SuperDrive vehicle intelligent solution. Horizon has also laid out a development tool chain, an AI development tool platform, and a real-time security vehicle operating system.

However, at present, Horizon’s software platform, operating system, and AI tool chain are mainly concentrated in the dimension of automobiles. The advantages of Cambrian are in broader dimensions such as cloud, edge, and end.

The intelligent networked car is no longer just an independent car, especially under the technical route of vehicle-road coordination. It is precisely because of this that full-stack capabilities have become another focus of competition among automotive “core” providers. The prelude to autonomous driving has just begun. Last year, the penetration rate of L2-level autonomous driving in my country reached 15%, that is, nearly 4 million new cars were equipped with L2-level autonomous driving systems. It is estimated that by 2030, self-driving cars will account for more than 40% of the total travel mileage, and the penetration rate of fully self-driving new cars will reach 10%. This is a huge market, and it may not only be an opportunity for the Cambrian, but also for many chip startups that have not yet entered the market.