Mechanical LiDAR relies on a motor-driven turntable or prism to achieve 360-degree rotation scanning of the laser beam; semi-solid-state LiDAR has a stationary receiving or transmitting module, with only the scanning components (such as rotating mirrors or MEMS micro-mirrors) undergoing mechanical movement; while all-solid-state LiDAR completely eliminates any mechanical moving parts, relying on optical phased array (OPA) or flash (floodlight array) technology to achieve instantaneous laser scanning of the entire detection scene. Each of these three architectures has its advantages and disadvantages in terms of technological maturity, cost, size, and lifespan. For different autonomous driving scenarios and levels of requirements, a comprehensive trade-off is needed based on performance indicators, mass production feasibility, and automotive-grade certification requirements.
Comparison of different types of LiDAR
The core working principle of LiDAR can be summarized as follows: the transmitter emits laser pulses, and the direction of the pulses is controlled by an optical system; when the laser encounters a target object, it is reflected back to the receiver. The receiver converts the reflected light signal into an electrical signal, and the back-end processing unit analyzes the round-trip time (ToF) and intensity information of the light pulse to obtain the distance between the target and itself and the reflection characteristics. Based on the same principle, different types of LiDAR differ significantly in optical scanning methods, the integration of transmitting and receiving modules, and the number of mechanical parts. Mechanical products, which follow the traditional three-module structure of "rotary disk + laser emitter + receiver array", have a mature and stable technical route, but this also results in problems such as large size, high cost and limited lifespan. Semi-solid-state products achieve the separation of scanner and target material by restricting mechanical movement to smaller components (such as rotating mirrors or MEMS micromirrors), thereby reducing size and cost. All-solid-state products resolve the depth information of the entire scene through optical phased arrays or floodlight arrays without any mechanical moving parts. They have the characteristics of small size, high reliability and strong mass production potential, but the technology is still in a rapid evolution stage and has not yet fully met the requirements of automotive-grade large-scale mass production.
Mechanical LiDAR
Mechanical LiDAR, as the earliest commercially available LiDAR architecture, is typically implemented using a motor-driven rotating bracket (or a rotating disk containing multiple reflecting prisms). This allows multiple laser emitters and receivers fixed to the bracket to scan the surrounding environment horizontally in 360 degrees. A representative example is Velodyne's early HDL-64E, which employs a 64-line multi-emitter, multi-receiver design. Through rotating scanning, it generates high-density point cloud data in real time, enabling accurate object detection and classification around vehicles both day and night. Mechanical LiDAR technology is relatively mature, with long detection ranges (often exceeding 200 meters, and up to 200 meters or more for targets with 10% reflectivity) and high point cloud density, meeting the long-range, high-precision perception requirements of L4 and higher-level autonomous driving in complex environments. However, due to the need for sophisticated mechanical structures and the coordinated operation of multiple optical components, its cost is high, its structure is large, its power consumption is high, and its heat dissipation is challenging. Furthermore, the lifespan of rotating components is difficult to fully match the requirements of automotive-grade applications.
Semi-solid-state lidar
In contrast, semi-solid-state LiDAR relies significantly less on mechanical components. It typically retains only one or more small scanners (such as rotating mirrors or MEMS micromirrors) to guide the laser beam across a specific angle range in the horizontal or vertical direction, while the receiving module acquires the echo signal through a fixed optical array. Depending on the type of scanner, semi-solid-state LiDAR can be further divided into two typical approaches: rotating mirror type and MEMS galvanometer type. Rotating mirror type semi-solid-state LiDAR usually uses a single-axis or dual-axis motorized rotating mirror, controlling the laser emission direction by high-speed rotation or oscillation of the mirror surface. It has the advantages of moderate cost and relatively low mass production difficulty. MEMS galvanometers utilize micro-electro-mechanical systems (MEMS) technology, integrating tiny mirrors onto the surface of a photolithographically engineered chip. Electrostatic driving signals cause the mirror surface to oscillate rapidly within a micrometer scale, thereby enabling the laser beam to sweep across a predetermined angle range. Whether rotating mirror or MEMS version, semi-solid-state LiDAR is significantly superior to mechanical products in terms of size, power consumption, and production cost, and is relatively easy to obtain automotive-grade certification.
Taking rotating mirror semi-solid-state LiDAR as an example, the current mainstream level in the market is concentrated on the "hundred-line level" design. For example, the 760 ultra-thin automotive LiDAR launched by Wanji Technology adopts a rotating mirror + multi-transmitter multi-receiver solution to achieve true 192-line scanning. The detection distance of 10% reflective surface can reach 200 meters, and can be extended to 300 meters. The horizontal field of view can reach 120°, the vertical field of view is 25°, the vertical resolution is 0.15°, and the horizontal resolution is 0.13°. The overall body thickness is only 24-30 mm. It has excellent point cloud quality and automotive-grade reliability, and is mainly aimed at advanced driver assistance systems (ADAS) scenarios. The above excellent performance makes rotating mirror semi-solid-state LiDAR one of the mainstream choices in the current autonomous driving market, and it has been successively applied to many mass-produced models.
MEMS galvanometer-based semi-solid-state LiDAR offers significant advantages in scanning accuracy and size. It achieves precise and rapid scanning of the laser beam by miniaturizing a high-speed vibrating micromirror array on a silicon substrate. Compared to rotating mirror solutions, the MEMS galvanometer structure avoids the inertia and scanning speed limitations imposed by large mechanical components, while further reducing the radar's thickness and overall size, making it easier to integrate with vehicle body designs. This architecture has garnered attention from numerous domestic and international LiDAR companies. For example, RoboSense's M1 series, based on MEMS galvanometers, achieves a wide horizontal and vertical field of view, high point cloud density, and stability, and has already secured orders from several automakers. Due to the unique nature of MEMS technology, the mass production cost of these products is expected to decrease significantly as the technology matures, thereby promoting the further application of semi-solid-state LiDAR in L2+ and L3 level autonomous driving mass-produced vehicles.
All-solid-state LiDAR
Mechanical and semi-solid-state LiDAR each have their own characteristics, but they both face compromises in terms of size and reliability. To completely eliminate the wear and lifespan limitations caused by moving mechanical parts and further reduce costs, all-solid-state LiDAR has become the industry's recognized "ultimate form." Typical technologies for all-solid-state LiDAR mainly include optical phased arrays (OPA) and Flash. OPA technology is similar to electronic scanning arrays in the radar field. By controlling the phase difference of the phase modulator, the laser beam at the transmitting end can scan in a preset direction or pattern without moving any physical components, and the echo signal is used in the phased array at the receiving end to perform corresponding phase calculations to recover three-dimensional depth information. Since no mechanical movement is required, OPA-type LiDAR has optimal reliability and potentially the smallest size, theoretically significantly reducing mass production costs. However, its core challenges lie in the manufacturing yield of the phased array chip, phase modulation accuracy, and the complexity of large-scale integrated optoelectronic device packaging. Currently, OPA-type products still face certain technical obstacles before large-scale automotive-grade mass production.
Flash-type all-solid-state LiDAR employs an instantaneous full-field illumination method similar to a camera flash. This involves emitting a laser pulse across the entire detection scene at once, then simultaneously acquiring the reflected echoes through a large-area receiving array to obtain a depth map of the entire scene. Compared to OPA technology, the Flash architecture eliminates the need for phase modulators or complex micro/nano-scale optical cavities. It only requires optical beam expanders and homogenizers along the optical path, along with a large-scale array of photodetectors, to achieve instantaneous, large-field-of-view depth acquisition. This architecture excels in short-range scenarios (such as intelligent parking and unmanned delivery robots), enabling millisecond or even sub-millisecond-level full-scene depth imaging. However, for high-speed automotive scenarios, challenges remain regarding signal-to-noise ratio and receiver sensitivity. Furthermore, the dual requirements for high-power laser emission and high-density receiving arrays in long-range detection necessitate further breakthroughs in mass production costs, thermal management design, and optical system stability for Flash-type all-solid-state LiDAR.
What are the advantages and disadvantages of different types of lidar?
From a performance perspective, mechanical LiDAR has long maintained a leading advantage in detection range and point cloud resolution, supporting accurate ranging of common targets with 10% reflectivity at distances exceeding 200 meters. Its 360-degree horizontal and vertically balanced omnidirectional field of view provides more complete environmental information for perception algorithms. However, its unit price often reaches tens of thousands of US dollars, and its size and height reach tens of centimeters, significantly impacting vehicle integration and wind resistance. In contrast, semi-solid-state LiDAR typically has a detection range of 100-200 meters (with rotating mirror designs even exceeding 200 meters), a vertical line count between 32 and 128 lines, and a horizontal scanning field of view that can be flexibly set via mechanical or MEMS micromirrors. However, it cannot achieve true 360-degree surround view and usually requires deep fusion with other sensors (such as cameras and millimeter-wave radar) to fill blind spots. The cost of semi-solid-state solutions has dropped to the thousand-dollar level, and the overall size can be compressed to a thickness similar to that of a camera at radio height, making it easier to achieve hidden vehicle body integration.
All-solid-state LiDAR theoretically possesses the potential for the smallest size, lowest cost, and highest reliability. If mass production yield reaches the target, its unit price can be controlled to a few hundred dollars or even lower. Simultaneously, eliminating mechanical parts significantly extends its lifespan, enabling tens of thousands of hours of trouble-free operation, which is beneficial for true automotive-grade mass production applications. Furthermore, besides emphasizing the feasibility of mass production, it is also crucial to address issues such as scattering attenuation, noise interference, and thermal effects during long-range detection. OPAs, limited by the phase control accuracy of the phase array, exhibit speckle and sidelobe noise during actual ranging, impacting detection accuracy and anti-interference capabilities. To meet the detection requirements of high-speed driving scenarios, Flash architectures require coordinated optimization in three aspects: laser emission power, receiver sensitivity, and the synchronous readout circuitry of the large-scale pixel array. This places higher demands on the manufacturing processes of current CMOS or Geiger mode avalanche photodiode (GPD) technologies.
How should I choose?
To meet the needs of different levels of autonomous driving, mechanical, semi-solid-state, and all-solid-state LiDARs can be functionally matched. For L4 and L5 ultra-high-level autonomous driving, vehicles require no human intervention in most scenarios, placing extremely high demands on the performance and redundancy of the perception system. High line count, high resolution, and long-range detection become essential to ensure timely detection of pedestrians, bicycles, other vehicles, and road obstacles in complex urban road environments or highway scenarios, and to perform accurate trajectory prediction and path planning. Mechanical LiDARs, with their mature and stable performance, were once the first choice for L4/L5 level test vehicles, but their high cost and automotive-grade compatibility limited the possibility of mass production applications. Therefore, although major autonomous driving companies initially equipped their test vehicles with a large number of mechanical Si-LiDARs, such as Waymo and Cruise's Velodyne HDL and Quanergy M8 products, they were gradually replaced by semi-solid-state arrays in mass-produced vehicles.
For Level 3 and below autonomous driving (including Level 2 Advanced Driver Assistance Systems (ADAS), Level 2+, and Level 3 semi-autonomous driving), semi-solid-state LiDAR is a more reasonable choice due to its better balance between performance and cost. Taking the currently mainstream rotating mirror semi-solid-state product as an example, it can provide sufficient detection range (generally within 150-200 meters, achievable for targets with 10% reflectivity) and a high vertical line count (64,128 lines). Combined with 360-degree or partial multi-sensor stitching, it can meet the functional requirements of lane keeping and automatic emergency braking (AEB) in urban or highway scenarios. MEMS galvanometer semi-solid-state LiDAR, with its smaller size, allows for concealed installation inside the vehicle, reducing wind resistance and visual intrusion, making it suitable for mass-produced vehicles with higher styling requirements. When semi-solid-state LiDAR is combined with cameras, millimeter-wave radar, millimeter-level high-precision maps, and high-performance domain controllers, a more redundant and secure multimodal perception system can be built, providing sufficient performance guarantees and cost-of-ownership control for Level 3 autonomous driving.
In the past two years, numerous domestic and international automakers and LiDAR companies have engaged in in-depth cooperation and mass production deployment around semi-solid-state and all-solid-state technologies. For example, the 96-line semi-solid-state LiDAR jointly released by Huawei and Jifo has undergone large-scale mass production testing on the Jifo Alfa S Huawei HI version; XPeng Motors' G9, in cooperation with RoboSense, uses the M1 semi-solid-state LiDAR with a MEMS galvanometer solution; GAC Aion, WM Motor M7, and other models have also successively equipped themselves with rotating mirror semi-solid-state LiDAR. Looking at all-solid-state LiDAR, Wanji Technology's 750 all-solid-state blind spot radar has been practically applied in several low-speed autonomous driving scenarios in China (such as sanitation vehicles and delivery robots), demonstrating the feasibility of all-solid-state technology in short-range scenarios. Furthermore, many startups such as Innoviz, Aeva, Hesai, and domestic companies like RoboSense and LeiShen Intelligent are conducting product development and process iteration around OPA and Flash technologies, striving to push all-solid-state LiDAR to the commercialization stage with automotive-grade mass production capabilities by 2025-2026.
Mechanical lidar, relying on precision machining and high-end optical components, remains difficult to reduce to automotive-grade prices at mass production scale. For example, Velodyne's early 64-line mechanical lidar cost over $50,000 per unit; the 32-line model reached over $20,000, which is clearly unsustainable in mass-produced vehicles. Semi-solid-state lidar, by reducing mechanical components and improving optical and electronic integration, has achieved a unit price of around $1,000, with costs continuing to decrease as shipments increase. Currently, major suppliers are collaborating with Tier 1 suppliers and OEMs to continuously optimize automotive-grade certification, reliability testing, and supply chain construction, striving to reduce the cost of semi-solid-state lidar to the $500-$800 range. Simultaneously, with the maturation of MEMS manufacturing processes, its monolithic integration and yield rates are further improving, and further cost reductions are expected within the next two years.
All-solid-state LiDAR theoretically possesses the greatest potential for cost optimization. If high-yield chip manufacturing and modular packaging can be achieved, its overall cost will be far lower than semi-solid-state or even mechanical solutions. However, from the current technological maturity perspective, all-solid-state still faces key bottlenecks in mass production scale, automotive-grade reliability, and a large-scale supporting ecosystem. The OPA architecture is limited by the speckle effect and temperature drift issues of phased array chips, requiring more R&D investment in chip process and system thermal design; the Flash architecture is affected by optical power divergence and receiving noise, requiring higher-performance high-speed AD (analog-to-digital converter) and large-scale pixel-level readout circuitry. Currently, the typical road test deployments of both OPA and Flash-type all-solid-state LiDAR are mainly concentrated in low-speed logistics, unmanned delivery, and intelligent manufacturing scenarios. To achieve large-scale application in L3+ or L4 level autonomous driving scenarios, breakthroughs are still needed in reliability verification, automotive-grade temperature adaptability, and high-power heat dissipation management.
Future Trends
In the future, the development of LiDAR technology will continue to advance towards the goals of "higher performance, lower cost, smaller size, and higher reliability." In terms of performance, high-line-count (e.g., thousand-line-count) high-precision designs will become possible, further increasing point cloud density and reducing noise, helping autonomous driving systems more accurately identify minute targets and texture features in extremely complex environments. Regarding cost, deep integration with CMOS processes will allow laser emission and reception modules to be integrated onto the same silicon substrate, leveraging the production capacity and mature packaging and testing processes of semiconductor giants to achieve relatively controllable mass production prices. In terms of size, with the increasing maturity of technologies such as optical integration, electronic integration, and optimized heat dissipation solutions, the size of all-solid-state modules is expected to shrink to a level comparable to that of a vehicle's front-facing camera module, greatly improving the flexibility of appearance design. In terms of reliability, high-temperature resistance and high-vibration-resistance automotive-grade certification will become the new industry benchmark, requiring companies to make more comprehensive investments in environmental adaptability testing, EMC (electromagnetic compatibility) testing, and long-term stability verification.
Furthermore, the deep fusion of LiDAR with other vehicle sensors (cameras, millimeter-wave radar, ultrasonic sensors, etc.), as well as its collaboration with high-precision maps, edge computing platforms, and V2X (Vehicle-to-Everything) communication, will jointly drive the overall upgrade of the perception layer of autonomous driving systems. In the future autonomous driving architecture, sensor fusion will no longer be just simple redundancy, but rather a deep association and joint reasoning of multi-dimensional and multi-modal information. For example, in extreme weather conditions such as low light or rain and fog, although camera performance degrades, millimeter-wave radar and LiDAR can still maintain strong detection capabilities; in high-speed scenarios, high-line-count mechanical or semi-solid-state radar can provide long-range warning information; in low-speed scenarios such as close-range obstacle avoidance and parking, all-solid-state Flash radar can quickly complete short-range depth reconstruction; and phased array technology can further improve anti-interference capabilities and target classification capabilities in the future, providing more accurate and reliable depth information for autonomous driving systems.
Mechanical, semi-solid-state, and all-solid-state LiDARs each represent different stages and focuses in the development of LiDAR technology. Mechanical LiDARs are characterized by mature technology and high detection accuracy, but due to their large size, short lifespan, and high cost, they are gradually transitioning to semi-solid-state and all-solid-state LiDARs. Semi-solid-state LiDARs, on the other hand, have become the mainstream choice for mass-produced vehicles due to their moderate cost, controllable reliability, and performance sufficient to meet the requirements of L2+ to L3 level autonomous driving. All-solid-state LiDARs, with their minimal size, optimal reliability, and strongest cost optimization potential, have become the industry's ultimate goal, but key technical challenges still need to be overcome in areas such as phased array chips, Flash detection device processes, and system integration.
When selecting a LiDAR solution, OEMs should comprehensively consider factors such as target autonomous driving scenarios, functional requirements, and cost budgets. Evaluations should be conducted across multiple dimensions, including performance indicators, automotive-grade certifications, mass production feasibility, and supply chain maturity, to ensure the most suitable LiDAR solution is matched for different vehicle series and levels of autonomous driving products. With technological advancements and the improvement of the industry chain, it is expected that between 2015 and 2026, semi-solid-state and all-solid-state LiDAR will accelerate their penetration into multi-level autonomous driving vehicles, ultimately achieving a complete transformation from high-cost experimental products to low-cost, mass-produced components.
