Author: Su Qingtao
In an article published at the end of August last year titled “What is the “engine technology” of LiDAR? A comprehensive explanation of industry technology barriers“, Jiuzhang Autopilot raised the point that the real barrier of LiDAR lies in its laser transmission and reception system, rather than the scanning systems represented by terms such as “mechanical rotation,” “semi-solid-state,” “rotating mirror,” “prism,” and “MEMS.”
As market competition enters a more intense phase, the terms “mechanical rotation,” “semi-solid-state,” “rotating mirror,” “prism,” and “MEMS” that revolve around scanning methods have lost their novelty and can no longer spark interest in discussions. Thus, laser transmission and reception systems have become the new focus of the market.
Based on different degrees of system integration, the laser transmission systems of TOF LiDAR in mass production can be divided into EEL and VCSEL, while the reception systems can be divided into APD and SPAD. According to differences in the wavelength of the light source, laser transmission and reception systems can mainly be divided into two categories: 905 and 1550. Due to their impact on key factors such as “human eye safety,” “detection distance,” “power consumption,” and “cost,” the classification based on 905 and 1550 is worth discussing.
905nm and 1550nm are both wavelengths of light, each with its own characteristics, such as receiving characteristics, transmitting characteristics, susceptibility to interference, and impact on human eyes. Actually, there is no real distinction between them, but in the application of LiDAR products, they each have their own “advantages and disadvantages” depending on different aspects.
At present, both 905 and 1550 LiDAR have started to enter the mass production and delivery stage, and the LiDAR manufacturers each have their own opinions on which technology route is “better.” If only viewed from a single standpoint, their claims might often seem “completely true.” However, according to the head of a certain LiDAR manufacturer, “car companies are not particularly satisfied with either 905 or 1550.”
From a third-party perspective, it is irresponsible to make simple and hasty statements such as “who is better than whom.” Instead, it is more valuable to clear up misconceptions and doubts through layer-by-layer analysis. To further explore the truth, during the writing of this article, Jiuzhang Autopilot interviewed over 15 people, including manufacturers mainly focused on 905 LiDAR and three manufacturers exclusively focused on 1550 LiDAR. In fact, this article has become a “debate platform” for the two sides.To avoid any technical bias caused by individual interviewees or being misled by one-sided opinions, the author has confirmed the initial draft of this article with multiple different lidar manufacturers in various technical fields. This not only provides “attacked parties” with a chance to defend themselves, but also makes the content of the article as objective as possible.
Note 1: In FMCW, 1550 is a better choice than 905, which has become an industry consensus. Therefore, the comparison between 1550 and 905 in this article only refers to the time-of-flight (TOF) technical route.
Note 2: Interviewees in this article include but are not limited to, the technical director of HESA, CEO of TDT Bo Junwei, vehicle business director of ANNER Technology Party Na, chief engineer of Wanjie Lidar Hu Panpan, Liang Renliu from Hawk Semiconductor, assistant to the chairman of Fu Shi Technology Wang Lizidong, market director for Ouster China Liu Zhigang, Wu Jiaze from Hamamatsu Photonics, optical business manager of Zonghui Xin Guang Huang Xinfeng, Li Dan from Xiant Technology, Liu Shaodong from Yingxun Xin Guang, and optical communication expert Kuang Guohua.
Note 3: In some cases, this article quotes the opinion of only one interviewee. This opinion may not be agreed upon by other interviewees or may even be “completely disagreed with” by others. Similarly, in some cases, the author believes that most interviewees have reached a “consensus” on a certain opinion, so the specific source of the opinion has not been indicated. However, there is still a possibility that some interviewees will “completely disagree” with such opinions.
The Advantages of 1550 Go Beyond “Detection Range”
When it comes to 1550 Lidar, people’s first reaction is that compared with 905, the most obvious advantage of 1550 is the ability to operate at higher optical power while ensuring eye safety, thus achieving longer detection range.
Bo Junwei, CEO of TDT, believes that for vehicle safety, a detection range of 250 meters is the “passing score” for lidar, but currently, “only 1550 can achieve this level.”
Bo Junwei said:
“Usually, when people discuss the detection range of lidar, they are referring to targets with a reflectivity of 10%. However, in some extreme cases, small targets with a reflectivity of 5% should also be visible – not only visible, but also not as single point.”
“Based on our testing results and those of some customers, although some 905nm lidars have a detection range of 150 meters for very large targets with a 10% reflectivity and a detection range of approximately 100 meters for relatively dark targets with a 5% reflectivity, their detection range for small targets with a 5% reflectivity is only about 40 meters due to divergence angle and other reasons. However, TDT’s 1550 Lidar has been extensively tested for targets with a 5% reflectivity and a size of 20-30cm.”And what about other aspects?
Before answering this question, we need to first popularize a basic knowledge that is easily overlooked. The application of the 1550nm band in the field of LiDAR came earlier than 905nm, such as in the high-end surveying and mapping field, which uses the 1550nm LiDAR, which started two or three decades ago.
The reason why the high-end surveying and mapping field tends to use 1550nm is not primarily because of the “detection distance”, but because the divergence angle of 1550nm is much smaller than that of 905nm, as shown in the figure below.
A larger divergence angle means that as the beam propagates further, the spot size becomes larger, which may result in insufficient energy returning from small objects far away that are illuminated by the beam, thereby causing measurement failure. In addition, it also affects the accuracy of point clouds in complex scenes. If the spot is large enough to reach two objects that are very close together in front and back at the same time and cannot be decoded and distinguished within the LiDAR, then a “false point” may be formed between the two objects. When encountering objects with a height difference, such as zebra crossings or checkerboards, it may even measure “wave lines”.
According to the explanation of the person in charge from Hesai Technology, the emission surface of the 905nm laser has several hundred micrometers, while the emission surface of the 1550nm fiber laser is only about 10 micrometers. The smaller the emission surface, the smaller the far-field divergence angle of the spot can be suppressed. This means that compared with 905nm, 1550nm can provide better measurement for small objects at far distances in terms of laser divergence angle.
In addition, many companies believe that the anti-interference ability of 1550nm is much better than that of 905nm.
1550nm is a fiber laser, and the light energy is emitted from very thin single-mode fiber laser waveguide heads, so the light energy density is high with extremely high brightness. The light emitted by the LiDAR needs to overwhelm external light sources in brightness to resist their interference. Therefore, the light emitted by the 1550nm fiber laser has a brightness advantage over 905nm, which also means that its ability to resist interference from sunlight and other sources is stronger.
However, Hesai Technology stated that this view is one-sided, and whether it is resistant to environmental light interference mainly depends on the size of the detector, so it cannot be concluded that “1550nm is more resistant to interference”.
Is there a solution to the “easily absorbed by water” problem of 1550nm?
Here is the answer: This problem objectively exists, and there is no solution, but it is not serious.
In September of last year, when the author was conducting research on FMCW companies, most manufacturers mentioned that the 1550nm light waves are easily absorbed by water droplets (a few millimeters of water layer can absorb all the energy of the 1550nm laser), making it difficult to be reflected back, therefore, it is difficult to work normally on rainy days.At that time, an unnamed founder of an FMCW LiDAR company said, “People in our industry often exaggerate the capabilities of 1550nm LiDAR, claiming it has good penetration and anti-interference abilities. But I have conducted experiments by splashing water on a wall, and in the eyes of 1550nm LiDAR, the wall appeared as a dark, motionless surface until the water flowed down. Numerous experiments have shown that our 1550nm LiDAR becomes blind when encountering a thin layer of water only two millimeters thick. This is an unsolvable issue. Therefore, 1550nm LiDAR can only be used on cloudy, sunny, or light rainy days, and not during heavy rainfall.”
Meanwhile, the head of a TOF LiDAR company stated that the reason why 1550nm has minimal impact on human eyes is due to its easy absorption by liquid water. The laser is absorbed by the water in the eye before it reaches the retina. However, the same property that makes it easy to absorb by liquid water also makes 1550nm LiDAR unable to measure when there is water on the ground or decreases its detection capabilities.
However, a VP of a commercial autonomous driving company argues that 1550nm LiDAR does not become “invisible” when encountering rainwater. Rather its detection distance becomes shorter after a portion of its energy is absorbed.
According to a representative from a 1550nm LiDAR manufacturer, the issue of 1550nm being easily absorbed by water can be overcome by increasing the laser’s power output. “If half of your power is absorbed by water, then doubling the power output would overcome the problem.” However, increasing power output also means greater heating challenges, lower reliability, and a shorter lifespan.
Donna, the person responsible for the Ouster automotive LiDAR, believes that increasing power output is not cost-effective. “For example, even if you increase the output by 10W, the difference between that and increasing the output by 1W is not significant, which means the input-output ratio is not proportional and the effect is not significant.”
Why? Donna explains that the characteristics of the 1550nm wavelength are easily absorbed by water. Even if you increase the power output, most of the light waves emitted will still be absorbed by water.
The technical director at Hesai Technology also agrees that increasing laser power output would help with the absorption issue, but this is only a palliative measure. “Water has strong absorption properties, and several millimeters of water can absorb over 90% of the 1550nm light energy.”
According to Dr. Hu Panpan, the chief engineer of Velodyne Lidar, increasing power output is just one aspect. There are also signal processing strategies and algorithms that can be used to improve performance. “But for extreme fog or heavy rain, no amount of enhancement will suffice.”
However, in the opinion of Tooda CEO Bao Junwei, the issue of 1550nm being absorbed by water has been exaggerated. In reality, it is not a severe problem.The rain can be divided into two types: “raindrops” and “water curtains” or “water puddles” in a sheet-like shape. Among them, “raindrops” do not generally have the ability to completely shield the 1550 wavelength of light (on one hand, the probability of laser beam hitting raindrops on a normal rainy day is not that high; on the other hand, the spot size is much larger than the size of the millimeter-sized raindrops). Only sheet-like “water curtains” or “water puddles” can absorb it.
However, usually, the “rainwater” encountered by the forward-facing lidar is airborne “raindrops”, and there are few situations where the rainwater in front is tilted downward like a waterfall in a “water curtain” shape. When the rain is so heavy that the detected object’s entire surface is covered with as much as 2 millimeters of water “waterfall”, it is impossible for humans to drive at normal speed, and it is not realistic to expect autonomous driving to handle this kind of situation perfectly in the next few years.
Rainwater on the ground in a sheet-like shape, that is, “water puddles”, can indeed avoid trouble, but not just 1550, encountering this kind of rainwater, 905 will also be “blinded”.
Dang Na also holds a similar view. “There is not much difference between 1550 and 905 in the face of forward-facing rainwater.”
Is there a solution to the “heat dissipation difficulty” of 1550?
The detection distance of 1550 is longer than that of 905, at the cost of higher emission power. The typical power consumption of 905 is about 20W, while the typical power consumption of 1550 is more than 30W. High power means high power consumption and even means that heat dissipation is more difficult. Since the heat dissipation problem is not easy to handle, the automakers hope that Tier 1 can solve it, and Tier 1 hopes that lidar manufacturers can solve it by themselves.
The electro-optical conversion efficiency of 1550 is lower
Many experts interviewed pointed out that if the power is controlled to the same level as 905, 1550 has no advantage in detection distance, and is even worse than 905. Because overall, the electro-optical conversion efficiency of 1550 is lower than that of 905.
According to the data provided by Velodyne, the electro-optical conversion efficiency of 905’s VCSEL laser may reach 40% in consumer electronics, and close to 20% in lidar, while the electro-optical conversion efficiency of 1550 is only slightly over 10%.
The so-called “electro-optical conversion efficiency” refers to the efficiency with which a lidar converts electrical energy into light waves. During this conversion process, there is a certain degree of energy loss, and the amount of loss determines the high or low electro-optical conversion efficiency.
According to Kuang Guohua’s explanation, the longer the wavelength of light, the lower the energy of each photon. Correspondingly, carrier absorption will be relatively more, and more carrier absorption means greater losses. Therefore, the energy efficiency of 1550 is naturally lower than that of 905.In addition, Kuang Guohua, the technical leader of Hesai, Bao Junwei, Hu Panpan, and Dan Na pointed out that 905 uses semiconductor lasers, while 1550 uses fiber lasers. The difference in principle between these two lasers also affects the electro-optical conversion efficiency.
In simple terms, the light emitting mechanism of semiconductor lasers is the generation of photons by particle transition between the conduction band and the valence band, achieving direct electro-optical conversion; while fiber lasers have a seed laser (still a semiconductor laser), which first converts electricity into a 1550-nanometer seed optical pulse, and a high-power 940/980-nanometer pump laser is used to strike the gain fiber as the gain medium to amplify the seed light, adding a process of “light-to-light conversion” which inevitably causes energy loss.
The fiber laser system is more complex, with the electro-optical conversion efficiency of pump lasers in the fiber amplification process being around 30% and fiber coupling and amplification losses being around 40%, resulting in an electro-optical conversion efficiency of only about 12%. Hesai said: “In the industry, the electro-optical conversion efficiency of fiber lasers is generally only slightly over 10%.”
Dr. Hu Panpan said: “Overall, the electro-optical conversion efficiency of 905nm lasers is definitely higher because there is one less fiber-coupling and amplification process than in fiber lasers. The electro-optical conversion efficiency of semiconductor lasers depends on the development level of the entire material and semiconductor process, while the electro-optical conversion efficiency of fiber lasers depends more on the coupling process, fiber amplifier design and implementation, and improvement of the material properties of special fibers.”
So, can 1550 be made into a semiconductor laser to eliminate the “fiber coupling and amplification” process? Yes, but according to Dan Na, if this is done, there will only be a seed laser, which is far weaker in emitting power compared to 905, and the detection range will correspondingly be far less than 905.
An Indirect Reason: Too Few 1550 Lasers
In Hesai’s 905+VCSEL LiDAR, the number of lasers often corresponds to the “number of lines”, such as 128 lines has 128 lasers. However, according to publicly available information, although Luminar’s 1550 LiDAR claims to have 640 lines, it only uses one laser. Some 1550 LiDAR units use two lasers.
According to Hesai, because the whole unit only has 1-2 lasers, in order to achieve the effect of “128 lines” and “640 lines”, the mechanical movement frequency of the scanning component of the 1550 LiDAR naturally has to be much higher than that of 905 LiDAR with many lasers. Therefore, the aperture of the scanning mirror needs to be especially small. As a result, it is more difficult to receive photons, and in order to reduce the difficulty of receiving photons (i.e., to improve the light-receiving efficiency), only the power of the emitting end can be increased.# Translation in English Markdown text with HTML tags, preserving the original formatting
Party Nan also holds a similar view.
However, Dr. Hu Panpan’s statement is:
“If it is a coaxial optical path, this statement is true, but if a non-coaxial optical path is used, it may not be true- the receiving aperture can be enlarged separately. The difference between coaxial and non-coaxial optical paths is that the optical path design is different, and the non-coaxial optical path can improve detection efficiency by increasing the receiving field of view.”
As mentioned above, whether there is a possibility for 1550nm to have as many lasers as Velodyne 128, given that there are only 1-2 lasers and 1550nm is forced to work at higher power, the answer is: It is not feasible from both a cost and a technical engineering standpoint.
Firstly, the cost of 1550nm semiconductor lasers is usually several times that of 905nm semiconductor lasers. Therefore, if a high-line 1550nm lidar adopts the same number of lasers as 905nm, it cannot afford the cost.
Secondly, at the technical engineering level, fiber lasers are large in size and have high power consumption. If many lasers are used, not only the integration difficulty will be particularly high, but it will also cause the whole laser radar to be too large to be installed on the whole vehicle.
Another indirect reason: detection end
Lidar is a system engineering, so power consumption is not only limited by the laser emission end but also by the detection end.
Currently, there are already mature single-photon detectors for 905 detection ends (only a few photons are needed for detection), which are silicon-based SPADs. In contrast, the 1550nm detection end uses indium gallium arsenide (InGaAs), which has a lower photosensitivity than silicon-based materials (it needs hundreds of photons for detection). Therefore, to achieve relatively long detection distance, the power of the emission end needs to be much higher than that of 905.
Based on the above three points analysis, it is almost impossible to solve the problem of high power consumption and difficult heat dissipation of 1550 lasers, thus power consumption cannot be reduced to the same level as 905.
Single Laser Solution & Multiple Laser Solution
As mentioned in the previous section, the VCSEL of 905 can have “as many lines as there are lasers”, while the 1550 generally has only 1-2 lasers and is unlikely to have a multi-laser solution in the short term.
Bao Junwei and other 1550nm manufacturers believe that 1550nm fiber lasers themselves have higher pulse power and repetition frequency, and only one laser transceiver is needed to achieve high-repetition-rate detection.
However, some 905 Lidar manufacturers pointed out that when 1550nm works at higher pulse power and repetition frequency, there are two drawbacks:The cost of greatly increasing the power of each point is that the number of points per second that can be achieved by 1550 is much less than that of 905, which in turn affects the resolution.
In order to enlarge the field of view and the point cloud quantity, the basic solution for a single laser is to use a two-dimensional mirror, which makes its structure much more complex than the multi-laser solution (usually a one-dimensional mirror). Correspondingly, in order to maintain performance and stability, higher research and development costs are needed.
However, in response to the concerns of some 905 lidar manufacturers that a single 1550 laser “overload (several or tens of times the load of a 905 laser) will affect its life, reliability and robustness,” Bo Junwei’s response is: currently, even if the point frequency increases by more than ten times, the required load of the lidar on the 1550nm fiber laser is not considered “overloaded”. As long as the design process is good, this load is not a problem.
Can 1550 be made into a VCSEL chip system?
It is repeatedly mentioned in the previous text that 1550 is too large in volume, which prevents the number of lasers from increasing. So, can 1550 be made into a VCSEL chip (using multiple lasers) system to achieve the same high integration degree as 905?
The answer is no.
According to Hesai:
“Not everything can be made into a chip system just by thinking about it. A necessary condition is that the non-chip system is already a mature and stable system that has been verified, but the 1550nm system is not mature enough today, and its reliability and performance have not reached requirements. Therefore, it is unlikely to be achieved by chipification.”
According to different manufacturing processes of the resonant cavity, the semiconductor laser chip can be divided into two types: Edge-Emitting Laser (EEL) and Vertical-Cavity Surface-Emitting Laser (VCSEL). EEL forms a resonant cavity by coating an optical film on both sides of the chip, and emits laser parallel to the substrate. On the other hand, VCSEL coats an optical film on both sides of the chip and emits laser perpendicular to the surface of the chip. VCSEL has many advantages such as low threshold current, stable single wavelength operation, high-frequency modulation, easy two-dimensional integration, no cavity surface threshold damage, and low manufacturing cost. However, its output power and electro-optical efficiency are lower than that of EEL.
According to Hu Panpan, some research institutes and manufacturers are indeed working on 1550 VCSEL, and samples have been produced, but the power is relatively small.
However, most respondents believe that it is unlikely for 1550 to be made into a VCSEL.
A researcher from a certain 1550 lidar manufacturer said, “1550 uses a fiber laser, which requires a process of gain fiber amplification. Its output power is much higher than that of VCSEL, which is just a laser diode, and most of them are spatially directly coupled in array form, without amplification.”According to the opinion of Guohua Kuang, an expert in the field of optical communication and operator of the self-media “Optical Communication Women”, fiber laser has a high power and large volume, and making it as VCSEL surface emission will face a limitation in the electronic circuit, which will lead to reliability risks.
In addition, in his article “Why 1310/1550 VCSELs are Rarely Seen” written in 2019, Kuang Guohua made a detailed explanation of this topic. The core points of his views are as follows:
VCSEL’s resonant cavity relies on the DBR reflector mirror, which is composed of two materials with high and low refractive indexes. (The same materials, GaAs and AlAs, are used in the reflector mirror of VCSEL, and different wavelengths can be adjusted by changing the thickness.)
To achieve higher total reflectivity, if there is a large difference in refractive indexes between the two materials, only a few layers are needed; if the difference in refractive indexes is relatively small, many layers are needed.
The materials (AlGaAs with different proportions) with a luminescence wavelength between 850-905 have two materials with relatively large differences in refractive indexes, while the materials of the indium phosphide series at the 1550 band have very small differences in refractive indexes (3.21, 3.37, 3.35). This means that if 1550 is used to make VCSEL, the demand for materials will be much higher than that of 905, leading to huge costs.
In addition, the thermal conductivity of several 1550 materials (indium gallium arsenide or aluminum gallium indium arsenide) is much lower than that of 905 materials. If they are made into VCSELs, they are prone to burn out upon powering on.
What does this mean? If 1550 is chosen, its luminescent material cannot use the gallium arsenide series, and it must use InP, such as indium gallium arsenide or aluminum gallium indium arsenide. However, the difference in refractive indexes of indium gallium arsenide is very small and is not very suitable for DBR. If DBR is needed, because its difference in refractive indexes is relatively small, its single-layer reflectivity is relatively low, and it needs to be stacked with a large number of layers to achieve near 100% reflectivity. In other words, if the indium gallium arsenide material is used to make a reflector layer in combination with the gain layer of 1550, then it is difficult for heat to dissipate.
According to Na Dang’s statement, Angora also tried the 1550 VCSEL scheme but found it impracticable and abandoned it.According to Hu Xiaobo, the CEO of Raysea AI, making VCSEL at 1550nm “doesn’t make sense” because “the beam quality of the 1550 fiber laser is very good, and we mainly value the beam quality of the fiber laser when using 1550nm laser as the light source. If it’s made into VCSEL, it’s not much different from the 905nm semiconductor laser.”
Improving the detection end is harder for 1550nm than for 905nm. As mentioned before, one reason for the higher transmission power of 1550nm is the lower sensitivity of the detection end, which requires higher power than 905nm to achieve a longer detection distance. This means that one way to lower the power consumption of 1550nm is to increase the sensitivity of the detection end. However, since the silicon-based detector of 905nm cannot absorb 1550nm light, it is necessary to pair the 1550nm indium gallium arsenide laser with an indium gallium arsenide detector. It is already mature to use 1550nm detectors in optical communication industry. The focus now is to expand its temperature range and improve its reliability to better meet the needs of automotive scenarios.
Currently, even if we ignore the factors of yield and cost, in terms of detection efficiency, response speed, and operating temperature range, 905nm silicon detectors have obvious advantages over 1550nm. However, Hesai believes that although the power of 905nm laser is ultimately limited by eye safety, longer detection distances can be achieved by increasing the efficiency of the detector. At present, the mainstream 905nm detector has started to use SPAD or SPiM with higher photosensitivity instead of APD. In the future, continuous improvement of single-photon detection efficiency is the main way to improve 905nm on detectors.
SPAD or SPiM are both “single-photon detectors”. The concept of “single-photon detector” is in contrast to detectors such as photodiodes (such as avalanche photodiodes APD) and CMOS image sensors, which have a detection limit of tens to hundreds of photons. Compared to the latter, the most significant advantage of single-photon detectors is their higher photosensitivity.Before explaining the detection efficiency, let’s first understand the working principle of a single photon detector. SPAD and SPiM are essentially photodiodes (PD). By continuously increasing the reverse voltage, PD enters the avalanche mode (also known as “Geiger mode”). In avalanche mode, one incident photon on the detector can generate a gain of 10 to the power of 6. Therefore, the effect of single photon detection can be achieved.
The detection efficiency of a single photon detector is a probabilistic parameter, meaning that when I receive a photon, there is a probability of triggering avalanche or no reaction. For example, if the detection efficiency is 10%, it means that out of 100 incident photons, 10 of them can trigger the avalanche.
Detection efficiency = quantum efficiency * photon avalanche probability * fill factor.
For silicon detectors, the quantum efficiency is basically higher than 90%; photon avalanche probability is related to the production process; and the fill factor of Hamamatsu’s detectors is more than 70%.
According to Wang Lizidong, the assistant to the chairman of Fortrend Technology, “to improve detection efficiency, one is to improve the probability of capturing photons through the design of SPAD devices and two is material doping and process improvement. However, the two are not immediately effective because with the improvement of efficiency, noise will also be amplified. Therefore, most of the time, efficiency and noise should be considered comprehensively. Higher efficiency is not always better. For example, a certain big company in Japan has very good noise suppression, even if the detection efficiency is a bit lower, it does not affect anything.”
Currently, the detection efficiency of single photon detectors is around 10%, which means that one avalanche is triggered per 10 incident photons. According to HeSai, this value will gradually increase to 20% and 30% in the future.
However, according to HeSai, the main beneficiaries of the single photon detector are 905, and it is “extremely difficult” to make 1550 into a single photon detector that can be used normally in a car environment (currently, PD and APD are mainly used for the 1550 detection end).
HeSai said: “The indium gallium arsenide material used in the 1550 detection terminal has a large amount of noise, especially when it is made into a single photon detector, it needs to be cooled to a temperature below minus 20 degrees to suppress the noise generated by the material itself. It is difficult to imagine a detector that needs to be cooled to below minus 20 degrees can survive in a car environment. This is why single photon detectors are rarely used for the detection end of 1550 now.”
In comparison, silicon single photon detectors used in consumer electronics have been fully verified.
At the end of April this year, the head of HeSai Technology posted a message on WeChat, saying that in addition to mentioning that it is difficult to make 1550 into a single photon detector suitable for a car environment,
He also said:> “The reason why we ultimately abandoned 1550 is that, through our analysis of future technology and industry chains, we believe that even in terms of performance, 1550 will lose out to 905 in a few years, and once 905 surpasses 1550 in performance, the gap will continue to widen. Hence, we redirected almost all of our research and development resources to the silicon-based route, focusing on the development of upstream core components and chips of the 905nm LiDAR.”
>
Regarding this view, a researcher at the Chinese Academy of Sciences said:
“Modern integrated circuits are based on silicon technology, which has led to the development of advanced technology and a more developed industrial chain for silicon-based materials. Therefore, the 905nm detector based on silicon technology has a very good technological basis for iterative upgrading to improve performance and reduce costs. At the same time, the 905nm detector can be directly integrated with integrated circuits and can therefore rely on the computational power of integrated circuit digital chips to improve performance.”
“The 905 detector is based on silicon-based or CMOS technology, and there is an unwritten industry rule that any application scenario accessible to silicon-based or CMOS technology will gradually be eroded by silicon-based or CMOS technology.”
“To use an analogy, silicon technology follows Moore’s Law and has developed a ladder to reach the sky for decades. 905nm can use this ladder to climb up quickly and rely on it to develop.”
However, Bao Junwei believes that high photosensitivity is a feature of single photon detectors, “but this may not be an advantage in the practical application of LiDAR for autonomous driving, depending on the specific system design.”
Can 1550 be used in Flash LiDAR?
In interviews over the past year or two, the industry generally believes that the ultimate goal of TOF LiDAR is Flash LiDAR. Currently, the lasers used in Flash LiDAR are mostly 905nm (Ouster uses 865), so is it possible to make Flash with 1550 in the future?
Bao Junwei and Ouster’s China Market Director Liu Zhigang both believe that it is technically feasible, but it is not necessary. Bao Junwei said that the cost would be too high, Cheng Zhengxi said that this is an “over-designed” feature that nobody would pay for, and Liu Zhigang believes that this is a “useless overkill,” as “the 1550 divergence angle is small, and the beam quality is high, which is very suitable for point sources of light, but FLASH is a surface light source that does not require any movable scanners.”
Production and Manufacturing
In addition to the performance and reliability discussed above, production and manufacturing are key to the mass production and delivery of LiDAR.
Velodyne states:”For 905nm devices, the industry scale of $300 billion CMOS image sensor can be utilized for larger wafer sizes, smaller pixel sizes, and, more importantly, there are over 20 mature super factories worldwide employing this production process. On the other hand, for 1550nm devices, the dependent optic communication industry chain is less than 1% of that for 905nm devices, thus production processes and manufacturing levels are limited. Currently, smaller wafer sizes and lower pixel sizes are adopted for 1550nm devices. As the application of 905nm devices continues to expand, the difference in production and manufacturing between the two will only widen further.”
In response to the question of whether the cost of 1550nm devices can be reduced to the level of 905nm devices as demand increases, a spokesperson for a certain LiDAR manufacturer said, “As demand rises, the price of a gold ring will decrease, but no matter how much it decreases, it cannot become as cheap as a copper ring. The BOM cost of raw materials is there for everyone to see.” However, Bao Junwei’s opinion is, “It depends on how much the price difference is, just like how few people currently buy a black-and-white screen smartphone to save a few hundred yuan.”
According to Jiuzhang Autonomous Driving, although most LiDAR manufacturers have reserved 1550nm technology, they still believe that 905nm will be the mainstream for a considerable period of time. Traditional Tier 1 LiDAR manufacturers such as Bosch, Denso, and Valeo all promote 905nm LiDAR. Of course, Bao Junwei’s opinion is, “Except for Valeo, who fixed on 905nm technology early on, the other companies are not the mainstream LiDAR providers, and Valeo’s application cannot be considered successful.”
In particular, Huawei has deep accumulations in the optical communication field and is considered by peers to “understand 1550 the most.” Logically, they would bet on 1550nm, but in reality, according to insiders, Huawei’s LiDAR mainly uses 905nm. “They must have made a very detailed comparison before coming to this conclusion.”
Some people say that over 60% of the funding in the LiDAR industry is going towards 905nm, and with more money, the maturity of the industry chain will improve quickly.
The head of Hesai Technology believes that 1550nm solutions are inferior to 905nm in terms of performance, reliability, and cost, so 1550nm will not be the future trend but only a transitional solution with relatively certain advantages in single-point ranging capacity in the past two years. As the performance of the silicon-based single-photon detector for 905nm improves, the advantages of 905nm over 1550nm will become greater over time.
Bao Junwei’s opinion on this matter is:> “The 1550nm route has been able to achieve stable and reliable results of 250 meters @10% for several years and has been put into mass production. It is relatively easy to further improve its performance to meet the needs on OEM roadmap. However, the 905nm route currently requires a lot of effort to achieve 150 meters @10%, and only reaches 200 meters by sacrificing some important indicators such as detection reliability in automatic driving outdoor scenes. Its further improvement will be even more challenging. Therefore, the gap between 905 and 1550nm will gradually widen.
Of course, the above discussion is all about TOF LiDAR. When the FMCW technology route matures, a situation of 905 and 1550 competing equally will emerge. Because FMCW has speed information and much stronger anti-interference capabilities than TOF, these are the advantages that 905’s TOF does not have. And FMCW is naturally 1550nm.
References
Why is it rare to see 1310/1550 VCSEL?
https://mp.weixin.qq.com/s/x7JRJTVnH4EX1FE3C9O0Sw
Wavelength-stabilized technology based on 905nm edge-emitting lasers helps to reduce cost and miniaturize LiDAR systems
https://www.ledinside.cn/qiye/20211222-51334.html”
This article is a translation by ChatGPT of a Chinese report from 42HOW. If you have any questions about it, please email bd@42how.com.