On August 9, US President Biden officially signed the Chip and Science Act (hereinafter referred to as the "Chip Act"); on October 7, the US Bureau of Industry and Security (BIS) announced the latest sanctions bill, "Imposing New Export Controls on Advanced Computing and Semiconductor Manufacturing Items Exported to China".
Foreign media also claimed that the US wanted to send China's semiconductor industry back to the "Stone Age" with this ban.
The latest bill shows that the scope of the US restrictions on Chinese chips is very narrow, but very precise, with the aim of limiting the development of advanced chips and supercomputers in China.
If we were still in the era of gasoline-powered cars, the impact of chips on automobiles might have been negligible. However, in the era of electrification and intelligence, autonomous driving chips and smart cockpit chips, as the core chips for automotive intelligence, will inevitably advance in terms of process technology and computing power. It is only a matter of time before they fall under the restrictions imposed by the United States.
If we follow the intended purpose of the bill, the United States hopes that Chinese AI chip design companies will face the embarrassing situation of having no tools available and no foundries to process their chips, thereby ultimately forcing them to withdraw from market competition.
Electric Vehicle Observer consulted several industry experts and technical leaders of leading automotive electronics solution suppliers. When discussing the impact on chips, they were divided into two camps: one was pessimistic, believing that China's smart electric vehicles may lose competitiveness in the future; the other was optimistic, believing that chip blockades would force industrial development, and that mainland China might be able to establish its own high-end chip supply capabilities within ten years.
01、
The significance of chips for intelligent electric vehicles can be best summarized by Li Jun, an academician of the Chinese Academy of Engineering and chairman of the China Society of Automotive Engineers: "Software-defined vehicles, data-driven vehicles, and chip-manufactured vehicles."
In a dialogue between Zhao Fuquan, professor at the School of Vehicle and Transportation Engineering and director of the Institute for Automotive Industry and Technology Strategy at Tsinghua University, and Li Jun, Zhao Fuquan further elaborated on the significance of chips to automobiles: "Chips carry the ability to process data and run software; otherwise, functions such as autonomous driving would be impossible, and even the best automotive hardware would not be able to play a significant role. As the brain of intelligent vehicles, chips will become the core of future automotive products. At the same time, chips will also become a key link in intelligent automotive manufacturing."
If traditional cars are similar to traditional feature phones, then smart cars are equivalent to the widely used smartphones of today, requiring software iterations and complex logical data processing capabilities.
Perhaps it can be summarized in simpler terms: "No chip, no intelligence."
Intelligent driving has led to a surge in demand for semiconductor devices, including vehicle controllers, analog chips, main control chips, power semiconductors, and memory chips.
Firstly, regarding the controllers, according to GAC R&D Center's predictions, traditional automobiles typically have 40-70 controllers and 400-700 chips. New energy vehicles are expected to have 45-80 controllers and 500-800 chips.
Secondly, the demand for analog chips is increasing. Intelligent driving requires the acquisition of large amounts of data through sensors. Real-world signals are extracted by sensors and converted into analog signals. These analog signals need to be processed by analog chips before they can be used by digital chips. The role of analog chips is to convert analog signals into digital signals, or vice versa.
A recent analysis by Deloitte shows that Level 4 autonomous driving can require up to 29 sensors, while Level 5 will require up to 32, thus significantly increasing the demand for analog chips.
Third, the computing power of the main control chip is improving rapidly. With the increase in intelligent sensors, the amount of data collected in the future will be enormous. From low-level L1 and L2 to high-level L3 and L4, the computing power requirements are increasing exponentially, and some new car models have already pre-installed computing power of over 1000 TOPS.
Fourth, in the field of power semiconductors, the value is also doubling. Data from TF Securities shows that the value of power semiconductors per vehicle in traditional fuel vehicles reached US$87.6, while the value per vehicle in new energy vehicles reached US$458.7, representing a more than fourfold increase.
In addition, the demand for memory chips is growing rapidly. As automotive storage systems become more intelligent, their capacity and performance will also experience rapid growth.
02、
After communicating with chip design company personnel and several industry experts, Electric Vehicle Observer learned that the US chip restrictions will not have a significant impact on China's automotive industry in the short term, but the long-term impact cannot be ignored.
To analyze the magnitude of the impact, we must first examine what specific restrictions the United States imposed.
According to the minutes of a conference call held on October 13 by the U.S. Department of Commerce’s Bureau of Industry and Security (BIS), the U.S. government is seeking a “leadership” in advanced logic and memory chips, rather than the previous “relative advantage.”
To maintain this "leading advantage," a series of specific measures have been taken, including restrictions on high-performance chips, supercomputer chips, and semiconductor manufacturing equipment.
In addition, there are many restrictions on chip companies and employees. Providing restricted services to China is essentially a presumption of guilt. Americans are not allowed to participate in business related to China's chip industry, including not only American citizens, green card holders, and asylum seekers, but also people on the payroll of American companies.
For example, the US has previously required Nvidia to suspend shipments of its A100 and H100 GPUs to mainland China, and AMD to suspend shipments of its MI200 GPUs. Furthermore, all other chips with peak computing power and inter-chip I/O performance equal to or greater than the A100, as well as any systems containing these chips, will be subject to export controls unless an export license from the US Department of Commerce is obtained.
NVIDIA A100 GPU
For example, chip design companies that design supercomputing chips in mainland China, wafer foundries that manufacture supercomputing chips, and overseas wafer foundries that use US technology to manufacture chips for Chinese supercomputing chip design companies will all be subject to US export controls.
For example, all Chinese mainland chip manufacturers will be subject to restrictions, and the scope of these restrictions will be expanded from 10nm and below to 16nm or 14nm logic chips. SMIC's expansion of its 28nm and above process technology will also require the procurement of US semiconductor equipment, which will need to undergo a rigorous US review process and obtain a license before it can proceed.
Currently, the mainstream size of automotive chips is generally between 14-40nm, which is basically within the range of mature manufacturing processes.
Analog chips, in particular, are mostly manufactured using mature processes, and their technology and products are relatively mature. Although many of them still rely on companies such as Samsung and TSMC for production, they are not within the scope of this US restriction.
However, digital chips, which are most closely related to intelligence, are the fastest-iterating products and have high requirements for computing power and manufacturing processes.
Li Zhaolin, a tenured professor in the Department of Computer Science and Technology at Tsinghua University and chief chip expert at the National New Energy Vehicle Innovation Center, told Electric Vehicle Observer that the current common manufacturing process for autonomous driving SoCs is 16/14nm, while smart cockpit SoCs are mostly 7-10nm.
Most domestic companies' intelligent driving chip products are still within the mature process range. Companies like Horizon Robotics have J5 chips manufactured at 16nm, which is below the US limit and can still be produced and used normally.
Even for high-end chips, the computing power of mass-produced products has not yet reached the range restricted by the United States, and therefore they are temporarily safe.
According to Li Zhaolin, the US bill currently does not directly affect automotive MCUs, autonomous driving SoCs, and other chips. However, in the long run, the bill will have a significant impact on the development of intelligent and connected vehicles, since the function and performance of chips have an increasingly important influence on the development of intelligent and connected vehicles.
03、
Smart cockpit SoCs and autonomous driving SoCs conform to the characteristics of digital logic chips.
The characteristic of logic chips is that they emphasize the ratio of computing speed to cost, and new designs or processes must be constantly adopted.
Generally speaking, the manufacturing process affects the chip area, and thus directly affects the chip price. By increasing the chip area, more transistors can be placed in a chip.
SoC chip
According to Intel's Moore's Law, the strength of a chip depends on the number of transistors it has. The more transistors there are, the stronger the circuit's logic control over current becomes, and the more and more powerful the chip can perform.
Furthermore, chips composed of the same number of larger transistors have larger wiring connections due to their larger area, which increases signal transmission delay and thus reduces computing efficiency.
Therefore, for every 1nm improvement in chip size, the overall performance increases by approximately 30%.
Besides performance improvements, advancements in manufacturing processes also reduce manufacturing costs. It's important to note that XXnm refers to the gate width of transistors. Higher process technologies allow for smaller gate widths, enabling transistors to be denser, resulting in a smaller chip area and less wafer space required. This further saves wafer space, reduces chip manufacturing costs, and ultimately increases the net profit of chip manufacturers.
Therefore, both chip designers and chip manufacturers will try their best to improve chip manufacturing processes.
Logic chips have the highest market share, the fastest product updates, and are also at the forefront of manufacturing processes.
The latest high-end SoC chips are manufactured using a 5nm process, while dedicated SoC chips, such as smart audio chips, are generally manufactured using a process between 16-55nm.
Currently, Horizon Robotics will be directly affected, as its next-generation J6 chip, with a computing power of 1000 TOPS and using a 7nm process, has the potential to become a central computing platform for vehicles. However, whether this chip will actually be released in the future remains highly uncertain.
Is it impossible not to improve the manufacturing process?
No. Because a more advanced manufacturing process means lower manufacturing costs and stronger performance. In the words of Zhou Yanwu, Research Director of Zoson Technology, "(If someone's process technology is 10 times more advanced than yours, but the price is 1/10 of yours, what can you do?)"
In other words, while there is still room for improvement in the manufacturing processes of overall intelligent driving SoC and cockpit SoC, the stagnation of China's chip manufacturing process means that Chinese chip products will lack international competitiveness and will be eliminated.
Li Zhaolin stated that China already possesses manufacturing processes such as 142840nm, and these processes will continue to be the primary technologies used in automotive core chips for the next 3-5 years.
Of course, the expansion of 14nm production capacity has also been limited.
According to Li Zhaolin, autonomous driving SoCs will not blindly pursue more advanced manufacturing processes because, in addition to performance factors, factors such as cost, reliability, and safety must also be considered.
From another perspective, this window of opportunity to balance performance, security, cost, and reliability is also a golden window of opportunity for China's chip industry to catch up and surpass others.
04、
Next, let's look at the impact of limiting Nvidia's A100 and H100 on smart electric vehicles.
The Nvidia A100 and the soon-to-be-shipped H100 chips, whose exports are restricted, are primarily deployed in high-performance data center servers for large-scale AI training and other high-computing applications in the cloud, and are crucial for AI system training. A supply disruption of these high-end chips would slow down AI training, significantly impacting fields like autonomous driving that demand high AI computing power.
Reports indicate that Chinese new energy vehicle and autonomous driving technology companies such as NIO, XPeng, Li Auto, WM Motor, SAIC, Jideo, and Pony.ai are all using NVIDIA's AI chips.
However, in Zhou Yanwu's view, most car companies do not need to build their own data centers, as the cost is too high. Using public cloud or hybrid cloud models is more cost-effective.
Of course, data security might be another issue at this point.
Moreover, AI training chips are not short-term consumables; stockpiling more is sufficient to meet expansion needs for several years. For example, the chairman of XPeng Motors believes that the company's current inventory can meet its needs for the next few years.
Furthermore, Zhou Yanwu also stated that there is no shortage of relevant alternatives on the market.
Nvidia itself has already released a replacement. On November 8, Nvidia announced that it will launch a new A800 GPU in China to replace the A100. This chip complies with US export control regulations.
To comply with recent U.S. export control regulations, the performance of the A800 is certainly far inferior to that of the A100. The A800's chip interconnect data transfer rate is 400 GB per second, lower than the A100's 600 GB per second. For large-scale data centers, the operating speed will be reduced.
A domestic company also had a chip that was expected to replace Nvidia's A100, but production was halted. This chip is the BR100 from Biren Technology. The discontinuation of this chip is indirectly affected by the US "chip law": TSMC decided to suspend the production of advanced chips for the Chinese startup Biren Technology to ensure compliance with new US regulations.
Biren claims that its 7nm AI graphics processor, BR100, uses Biren's original "Biren" chip architecture, which contains nearly 80 billion transistors and is three times more powerful than NVIDIA's A100 chip, achieving one trillion calculations per second.
According to Taiwan's Liberty Times, Shanghai Biren Intelligent Technology, which was suspended from supply by TSMC, has been severely impacted and is reportedly planning to lay off one-third of its workforce.
Therefore, it is clear that chip companies are the most directly affected, while the impact on car companies is not yet significant.
05.
However, in the long run, it will inevitably have an impact on China's intelligent electric vehicle industry.
When discussing the impact on China's intelligent electric vehicle industry, experts' attitudes are divided into two camps: pessimistic and optimistic.
Pessimists believe that for Chinese intelligent electric vehicle manufacturers, the lack of an independent and controllable supply chain means another bottleneck. Companies that want to develop in the field of intelligent technology will face the problems of being controlled by others in the industrial chain and high costs, thus losing their international competitiveness in intelligent technology.
"The United States could cut off your supply at any time through TSMC, or postpone your supply," an industry insider said.
Because China's chip industry chain is positioned at a relatively low end, its dependence on foreign countries is very high in areas such as chip manufacturing equipment, EDA tools, and chip IP cores. Achieving self-sufficiency and control over the industry chain is not easy. "China faces the risk of a generational gap in its higher-process chip technology."
However, Mr. Huang, the technical head of a leading automotive electronics solutions supplier, is relatively optimistic. Mr. Huang believes that while the blockade will force development, short-term pressure is unavoidable, but it will also help domestic companies create a complete closed loop. When downstream users begin to choose chips designed and manufactured in China for supply security, it will greatly promote the development of the industrial chain and technology.
In Mr. Huang's view, chip manufacturing is actually an engineering problem. "We've already overcome the Long March, the Nanniwan Campaign, and the Shangganling Campaign. Engineering problems can definitely be solved, maybe in less than ten years."
The long-term blockade imposed on China by the West has made China the only country in the world with a complete industrial system. The US blockade of high-end chips in China might actually help China establish a complete chip industry chain.
