Black Sesame Smart Yang Yuxin: Self-driving cars accelerate mass production, and large-scale in-vehicle computing chips need to be “licensed” to board the car

In the first half of this year, the field of car manufacturing has been very lively. On the contrary, the auto industry has been facing the pain of “core shortage” since the second half of last year.

Now that 2021 has passed halfway, what is the “core shortage” situation of the domestic auto industry?

With this question in mind, at the recently concluded 2021 WAIC World Artificial Intelligence Conference, Meike.com interviewed Mr. Yang Yuxin, the company’s CMO, at the Black Sesame Intelligent Technology booth focusing on providing car-level large computing power chips.

In Yang Yuxin’s view, the current “lack of cores” caused by the shortage of production capacity is only temporary, and the “lack” of domestic autopilot-related chips will cause a mid-to-long term “lack of cores” in the domestic automobile market.

Car-level chip certification is strict, and domestic “car core shortages” exist in the mid-to-long term form

As we all know, if a chip wants to be “on the car”, it must pass the “car-level” certification.

But what is a car-level chip? Faced with this problem, Yang Yuxin first pointed out that, unlike ordinary chips, if a company wants to design car-level chips, it needs to “hold a certificate.”

“The team is going to design automotive-grade chips. First, it must go to the International Automobile Electronics Association for design training, and then get a certificate to prove that the team has the ability to design automotive-grade chips.”

After this, the type of IP adopted by the chip, the production and packaging of the chip, all need to adopt products and production lines with car-level certification.

Among them, in the field of IP, in addition to outsourcing IP authorization, Black Sesame Intelligence has also rarely selected two self-developed IPs in the industry-car-grade image processor NeuralIQ ISP and DynamAI NN car-grade low-power neural network acceleration engine.

“We believe that these two types of IP are the core IPs of autonomous driving chips,” Yang Yuxin said. “For example, in image processing, automated driving cars must first be clearly understood before they can understand them. It is also getting higher and higher. When the amount of data becomes larger and larger, we have to ensure the accuracy of the perception results, so we must improve the quality of the data.”

At the same time, “the car’s’understanding’ means that it has to make inferences based on massive amounts of data, so the neural network accelerator must also be powerful enough, and the car has the highest requirements for inference calculations.”

Based on the above two considerations, Black Sesame Intelligence chose a self-developed IP, which is a certified car-level IP.

After the release of the Black Sesame Intelligent Huashan No. 2 A1000 autonomous driving chip, it passed the ISO26262 functional safety product ASIL B Ready certification in July 2020, and passed the functional safety process certification of the highest automotive safety level ASIL D with zero deviation in June this year.

Huashan No.2 A1000 is currently the first domestically produced high-performance autonomous driving chip based on a mature vehicle-specific functional safety system. A single chip meets ASIL B level requirements and supports ASIL D level safety islands.

It is reported that ISO 26262 is a globally recognized automotive functional safety standard. It evaluates the entire life cycle of a product. It is one of the entry barriers for global Electronic component suppliers to enter the automotive industry, and it is also one of the necessary conditions for the development of automated driving mass production. .

Only by obtaining this type of certificate does it really mean that this chip can be called a “car-level chip.” Prior to this, the chips developed and produced in accordance with the car-level standards can only be said to have the qualifications for the certification of the car-level standards.

However, in the context of the surge in orders for chips from industries such as global automobiles and consumer electronics, foundries led by Taiwan semiconductor Manufacturing Co. have already been overloaded. In the chain reaction, there has also been a “global core shortage” situation in the short term.

In this regard, Yang Yuxin said that Black Sesame Intelligence currently does not have a deep sense of the threat of “core shortage”.

But for this phenomenon, he added that the “core shortage” caused by insufficient production capacity is only a short-term phenomenon, and from the perspective of the domestic automotive chip industry, there is also a long-term “core shortage”.

How do you say this?

“What is lacking right now are ready-made mature process chip products that have been developed, similar to single-chip microcomputers, which are short-term. What is lacking in the mid-to-long term is the autopilot chip that is not available now and that everyone will need in the future.”

Yang Yuxin pointed out that China’s smart car industry is now at the forefront of the world, but it lacks new chips that support Chinese automakers to quickly develop smart driving technologies, such as autonomous driving chips and smart cockpit chips.

Behind this, there are both a long time for certification of car-level chips (at least two years from design to certification) and a small number of domestic car-level chip manufacturers. “At present, there are only a handful of companies in China that are really doing self-driving chips.”

Yang Yuxin further added: “As the first domestic company to pass the ASIL D-level autopilot chip, Black Sesame Intelligence has the world’s most advanced autopilot and car-regulated chip technology. We provide better technology and more flexible support. The system helps domestic automakers to apply the world’s leading autonomous driving technology to their products better and faster.”

In this way, when emergencies such as epidemics and disasters occur again later, “Chinese automakers do not have to worry about the lack of cores because they have an independent supply chain system.”

Self-driving cars require large computing power, and it will take at least 5 years to achieve real L4

After participating in WAIC, Meike.com can clearly feel that the proportion of self-driving cars on Display is gradually increasing.

Once upon a time, most of the booths were arranged by autopilot manufacturers outside the exhibition hall. In this exhibition, it can be said that they occupy at least a quarter of the entire exhibition venue, ranging from clean cars to large trucks.

At the same time, Meike.com also learned after talking at the major booths that “realizing L4 autonomous driving” has almost become the external “standard caliber” of autonomous driving solution manufacturers, whether it is for autonomous driving rental for public travel. Vehicles are also self-driving trucks used in the logistics and transportation industry.

However, in Yang Yuxin’s words, the current solution does achieve L4 level autopilot functions, but there is still a long way to go before “really on the road”.

Of course, we can also see that on some urban roads, including Baidu, many manufacturers have launched their own autonomous driving fleets and have conducted open tests in some areas.

“L4 level involves open roads and needs to cover a variety of different scenarios, and the pilot is only running in a specific area. After all the high-definition maps are completed, the autonomous driving fleet will adapt to the environment of this area to respond to each A possible scenario to ensure the safety of unmanned driving in the pilot area.” Yang Yuxin said.

As for when the L4 level can actually be implemented, Yang Yuxin also gave his own prediction: at least 5 years.

For high-level autonomous driving such as L2+/L3/L4, Black Sesame Intelligence has released the Huashan series of large computing power autonomous driving computing chips, including the Huashan No. 2 A1000, and the enhanced version of the large computing power chip Huashan No. 2 A1000 released in April this year. Pro.

The Huashan No. 2 A1000 Pro has a computing power of 106-196 TOPS and a power consumption of only 25w. The computing power can support high-level automatic driving functions, seamlessly connecting from parking, urban interiors to high-speed scenes. Just a few days ago, the A1000 Pro autopilot chip has also been successfully taped out.

Yang Yuxin said that according to the industry’s estimation of the computing power requirements for each stage of autonomous driving, the L2 level requires about 10+ TOPS, the L3 level requires 100+ TOPS, the L4 level is expected to require 500+ TOPS, and further upwards may require 1000 TOPS. The maximum computing power that physics can currently achieve.

However, he also said to Meike.com, “For example, the 500 TOPS required for L4 is based on the estimated value of the current coverage scenario demand. It does not mean that when the autonomous vehicle drives to a fully open road, 500T Will suffice.”

Under this premise, all manufacturers can do is to increase the computing power of the chip as much as possible.

finally

Autonomous driving has become an inevitable trend of automobile revolution. Visible to the naked eye, whether it is a new energy vehicle or a traditional gasoline vehicle, whenever a new car is launched, the intelligent attributes represented by autonomous driving are one of the main functions.

In this regard, Yang Yuxin also pointed out that in view of factors such as laws and regulations, determination of the person responsible for the accident, although some functions of these auto products have reached L3 in function, the actual sales process is still L2.5 and L2+.

“At the moment, everyone is still polishing their product solutions in accordance with higher-level technologies.”

At the customer level, Black Sesame Intelligence has reached cooperation with FAW, Dongfeng, SAIC, and Guangzhou Automobile. Yang Yuxin also revealed that as soon as the second half of 2022, there will be mass-produced models equipped with black sesame intelligent car-level autonomous driving computing chips on the market.