ALLPCB
Introduction
Advances in silicon photonics enable integration of active and passive devices of LiDAR transmit and receive modules on a single chip, reducing volume, improving stability, and lowering cost, which supports wider adoption in applications such as autonomous driving.
According to Mems Consulting, the research team led by Chen Xiaolin at the Southwest Institute of Technical Physics reviewed this field, covering LiDAR fundamentals and common ranging principles, analyzing typical on-chip LiDAR scanning schemes, and discussing current challenges and development directions. The review was published in the journal Progress in Laser and Optoelectronics under the title "Review of On-Chip Silicon LiDAR Technologies."
LiDAR Fundamentals
Laser Wavelength
Considering atmospheric transmission windows, eye safety, and available laser and photodetector types, LiDAR wavelengths commonly range from 0.8 to 1.55 μm. Table 1 in the original review lists corresponding laser and detector types. For automotive LiDAR, ambient temperature variation means the laser emission wavelength must remain within the passband of any optical filters used to suppress background light, which is an important consideration when selecting the light source.
Mainstream LiDAR wavelengths are 905 nm and 1550 nm. Pulsed LiDAR often uses 905 nm lasers; an advantage is the availability of less expensive silicon-based detectors and lower optical loss compared with 1550 nm in some contexts. However, 905 nm light can penetrate the ocular media to the retina, so peak power is limited for eye-safety reasons. The 1550 nm band is suitable for long-range continuous-wave LiDAR because light is absorbed in the anterior eye and does not significantly reach the retina, allowing higher emitted power. This wavelength is widely used in communications and benefits from mature, low-cost continuous-wave laser sources.
Detection Range
Detection range is the maximum distance at which a LiDAR can detect a target and is primarily constrained by emitted optical power. For pulsed LiDAR, range also depends on the pulse repetition period T, since echoes must return within T to resolve correct distances; echoes returning after T can cause range ambiguity. For FMCW LiDAR, the laser linewidth Δν affects the coherent length Lc: narrower Δν yields longer Lc, and targets beyond the coherence length produce much lower echo SNR and may not be detected. Range specifications are typically given for targets with low reflectivity (for example 10%), and actual performance depends on environmental and surface conditions.
Field of View
Field of view (FOV) denotes the angular region a LiDAR can sense, usually expressed in degrees. Automotive applications require both horizontal FOV (HFOV) and vertical FOV (VFOV). Larger FOV increases angular coverage and situational awareness.
Measurement Accuracy and Resolution
Measurement accuracy combines precision and trueness. Precision refers to the repeatability of distance measurements under the same conditions, while trueness refers to how close the mean measured value is to the true distance. These are affected by random and systematic errors in the measurement process, respectively. Distance measurement trueness is mainly influenced by optical signal generation and measurement electronics; angular trueness depends on the accuracy of beam steering at the transmitter.
Measurement resolution includes range resolution and angular resolution. Range resolution is the minimum distinguishable distance between two targets in a single measurement. For pulsed LiDAR, narrower pulse width improves range resolution. For FMCW LiDAR, range resolution is inversely proportional to modulation bandwidth and depends on frequency linearity. Angular resolution is the minimum distinguishable angle and can be improved by increasing the aperture size.
Ranging Principles for On-Chip Silicon LiDAR
Pulsed Time-of-Flight (TOF)
The pulsed TOF method uses direct detection: measuring the time for a laser pulse to travel to a target and back. TOF LiDAR has a relatively simple ranging principle and system design. Pulsed LiDAR can suffer from range ambiguity when the echo is shifted by one or more pulse periods relative to the transmit pulse. Scattering and propagation losses reduce received pulse energy and SNR, limiting range. To increase effective range while respecting eye-safety limits, pulse bursts can be used to reduce per-pulse peak power while integrating received energy over multiple pulses to improve SNR and accuracy. Despite limitations, pulsed LiDAR remains competitive due to its implementation simplicity.
Amplitude-Modulated Continuous-Wave (AMCW)
AMCW, also called indirect TOF, modulates the emitted laser amplitude with a period longer than the round-trip time and compares phase between received and emitted signals. Range resolution depends on modulation frequency and phase measurement resolution; higher modulation frequency improves resolution. As with pulsed methods, phase repeats every 2π, reducing unambiguous range. A common solution is to use multiple modulation frequencies: a high-frequency carrier for fine range measurement combined with one or more lower-frequency channels for ambiguity resolution.
Random Modulation Continuous-Wave (RMCW)
RMCW encodes the emitted light with a pseudo-random bit sequence (PRBS) in amplitude or phase and estimates time-of-flight by correlating the received signal with the original PRBS template using matched filtering. Because PRBS correlates only with itself, RMCW is robust to sunlight, artificial lighting, and other LiDAR signals. However, RMCW is sensitive to relative target velocity, laser phase noise, and speckle, presenting technical challenges. Companies such as Baraja have implemented RMCW combined with prism-dispersion-based spectral scanning in commercial prototypes.
Frequency-Modulated Continuous-Wave (FMCW)
FMCW LiDAR uses a time-varying frequency modulation (triangular, sawtooth, sinusoidal, etc.). The received light is mixed with a local oscillator; the beat frequency of the intermediate-frequency (IF) signal yields target range. Triangular modulation can simultaneously provide range and velocity via Doppler shift, while sawtooth is typically used for range only. Sinusoidal modulation requires frequency offset adjustment and is generally applicable to single-target scenarios. When a target moves, Doppler shift causes different beat frequencies on the rising and falling linear-frequency segments.
Compared with pulsed ranging, FMCW offers several advantages: coherent detection that is less susceptible to sunlight and nearby LiDAR interference, direct velocity measurement via Doppler, range precision determined by modulation bandwidth rather than receiver bandwidth, and higher sensitivity that allows lower optical power and avoids nonlinear effects in waveguides. FMCW systems can use relatively low-frequency receiver electronics and often do not require avalanche photodiodes or ultrafast detection circuits, which favors silicon photonic integration. The main technical challenge is generating highly linear frequency sweeps; solutions include interferometric calibration, optical frequency combs, microresonator references, or external cascaded optical modulators.
On-Chip LiDAR Scanning Implementations
Flash Array LiDAR
Flash pulsed LiDAR operates like a camera: a high-repetition-rate pulse illuminates the entire FOV, and a detector array measures time-of-flight per pixel to produce 3D images. Examples include a SPAD-based chip from Delft University of Technology built in 0.18 μm CMOS with 252 × 144 pixels and 1728 12-bit time-to-digital converters (TDCs), with on-chip partial-histogram readout to compress data. Other implementations include block-illumination strategies and compact 0.18 μm HV CMOS flash LiDAR chips that demonstrated indoor 3D imaging at 4.5 m and outdoor ranging beyond 20 m with relative error under 0.35% in reported experiments.
Flash LiDAR has no moving parts, offering vibration resistance, small size, and low cost. Point clouds are generated from single pulses rather than point-by-point scanning, simplifying synchronization. Drawbacks include the need for high peak power to illuminate a large area, relatively low SNR and limited range/FOV, and sensitivity to target reflectivity. Range and angular resolution are constrained by detector array scale and performance; large arrays increase data processing load, requiring trade-offs between imaging resolution and speed. Weak echoes often necessitate expensive SPAD detectors.
Optical Phased Array (OPA)
An OPA typically consists of a splitter network, phase shifters, and optical antennas. Phase shifters introduce controlled phase delays (thermal or electro-optic), and antennas couple light to free space using grating couplers, edge couplers, or end-fire emitters. Beam steering is achieved by controlling the phase across waveguides to shape the wavefront.
In 2017 the MIT Electronic Research Laboratory demonstrated an on-chip FMCW LiDAR using triangular-wave modulation. Analog Photonics demonstrated a coherent 2D solid-state LiDAR prototype using two large OPAs, showing real-time 3D coherent imaging with sufficient detail to resolve limbs of a person at 7 m. Samsung has demonstrated a 32-channel silicon photonics OPA with integrated semiconductor optical amplifiers (SOAs) and later integrated a tunable laser, SOA, and 32-channel OPA on an 8.7 mm × 3 mm chip using III-V-on-Si processing. Research groups have also produced large-channel-count OPAs on SiN-SOI platforms.
OPA is compact and compatible with low-cost fabrication and has no moving parts, enabling rapid random pointing with high directional gain. Scaling OPAs is challenging because all antennas require precise amplitude and phase control; most large-FOV OPAs are effectively one-dimensional arrays, with the orthogonal scan axis achieved by tuning the laser wavelength. OPAs also face challenges including optical loss, array crosstalk, thermal stability, and limited on-chip optical power, which constrain long-range detection.
Lens-Assisted Beam Steering (LABS)
LABS combines an on-chip switch/antenna array with an on- or off-chip lens, aligning the array with the lens focal plane. Switches (Mach-Zehnder interferometer, MEMS, or ring-resonator switches) route light to specific antennas; the lens collimates and steers the selected beam into free space. Only one emitter is active at a time, so scanning is discrete.
Demonstrations include a solid-state pulsed LiDAR based on LABS and integrated 2D on-chip transceiver arrays supporting beam transmit, steering, and receive. A high-resolution chip LiDAR using a MEMS-based 128 × 128 focal-plane switch array integrated on a silicon photonics chip has been reported. LABS implementations can achieve fast random scanning, binary-switch control simplicity, flexible antenna layouts for high pixel density, and true 2D steering. Limitations include the need for many switches to reach high spatial resolution, lens aberrations that can distort beams, and potential blind spots when steering step sizes are large.
Slow-Light Grating (SLG)
SLG emits a guided mode into free space and controls the beam angle by changing wavelength or effective refractive index, exploiting slow-light effects to increase beam-angle sensitivity. A 2022 demonstration combined SLG with Ge photodiodes to implement an on-chip FMCW LiDAR system using zero-difference detection.
Compared with OPA and FPA, SLG reduces the burden of large-scale antenna integration and phase calibration, and large-range, high-resolution scanning can be achieved by thermo-optic tuning of a fixed laser wavelength. Limitations include sensitivity to edge temperature nonuniformity when using thermal tuning, slower scan rates due to heating, potential need for temperature control systems, high demands on etch uniformity, and currently higher optical loss and noise that limit detection range. SLG is still at an early research stage and requires further validation for practical use.
Other LiDAR Architectures
If only fixed-angle ranging is required, beam steering can be omitted and the system simplified. For example, Thales demonstrated an FMCW LiDAR on a silicon photonics platform. Such fixed-angle architectures can be simple and low-cost but may introduce complexity due to requirements for circulators and collimators at the output, limiting some on-chip advantages.
Summary and Outlook
LiDAR offers high precision and robustness to interference, making it a key sensor for vehicle perception and autonomous driving applications. Traditional mechanical scanning LiDAR faces cost and reliability constraints, while solid-state LiDAR approaches have strong development potential. Companies in China and abroad have invested in product development and industry deployment for solid-state LiDAR.
Silicon photonics has matured across materials, equipment, fabrication, packaging, and testing, providing a promising integration platform for LiDAR. This review focused on four solid-state on-chip scanning approaches: flash, OPA, LABS, and SLG, summarizing research progress over recent years and highlighting technical characteristics. Flash LiDAR is simple and relatively mature but limited in range and precision, so it is currently suited for low-speed or lower-precision scenarios. OPA offers compactness, fast scanning, and potential for low-cost mass production, and it is a major current research direction despite remaining challenges in the supply chain and manufacturing. LABS and SLG are less mature and require more demonstration and technical accumulation before commercialization.
Beyond scanning subsystems, other LiDAR components such as lasers, modulators, amplifiers, and photodetectors have seen chip-level progress, but discrete chips still require fiber or free-space coupling that introduces challenges in power consumption, size, and stability. With continued development of heterogeneous integration platforms, it is foreseeable that future on-chip solutions will integrate the active and passive components needed for complete LiDAR systems, improving stability, simplifying manufacturing and installation, and reducing size and cost, thereby enhancing LiDAR competitiveness in autonomous driving and related fields.
