betting on high-level autonomous driving, car companies build chips and lay out software and hardware integration
2024-09-14
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
source: securities times
on august 27, at the 10th anniversary event of xpeng motors, he xiaopeng, the founder of xpeng motors, launched the company's self-developed chip "xpeng turing".
he xiaopeng said that the chip has recently been successfully taped out (trial production) and can be used in robots, ai (artificial intelligence) cars and flying cars.
coincidentally, at the nioin2024 event held at the end of july, nio founder li bin also "personally endorsed" and announced the successful tape-out of the company's independently developed world's first 5nm smart driving chip, the shenji nx9031.
unlike the panic stockpiling and investment layout of automakers caused by the shortage of automotive chips a few years ago, this round of chip manufacturing by automakers is more proactive and mainly focused on the field of autonomous driving chips.
layout ideas show differences
it is not a new topic for car companies to make their own chips. in 2020, the global chip supply was tight, causing some car manufacturers to be unable to deliver products on time due to chip shortages.
the short-term imbalance between supply and demand has alerted more and more automakers. if the independent control of key components such as chips is ignored, the industrial chain will face the problem of "stuck neck" in the medium and long term.
based on this, starting from 2021, automakers such as saic, dongfeng motor, geely auto, and great wall motors began to invest in the chip industry on a large scale, and their products covered body control chips, smart cockpit chips, and autonomous driving chips. among them, there are many automakers that invest in control chips such as mcu, igbt, and sic.
the entry of car companies has stimulated the enthusiasm of many start-ups to make chips. from the shortage of supply to the rush of companies to layout, automotive chips have become a hot track.
yuan chengyin, director and general manager of the national new energy vehicle technology innovation center and secretary general of the china automotive chip industry innovation strategic alliance, told the securities times reporter that although there are many players, few have achieved large-scale development. the advantage of car companies making chips is that they can better combine application scenarios and make targeted chips.
in fact, this trend of building "chips" around application scenarios is becoming more and more obvious. some car companies have begun to shift from investing in control chips to specializing in autonomous driving chips.
tesla is a pioneer among many automakers in developing its own autonomous driving chips. before 2016, tesla's smart driving hardware system mainly used chips from mobileye and nvidia. however, tesla ceo musk believes that the power, performance, cost and delivery speed of the products of the above two companies cannot meet tesla's long-term needs.
in 2019, tesla's self-developed fsd chips were installed in large quantities in its vehicles, which not only reduced the company's dependence on nvidia, but also made the hardware more compatible with the self-developed software algorithms, providing better compatibility and matching, further improving the autonomous driving level of tesla's products.
at present, the self-developed chips of nio, xpeng, and ideal are all following tesla's ideas. in july this year, nio announced that its self-developed smart driving chip shenji nx9031 was successfully taped out. this is the industry's first high-end smart driving chip manufactured using 5nm automotive-grade technology, with more than 50 billion transistors, stronger comprehensive capabilities, and higher execution efficiency.
the "xiaopeng turing" chip just released by xiaopeng motors is also adapted to the vehicle's autonomous driving needs, can meet l4 autonomous driving, and can provide computing support for large models. its computing power is three times that of existing chips.
in addition, byd, which has been deeply engaged in the field of power semiconductors and automotive-grade chips for many years, has also deployed autonomous driving chips through investment in recent years.
what is the input-output ratio?
compared with outsourced chips, do car companies' self-developed chips reduce costs or increase costs? there has been controversy in the industry about the above question.
for example, leapmotor, which once jointly developed the lingxin 01 chip with dahua technology, has adjusted its strategy of developing its own chips. the core reason is that the input and output are not proportional. it is reported that the lingxin 01 chip launched by leapmotor in 2020 has been mass-produced on c01 and c11, and can meet the needs of l3-level autonomous driving functions.
although the product has been put into use in cars, leapmotor founder zhu jiangming believes that the current ai chip industry is gradually maturing, and chip manufacturers should be allowed to build chips, while car companies should focus on developing algorithms.
yang yuxin, cmo of black sesame intelligence, also believes that car companies must make a trade-off, or even a game, between technological leadership and cost when developing their own chips. when they need to highlight differentiation, car companies can choose to develop their own chips; but from the perspective of value measurement, purchasing third-party chips has more cost advantages, and it is not cost-effective for car companies to do it themselves.
car companies that firmly advocate self-developed chips do not agree with the above statement and believe that the output brought by self-developed chips cannot be measured simply by monetary value.
"basic research and development also requires accounting, there must be a return, and you can't do research and development for the sake of research and development." li bin, founder and chairman of nio, told reporters that the research and development of the shenji nx9031 chip cost a lot of money. but nio still insists on self-development. the core reason is that it is very clear that smart electric vehicles will once again become the commanding heights of technological innovation, and ai will become the core basic capability of smart electric vehicle companies.
in his opinion, the automotive industry will fully enter the ai era, and smart electric vehicles will evolve into ai intelligent entities with the ability to perceive, think and control execution.
therefore, whether it is developing self-developed chips or developing a full-domain system, the core goal of car companies is to seize the new track of ai cars.
li bin revealed that the r&d cost of chips is very high, but as long as they can be mass-produced, the cost of a single chip will gradually decrease. therefore, high r&d investment can also bring back considerable gross profit. it’s just that the effect of “r&d for gross profit” is slow, but nio believes that this idea is an important way to increase gross profit.
he xiaopeng's idea is surprisingly consistent with li bin. "in the future, companies that are determined to make a difference in ai may have non-universal chips, that is, proprietary ai chips like xiaopeng turing chip." he xiaopeng told the securities times reporter.
self-developed chips do not mean giving up external sourcing. next, xpeng motors will use nvidia and self-developed chips to jointly support xpeng motors' hardware system.
the trend of software and hardware integration is obvious
with the in-depth evolution of the trend of automobile intelligence, the trend of software-defined cars has become more and more obvious. however, with the accelerated implementation of autonomous driving, the industry has increasingly felt the demand for advanced hardware for high-level autonomous driving.
"for l4 autonomous driving functions, software and hardware are equally important. just like if you buy the game "black myth: wukong", you still need a high-configuration computer to play it smoothly." an industry insider used a metaphor to describe the important role of hardware to the reporter. in his opinion, as chinese car companies have rolled up the urban noa competition, bev+transformer will lead the perception paradigm of autonomous driving, and all of this requires strong computing power to support it. in addition, autonomous driving chips with greater computing power will bear the heavy responsibility of the technical foundation.
the above-mentioned person believes that in the next few years, a number of autonomous driving chips with computing power exceeding 1000tops will emerge in the industry, and the chip process will become smaller and smaller.
although high computing power is needed by the market, the industry generally believes that the chips that are really needed for high-level autonomous driving are those that can be better integrated with software. li bin told reporters that the software and hardware of automobiles are updated very quickly at present, and nio hopes that each generation of platform products can adopt an intelligent system, including software and hardware, which is more conducive to long-term maintenance by enterprises through an integrated approach.
the "autonomous driving software and hardware integration evolution trend research report" released by chentao capital also believes that software and hardware integration is the future trend. the above-mentioned "report" points out that for low-level autonomous driving, car companies often directly adopt the supplier's software and hardware integration solutions, but for key capabilities such as high-level autonomous driving algorithms, the proportion of self-developed by car companies will increase.
it can be seen that car companies' self-developed chips are not only a strategic consideration, but also a pragmatic decision made based on the trend of technological evolution.
however, some industry insiders have expressed different views. zou guangcai, deputy secretary-general of the china automotive chip industry innovation strategic alliance, believes that the self-developed chips of automakers will also break the division of labor in the industry to a certain extent and reduce the living space of parts suppliers. in the future, the issue of the division of labor mechanism in the automotive chip industry will require joint efforts from upstream and downstream. (reporter han zhongnan)
