(Report Producer: Changjiang Securities)
LiDAR: The technical side is still water flowing deeply, and the market layer is competing for flow
Laser detection has many advantages and long-term technological development
Lidar is a detection device that uses a laser as a signal wave. LiDAR (Light Detection and Ranging) is an abbreviation for LiDAR, which combines lasers, global positioning systems (GPS), and inertial navigation systems (INS). Similar to ordinary radar, lidar measures by detecting signal waves bounced back by objects, except that it uses lasers as signal waves. Because the laser has high brightness, high coherence, and good monochromaticity and directionality, lidar often has the advantages of accurate measurement and less disturbed.
From the structural point of view, lidar is divided into four major parts, namely the transmitting module, the receiving module, the scanning module, and the control module. When the lidar is working, the transmitting module is responsible for transmitting the laser, the scanning module is responsible for scanning a specific area, the receiving module detects the return light, and the control module processes the point cloud map to finally complete the detection.
Lidar has a long history and has developed rapidly in the field of autonomous driving in recent years. In 1960, the first laser was born. In 1968, Hickman and Hogg of Syracuse University in the United States built the first laser seawater depth measurement system. In the 1990s, lidar was used for terrain surveys. In 1990, Professor Ackermann of Stuttgart University in Germany developed the first laser cross-sectional measurement system, forming an airborne laser scanner. Since then, the technology of spaceborne lidar has gradually matured. In 2003, NASA proposed to use it to measure changes in the ice surface of the poles, officially adding geo-laser altimeters to earth observation systems. In recent years, lidar has been used in the field of autonomous driving, ushering in a new opportunity for development.
Lidar is widely used downstream, and civil scenarios are gradually broadened, including mining, forestry, archaeology, geology, seismology, topographic surveying, forestry survey, disaster early warning, AR/VR, unmanned driving, Internet of Things and other scenarios. From the downstream point of view, due to the high cost and huge volume, lidar is mainly used in the military or public domain. However, in recent years, with the development of technology and the maturity of the industrial chain, lidar has gradually expanded in the civilian field, such as the acceleration of popularity in mobile phones, AR/VR, automatic driving and other fields.
There are many classification scales and complex technical principles
At the scale of the scanning module, lidar can be divided into mechanical, semi-solid and solid-state. Mechanical LiDAR: With a mechanical turntable, a 360-degree horizontal measurement angle can be achieved by turning the transmitter module. And mechanical lidar scanning speed is fast, strong anti-jamming ability. However, mechanical lidar relies on mechanical structure rotation to achieve scanning, which has the disadvantages of serious physical wear, high cost, and bulky volume. In order to draw a more detailed point cloud map, mechanical lidar is often equipped with multiple transmitters and receivers, often referred to as 16-wire, 32-wire, 64-wire, etc. Multi-beam lidar has a better angular resolution and can capture smaller objects at a distance.
Semi-solid-state lidar has a variety of solutions. The advantages of the MEMS scheme are lower cost and small size, but the micro-galvanometer is fixed by a monocrystalline silicon cantilever and the system is relatively fragile. And there are also certain limitations in terms of signal-to-noise ratio, field of view, detection distance, etc. In the rotor solution, the lens rotates around the center of the circle, so it has low power consumption, strong durability and easy access to the car rules. However, the rotating mirror scheme also needs to be improved in terms of signal-to-noise ratio, field of view, detection distance, etc. The prism scheme can increase the laser wiring beam to improve the precision and detection distance, but the center point cloud is dense, the edge is sparse, the scanning pattern is complex, and the back-end algorithm maturity is low. From the perspective of terminal applications, various schemes have more mature players. MEMS program players include Innoviz, Hesai Technology, etc.; The main players of the rotary mirror scheme are Huawei, Valeo, Luminar, Innovusion, etc.; Prism solutions are the first of its kind in DJI Livox.
From the perspective of automotive-grade applications, solid-state lidar is the ideal choice, and semi-solid-state lidar is a transitional product. Although the mechanical lidar has a large FOV and excellent performance, there are also many fatal defects: 1, the rotation of the mechanical turntable brings physical wear, the equipment loss is serious, and the average failure time is only 1000-3000 hours, which is difficult to meet the life requirements of the vehicle specification level. 2. The turntable requires no obstruction around it, so it must be installed on the roof. On the one hand, it affects the aesthetics, on the other hand, it is difficult to protect the sun, rain and high-speed flowing air during driving. 3. The cost is expensive, including manufacturing costs and commissioning costs. Mechanical lidar is expensive to manufacture, and requires manual commissioning, and the delivery cycle is long, which is obviously unbearable for civilian vehicles. In the long run, solid-state lidar does not have mechanical movement, service life and volume problems can be solved, and with large-scale mass production, the marginal cost can be reduced to a very low, especially the OPA solution of solid-state lidar is the ideal solution for vehicle-grade laser radar. Semi-solid-state lidar is a compromise between mechanical lidar and solid-state lidar, there is still a small amount of mechanical movement, the current semi-solid-state schemes (prism, rotor, MEMS) have vehicle-level products, and gradually realize the car.
The 905nm lidar detection distance can reach about 150m, which can basically cover the daily driving scene. Through public data, the adhesion coefficient of dry cement pavement is about 0.7~1.0, the adhesion coefficient of wet cement pavement is about 0.4~0.6, and the adhesion coefficient of about 0.3~0.4 at the beginning of rain. Assuming that the reaction time is 0.5 seconds and g is 9.8m/s^2, the braking distance can be calculated according to the mathematical path formula. At a speed of 100Km/h, the braking distance is less than 150m on all three surfaces, while at a speed of 150km/h, the braking distance is less than 150m only on dry roads. According to the Regulations for the Implementation of the Road Traffic Safety Law of the People's Republic of China (promulgated in 2017), highways should indicate the speed of the lane, the maximum speed shall not exceed 120 kilometers per hour, and the minimum speed shall not be less than 60 kilometers per hour. Therefore, the detection distance of 150m can cover almost all driving scenarios, and then by setting up an intelligent driving system, the speed is automatically attenuated when it rains.
Based on the detection principle as the scale, lidar can be divided into triangular ranging method, time-of-flight method (TOF), continuous frequency modulation method (FMCW) and so on.
Triangular ranging method: The light returned by the laser after being scattered by objects in different positions is focused by the lens on different positions of the optical device, and the distance between the measured objects can be obtained by calculation. Common triangulation methods include direct beam type and oblique type, etc., direct beam type refers to the laser perpendicular to the surface of the object to be measured, oblique type refers to maintain a certain tilt when the laser is incident. The resolution of the triangulation method fluctuates with distance. And because the use of geometric relations to achieve the purpose of detection, each change requires the optoelectronic device to read out the position information, so the response speed is more difficult to improve.
Time-of-flight method: That is, TOF (Time of Flight), the laser is emitted from the transmitter and bounced back by the measured object and then received, because the speed of light is determined, so the time difference of laser flight contains the position information of the object. Among them, the TOF scheme can be divided into two types: dTOF and iTOF, where dTOF directly measures the flight time, and iTOF indirectly obtains the flight time by measuring the phase. In contrast, dTOF has better immunity to interference and a longer effective measurement distance, while iTOF has a higher graphic resolution. Overall, the TOF method has a fast response speed and high detection accuracy. However, the working conditions are more demanding, such as a large peak pulse transmission and a weak echo signal. The TOF method is relatively mature and is the mainstream technology of the current vehicle-grade lidar.
FMCW method: the optical frequency emitting the laser is modulated, the frequency difference can be obtained by coherence between the echo signal and the reference light and the mixing detection technology, and the time of flight is indirectly obtained to calculate the distance of the target, if the measured object is moving, the speed of the object can be measured in combination with the Doppler effect. The FMCW method has a strong anti-interference, but due to the high technical difficulty, it is currently under research.
The TOF solution is superior to the triangulation method and is currently the mainstream of vehicle-grade applications. Triangulation is less expensive, but has poor performance. The TOF scheme can emit a large number of lasers in a short period of time and receive return light for analysis, which has the advantages of fast scanning speed, wide scanning area and high accuracy. The FMCW solution has superior performance, but the technical difficulty and high cost are high, and most of them are currently under research. The TOF scheme cannot directly obtain the speed of the measured object, and it is susceptible to interference in rain and snow, and the problem of crosstalk between vehicles cannot be ignored. The FMCW method can directly measure the speed of movement of the object and is not easily disturbed, so it has a high signal-to-noise ratio and advantages in terms of sensitivity. However, the FMCW method is difficult to integrate and relies heavily on discrete devices. Nevertheless, with the development of semiconductor technology, the degree of FMCW integration can be more thorough and thus enjoy the dividends of Moore's Law. In addition, FMCW has higher requirements for channels, which directly makes its cost high and difficult to reduce in the short term. In the long run, cost reduction requires large-scale mass production and mature supply chains. In summary, from a technical point of view, the 1550nm+FMCW+OPA scheme is the ideal solution for lidar, but in the short term, the 905/1550nm+TOF+ semi-solid-state/Flash scheme will occupy the mainstream market. With the development of automotive intelligence, the crosstalk faced by the TOF solution will become increasingly prominent, the advantages of FMCW will gradually be revealed, and the FMCW solution is naturally suitable for OPA. Of course, these are all under ideal circumstances, and conditions must be met to achieve mass production: (1) the technical maturity is improved. (2) Large-scale mass production to reduce marginal costs.
The market space is vast and the competition pattern is relatively scattered
The lidar market size will reach $13.54 billion in 2025 and a CAGR of 64.5% by 2019-2025. In general, autonomous systems need to monitor external changes and make decisions, and the way to obtain information is through a variety of detectors. As an important member of the detection family, lidar has an irreplaceable advantage. With the rapid development of autonomous driving, Internet of Things, drones, AR/VR and other fields, the demand for lidar has ushered in explosive growth. According to Sullivan's research, the lidar market will reach $680 million in 2019 and $13.54 billion in 2025, with a GAGR of 64.5%.
The threshold of lidar technology is high, but the market competition pattern is relatively scattered. Lidar technology barriers are high, but the market is more fragmented except for Valeo. According to Yole's calculations, Valeo has an absolute advantage in the automotive and industrial lidar market, with a market share of 28%, mainly due to its accumulation of many years. However, in addition to Valeo, the market concentration is not very significant, especially in recent years, the rapid development of automobile intelligence has brought opportunities to many new players.
In terms of technology, 905nm and mechanical lidar are currently the mainstream of the market. From the perspective of laser wavelength, 905nm lidar is the mainstream of the current market, with a share of 69%; From the scanning method, mechanical lidar is the mainstream of the current market, with a market share of 66%, followed by MEMS lidar with a market share of 17%.
Downstream applications: New forces grow to build LiDAR certainty
The classification of automatic driving levels is gradually clear, and L3 and above can be called automatic driving. In response to the new changes in the autonomous driving industry, the American Society of Automotive Engineers (SAE) revised the previous classification of autonomous driving in 2021, and in the new standard, autonomous driving is divided into 6 levels (L0 to L5). L3 and above functions can be called automatic driving, L0 ~ L2 is only a driver support function. In self-driving cars, the car's intelligent system is the actual driver of the car. However, the human driver in L3 needs to provide a takeover as needed; The realization of the autonomous driving function in L4 has specific conditions (such as environment, etc.); The L5 enables full, unrestricted autonomous driving. The Ministry of Industry and Information Technology of the Mainland has also divided the level of autonomous driving according to local conditions, which is not fundamentally different from SAE's standards.
Environmental perception mainly relies on cameras, ultrasonic waves, lidar, millimeter wave radar, etc., and is the eye of automatic driving. From the perspective of hardware architecture, there is a short board effect on the real landing of automatic driving. Perception, decision-making, and control of any link that drops the chain will directly affect the final success or failure. As a link in information collection, the perception layer is the basis for the decision-making level to make correct decisions.
Purely from the perspective of sensors, various sensors have advantages and disadvantages.
Camera is the most certain sensor of market growth, throughout the autonomous driving car companies, cameras have shown a trend of increasing numbers and pixels. The light is focused on the photoelectric device by the optical lens, and the photoelectric device in the camera often uses a CMOS sensor, and the CMOS sensor converts the light into an electrical signal, and the electrical signal can form an image after a series of processing such as filtering and amplification. The advantage of the camera is that it can obtain image information, thus simulating the most realistic driving state of the human driver. The disadvantage is that the detection distance is short, it is easy to be affected by rain and snow and other weather, and the recognition ability at night is greatly reduced.
Millimeter wave radar: refers to radar that operates in the millimeter wave band [30~300GHz frequency domain (wavelength 1~10mm)]. Millimeter waves have strong penetration, are less affected by factors such as rain, snow, and ash, and have the advantage of all-weather. In addition, millimeter-wave radar has the advantages of strong anti-jamming ability, small size, easy integration, and low cost, but millimeter-wave radar has poor recognition ability for stationary objects and low spatial resolution accuracy.
Ultrasonic radar: Ultrasonic waves are used as signal waves for measurement. The advantages of ultrasonic radar are strong penetration, simple ranging method and low cost. The disadvantages are that the detection distance is close, the accuracy is low, etc.
The new forces have always been the pioneers of autonomous driving, including Tesla, Weilai, Ideal, Xiaopeng, etc., and their models continue to exert efforts in autonomous driving. We think there are at least a few reasons why new forces are investing heavily in autonomous driving. (1) Autonomous driving represents a new round of technological change, which is the performance of the rapid development of artificial intelligence, 5G, Internet of Things and other technologies in the field of passenger cars in recent years, and the technological change is in line with the brand positioning of the new forces. (2) The automobile architecture of traditional oil vehicles has matured and has strong inertia, and exerting efforts on automatic driving involves changing the electronic and electrical architecture, oil-electricity conversion, software and hardware coupling and other issues. Under the dual-carbon policy, the government subsidizes trams a lot, and the products of the new forces are all trams, which can use the advantages of energy to design a suitable electrical architecture to support the operation of intelligent driving systems, so as to quickly seize the market for traditional fuel vehicles. (3) Capture the psychology of consumer curiosity in commercial publicity, especially the young people's pursuit of fashion, trendy and innovative mentality, in order to achieve double the publicity effect with half the effort. Lidar is the focus of controversy over the route of autonomous driving technology, which differentiates between the visual school and the lidar school. The visual school, represented by Tesla, insists on not using lidar; The lidar faction has absorbed most of the new power manufacturers and believes that the use of multi-sensor fusion, including lidar, is the right way. Judging from the mainstream models of the current new forces, the number of cameras equipped with them is 8 to 13, ultrasonic radar is about 12, millimeter wave radar is about 1 to 5, and lidar is gradually getting on the car, about 0 to 2.
The-for-tat confrontation between the visual and lidar factions has a long history, and this report does not intend to score a single point. However, by combing through the technical paths and development histories of the main members, we can draw the following conclusions. 1. Tesla adheres to the pure visual route has its natural advantages and development inertia, and will not change the technical route to the laser radar faction in the short term, and Tesla's leading position in the new forces is difficult to shake in the short term. 2, Tesla's business model is difficult to copy, lidar faction does not have Tesla's same massive data and supercomputing center. Strong perception schemes can reduce the dependence on algorithms and integrate supplier advantages to seek business breakthroughs. This has led to deterministic growth in the lidar market.
Visualism: Tesla rejects the underlying logic of "lidar."
Tesla is a typical representative of the new power of the automobile, founded in 2003. In the early years, Tesla released its first car, the Roadster, and in 2010 the company was listed on the NASDAQ in the United States. In the following years, Model S, Model 3, and Model Y were released. In 2022, Tesla's Berlin plant officially started construction.
Tesla's production and sales are booming, and its market value has far left Toyota and become the world's largest car company. In terms of sales, Tesla's global sales have been steadily increasing. Tesla sold 936,000 units worldwide in 2021, +87% year-on-year. In 2022, under the background of local epidemic fluctuations in the Chinese market and the Russian-Ukrainian war in the European market, Tesla delivered about 564,000 vehicles worldwide in the first half of the year. In terms of production capacity, Tesla currently has six major factories, namely the California factory, the Shanghai factory, the Berlin factory, the Texas factory, the New York factory, and the Nedahua factory. In terms of autonomous driving, Tesla is the only company in the world that has full-stack self-research in its core field. Chip layer: Tesla cooperated with Mobileye, NVIDIA and other companies in the early days, and then embarked on the road of self-research. In 2019, Tesla released HW3.0 is equipped with a self-developed FSD chip, and the computing power has risen all the way to 144TOPS. Perception layer: Compared with more than a dozen domestic camera manufacturers, Tesla's sensors are more conservative. From the HW 2.0 released in 2016 to the current HW 3.0, it is equipped with a three-way camera, five surround view cameras, 12 ultrasonic radars and a millimeter wave radar. In 2022, Tesla announced that the Model 3 and Model Y, which began selling in the U.S. market in May, will eliminate the front millimeter wave radar and replace the driver assistance function with a pure vision solution provided by the camera.
"Since all kinds of sensors have advantages and disadvantages, then using multiple sensors in the system, through the complementarity of sensors can achieve automatic driving?" In fact, this may simply be a layman's inertia of thought, or it may not be the only path to Rome. The operation of automatic driving is very complex, not a simple addition, as mentioned earlier in this article, the perception layer acquires the data and then makes instructions through the decision layer operation. Therefore, the additional data may improve the correct rate of decision-making, and may also introduce noise to the decision-making and affect the correct decision-making.
We can give a simple example of how additional sensors do not necessarily have a gain effect on the system, and may even drag down the system. In the post-fusion algorithm, where the result produced by different sensors is arbitrated, we assume that there are two sensors in the system that trigger an emergency braking when the probability in the result is greater than 0.8.
Under normal circumstances, the probability of sensor B being identified as a dangerous target is 85%, and if there is only sensor B in the system, then due to the result greater than 0.8, an emergency braking is triggered. When sensor A is added, the more weight given to A when arbitrating the outcome due to the poor performance of sensor A, the less reliable the system's results are.
In extreme cases: Sensor B does not identify a dangerous target (probability value 70%), and the system does not trigger an emergency system. The addition of sensor A performs well, but due to insufficient weighting, the results of A do not work.
The example we give is based on a post-fusion algorithm. Pre-fusion algorithms may improve this problem. But we cannot ignore the other problems of the pre-fusion algorithm. The model in the pre-fusion algorithm requires a lot of data to train, and the requirements for hardware such as chips and communications are very high, and they will also face new problems, such as coordinate alignment. The current ADS algorithm is also transforming from a post-fusion algorithm to a feature-level pre-fusion algorithm, using a combination of CNN, RNN, Transformer, GNN multi-network structure, and the algorithm is very complex. Tesla adopts the idea of "pre-fusion", directly fusing the data obtained by multiple cameras around the body, and then mapping the 2D image information into a 3D space through a neural network.
Tesla's reason for rejecting lidar is two: Tesla's natural advantage: massive data + super computing power. The establishment of the database is a prerequisite for success. We can understand the computer as a diligent stupid child, and the model training of machine learning is to use data to constantly tell the stupid child what kind of scene to do, but despite this, when encountering a new scene, the stupid child will still be confused. Therefore, the collection of road condition information is a prerequisite for success. As mentioned earlier, Tesla will sell 936,200 cars worldwide in 2021, and deliver more than 560,000 vehicles in the first half of 2022 under the pressure of the general environment. Such huge sales are far from being comparable to other new forces. These cars travel all over the world, collecting massive amounts of road condition data for Tesla and building a huge database system. Tesla's "shadow mode" has achieved effective data collection for the first time. Shadow mode means that the system and sensors are still working regardless of whether the driver has the auto-driving function enabled or not. The system algorithm has been simulating the driving decision of the car, and once the driver's operation contradicts the decision made by the system algorithm, the scene will be defined as an extreme road condition, and the data will be transmitted to Tesla's server correction model. Super computing power is the basis for training high-quality models, and Tesla has built the world's fastest AI training computer, ExaPOD. Tesla released its self-developed AI training chip Dojo D1 in 2021. The chip uses nanotechnology, 50 billion transistors are integrated on the chip, and the hash rate of a single chip FP32 can reach 22.6TOPS, and the BF6/CFP8 hash rate can reach 363TOPS. On the basis of D1, Tesla designs training modules at the system level, and then 120 training modules together form the ExaPOD supercomputer. With a final strength of 1.1EFLOPS, the computer is the fastest AI training computer in the world.
From data collection to model training, from perception to decision-making, Tesla has built a wall that is very high, and competitors are difficult to cross in the short term. The data collected by the pure vision scheme, supplemented by the model training of the super-powerful computer, is still in a leading position in the field of autonomous driving.
Tesla rejects lidar for three reasons: cost-effective, lightweight business model. The first principle refers to the algorithm that directly solves the Schrödinger equation after some approximations according to the principle of the interaction between the nucleus and the electron and its basic laws of motion, starting from the specific requirements. The popular understanding of "first principle" is to lose money for the sake of daily loss and return to the original. In the development process of Tesla, the "first principle" shines brightly, in short, it is to reduce costs and increase efficiency, such as simplifying design and eliminating unnecessary devices. On the issue of lidar, the price of mechanical lidar once cost tens of thousands of dollars made civilian vehicles unattainable. In recent years, with the gradual commercialization of semi-solid-state solutions, the price of lidar has plummeted, but so far, the price of lidar is still very high relative to the camera, and the current price of lidar on the market is 1000 US dollars, while the price of a car camera is only tens of dollars. The installation of lidar will increase the cost of bicycles, hindering the process of reducing their weight and popularization. Therefore, if lidar cannot match the back-end algorithm to make a large performance leap, blindly carrying it may not be a wise choice.
Lidar faction: The cost gap is gradually being filled, and it is time to land on the car
Why have many new forces outside Tesla and traditional OEMs joined the lidar faction? We believe that there are at least three main reasons:
Reason one: the defects of the pure visual solution in the perception layer are very obvious, and Tesla cannot solve it in the short term. There are many flaws in the pure visual solution, and not all car companies have a strong background like Tesla. There is no doubt that cameras are indispensable, because cameras can simulate the real state of human drivers, obtain traffic signs, traffic lights and other color information, which other sensors can not do. However, the shortcomings of the camera are also prominent, such as being greatly affected by rain, snow and fog weather, and the performance at night is attenuated in a straight line. In many cases, the pure visual scheme may be due to the lack of location information caused by failure, for example: Scenario 1: The pure visual scheme may identify the advertising person pasted outside the car as a pedestrian, thereby triggering emergency braking, resulting in the occurrence of "ghost braking". Scenario 2: When a large amount of white background appears, it will bring difficulties to the algorithm. For example, Tesla once crashed into a white van parked on the side of the road, probably because the vision system recognized it as an object such as a white cloud. The flaw in the pure vision scheme boils down to the lack of position information, which lidar can obtain. Laser radar collects back light to draw a point cloud map, which can directly obtain the distance information of the object. For example, in case one, since the person in the advertising picture is flat, it is not possible to be a pedestrian. In scenario two, the vehicle itself is a three-dimensional object, which is different from the environment. Therefore, if lidar informs the intelligent system of location information, these errors may be remedied.
Reason two: The cost of lidar is the biggest obstacle to popularization, but it has now plummeted to the green light area. Early lidar was expensive, with Velodyne's 64-line mechanical lidar selling for about $80,000 and 32-line lidar for about $20,000. With the development of technology, lidar has gradually transitioned to semi-solid or solid-state solutions, and the current price of lidar has been reduced to less than 1000 US dollars, which has entered the green range acceptable to automotive OEMs. According to public information, the lidar of Ouster, Luminar, Sagitar Juchuang, Huawei, DJI Livox, Tudtong and other companies has dropped to less than $1,000 and is gradually mass-produced.
Reason three: Tesla's business model is not replicable, equipped with lidar differentiated competition, may be the best opportunity to overtake in curves. Tesla's full-stack self-development model is not achieved overnight, but has been gradually established over many years. In the early stage of Tesla's development, it was limited by funds and research and development strength, and the chip was also purchased from Mobileye. Mobileye is the world's leading manufacturer of autonomous driving solutions, using a "black box" model when delivering, that is, the integration of algorithms and chips, which allows Mobileye to grasp the initiative and leaves little space for car companies. Tesla then threw itself into the more liberal NVIDIA camp, and finally began to develop itself and build its grand database and huge data center. Tesla's sales are several times that of other new forces, and its data collection capabilities are far from being comparable to other new forces. In China alone, Tesla sold 147,000 units/323,000 units in 2020/2021, several times more than Weilai, Ideal, and Xiaopeng. In terms of models, Tesla's main models are Model Y and Model 3, with sales of 171,000 units and 151,000 units in 2021, respectively. Tesla in the Chinese market alone has opened up a gap with other new forces, looking at the world, Tesla's global sales of 936,000 vehicles in 2021 are even more superior, and the high wall is difficult to overcome.
Tesla has the world's top supercomputing center and leading image processing capabilities, which is also a gap between other new forces. For new forces and traditional OEMs that want to break through in autopilot, compared to Tesla, which has lost its first-mover advantage, it is expensive to copy Tesla's business model. Most car companies do not have Tesla's strong capital, technology and talent strength, even if they spend a lot of money to establish Tesla-like supercomputing center, the richness of the database is also very different from Tesla, after all, Tesla has millions of vehicles running around the world, these vehicles transmit data back to Tesla's servers every day, and Tesla has gathered experienced engineers on the autopilot model for many years. Therefore, Tesla's business model is not replicable, for the vast majority of car companies, in the early stage of development of the algorithm and chip research and development to the automatic driving program manufacturers, the use of relatively mature automatic driving program providers to make up for their shortcomings, gradually accumulate funds, technology, establish a database, and slowly strive for the initiative is the best choice. The choice of a "strong perception scheme" equipped with lidar can reduce the dependence on the algorithm of the decision-making level in order to overtake in curves. Tesla is not good at using lidar to compensate for the problem of pure vision solutions, in addition to achieving differentiation in publicity to capture consumer psychology, perhaps it can really explore a new path in technology. In summary, non-Tesla manufacturers will actively or passively join the lidar faction, thereby pulling the demand for lidar.
2022 is the first year of lidar on the car, and the blue ocean has since opened. The lidar produced by various manufacturers has been put on the car one after another, and the models equipped with it include Weilai ET7, Feifan R7, Xiaopeng G9, Ideal L9, Polar Fox Alpha S Huawei HI Edition, Nazha S, Sharon Mech Dragon, Weima M7, Avita 11, Zhiji and so on. From the perspective of the number of mounts, if the laser radar performance is high, it is equipped with 1 (such as WEILAI ET7, Extraordinary R7), and if the lidar is slightly inferior, it will win by volume, carrying 1 to 4 pieces. The development trend of autonomous driving is certain, and L4 is expected to be achieved in 2030, and the lidar market is growing rapidly. At present, the mainstream models in the automotive market are mostly below L3, and with the development of autonomous driving, the lidar market will benefit deeply.
Upstream devices: foreign leaders dominate, and domestic manufacturers develop rapidly
From the perspective of the industrial chain, as mentioned above, lidar is mainly composed of four parts: transmitting module, receiving module, scanning module, and control module, and the upstream is mainly optical and electronic components, including collimators, diffusion plates, beam splitters, narrowband filters, lasers, photodetectors, scanning mirrors (if any), FPGAs, etc. The midstream is mainly responsible for integration, and downstream applications include automotive, robotics, industrial, surveying and mapping, military and other fields.
Semiconductor lasers: EEL and VCSEL market mainstream
Laser principle: Microscopic particles have a specific energy level structure, when interacting with photons, particles will absorb or radiate photons accordingly. Lasers are "light radiated by particles", and the electrons in atoms absorb energy and jump from low energy levels to high energy levels, and then fall back from high energy levels to low energy levels to release energy in the form of light. The photons released during this process are highly consistent, so monochromaticity, coherence, and directionality are excellent. The laser is mainly composed of four parts: optical system, power supply system, control system and mechanical structure. Among them, the optical system is mainly composed of pump source, gain medium, resonant cavity and other optical materials. Particles in the gain medium are excited by the pump source and change from a ground state to an excited state, and since the excited state is an unstable state, energy is released back to the ground state again. In this process, energy is released in the form of photons, thus forming a laser.
Semiconductor lasers are developing at a rate close to Moore's Law, with performance and cost improving more than tenfold every decade. In 1962, a team led by Robert Hall of the General Electric Company demonstrated the infrared emission of a gallium arsenide semiconductor, the world's first laser semiconductor. The most advanced semiconductor lasers of 1985 coupled 105 milliwatts of power into a 105 micron core diameter fiber. Today's most advanced semiconductor lasers can produce more than 250 watts of 105 micron fiber with a single wavelength, which means a 10-fold increase every eight years. It is very coincidental that this is closely consistent with the speed proposed by Moore's Law, that is, high-power semiconductor lasers integrate photons into optical fibers at speeds similar to Moore's Law.
The laser market is growing steadily, and on-board lidar is driving demand. According to Laser Focus World, the market size of lasers in China will reach $14.74 billion in 2022, +16.25% year-on-year. From the perspective of downstream proportion, the main demand for lasers in 2020 comes from the material processing and lithography market, communication and optical storage market, scientific research and military market. As lidar gradually gets on the car, laser demand is expected to rise rapidly.
Foreign enterprises started earlier and have a leading edge in technology. Semiconductor lasers and chips to The two lu group, Longmeitong, Enai Group, IPG optoelectronics and other foreign enterprises, the domestic strength of the strong enterprises are Wuhan Ruijing, Torch Technology, Changguang Huaxin, KaiPrin, Xinghan Laser and so on.
Photodetectors: led by foreign manufacturers
Photodetectors need to be used with light sources, and current detectors are mainly based on silicon (Si) and indium gallium arsenic (InGaAs) substrate materials. Si belongs to the first generation of semiconductors, the process maturity is slightly higher, often with 850nm, 870nm, 905nm, 940nm and other bands. From a technical point of view, lidar detectors mainly include PD, APD, SiPM/MPPC, SPAD and so on. SPAD, APD, and PD are of the same family, but operate in different voltage ranges and have different gains. PD is applied to a smaller reverse voltage, so there is no gain. The APD operates in a linear range with a gain of around 100x. The SPAD operates in the Geiger range, and since the photons do not automatically stop after receiving an avalanche at this time (requiring a quenching circuit), the theoretical gain can reach infinity, that is, it can detect a faint light in the distance. SiPM/MPPC, or Silicon photomultiplier/MPPC (also known as multipixel photon counter according to the principle), consists of SPAD series quenching resistors connected in parallel. The pixel area per UNIT of SPAD is higher than that of MPPC, but the MPPC signal can reflect the strength or weakness. Therefore, the SPAD resolution is higher and the MPPC frame rate is faster.
The performance of various detectors varies greatly, with SiPM and MPPC having the highest gain. According to Hamamatsu' data, the ranging range of PD, APD, and MPPC increases sequentially. In terms of gain, the PIN has no gain, the typical gain of the APD is 10 to 100, and the gain of the MPPC can reach 10^5. In terms of working voltage, THE PIN works at about 10V, the APD works at 100~200V, and the working voltage of MPPC can reach dozens of Ford. Foreign manufacturers lead the way, and domestic manufacturers compete for layout. At present, the domestic manufacturers with layout of lidar detectors are Lingming Photons, Jingbang Technology, Fushi Technology, Core Vision, etc., while foreign manufacturers are Sony, ON Semiconductor (acquisition of SensL), Hamamatsu, Canon, etc., especially Sony and ON Semiconductor are the strongest. APD is the mainstream of the current lidar market, SPAD/SiPM is an important development trend, but the process difficulty is higher, and there are fewer vehicle-grade products.
MEMS micro-galvanometer: domestic enterprises Are up
MEMS lidar detects by working with lasers and MEMS micro-galvanometers. MEMS lidar only needs to scan the pulses emitted by a single laser into a multi-line multi-beam effect of mechanical lidar, and does not need to be equipped with additional transceiver modules, so the cost plummets. Further, MEMS micro-galvanometers use semiconductor processes with strong scale effects, so the cost can be further reduced after mass production.
Technological change is accelerating, and the market size is growing steadily. According to Yole, the MEMS market size of $12.1 billion in 2020 will reach $18.2 billion in 2026, with a CAGR of 7.2%. In terms of segmentation, consumer electronics is the largest market for MEMS sensors, with a market size of $7.13 billion in 2020 and $11.27 billion in 2026, with a GAGR of 7.9%. The automotive market will be $2.03 billion in 2020 and $2.86 billion in 2026, with a CAGR of 5.8%. In terms of the competitive landscape, the United States, Japan and Germany stand on three feet. According to Yole, the highest-grossing MEMS companies in the world in 2020 are all European, American and Japanese companies, including Bosch, Broadcom, Qorvo, STMicroelectronics, TI, Goermicro, HP, Knowles, TDK, Infineon, etc. Companies that are currently capable of producing MEMS micro-galvanometers include Bosch, Infineon, Hamamatsu, STMicroelectronics, Mirrorcle, and others.
Domestic enterprises are rising, and mass production and manufacturing still rely on OEM. The development of MEMS sensors is divided into three links: design, manufacturing, and packaging and testing, and an industrial chain has been gradually established in China. In the field of scientific research, there are Suzhou Institute of Nanotechnology, Institute of Electronics, Chinese Academy of Sciences, etc. In the market, high-quality enterprises such as Goertek Shares, Minxin Shares, and AAC Technology have emerged. MEMS micro-galvanometer technology has long been around, and in recent years it has exploded due to unmanned driving. At present, the enterprises that can provide MEMS micro-galvanometers include Xi'an Zhiwei Sensing, Wuxi Weiao Technology, Changzhou Chuangwei Electronics, Shanghai Microtechnology Institute of Industry and Research, etc. At the same time, Minxin Shares and other enterprises have also begun early research and development, Intang Intelligent Control in 2022 plans to issue stock financing for MEMS micro-galvanometer research and development and industrialization projects, the total investment of the project is expected to be 251 million yuan.
Terminal: Rising stars take advantage of the momentum and recommend paying attention to domestic forces
Velodyne: The industry's veteran leader, the strength can not be underestimated
Velodyne began in 1983 when founder David Hall founded Velodyne Acoustic to design and produce subwoofers. In 2004, Velodyne opened the way for autonomous driving by participating in the Defense Advanced Research Projects Agency-funded Unmanned Challenge, inventing real-time 3D lidar the following year. In 2007, the HDL 64E became the first commercial, mass-produced real-time 3D lidar. Velodyne's product matrix has since been refined, with the launch of Puck in 2014; Launched three Puck series products in 2016; Launch of Alpha Puck in 2017; Launch of VelaDome in 2019; Launch of Velabit, Velarray M1600, Velarray H800 in 2020; Launch the next generation of Velabit sensors in 2021.
The product matrix is perfect, and both software and hardware are available. Velodyne products have four first-level classifications, including mechanical radar (orbit radar), solid-state radar, infrastructure, radar software, and more. Among them, the orbit radar is mainly used in automobiles, including Alpha Prime, Ultra Puck, Puck, Puck Hi-Res, HDL-32E, etc.; Solid-state radar is mainly used in robotics, ADAS, etc., including Velarray M1600, Velarray H800, Velabit, etc.
Velodyne lidar is mainly divided into mechanical and solid-state types. The early HDL-64/HDL-32 was a mechanical lidar with higher performance, but the product was expensive and bulky, mainly used in the field of surveying and mapping. At present, the mechanical lidar on sale on the official website also has Alpha Prime, Ultra Puck, Puck, Puck Hi-Res, etc., with wiring harnesses ranging from 16 lines to 128 lines, and laser wavelengths in the near-infrared region. Solid-state lidar includes the Velarray M1600, Velarray H800, Velabit, and has a detection range of up to 200m.
The company's performance declined, but it still has an absolute advantage in the field of lidar. Operating revenues for 2019, 2020 and 2021 were $101 million, $95 million and $62 million, respectively, while gross profit was $30 million, $0.25 billion and -$0.06 billion, respectively. In terms of regions, North America, Asia-Pacific, Europe and Africa have declined to varying degrees. The decline in the company's performance is partly affected by the epidemic and the supply chain, but the company's market share is still very high. According to the annual report information, Velodyne's 21-year loss in addition to the reasons mentioned above, but also because of the reduction of the price of lidar, and increased investment in research and development, for the large-scale commercial use of lidar efforts.
The customer resources are strong, and the AwV project is supportive. Velodye has developed over the years and has accumulated profound technical background and customer resources. Velodyne's Automated with Velodyne project is dedicated to commercializing next-generation automation solutions based on Velodyne laser radars with more than 100 partners, including NVIDIA, Siemens, LG and other well-known vendors. With Velodyne's excellent position in the industry, the company is expected to enjoy the development dividend of the industry and brave the trend of automotive intelligence.
Ibeo: The industry's first 4D lidar, mass production will begin in 2022
Ibeo Automotive Sensing was founded in 1998, and sick AG became a major shareholder in 2000. 2010 Management completes the acquisition and formation. In 2016 ZF Friedrichshafen became the company's majority shareholder. In 2017 and 2019, the company established subsidiaries in the Netherlands and Detroit, USA. In 2021, Ibeo partnered with AAC Technologies.
Projects: Ibeo participated in the DARPA City Challenge in 2007, working with Rinspeed on multiple projects in 2008, 2015, 2019, and Local Motors in 2016. In 2020, Ibeo took the independent package delivery research project and was nominated by Great Wall Motor as the first LiDAR series supplier. Products: In 2000, the company introduced LD-ML LiDAR and ALASCA LiDAR, and in 2005 the company introduced ALASCAXT, the first Ibeo sensor with an FPGA. IbeoLUX was launched in 2008 and ibeoNEXT in 2019, and its samples entered the market the following year. Launch of ibeoNEXT SOP in 2022.
The company currently offers two LiDAR, ibeoNEXT and ibeo LUX. Among them, ibeo LUX was launched earlier, there are ibeo LUX 4L, ibeo LUX 8L, ibeo LUX HD three, 10% reflectivity detection distance in the range of 30 ~ 50m, the horizontal FOV is 110 °. The radar uses a 905 nm laser with a distance resolution of 4 cm.
ibeoNEXT is the company's flagship product, without any mechanical structure, is a solid-state lidar. ibeoNEXT is the industry's first 4D sensing system that creates a 3D point cloud plus an image representation intensity with an angular resolution of 0.05° in space. At the same time, ibeoNEXT uses continuous flash technology to scan the environment, each scanning thousands of measurements. In addition, ibeoNEXT has a powerful modular function, according to different requirements in the development to select different parameters such as field of view, detection distance and so on (by matching different optical components). For example, the selectable levels of FOV are 11.2°, 32°, 60°, 120°.
Luminar: Genius founder pilot, chasing light 1550nm
Luminar successfully registered on nasdaq in december 2020, becoming the second listed lidar stock in the world after Velodyne and surpassing Velodyne in market capitalization. Founder Austin Russell, who is 26 years old (2022), developed his first vision system at the age of 11, built a supercomputing prototype as well as an optical system, and thought about how to put it into practice. He applied for his first patent at the age of 12 and later entered Stanford University before dropping out to start a business. Luminar's Iris is a high-performance product for 1550nm+ galvanometers. Luminar's lidar Iris will be delivered in series production in 2022. Iris lidar has excellent performance, thanks to the use of 1550 nm laser, so it can achieve a longer detection distance while protecting the human eye, with a detection distance of up to 250m at 10% reflectivity. In the choice of scanning mirror, Iris uses a two-dimensional galvanometer with a horizontal FOV of 120° and a vertical FOV of 20°. On the receiver device, the ASIC used by Iris is superior and less expensive than the ADC. In terms of tracking, Iris can detect 80m of roads and maneuverable space, 150m of lane markings, 250m of vehicles, etc. Customer development is smooth and revenue growth is rapid. On the customer side, Luminar has reached eight of the world's top ten OEMs, and its lidar has landed on the ground. In terms of revenue, the company's revenue in 2021 was $32 million, +128.97% year-on-year, of which North America was the mainstay.
Sagitar Juchuang: The first vehicle-grade solid-state lidar preemptive mass production
Founded in 2014, founder Qiu Chunxin holds a Ph.D. in control science from Harbin Institute of Technology and has published many papers in well-known journals at home and abroad. After completing the research on related topics, he realized the bright prospects of lidar and founded Sagitar Juchuang, after which the company has mass-produced a large number of high-quality products, and has won the favor of industry and capital along the way. In 2018, it obtained strategic financing from Ali Cainiao, SAIC and BAIC, and in 2022, it obtained strategic financing from BYD, Yutong, Hong Kong Lixun, Xiaomi Yangtze River Industry Fund and so on.
Sagitar Juchuang's products are mainly mechanical, but its solid-state liDAR RS-liDAR-M1 is the world's first mass-produced vehicle-grade solid-state lidar. The mechanical type occupies half of the product map of Sagitar Juchuang, and the wiring harness is not equal from 16 to 128 lines, and the detection accuracy, FOV, angular resolution and other aspects have their own advantages and disadvantages. From the technical characteristics, the wavelength of Sagitar Juchuang LiDAR is 905nm (except for the earlier Seeker), and the detection principle is TOF. The RS-liDAR-M1 is the world's first lidar production SOP, and models such as the GAC AION LX Plus and lotus ELETRE equipped with the M1 have also been launched.
Although Sagitar Juchuang is a young player, but the rapid development, preemptive release of mass production of solid-state vehicle lidar, has established cooperation with a number of famous enterprises, including SAIC, Geely, FAW, GAC Aean, BYD, ZEEKR, WM Motor, Yutong Automobile, Horizon and other machine manufacturers and solution manufacturers, is expected to achieve rapid growth in performance.
Hesai Technology: MEMS has passed the car regulations and the chip development route
Hesai Technology was founded in Shanghai in 2014 and began to develop lidar independently in 2016. In 2017, the company received a Series B financing led by Baidu, a Series C financing led by Bosch in 2019, and a Series D financing led by Hillhouse, Xiaomi, and Meituan in 2021, with a cumulative financing of more than $500 million. The founder graduated from Tsinghua, Stanford and other first-class schools at home and abroad, in addition to having strong technical strength, he also has a strong large-scale production capacity of vehicle regulations. In 2022, the "Maxwell" super intelligent manufacturing center will be put into production.
Hesai Technology LiDAR has four series, namely AT series, Pandar series, QT series, and XT series. Among them, there is only one product in the AT series, AT128, which is a semi-solid-state lidar based on MEMS micro-galvanometers, focusing on thin and light shape and excellent performance. There are four products under the Pandar series, all of which are mechanical lidar, designed for unmanned main radar, the detection distance is 120~200m at 10% reflectivity, the number of lines is 40~128 lines, accuracy, FOV and so on have their own characteristics. QT series features ultra-wide angle, vertical FOV up to 105.2°; The XT series features high precision with a horizontal angle resolution of 0.09°.
Technological innovation has developed in an all-round way There are four main development directions of Hesai Technology in the future, namely chip research, FMCW development, test system and algorithm research and development. Among them, in the company's chip development plan, on the transmitter, Hesai Technology will replace the EEL with VCSEL, and the VCSEL will develop from linear array to area array; On the receiver, Hesai Technology has established the development route of APD-SiPM--SPAD, focusing on the high performance of lidar while exerting the advantages of integration and scale.
Customer cooperation is gradually getting better, landing on the car and moving forward. Hesai Technology has established cooperative relations with a number of companies, including OEMs, Tier1 suppliers, and autonomous driving solution providers. Hesai Technology's lidar has also landed on the car, including ideal, set, high-hehe, Aichi, zero-run and other models.
DJI Livox: The first prism solution
As mentioned earlier, the Prism scheme was originally created by DJI Livox. Because the reliability of the equipment is negatively correlated with the complexity of the movement mode, and the traditional mechanical movement is complex and there are more moving parts, DJI hopes to make a breakthrough in this direction, so as to provide partners with a mass-produced, stable and reliable vehicle-grade radar, so the prism scheme came into being. The prism scheme changes the optical path by driving the motion of the wedge by electromagnetic force, in which the electronic device is completely unaffected in the motion.
The prism scheme performs better than traditional lidar. DJI's LiDAR uses a non-repetitive scanning method, with trajectories superimposed on each other. In general, the average sampling rate in the unit stereo angle = the full FOV internal sampling rate / FOV coverage of the cube angle, after calculation can be seen that the Mid series in the integration time of 100ms, the field of view coverage can be similar to the 32-line traditional laser radar. Horizon and Tele have higher field of view coverage.
DJI Lanwo currently has a total of 6 products on sale on its official website, and HAP performance is superior. The products on sale on the official website include HAP, Landao-70, AVIA Autoshade, Haojie Horizon, Thailand Tele-15, and Mid-40/Mid100. DJI's products all use 905nm lasers, with detection distances ranging from 90 to 320m and a distance error of less than 2cm. In terms of price, the cheapest Mid-40/Mid100 is only available for 3999 yuan, while the hap of the car grade is priced at 7999 yuan. In terms of performance, HAP can now detect black vehicles 150 meters away, as well as pedestrians wearing black clothes 120 meters away, which can meet the needs of current automakers. DJI's partners include Zhitu, Xiaopeng and Yutong Bus.
Industry chain: technology-based expansion of customers, first-mover advantage is strong
As far as the industrial chain is concerned, most of the listed companies in the domestic lidar business are engaged in optical components, lasers and other related businesses, because the industry is still in the early stage of development, the technical path is not unified, most of the companies in the industry cooperate with downstream customers to do technology iteration, and the evaluation dimension of early manufacturers is relatively single in whether the technology can match, that is, the technology leadership can obtain manufacturer orders. In the future, it is imperative to reduce the cost of lidar or will require upstream suppliers to further do a good job in cost management and test the mass production and manufacturing capabilities of manufacturers. We believe that at the investment level, we can preferably select manufacturers with high technical barriers, large customers and system solution capabilities, such as Shunyu Optics, Juguang Technology, Changguang Huaxin, etc.
Sunny Optics: The leader in the field of optics, with profound technical heritage
A leading manufacturer of optics, it has a long history since 1984. The world's leading manufacturer of integrated optical parts and products. Founded in 1984, the company's main business is optical lenses and imaging modules, listed on the Hong Kong Stock Exchange in 2007 (stock code: 2382. HK）。 The company's downstream covers mobile phones, automobiles, security, VR/AR, microscopic instruments, industrial and many other fields, with globally renowned customers in the industry and stable cooperation. The company has accumulated profound technology in the field of optics, positioned itself in the design experts to cooperate with customers for product innovation, and has become a "famous supporting role" for large customers.
Sunny's revenue fell slightly in 2021, but profits did not fall but increased. In 2021, the local epidemic rebounded, superimposed on the lack of core, macro economic pressure, bulk material costs and other factors, Sunny's operating income was 37.497 billion yuan, a slight decrease of 1.33%, but the attributable net profit was 4.988 billion yuan, an increase of 2.39% year-on-year. In recent years, the company's gross profit margin and net profit margin have increased steadily, and the gross profit margin and net profit margin in 2021 are 23.3%/13.5%, respectively, which is +0.4pct/0.5pct year-on-year.
Production bases/R&D centers are located at home and abroad, and the global layout is steadily advancing. At present, the company's domestic production bases are distributed in Yuyao City, Zhejiang Province, Zhongshan City, Guangdong Province, Shanghai City and Xinyang City, Henan Province. At the same time, in order to promote the global layout, the company has set up production bases in India and Vietnam, and the production of the Indian base is planned, and the first phase of the construction of the Vietnam base has been basically completed and gradually put into production. In terms of R&D centers, the company has set up R&D centers in China, the United States and South Korea to provide technical assistance. Sunny has a deep technical heritage and rapid development of vehicle-related products. The company's automotive business includes automotive modules, lidar lenses, etc. In 2021, the company's vehicle lens shipments increased by +21.0% year-on-year to 67.98 million pieces, and the research and development of 2 million and 3 million pixel glass-plastic hybrid lenses containing multiple plastic lenses has been completed. A number of 8-megapixel ADAS in-vehicle lenses based on NVIDIA, Qualcomm, and Horizon platforms have been certified and have received platform-based projects from many automakers, some of which have been mass-produced. In addition, Sunny's self-developed ADAS lens with defrost and defogging functions has been acquired as a platform project for L3/L4 autonomous driving in several depots. In terms of lidar, the company provides optical parts such as transceiver components, transceiver modules, optical windows, and polygon prisms, and more than 20 fixed-point cooperation projects in 2021, two of which have been mass-produced. In-vehicle revenue has grown rapidly, and the layout of lidar has gradually entered a better state. From the perspective of revenue, the company's revenue from in-vehicle-related products in 2021 was 2.961 billion yuan, +17.87% year-on-year. From the product point of view, the company's main products in lidar include lenses, optical windows and so on. In terms of lenses, the company has many years of R & D design experience and strong engineering manufacturing capabilities, can provide lens solutions based on various lidar solutions, optical windows, the company can produce various shapes of windows, with good opaque, high hardness impact resistance, high transmittance, heating defrosting and fogging, electromagnetic shielding and other properties.
Torchlight Technology: Laser layout is earlier, and customers cover well-known foreign enterprises
In the field of laser, deep ploughing and meticulous cultivation, strong technology and high-quality products. Torch Technology was incorporated in 2007, and the following year the company produced the first batch of semiconductor lasers. In 2009, the company passed ISO9001 quality system certification and delivered the first 100-kilowatt area array semiconductor laser. Subsequently, the company's product matrix gradually expanded, winning a series of projects and awards. In 2015, the company led the development of two national standards for semiconductor lasers. In 2017, the company acquired LIMO, the world's leading field in micro-optics. In 2019, the company established the Automotive Division and officially entered the field of lidar. In 2020, the company's monthly output of fast-axis collimators exceeded 10,000, and it landed on the science and technology innovation board in 2021.
From "generating photons" and "regulating photons" to "photonic application modules and systems", the company's business continues to expand in the industrial chain. TorchLight Technology is mainly engaged in the research and development, production and sales of semiconductor laser components (photon generation) and laser optical components (regulated photons) in the upstream of the laser industry, and is expanding the "photonic application module and system" business in the middle reaches. Downstream applications are widely used, including advanced manufacturing, medical health, scientific research, automotive applications, information technology, optical communications, optical storage and so on.
The company's product genealogy is comprehensive and the engineering experience is rich. Torch technology main business includes semiconductor lasers, laser optics, automotive solutions, system application solutions four categories, is the semiconductor laser industry vertical industry chain company, products including active devices, passive devices, advanced materials, fiber coupling materials, fast axis collimator, beam converter, lidar transmitter end, beam shaping solutions, cabin monitoring VCSEL launch module, etc. The company's core technology involves two major fields of semiconductor laser and laser optics, including eutectic bonding technology, thermal management technology, beam conversion technology, light field homogenization technology, etc.
Research and development has entered a stable period, and the three rates have gradually stabilized. The company's research and development expenses in 2021 were 68 million yuan, an increase of 3.02% year-on-year. In 2021, the R&D expense ratio was 14.25%, -5.18pct year-on-year, and the rapid reduction of the R&D expense ratio was not only the gradual stabilization of the company's research and development, but also due to the rapid growth of revenue, the denominator was raised. As of May 2022, the company has 113 overseas patents, 123 domestic invention patents, 150 utility model patents and 32 design patents. In terms of three rates, the company has gradually stabilized its three rates in recent years. In 2019, 2020 and 2021, the company's three rates were 32.27%, 24.13% and 24.34%, respectively.
The layout of lidar-related products is earlier, and the products have entered many well-known enterprises. Torchlight Technology began to develop the high peak power lidar area array light source that began in 2016 and has signed contracts with automotive customers and has now entered the mass production stage. Juguang Technology products have passed the IATF16949 quality management system certification, the German Association of the Automotive Industry VDA6.3 process audit, in the design and development of vehicle-grade products and engineering manufacturing has accumulated a lot of experience, customers cover Velodyne LiDAR, Luminar, Argo AI and other well-known manufacturers. A variety of linear spot lidar launch module products released by the company in 2021 have been sent samples for a number of customers, and a domestic lidar head customer project is scheduled, and it is expected to be mass-produced at 22Q3.
Changguang Huaxin: high-quality light source enterprise, horizontal expansion and vertical extension
Changguang Huaxin was established by the returnee team in 2010, and in 2012, it was formally established relying on Changchun Optical Machinery Institute and Strategic Investor Company. In 2013, high-energy laser chips, modules, arrays, systems and other products were fully put into production. In 2017, the company pioneered and implemented a 976nm fiber laser pumping solution. In 2018, the VCSEL Division was established and a 6-inch line of VCSEL chips was established; In the same year, the Laser Systems Division was established and a number of products were launched. In 2019, it introduced a 15w high-power semiconductor single-tube chip, a 976nm fiber-coupled module and various series of direct semiconductor lasers. In 2020, 18w and 25w high-power semiconductor single-tube chips and VCSEL surface emitted semiconductor laser chips were launched, and in-house InP optical communication chip manufacturing processes and production lines were introduced. Mass production of 30w high-power semiconductor single-tube chips will be achieved in 2021.
Scientific research experts sit in, and core technical personnel are senior executives of the company. Changguang Huaxin disclosed a total of four core technical personnel, namely Min Dayong, Wang Jun, Liao Xinsheng, Pan Huadong, graduated from Huazhong University of Science and Technology, McMaster University, Academy of Sciences, Fudan and other famous universities, has many years of scientific research or experience, of which Min Dayong enjoys state council subsidies, Wang Jun is a visiting professor of Sichuan University and a part-time professor of the University of National Defense Science and Technology, and Liao Xinsheng is a part-time professor of the University of National Defense Science and Technology. The four core personnel are all senior executives of the company, which shows the importance the company attaches to scientific research. The products are widely distributed and cover many fields. Changguang Huaxin products have optical fiber coupling modules, arrays, devices, direct semiconductor laser systems, high-power laser chips, optical communication chips and other categories, each category has a different series of products, widely used, covering multiple downstream.
The company's revenue has multiplied, and gross/net profit margin has grown rapidly. In 2021, Changguang Huaxin achieved a total operating income of 429 million yuan, +73.59% year-on-year; In 2021, the net profit attributable to it was 115 million yuan, a significant increase of +340.49% year-on-year. In terms of profitability, the company's gross profit margin/net profit margin in 2021 was 52.82% and 26.87%, respectively, compared with +21.46pct and +16.28pct year-on-year, respectively. The company's net profit margin has risen steadily in recent years, and the sharp decline in net profit margin in 2019 is due to the recognition of share-based expenses.
R&D investment is gradually rising, and the three rates are gradually stabilizing. The company's research and development expenses have grown rapidly in recent years, and the company's research and development expenses in 2021 will be 0.86 billion yuan, +42.43% year-on-year. In 2021, the R&D expense ratio reached 20.03%, a year-on-year -4.38pct. In terms of rates, there is a trend of gradual decline. Among them, the increase in the management expense ratio in 2019 was due to the recognition of share-based payment fees. Excluding the impact of share payments, the expense rates in 2018, 2019 and 2020 were 59.90%, 52.80% and 35.05% respectively.
The company pays attention to innovation, and has strong technical strength to help localization. The company has gradually realized the localization of high-power, high-reliability, high-efficiency, wide-wavelength range single-tube chips, and the technical level is benchmarked against foreign advanced manufacturers. The company successfully introduced mainstream lasers and equipment manufacturers such as Ruike Laser, Chuangxin Laser, han's Laser and so on. The electro-optical conversion efficiency of the high-power bar chip is more than 63%, and the wavelength includes 808nm and 940nm, which is widely used in solid-state laser pumping sources. At the same time, the company is surprisingly upright, continuously expanding the surface emission chip by the side emission chip technology, building a technical process platform of two structures of side emission and surface emission, and expanding the high-efficiency VCSEL laser chip and high-speed optical communication chip two major product chip platforms. Among them, VCSEL chips basically cover the mainstream chip market demand, including proximity sensors, structured light, TOF, etc., and the next generation of dTOF technology VCSEL products are being developed. In the field of optical communication chips, the company has the production capacity of wafer manufacturing, chip processing, packaging and testing.
The development path is clear, and the layout of lidar and other fields is laid out. After years of development, the company has established two major material systems of semiconductor laser chip gallium arsenide GaAs and indium phosphide InP, and built two major chip structure process technologies and manufacturing platforms for side emission and surface emission. Looking forward to the future, the company will adhere to the development strategy of "one platform, one fulcrum, horizontal expansion, vertical extension". A platform: Suzhou Semiconductor Laser Innovation Research Institute jointly built by the company and the suzhou high-tech zone government. A fulcrum: the company has the core technology and full-process manufacturing process of high-power semiconductor laser chips. Horizontal expansion: from high-power semiconductor laser chips to VCSEL chips and optical communication chips, the product application field expands to consumer electronics, lidar and so on. Longitudinal extension: Extends longitudinally to laser devices, modules and direct semiconductor lasers.
(This article is for informational purposes only and does not represent any of our investment advice.) For usage information, see the original report. ）
Featured report source: [Future Think Tank]. Future Think Tank - Official website