Nvidia forced away a car company, self -developed chip arrow on the string ｜ Focus analysis

Just as there is only one Apple in smartphones today, Tesla's self -developed chip success is difficult to imitate.

Wen ｜ Li Anqi

Edit ｜ Su Jianxun

Cover Source | Visual China

GM is not the first, nor will it be the last self -developed chip car company.

Recently, according to Reuters, Cruise, a self -driving company of GM, is developing four chips at the same time. It is expected that it will be carried on Origin (no steering wheel or pedal) designed for unmanned driving by 2025.

Earlier, the first car company in the industry to abandon chip manufacturers who moved to self -developed chips was Tesla. Now GM also hopes to start this script.

In China, self -developed chips in car companies have become more and more trend. Since 2020, car companies including BYD, Geely, Zero Run, Weilai, ideal, Xiaopeng, Xiaomi have entered a self -developed chip model.

Some people in the industry speculate that the driving force for self -developed chip by car companies comes from the core crisis of the lack of in the past two years. After all, in April and May of this year, many car companies cannot even determine how much car production capacity will be next month.

It is understood that the most important thing in this round of core lack is the traditional MCU (microcontroller or single -chip microcomputer) class, responsible for the power conversion IGBT (insulation grid bipolar transistor) class, but these chips are not difficult to produce. In other words, compared to high -end chips that are completely stuck in the neck, the above chips can only be mass -produced in two or three years.

What the car companies such as GM and Wei Xiaoli want to seize are high -end chips that are more closely related to automobile intelligence. In this field, players such as Nvidia, Qualcomm, Mobileye are mainstream, and chip companies such as Huawei, Horizon, Black Sesame, and Xinchi are also struggling to catch up.

Since GM CRUISE's autonomous vehicle has been equipped with Nvidia's Orin chip, why does GM choose to develop a self -developed chip?

"Core disease" still needs "core medicine" medicine

According to Carl Jenkins, the head of Cruise hardware, two years ago, Cruise spent a lot of money to buy GPUs from a chip supplier, but because the number of models was small, the two sides could not negotiate at all. "So we must master our own destiny."

Although it is not clearly pointed out, it can be seen from the past cooperation resume that the chip supplier is Nvidia. With the 200TOPS (computing power unit) and the 7nm process of the Orin chip, Nvidia has already mastered a considerable right to speak in the field of autonomous driving chips. According to the official data of Nvidia, as of March this year, the company's car customers' orders in the next six years have exceeded $ 11 billion. Last year, the number was about $ 8 billion.

However, GM's autonomous driving company Cruise has only hundreds of autonomous vehicles, which is difficult to talk about conditions with Nvidia. This also triggers the idea of ​​universal self -developed chips.

It is understood that the four chips that GM CRUISE is studying themselves include: a central computing chip named Horta, sensor data processing chip, radar chip, and an unreasonable chip.

Although the person in charge of Cruise does not clearly disclose the investment in chip development, it will be diluted by the production of all unmanned vehicle vehicles equipped with chip -mounted chips. Cruise CEO Kyle Vogt said that self -developed autonomous driving chips will make the cost of autonomous vehicles reaching the best in 2025, and it can produce the model on a large scale.

Of course, if it is only used on the unmanned vehicle, the cost may be difficult. GM CEO Mary Barra previously said in a public speech that GM will develop a "personal autonomous car" within 10 years. Therefore, it is not ruled out that the possibility of four chips put in other models of GM.

Cruise's chip leader Ann GUI said that the self -driving chip Horta has been developed two years ago. At present, Cruise is working with an Asian chip manufacturer to produce its chips with a large scale.

It is important that GM saw the hope of self -developed chips, and Tesla successfully escaped the control of the chip company.

It is understood that in 2014, Tesla's first -generation self -driving hardware used Mobileye Eyeq3 chips. In 2016 and 2017, Tesla used Nvidia's Drive PX 2 and PX 2 upgraded versions. Eventually in 2019, Tesla completely got rid of Nvidia and will use its own FSD chip in the future.

It is reported that Tesla's self -developed FSD chip has increased from 24TOPS to 72TOPS from 24TOPS to 72TOPS compared with Nvidia Drive PX2 chips, but the efficiency of the same model of the same model has increased by 21 times.

This is why car companies have developed their own chips: In addition to reducing costs to reduce costs, they can better meet the autonomous driving software algorithms of car companies and improve system performance.

Yang Yuxin, chief marketing officer of Black Sesame Smart Smart, recently said at the media communication meeting: AI chips have two technical directions: one is the general GPU architecture of Nvidia, the other is Tesla, Qualcomm, and Mobileye. Huawei follows. ASIC (special integrated chip) technical route. The former is a universal architecture solution, and the latter is specifically targeted at autonomous driving. It often takes special consumption of the power consumption and performance needs of autonomous driving.

A car algorithm engineer given 36 氪 Example: For example, the autonomous driving software of the car company uses a certain image processing algorithm, but there is no chip on the market to accelerate the image algorithm or the cost is expensive. If you develop a self -developed chip, you can customize the special algorithm to accelerate the hardware unit in the chip. In this way, you can run your home algorithm with the self -developed chip, which not only runs fast, but also takes less time. As the vehicle is intelligent, the complexity of production is getting more and more than the perception of car companies, and car companies are willing to make more investment in the chip field. The international consulting agency Gartner has made predictions: Due to the shortage of chip and the electrification and automation of automobiles, half of the top ten automotive host plants (OEM) will independently design chips in 2025 to enhance their route map and supply chain to their own product routes and supply chains. control.

How difficult is it to be copied by the self -developed chip road?

Of course, for Chinese car companies, self -developed chips also have some factors for international political disputes. A few days ago, Nvidia's AI computing power chip A100 was banned from selling in China and attracted great attention from the industry.

Automobile companies have self -developed chips, and they have already been on the string. There are already a group of car companies in the industry to try first.

36 氪 has been reported exclusively in a row. Xie Yan, vice president of Huawei Consumer BG Software Department, and Huawei Terminal OS Minister, joined the ideal car and is responsible for the research and development department of the competent system. Chip, etc. Weilai Automobile's self -developed chip team is located under Bai Jian, vice president of Smart Hardware, has recruited a large number of personnel from Huawei, Alida Academy, Nvidia and other people, and has a clear streaming plan. Xiaopeng Automobile also established a chip research and development team with a scale of about 10 people.

New Energy One Brother BYD is also preparing for the intelligent driving chip research and development team. The 500 -person autonomous driving team also covers the talents in the chip field.

Earlier, the core engine technology jointly established by Geely Automobile and ARM China plans to launch a high -end autonomous driving chip AD1000. The single -chip computing power meets the requirements of ADAS L3+. It is expected to be commercial in 2024.

It is not difficult to see that the chip is becoming a very core component in the field of intelligent electricity. If this part is subject to people, it may also repeat the path of "controlling" the fuel vehicle era by international giants. Regardless of whether it is active or passive, the head car company hopes to obtain sufficient security through self -developed chips.

But as there is only one Apple in smartphones, Tesla's self -developed chip success is difficult to imitate.

This is obviously a very high investment and a long cycle. According to data from Semiengingeering, a semiconductor technology research institution, the investment in the development of 28nm node chips is US $ 51.3 million, the development cost of 16nm node chips has doubled to $ 100 million, and the 7nm node chip has reached US $ 297 million. The high -end chip required by car companies in the moment is basically about 7nm.

At the same time, the car's car -level product attributes also determine the chip from design, research and development, streaming, and mass production, which takes several years. Tesla spent more than 3 years from proposing a self -developed chip to the real chip to get on the car.

Recently, Yang Yuxin, the chief marketing officer of the domestic car chip company Black Sesame Smart Smart Smart, also expressed the same point at the media communication meeting: Tesla and Apple are the same. It is difficult for these subverters to get the best or most adapted resources in the industry. Only self -research can achieve what they want.

"But after the direction is verified, all the best people and the best resources in the supply chain are inclined in this direction. At this time, when you make similar products, you can always find better, cheaper, and more more Fast things (to replace). "Yang Yuxin said.

He believes that at this point in time, the chip is developed by self -developed chips. In addition to spelling with the most professional people in the industry, fighting resources, team management, and definition of products in the industry, the most important thing is that regardless of whether the chip is sold in the later period One million pieces, the investment of hundreds of millions of dollars in the early stage is indispensable. Auto companies need to develop their own determination to invest hundreds of millions of dollars per year, and consider whether the shipment can support R & D costs.

All in all, whether it is the desire for the industry's right to speak for the core technology of smart vehicles or the potential impact of international disputes on the automotive industry, car companies hope to use self -developed chips to obtain more chips. And after Apple and Tesla, the answer is still a question mark.

Let's "share

Self -developed chip arrow on the string

- END -