Late last year, Chentai Technology, a leading millimeter-wave (mmWave) radar company, filed its prospectus. Its founding team, backed by the halo of Huawei, its deep integration with major customer BYD, coupled with bold and capitalized strategic losses, perfectly epitomizes the typical profile of today's NEV suppliers.
The mmWave radar is the main and only product of Chentai Technology. Its 5.5th-generation 4D high-resolution mmWave radar has successfully hitched a ride with BYD. In the first half of last year, 97.4% of Chentai's revenue came from BYD, which, rounding up, can be considered part of the FinDreams system [1].
Having secured orders from the big brother, it naturally has to take the bullets for them. Since 2022, as the automotive price war has raged on, the average selling price of Chentai's mmWave radars has been continuously dropping. In the first half of last year, the price of its outstanding forward-looking radar was already halved.
Correspondingly, the gross profit margin has also shrunk significantly, with the profit per unit falling below 20 RMB. Selling one unit only earns enough for a cup of milk tea.
In the current environment, a price war is not a good word, but it is ultimately a signal and an inevitable stage of market consolidation and clearance. From near complete barrenness to a fiercely competitive red ocean, domestic mmWave radar took only a single decade.
The Imperfect Sensor
Tesla, which has been vigorously promoting the pure vision route, also once flirted with mmWave radar in its early days.
In the famous car crash in 2016, a Model S with Autopilot engaged collided with an 18-wheel truck crossing the road, exposing the shortcomings of the vision-only solution for the first time. Due to the bright sunlight and glare, the camera failed to recognize the white truck trailer that 'blended in' with the sky, ultimately leading to the tragedy [2].
Following the accident, Musk immediately issued a decree to promote mmWave radar, changing the previous 'vision-primary, radar-secondary' sensor architecture, with the goal of strengthening the AEB function.
In other words, Tesla was not born purely as a vision advocate; it also took a detour and got distracted along the way.
For ADAS, the chip is the brain, and the sensors are the eyes, used to perceive surrounding environmental information. Mainstream sensors fall into four categories: ultrasonic radar, cameras, mmWave radar, and LiDAR.
Theoretically, cameras are the closest to the human eye, but they require algorithms to perceive accurately. Moreover, cameras have high requirements for lighting; both low light and strong light affect their 'vision'.
mmWave radar and LiDAR belong to active sensors, directly emitting electromagnetic waves for detection. There is no issue of 'whether they can see or not,' and they are less affected by lighting, which can compensate for the shortcomings of cameras.
Tesla ultimately chose mmWave radar, with the core reason being cost. At that time, a single LiDAR unit cost tens of thousands of dollars, whereas an mmWave radar could be installed for just a few hundred dollars.
The advantages and disadvantages of mmWave radar are very obvious. On the one hand, it is unaffected by light and weather, allowing it to be on duty without discrimination 365 days a year. Especially in rainy and foggy weather, where the detection performance of LiDAR significantly decreases, mmWave radar remains completely unaffected.
On the other hand, mmWave radar has a natural shortcoming in resolution, acting like a 'sensitive blind person'.
First, the lateral angular resolution is very low, meaning the emitted beam is relatively wide, making it difficult to distinguish closely adjacent objects. Once the distance increases, it can easily misidentify two motorcycles driving side by side as a single large truck.
Second, it lacks elevation angle information and height measurement capabilities. Objects at different heights are projected onto the same 2D plane, making it difficult to distinguish between a stationary obstacle vehicle and an overpass. Once the algorithm misidentifies the object, the vehicle either triggers 'phantom braking' or charges straight into the 9¾ platform.
Given this situation, mmWave radar typically filters out stationary objects as clutter to avoid false alarms. However, doing so inevitably leads to missed detections. Therefore, in most ADAS systems, mmWave radar has to follow the instructions of the cameras.
In contrast, the angular resolution of LiDAR is far superior to that of mmWave radar. Early mechanical LiDAR could use multiple internal lasers to perform 360-degree rotating scans to complete 3D modeling. The more laser lines, the denser the point cloud, and the clearer the imaging.
Musk's original idea was to use software algorithms to break through the performance limits of mmWave radar. However, radar resolution is tied to hardware configuration, and the coarse field of view is more of an inherent design issue. Software can change the world, but it cannot change physics.
In 2021, with major breakthroughs in Tesla's self-developed vision algorithms, mmWave radar became a sacrifice. Later, the exterior of new cars was left with only cameras standing firm as before.
While Tesla remains steadfast in its pure vision route for a hundred years without wavering, the abandoned mmWave radar has instead been rejuvenated in China.
The "4D" Metamorphosis
On the eve of Musk slashing mmWave radar, the latter completed a major technological upgrade, evolving from 3D to 4D.
3D mmWave radar only provides velocity, distance, and azimuth information, lacking height measurement capabilities. As the name suggests, 4D mmWave radar can additionally provide elevation angle information.
Within 4D mmWave radar, there is a category of imaging radar, characterized by a significant increase in resolution and denser point clouds. It can outline the contours of target objects, and its ability to distinguish between obstacle vehicles and overpasses has made rapid progress.
Therefore, 4D imaging mmWave radar is not simply adding a 'D,' but rather a structural improvement in perception capabilities.
Around 2020, 4D imaging radars from major manufacturers were released one after another. International Tier 1 suppliers such as Continental and ZF were busy delivering the first batch of designated projects. The industry as a whole was in a stage where the technology was not particularly mature and the costs were not particularly affordable, making large-scale application seemingly quite difficult.
However, Musk did not expect that 4D imaging radar would almost get the 'power fantasy' script of LiDAR. With continuous technological breakthroughs and rapid cost reductions, it took only a few years to make LiDAR feel extremely uneasy and treat it as a formidable enemy.
The performance of mmWave radar is inseparable from a key metric: angular resolution. Angular resolution is related to the antenna aperture; the larger the aperture, the clearer the radar can 'see.' It is similar to the satellite dishes from decades ago: the larger the dish, the stronger the signal.
However, cars cannot drive around with a giant dish on their roofs, so they have to find alternative ways to expand the aperture.
The mainstream solution is to cascade multiple RF chips, placing several identical chips on a single PCB, and utilizing MIMO technology to virtualize more antenna channels, thereby increasing the virtual aperture.
To put it simply, the antenna aperture is equivalent to the size of a mosaic puzzle, determining the theoretical upper limit of clarity. The antenna channels are the pieces that make up the puzzle; the more pieces there are, the richer the image details and the clearer the imaging.
For example, 4 transmit and 4 receive (4T4R) is equivalent to 16 virtual channels. By cascading four RF chips, theoretically, 16 transmit and 16 receive channels can be obtained respectively, increasing the number of virtual channels to 256, which virtually expands the antenna aperture.
Compared to the 12 equivalent channels of 3T4R in 3D radar, the number of available channels in 4D imaging radar is more than 10 times higher. The advantage is higher resolution, but the downside is increased cost and power consumption. The difficulty of acquiring reliable radar signals also increases, requiring an upgrade in the computing power of the backend processor, affecting the entire system.
Currently, 4D imaging radar can already rival 64-line LiDAR, though there is still a gap compared to high-line-count LiDAR. This has also caused the market demand for 4D imaging radar to diverge along two paths:
First, for cost-sensitive mid-to-low-end vehicle models, automakers are keen to see 4D imaging radar replace LiDAR to drive down the cost of ADAS. Given that LiDAR prices are dropping lower and lower, 4D imaging radar has no choice but to grit its teeth and follow suit.
Second, for flagship and high-end vehicle models where cost is of no concern, 4D imaging radar, cameras, and LiDAR are all fully equipped. Theoretically, the three complement each other and provide redundancy.
Amid the slogan of 'intelligent driving for all,' throughout last year, the pre-installed volume of 4D mmWave radar in passenger cars exceeded 15 million units, accounting for one-third of the total mmWave radar volume. Among them, the delivery volume of cascaded 4D imaging radar reached nearly 2.6 million units [4][5].
Back then, it took more than a decade for 3D mmWave radar to replace 2D mmWave radar (which could only measure distance and speed), simply because the former was expensive and exclusively available for luxury cars.
Today, 4D mmWave radar is beginning to replace 3D radar, but with the participation of Chinese companies, things are not that simple.
Taking a Shortcut is Not Shameful
In a vehicle, apart from the in-cabin mmWave radar, the out-cabin radars are mainly divided into forward-looking radars and corner radars. Corner radars are primarily used for blind-spot monitoring and lane-change assistance; forward-looking radars are related to active safety, with application scenarios including ACC and AEB. They have higher performance requirements, which also means stricter screening of suppliers.
In the forward-looking radar market, Bosch, Continental, and Denso together took about 70% of the share—which is the figure after being diluted by a large number of Chinese companies.
In the first decade of the century, the story of mmWave radar was concentrated outside of China. The entire market was firmly held by international Tier 1 suppliers, most notably represented by the 'ABCD' combination: Autoliv, Bosch, Continental, and Delphi (later split into Aptiv).
Bosch and Continental began developing mmWave radar in the last century. Relying on the approach of 'improving one generation, designing one generation, and pre-researching one generation,' they have long held the technological initiative.
Since radar spans multiple disciplines such as microwave communications, signal processing, antenna design, and algorithm development, and also has high manufacturing barriers, the returnee faction was an important branch in the first wave of domestic startups. Another batch consisted of startups with relevant technical backgrounds, and Chentai Technology was among them.
Around 2015, mmWave radar startups were established one after another. At that time, the wave of NEVs swept across China, and the authorities were also preparing to introduce the AEB project into the C-NCAP for new car safety testing, which to some extent ignited the startup wave for mmWave radar.
However, since high-quality resources were all controlled by international giants, domestic companies mostly had to start with corner radars—that is, start from 'marginal' functions like blind-spot monitoring, and even had to divert energy to develop security/traffic radars to make a living.
Unexpectedly, in 2021, the automotive industry faced an epic chip shortage, and mmWave radar was also affected.
At that time, there was a massive shortage of Bosch radar chips, affecting more than 10 domestic automakers. New forces like Li Auto and XPeng were forced to prioritize the delivery of models with lower radar usage, or deliver the cars first and install the radars later [7].
The massive explosion of the domestic NEV market, coupled with the global supply chain crisis, objectively yielded a large amount of incremental market share to local suppliers who were geographically and logistically closer, allowing the latter to secure precious orders and achieve self-sustaining growth.
Around 2023, the high-level intelligent driving market exploded. Highway and urban navigation assisted driving functions were widely installed in cars. Corner radars entered the stage of batch domestic substitution, and 4D imaging radar also handed in its first report card.
From the launch of the Li Auto L7 equipped with SenSmart's 4D imaging radar at the beginning of the year, to the debut of SailLeading's similar product in the NIO ET9 at the end of the year, the mass production timetable for domestic 4D imaging radar was unprecedentedly close to that of international Tier 1 suppliers.
It was also at this time that domestic mmWave radar companies began to make a name for themselves in the forward-looking radar market.
When Chinese companies and international Tier 1 suppliers bid for the same project, a price war becomes inevitable. Currently, dual-cascaded 4D imaging radars are hovering around the 1,000 RMB mark.
Looking across the NEV industry chain, the narrative of Chinese companies tearing open the monopoly of giants from the downstream and participating in global competition is common. The brutal clearance presented in the form of a price war is hard to avoid. For mmWave radar, this endurance race has just entered its most intense stage.
References
[1] Chentai Technology Prospectus
[2] Deadly Tesla Crash Exposes Confusion over Automated Driving, Scientific American
[3] Elon Musk is lying about Tesla’s self-driving and I have the DMs to prove it, electrek
[4] Nearly 900% Growth: 4D Radar Rigid Demand Explodes, Breaking Through Multiple Dilemmas of Regulations, Performance, and Cost, Gaogong Intelligent Automotive
[5] China Passenger Car ADAS Market Analysis, Gasgoo
[6] Radar Veteran Returns to China to Start a Business, Building 20 mmWave Radars to Compete Head-on with Bosch and Continental, CheDongxi
[7] XPeng P5 Forced to Adopt Radar Retrofit Delivery Plan, Chip Shortage Becomes the Culprit Again, Yicai