For the entire automotive industry chain, autonomous driving is both a red ocean and a blue ocean.
Research AndMarkets, a market research firm, expects the global advanced driver assistance systems (ADAS) market to grow at a CAGR of approximately 13% between 2021 and 2026. In the next five years, this market is expected to reach a scale of nearly 100 billion US dollars, with in-vehicle AI chips (SoCs) as the main incremental market.
In the context of the great change of automobile intelligence, the traditional distributed architecture is difficult to meet the upgrade requirements, and the centralization of automobile control is the trend of the times, and the product layout of domestic AI chip leaders is also the same. At present, manufacturers including Huawei, Horizon and Black Sesame Intelligent Technology have released their own automotive grade SoC chips.
Unlike MCUs that focus on CPU computing, autopilot SoC chips generally integrate components such as CPUs, GPUs, and neural network accelerators and MPUs for AI inference computing. In addition, based on various I/O interfaces, The SoC enables greater functionality.
Is computing power the first priority of the autopilot master soC?
Under the market competition situation of a hundred flowers, the computing power of a new generation of chip products is constantly improving. The market generally believes that only by leaving more redundancy in computing power can it provide more space for the subsequent needs of "hardware embedding and OTA upgrade".
It is reported that the Huashan No. 2 A1000 chip is currently in mass production in automatic driving applications and has the highest computing power of soC products. At the same time, the Huashan No. 2 A1000 Pro released by Black Sesame Intelligence last year, int8 computing power was further increased to 106 TOPS, INT4 computing power 196 TOPS, continuing to maintain the highest domestic hashing power autonomous driving chip position.
As autonomous driving continues to upgrade and the concept of "software-defined cars" becomes the consensus of the industry, OEMs are increasingly demanding SoC chips with higher computing power, higher throughput, lower latency, and low power consumption. In fact, the development trend of the industry has also forced chip manufacturers to step up the development of high-computing Power SoC chips.
With the continuous and in-depth integration of AI technology and the automotive industry, the era of co-evolution of autonomous driving SoCs and algorithms has arrived. In the short term, the improvement of chip computing power is still the first priority of the upgrade of the autopilot master soC.
Black Sesame Intelligence pointed out to Autolab that the co-evolution of chips and algorithms should not be a bundle relationship, but rather that chips give algorithms more degrees of freedom and liberate the shackles when the algorithm is implemented. If the chip itself has insufficient computing power and low openness, the synergy between the chip and the algorithm can only play a role in cutting enough.
Therefore, as the cornerstone of algorithm implementation, SoC chips should be based on large computing power and high compatibility, while providing an open and flexible development environment to minimize the time and cost of algorithm adaptation chips.
How difficult is it for the autonomous driving master SoC to achieve mass production of models?
According to such standards, what are the challenges of designing an autonomous driving AI chip that meets market demand?
"Chip car specification requirements, architecture planning, computing power planning, chip area, cost, power consumption, etc.", Black Sesame Intelligence pointed out for example, the lack and defects of any link may lead to the chip ultimately unable to be mass-produced or unable to meet the future application needs of customers after mass production, and the biggest challenge is still in mass production.
In order to achieve mass production of a vehicle-grade chip, in addition to the design process of the upstream chip factory, it depends more on the cooperation between the chip factory and the oem. Including: whether the car company is willing to adopt, how much the adoption rate is, and how effective the feedback is.
According to data, in the first eight months of 2021, more than 1.31 million Chinese self-owned brand cars were equipped with ADAS systems, an increase of 88.6% year-on-year, and the assembly rate reached 25.3%, an increase of 5.7 percentage points over the same period of the previous year. In terms of mainstream technology applications, the assembly volume and assembly rate of L2 ADAS increased by 153% and 6 percentage points respectively over the same period last year; the assembly volume and assembly rate of L2.5 ADAS increased by 227% and 0.8 percentage points respectively. In contrast, only 17,000 vehicles were fitted with L3-level ADAS solutions, with an assembly rate of 0.3 percent.
Under the premise that the technology meets the market requirements, the higher the market's acceptance of autonomous driving technology, the volume of the entire industry will increase significantly. In addition to the iteration of the product itself, factors such as supporting road end facilities, policy support and consumer purchase willingness are particularly important for the commercial landing of autonomous driving.
It is foreseeable that the domestic robotaxi industry is gradually entering a better situation, which is expected to bring new development windows to autonomous driving technology. It is revealed that the black sesame intelligent main control AI chip will achieve mass production of models in 2022, and the domestic autonomous driving supply chain system is gradually maturing.
Perhaps people are more concerned about how long it will take for drivers to completely free their hands than when cars will be able to fly, and how to solve the "last mile" problem of commercial landing of fully autonomous vehicles...
Is there a way out by taking the Tesla full-stack closed system self-developed route?
At this stage, perhaps no one can give a positive answer, in this new field, everyone is feeling the stones to cross the river. And in this process, there must be success and failure in the attempt.
At a time when the global automotive industry is facing a crisis of lack of cores, Tesla relies on its strong internal software research and development capabilities to quickly apply alternative components through reprogramming, ensuring the safety of the supply chain and the stability of delivery. This has also made more and more car companies begin to realize the importance of self-research.
However, Black Sesame Intelligence believes that the self-development of full-stack closed systems is difficult to be copied in every industry. Apple appeared in the mobile phone industry, Tesla appeared in the automotive industry, and the development model chosen by the two companies is very closely related to the stage of industrial development.
In cooperation with Qualcomm and Intel, Apple turned to self-developed chips not only to save costs but also to achieve higher performance, for the oligopolistic mobile phone / computer processor market, Apple is a well-deserved "game breaker". Similarly, Tesla has almost single-handedly thrown off traditional fuel vehicle companies and sat on the top of the new energy vehicle sales throne.
The above two classic cases have allowed more car companies to spontaneously try to establish a closed-loop supply system with themselves as the core. News about car companies setting up chip subsidiaries has been heard endlessly in the past two years.
However, unlike Tesla's "game-breaking", the automotive industry has developed in the direction of electrification, networking and intelligence, and the direction and goal of the entire automotive industry chain are clear and consistent. Therefore, "according to the logic of industrial development, it is basically difficult to appear full-stack closed self-developed car companies." ”
What is clear is that going it alone is no longer applicable to the current situation. In the face of rapidly changing market demand and technological upgrading, only cooperation can quickly respond and seize market opportunities in a timely manner.
As Black Sesame Intelligence said, even if 60% of the design of a SoC chip is generic, it still accounts for about 40% of the differentiated competition. Compared with self-research, oem and chip factories can promote product listing at a faster speed and lower cost through customized cooperation; for chip factories, they can also accumulate more experience in the car to accelerate product iteration and optimization, and further empower car manufacturers to upgrade products.
Write at the end
Standing at the cusp of autonomous driving, new opportunities are already in front of China's entire automotive industry chain. Represented by the new forces of car manufacturing, domestic electric vehicle brands are rapidly rising, and the intelligent change is also expected to become the "east wind" for domestic chip manufacturers to break through.
Facing the era of fully autonomous driving, Black Sesame Intelligent already has a layout of chips and related solutions, including chip computing power angle, architecture angle, software angle, and the ecosystem that needs to be cooperated with for automatic driving, such as sensor partners.
Black Sesame Intelligence believes that with the help of this wave of development opportunities, domestic manufacturers are expected to reach a leading position in the field of global intelligent new energy. To achieve this goal, it is necessary to cooperate with the upstream and downstream of the industrial chain, especially to obtain the active support of technological innovation enterprises. For the development and popularization of the entire autonomous driving technology, the interaction between upstream chip factories and downstream automakers and the common exploration and implementation practice at the technical level will play a very critical role.