For cutting-edge technology concepts, there are a lot of pits in A-share concept stocks!This article analyzes the current situation and future trends of autonomous driving chips to help you avoid the giant holes you may encounter.
First of all, the common pitfall is that the automotive chip is equivalent to the autonomous driving chip, and the chip dedicated to the car is also called the automotive machine chip. Some unscrupulous analysts will regard automotive-grade chips (to be precise, electronic components)**, such as IGBTs, MCUs, control domains, etc., as intelligent driving chip concept stocks. There is a huge gap between the technical content of the two.
The second is to mix the car machine SOC chip with the autonomous driving chip, the car machine SOC is similar to the smartphone chip, mainly to solve the car machine multi-** part of the function, the car machine chip players are mainly mobile phone chip players such as Qualcomm, there are also more domestic companies have launched their own car machine SOC chips, the gap is increasing. It will not be expanded here.
The most insidious thing is to fool the assisted driving chip as an autonomous driving chip, and this difference is confused even by many half-bucket water analyses. The country has clearly defined the difference between L1-L2 and L3-L5, where L2 and below are assisted driving, and L2 and above are autonomous driving. There is a huge gap between the two technologies, and the requirements for chip performance are also very different. Some of the assisted driving chips are not as good as the high-end graphics cards that play games. After the release of X**IER, NVIDIA found that the performance of 30TOPS (trillion calculations per second) was already stretched in the face of L3 level autonomous driving, and the computing power demand of various partners gradually exceeded 30TOPS. It is common to think that the market share of assisted driving chips is the main chip for autonomous driving, such as Intel's acquisition of Mobileye and Horizon Journey 2.
As mentioned in the series of articles, the algorithm is the soul and the chip is the brain in autonomous driving. The essence of autonomous driving is to observe the road conditions and then decide how to drive. Humans mainly rely on their eyes and ears to receive information, while smart cars rely on radar and cameras. Although the human eye is nowhere near as powerful as a combination of cameras and radar, the human brain is far superior to a computer. The human brain has a powerful ability to filter information. Imagine driving from Guangzhou to Shenzhen, how many cars and high-rise buildings along the way, but the human brain will forget most of the information. This is not the case with computers, Huawei self-driving ADS20 has 29 sensors, and the whole process is 2 hours (under traffic jams) They work almost indifferently, and the information flow generated may be enough to explode hundreds of gigabytes of hard drives. Autonomous driving is not the full name of monitoring, only need to refine the information related to driving safety, and the algorithm needs to put forward this useful information with the help of the chip. Autonomous driving involves a lot of graphical analysis, fusing the distance information provided by the radar, and then modeling the space to determine how the car will drive. This kind of operation is a kind of specialized graphics operation, which operates in a matrix at the bottom of the computer. Before the advent of chips for autonomous driving, GPUs were used to run similar operations. Therefore, NVIDIA, which has experience in GPU design, saw this opportunity, developed autonomous driving chips, and began to lead far ahead. It is worth noting that the underlying layer of AI computing also requires a large number of matrices to operate, so companies that can design AI chips can also start from the GPU.
The gap between specialized chips and general-purpose chips is huge. For example, the general-purpose chip is a stick, which can be used to carry burdens, dry clothes, and go into battle to kill enemies. The special chip is a spear, and the killing efficiency is not a level. What's more, after the division of labor between GPU and CPU, it has evolved to the level of laser cannons. In autonomous driving scenarios with extremely low latency requirements, general-purpose chips are completely insufficient. Traditional chip giants Intel, Qualcomm and other companies can only watch Nvidia present the spaceship of the times and leave them behind.
Algorithms are the soul of autonomous driving. Powerful chips are the guarantee of autonomous driving, but they are not decisive, and the greater the computing power, the better. If you can design a chip that fits your own algorithm, the specialization is more specialized, and the integration of software and hardware can even be dialed in four or two, just like Apple's mobile phone chip makes Apple's mobile phone system very smooth, but the power process does not necessarily exceed Qualcomm's chip.
There are two main types of autonomous driving chip companies, one is independent research and development, and the other is a third-party chip manufacturer. The most extreme case is that cars, algorithms, and chips are all developed independently. There is currently only 15, that is, Tesla and Huawei (count half). There are quite a few companies that integrate cars and algorithms. Major new forces and new brands, which are mainly due to the cost of chip design. According to some more reliable analysis, the design cost of 7nm chips is 2$4.5 billion, and the cost of a single tape-out is about $30 million. At present, the mainstream third-party autonomous driving chip Nvidia's Orinx is at $400. It may take at least 620,000 Orinx chips to recoup the design fee, without considering the manufacturing cost. For this new energy vehicle brand, which can achieve 300,000 sales, it is extremely luxurious to design chips independently in the initial stage. So only Tesla and Huawei dare to do such a thing. Choosing ready-made third-party chips and doing a good job in algorithm optimization is a more economical choice for car companies.
Therefore, there are two types of autonomous driving chip companies.
1. Independent research and development: Tesla and Huawei.
2. Third-party chips: NVIDIA, Horizon, Black Sesame. The latter two of these are Chinese companies.
For A-share investors, it is a sad story to want to invest in autonomous driving chips, because for the time being, we have no direct investment target. Among them, new companies such as Huawei and Changan have not injected autonomous driving chips and algorithms. If not, there is no target for A-shares to invest in again. Horizon and Black Sesame are not yet available, but even so we will introduce the main self-driving chips.
1. Tesla.
Tesla, which is ambitious, wants to learn from Apple to develop its own chips and break the ceiling of autonomous driving by itself. Tesla's previous chip was HW30, the world's leading autonomous driving system FSD uses 2 chips in the third generation, with a total computing power of 2*72TOPS, and only uses 14nm process, lagging behind NVIDIA. Transitioning to HW40, it has been upgraded to the 7nm process, and the computing power has become 2*216tops. Tesla lags behind Nvidia in computing power, does not blindly pursue advanced processes, and pays more attention to economic strength, so I think autonomous driving does not necessarily blindly pile up computing power (such as NIO ET7, stuffed with 4 Orinx). There are already rumors that Tesla is designing the HW50, but the specific participation will not be until 2025, and it is rumored to be at the 1000tops level, using the 4nm process.
2. Huawei. Huawei can not only design the Kirin series of mobile phone chips, but also design artificial intelligence chips at an early stage. Huawei, which has a strategic vision in intelligent driving, has long transformed the Ascend series chips into the field of intelligent driving, and released a platform called MDC for intelligent driving as early as 2020, and currently has three platforms with computing power. Among them, the 610 single-chip computing power is 200 tops, which is the first to be carried on Jihu Alpha, and the 810 single chip computing power is 400 tops. There is also 210 that can be used in low-end scenarios, which is not an autonomous driving chip.
In the early days, Huawei still had reports on the chip computing power process, after the sudden return of the mate60. The Huawei chip has become a secret, so the subsequent use of Huawei's autonomous driving chip in Wenjie M7 and other models has become a low-key treatment. According to the analysis, Huawei ADS2 will be installed after 20230 should all be equipped with at least MDC610 or above chips, otherwise it will not be possible to complete quasi-L3 level autonomous driving. The MDC610 chip is definitely inferior to NVIDIA's 7nm in terms of manufacturing process, and the gap in computing power is not big. The official chip manufacturing is still under secrecy, but autonomous driving is the same as Tesla, and the overall effect is relatively easy to surpass after the integrated design.
In the future, Huawei's chips will be mainly subject to the manufacturing process, and will begin to lag behind in single-chip computing power after Tesla and Nvidia enter the next generation. It needs to be made up in other ways. Fortunately, the power consumption and volume requirements of the car are not as high as those of the mobile phone, and it is completely possible to put a few more chips.
3. Nvidia.
After Nvidia developed a dedicated autonomous driving chip, it is currently the mainstream of autonomous driving chips. At present, it has experienced two versions of X**ier and Orin X, of which X**ier computing power has been eliminated. The mainstream is Orin X, which uses 7nm** and has a computing power of 256TOPS on a single chip, which is currently the most powerful autonomous driving chip. Nvidia has not stopped its leading pace, and has announced Thor, which will increase the single-chip computing power to 1000tops. In the case that Tesla's chips are not supplied, Nvidia has almost no powerful challengers until Huawei's chips can be produced domestically. From the perspective of model coverage, only the models that use NVIDIA and the models that do not use NVIDIA are in an absolute monopoly position. For example, our Xpeng, Weilai, Ideal, BYD, SAIC and other models. Despite their emphasis on autonomous driving, these companies have the short-term ability to design cheaper and better chips.
Fourth, the horizon.
Horizon is similar to Mobileye, starting with the driver assistance chip. Gradually began to launch an autonomous driving chip that can reach the L3 level - Journey 5. The single-chip computing power reaches 128tops, which has a certain gap with NVIDIA. However, as a second choice other than NVIDIA, it has been tried by many car companies. For example, BYD not only chose Nvidia, but also tried to use Journey 5. For example, the ideal L8 uses Journey 5, but the L9 uses NVIDIA. SAIC, Changan, etc. also have Horizon cooperation, but it is estimated that they are all alternative and sub-high-end models.
Horizon itself has closed a Series C funding round and is already valued at more than $5 billion. According to the urine nature of A-shares, the listing of such a company will be very slow and expensive.
5. Black sesame seeds.
Black Sesame is also a Chinese chip company that plans to be listed in Hong Kong, and has been released: Huashan No. 2 A1000, A1000L, A1000Pro. The most advanced is Huashan No. 2 A1000PRO only 106TOPS, and it has also launched a cost-effective Wudang series. For the time being, it is not known that familiar model companies adopt their chips.
Chips have the characteristics of winner-take-all, and other small chip companies are not optimistic except for Nvidia. At present, electric cars and autonomous driving are both key times to grab the market, and chips have a great impact on algorithms. There is no need for car companies to take risks for the hardware of chips. So there's not much room left for them. But the emergence of Huawei has changed this rule, because Huawei is not only a chip manufacturer, but also has its own algorithm and vehicle software design. Huawei's car itself has a market, so it also provides an opportunity for chips to test their strength. Plus autonomous and controllable.
The biggest suspense in the investment opportunity of autonomous driving is whether Huawei will inject chip assets, and of course, the rapid listing of companies like Horizon also gives investors more choices.