CEO Tips | Domestic Automotive-Grade Chips Are Striving to Catch Up from the "Chip Shortage", and the Future of Smart Cars is Full of Imagination.

In the era of electrification and intelligence, new energy vehicles have already become an inevitable trend. We all know that the important signs that distinguish smart cars from traditional cars are autonomous driving and smart cockpits. The realization of these functions is actually inseparable from chips as the underlying support. The demand for automotive chips in new energy vehicles is 1,600 per vehicle, which is three times that of traditional fuel vehicles. Why do smart cars need to use so many chips? Are more chips always better? Does the problem of "bottleneck" in domestic automotive-grade chips still exist? In the future, will a car be able to solve everything with just one CPU like a computer? For consumers, what revolutionary changes will the development of automotive chip technology bring to the driving experience?

With these questions in mind, the Hard Tech Special Session of the 36Kr "CEO Tips" live stream invited Zhong Jianfeng, the founder of Yitai Microelectronics, and Gao Feng, the co-founder of Ouye Semiconductor, to discuss together: What will the future smart car look like?

In this live stream, the two guests mainly discussed the following issues:

1. The demand for automotive chips in new energy vehicles is 1,600 per vehicle, which is three times that of traditional fuel vehicles. Why do smart cars need so many chips?

2. In the past few years, the market has suffered from a serious "chip shortage", but in recent years, Chinese chip companies have grown rapidly, and many excellent enterprises have emerged. What developments have domestic automotive-grade chips experienced? What stage are they currently at?

3. We know that there are many types of automotive-grade chips. Specifically, for the Ethernet chips and Soc chips that the two of you are working on, which parts of the car's functions are they controlling respectively?

4. One phenomenon is that new energy vehicles are definitely the strongest in our country, but in the fields of smart cockpits and smart driving, the core chips that are familiar to everyone are basically still using foreign ones. Does the "bottleneck" problem of domestic automotive-grade chips still exist?

5. As chip manufacturers, getting orders is particularly important. May I ask which domestic car companies the two of you are currently cooperating with? What is the specific form of cooperation?

6. Tesla's newly released Robotaxi does not have a steering wheel, accelerator pedal, or brake pedal. In the future, the car will be an intelligent carrier. With the development of chip technology, what will the future smart car look like?

7. Chinese car companies have made remarkable progress in electrification and intelligence, which has put forward many innovative requirements for the functions and performance of chips. In the future, will smart cars be able to solve everything with just one CPU like a computer?

The following is the conversation between the two guests and 36Kr, and some of the content has been sorted and edited:

36Kr: The demand for automotive chips in new energy vehicles is 1,600 per vehicle, which is three times that of traditional fuel vehicles. Why do smart cars need so many chips?

Zhong Jianfeng: Currently, the number of chips in traditional fuel vehicles is approximately between 600 and 700. In contrast, the number of chips in electric vehicles has significantly increased to 1,600, and the chip demand of advanced smart cars may even reach two to three thousand per vehicle. There are two main reasons for this increase in chip quantity: one is the growth brought about by the electrification of automobiles, and the other is the growth brought about by intelligence. Intelligence mainly includes two categories: intelligent driving and intelligent cockpits, and the underlying support of intelligent technology is realized through chips. Automotive chips can be regarded as the "brain" of the car, and their functions can be roughly divided into three categories: control, perception, and execution. Therefore, the increase in the number of chips in the car is mainly due to the complexity of the car's functions and the growth of intelligent needs.

Traditionally, the increase in the number of chips does indeed mean an increase in the car's functions. However, with the increase in the number of chips, some problems have also arisen. For example, the number of microcontrollers (MCUs) in the car has reached more than 100. We can regard the MCU as the "cerebellum" of the car. Humans have only one brain, but a car needs so many "cerebellums". This is due to the limitations of technological development and cost considerations. The current trend is to gradually reduce these "cerebellums" and instead use one or several larger "midbrains" or "brains".

To achieve these functions, computing power, sensors, and actuators are needed, and these are all realized through chips. From a technical perspective, we hope that the fewer chips the better, but the ultimate goal is to meet the needs of users.

Gao Feng: I would like to explore this issue from another perspective, that is, the value per vehicle. This value is a more fundamental change, and it is not limited to the number of chips, because there are many types of chips, including power analog, MCU, SoC, etc.

In traditional fuel vehicles, the total value of all semiconductors is approximately between 300 and 400 US dollars. However, with the electrification and intelligence of automobiles, this value is expected to reach the thousand-dollar level. The lowest prediction is also above 2,000 US dollars, and the higher prediction may reach more than 4,000 US dollars. This means that the value of the chips manufactured on a 12-inch wafer has increased from 300 to 400 US dollars to 2,000 to 4,000 US dollars, which is an order of magnitude increase. This trend is more important than simply looking at the number of chips. Why has this trend occurred? As Mr. Zhong said, the computing demand of automobiles has increased. Chips are essentially computing-driven, and the increase in computing has also driven the demand for passive components and storage. As the complexity of automotive computing increases, it consumes more semiconductors because the essence of chips is to provide computing functions.

Now, there are more and more intelligent sensors. A simple vehicle may only be equipped with one camera and a few radars, while a high-end vehicle may be equipped with more than a dozen cameras and multiple radars, and even more expensive lidar. These sensors, from perception to interaction, such as the cockpit, require a lot of computing. Traditional vehicles in the past were mainly mechanical products with the engine as the main body, and the computing required was very little, mainly mechanical control. After intelligence, in order to provide more experiences, including autonomous driving, the vehicle needs to perceive and compute independently, which makes the computing volume larger and larger.

This is the most fundamental change. Automotive chips are becoming more and more important in vehicles, changing from a relatively accessory product value form to a key factor that determines the vehicle's characteristics and even selling points.

36Kr: In the past few years, the market has suffered from a serious "chip shortage", but in recent years, Chinese chip companies have grown rapidly, and many excellent enterprises have emerged. What developments have domestic automotive-grade chips experienced? What stage are they currently at?

Gao Feng: We need to correct a misunderstanding, that is, the so-called "chip shortage" problem. In the past few years, what the automotive industry really lacked was not computing chips, but MCUs (Microcontroller Units) and some analog devices. This shortage was mainly due to the impact of the epidemic, which led to an increase in the requirements for inventory levels, as well as the sudden increase in the demand for consumer electronics and personal computers during the epidemic, which all led to changes in the inventory cycle. However, for the future-oriented automotive intelligence part, there is actually no "chip shortage" situation. If we roughly classify, the chips related to intelligence can be divided into two parts: one is the stock feature, mainly the MCU market; the other is the incremental market, mainly the intelligent computing market dominated by SoC (System-on-a-Chip).

In the era before intelligence, the chips that automobiles needed in large quantities were mainly MCUs. The global market pattern is very clear and stable, mainly composed of three companies from Europe and one each from the United States and Japan, such as NXP, ST, Infineon, Texas Instruments, and Renesas. The supply chain system of these companies is very stable. After the market stabilizes, they have entered a market feature similar to bulk commodities, which is why there are supply chain shocks and "chip shortage" problems.

On the other hand, the incremental market, that is, the intelligent computing chip market, is actually an incremental market. In the process of the value of automotive semiconductors increasing from three to four hundred US dollars to two to three thousand US dollars, most of the increment does not come from the increase in the use of traditional MCUs, but from the increase in the demand for new computing chips that carry on-board computing. Therefore, the bottleneck in this part is not a supply chain problem, nor is it the "chip shortage" problem we mentioned. Including the chips of some companies mentioned earlier, they are not to solve the so-called "chip shortage" problem, because the chips they need have never been unavailable.

Zhong Jianfeng: I would like to add some actual situations of the entire industry here. In fact, during the stage of traditional chip shortage, the core chips in intelligent driving and intelligent cockpits did not encounter a "chip shortage" situation.

Regarding the development stage of domestic automotive-grade chips, I would like to briefly add. We observe that the development of domestic automotive-grade chips can be divided into three stages:

Starting stage: At this stage, many domestic chip companies began to imitate foreign chips, including making some alternative products. This was a common practice in the past few years when the chip shortage was more serious.

Catching-up stage: With the accumulation of technology and the improvement of market awareness, domestic enterprises have begun to shift from imitation to creation, launching chips that can reach the same level as the industry in terms of performance, quality, and launch time.

Surpassing stage: At this stage, domestic automotive-grade chips will exceed the international level, lead the industry development, and even make breakthroughs in internationalization.

There are many types of automotive-grade chips, including those related to electrification, sensors, computing power, analog, and digital chips. According to statistics, in the domestic automotive chip industry, the localization rate of SoC (a computing power chip similar to the brain) is between 5% and 8%, the localization rate of MCU (similar to the cerebellum, a chip with less powerful technical computing power) is close to 10%, and the localization rate of power devices is 35%.

Specifically to the automotive Ethernet switch chips we are working on, the current localization rate is still zero. Therefore, in the entire domestic automotive-grade chip field, the stages of different types of chips vary greatly. Some are in the starting stage, some are in the catching-up stage, and some are about to enter the surpassing stage. Of course, more chips may still be in the starting stage.

36Kr: We know that there are many types of automotive-grade chips. Specifically, for the Ethernet chips and Soc chips that the two of you are working on, which parts of the car's functions are they controlling respectively?

Zhong Jianfeng: Let's talk about the user's experience of automotive intelligent driving. What users can feel is the transformation of the car from assisted driving to autonomous driving. For example, Tesla's recently released Robotaxi, which even does not have a steering wheel and brakes, has fully realized unmanned driving. For the product, the automotive Ethernet we are working on is actually building an information highway inside the car to provide high-speed data transmission for the car. Autonomous driving needs to process a large amount of data, including the transmission of sensor information and the execution of computing results, and these all rely on a high-speed information highway.

We can compare the automotive Ethernet chip to the information highway in the car. Without such an information highway, applications such as intelligent autonomous driving and intelligent cockpits cannot be supported. Therefore, we are actually providing the communication foundation. If we compare assisted driving to a channel, without this channel, instructions cannot be received in a timely manner, execution may be delayed, and even complete instructions may not be received.

Traditional cars also have communication technologies, but the bandwidth is low and cannot meet the large data volume required by autonomous driving and intelligent cockpits. This is like the 4G and 5G networks we are using now, while the communication technology of traditional cars may still be in the 2G era, obviously unable to meet the needs of modern cars. Now, automotive Ethernet technology enables our cars to achieve a communication level similar to 4G or even 5G, which is a huge leap from the 2G era to modern communication technology.

Gao Feng: Before explaining what SoC is specifically, I would like to put forward a point: Both Mr. Zhong and I used to be engaged in the IT industry, which shows that the essence of automotive intelligence is the integration of the IT industry and the automotive industry.

In the past, automobiles had a set of proprietary communication protocols and components, as well as an independent ECU hardware development model; while the IT industry focused on computing and communication. Now, automotive intelligence is to introduce the communication protocols and computing architectures of IT into the automotive industry. Simply put, the car will become a smartphone with wheels. Of course, this metaphor is not completely accurate, but it is accurate in terms of mobility and communication. Automobiles need communication and computing to support intelligent experiences. From this perspective, every smartphone needs an SoC, whether it is Kirin 9000 or Qualcomm 810, these chips determine the intelligence level of the phone. Automobiles may also need multiple SoCs to provide the various computing capabilities required, and this is the computing power basis we provide.

SoC is a collection of heterogeneous computing, which means that different computing tasks require different types of chip architectures to handle. For example, the CPU is a general-purpose computing mode, but it may not be efficient for specific tasks. This is why NVIDIA developed the GPU, which was originally used for graphics processing, but later found to be suitable for AI and deep learning computing. The GPU and CPU can work together and divide the work to improve computing efficiency.

There are many different computing cores on the SoC, such as CPU, NPU (for large-scale parallel computing for AI computing) and GPU (for graphics processing). These cores are integrated onto a single chip and effectively integrated through a bus configuration to form a system-on-a-chip, that is, SoC (System on a Chip). SoC is a computing system on a chip that can perform the most efficient computing for a specific scenario.

36Kr: One phenomenon is that new energy vehicles are definitely the strongest in our country, but in the fields of smart cockpits and smart driving, the core chips that are familiar to everyone are basically still using foreign ones. Does the "bottleneck" problem of domestic automotive-grade chips still exist?

Zhong Jianfeng: In recent years, we have made significant breakthroughs in many fields, especially in the automotive industry. Compared with three years ago, the entire vehicle industry has made tremendous progress. The penetration rate of new energy vehicles has reached 50%, while it was only about ten percent three years ago. This shows that the domestic vehicle manufacturing capacity has been enhanced with the rise of new energy vehicles.

In the chip field, significant progress has also been made in recent years. Many fields have achieved a transformation from nothing to something, and many chip manufacturers have emerged, launching many products. This is particularly evident in the two main fields of intelligent cockpits and autonomous driving. Some car manufacturers have launched commercial automotive-grade chips. For example, Horizon and Black Sesame in the field of autonomous driving, and Sinchip and Huawei in the field of intelligent cockpits, have all made certain breakthroughs.

Nevertheless, there are still some bottleneck problems, which are related to the overall development stage of domestic automotive-grade chips. Due to the wide variety of chips, there are significant differences in the competitive situation of different types of chips in the market. Some fields no longer have bottleneck problems, but in some fields, such as power control in the high-safety field, due to safety concerns, manufacturers will be more cautious when using them, and it requires a long time of accumulation, so the penetration rate is still low. In the communication field, due to the high technical threshold and the small number of foreign manufacturers, although the fields of new energy vehicles, autonomous driving, and intelligent cockpits are developing rapidly, in the field of automotive Ethernet, foreign chips are still being used. This indicates that in some specific fields, we still face certain bottlenecks.

Gao Feng: In the new field of automotive intelligent computing SoC, we first need to recognize that automotive intelligence is still in its infancy and far from the time when the market pattern is determined. Intelligence has just begun. Although the trend of electrification is relatively clear, including the motor, electronic control, and battery technologies are tending to be standardized, Chinese suppliers have achieved a good market position in this regard, such as CATL and Guoxuan High-Tech.

However, the field of intelligence has just started. We call it the first half of electrification and the second half of intelligence. The exploration of intelligence is still ongoing. Even a company like NVIDIA, which has not yet established a clear advantage. NVIDIA's chip sales in the automotive field last year were only more than one million, accounting for less than 3% of its total revenue, and its share in the global automotive chip market is also very small. The global annual production of 80 million to 100 million vehicles, NVIDIA's sales are still in a very early stage compared to this. Moreover, NVIDIA's automotive chips are more transformed from the data center shared architecture, and are not specifically designed for automobiles. Qualcomm's main business was previously in the mobile phone field, but with the shrinking of the mobile phone market, Qualcomm is also looking for new growth