Recently, the emergence of self-driving vehicles has revolutionized conventional travel and marked a new era of high-level artificial intelligence. As a remote sensing technology, light detection and ranging (LiDAR) systems with active real-world perception are essential to autonomous driving, which requires a 360° panoramic field of view (FOV) and quick target recognition [1–3]. Traditional LiDAR systems rely on mechanical components to steer beams [4,5], facing challenges in bulkiness, high cost, and low reliability. The optical phased array (OPA), an innovative beam control technology rooted in microwave phased array theory, enables beam steering through simply manipulating the phases of individual optical antennas [6–8]. It features non-mechanical operation, high flexibility, and rapid response, and could synergize with other photonic devices such as modulators and detectors, positioning itself as a promising solution for the development of all-solid-state chip-scale LiDAR systems. Nevertheless, most OPAs suffer from limitations in FOV due to unwanted grating lobes [9] and diffraction-restricted envelopes [10]. Grating lobes divert power from the main lobe, leading to aliasing, while the diffraction envelope affects the uniformity of the emitting power. Moreover, the characteristics of single-beam emission restrict the speed of target recognition [11]. These issues hinder the potential utility of OPAs in intricate LiDAR applications.

OPAs with 180° aliasing-free beam steering have been extensively studied. A prevalent approach is reducing the spacing between antennas to half-wavelength or less and introducing effective index mismatches [9,12–18]. This is widely recognized as an effective way and theoretically supports 180° steering range, as it ensures constructive interference in a unique direction. Typically, antenna arrays are designed with non-uniform widths [9,12–14] or curved geometries [15–18] to modify the effective index distribution for enhanced crosstalk suppression. Although this compact antenna configuration improves FOV, it is constrained by a narrow far-field envelope from the antenna aperture, resulting in notable power attenuation at the edges and shrinking the effective steering range. Alternatively, non-uniform antenna distributions can broaden FOV with wider spacing, but this increases background noise and complicates detection [10,19–23]. Consequently, current methods achieve a maximum practical FOV of only 160° [22]. In addition, existing methods for multiple beams predominantly rely on either systematically assembling sub-antenna arrays on a chip or cascading numerous chips in free space [24–27], which are complicated due to the usage of bulky elements. Overall, there remains a critical gap in compact OPA solutions that combine an ultra-wide FOV with multi-beam steering capability for panoramic vision and rapid target identification.

Here, we propose a silicon-based OPA that accomplishes an ultra-wide lateral FOV of 360° with parallel steering ability. A common two-dimensional (2D) grating serves as the optical antenna, and it is shared by four independent OPA units. These units enable simultaneous emission and manipulation of beams to establish four distinct sub-FOVs, thereby quadrupling the available steering range. By utilizing a non-uniform waveguide array with half-wavelength spacing, each unit experimentally provides a 120° coverage, collectively facilitating seamless 360° steering through the synthesis of sub-FOVs. The distinct transmission directions and orthogonal polarization states of beams from different ports help minimize mutual interference, allowing each OPA unit to operate individually and stably. Moreover, we apply a dissipative Kerr-soliton (DKS) microcomb with frequency-modulated continuous wave (FMCW) signals into the proposed OPA system to demonstrate parallel coherent detection. The comb lines are spatially dispersed with diffractive gratings and separately measure distances and velocities in a real parallel state. By using 10 comb lines combined with multi-beam characteristics of the system, simultaneous detection of 40 objects is achieved, resulting in a high ranging accuracy of 1 cm and a velocity measurement error of less than 0.5 cm/s. This innovative approach presents a promising solution for high-performance LiDAR systems.

Figure 1(a) illustrates the schematic diagram of the proposed silicon-based OPA. A 2D grating is incorporated as an optical antenna for light radiation, with all four input ports being effectively utilized and connected to components including waveguide arrays, beam splitters, and phase shifters, enabling the system to function as four independent OPA units. Light from an external laser is coupled into the chip via an end-face coupler. The beam-splitting network of each OPA unit, composed of cascaded multi-mode interferometers (MMIs), divides the light into multiple coherent channels. Each channel passes through a phase shifter for independent phase control to achieve beam steering. A non-uniform-width waveguide array with spacing of half-wavelength (775 nm) is employed to obtain a wide aliasing-free FOV with minimal crosstalk, and the selected waveguide widths are 580, 380, 560, and 400 nm, as shown in Fig. 1(b). When the laser is coupled into one of the waveguides, the simulated optical field distribution is displayed in the upper image of Fig. 1(c), and the corresponding crosstalk between adjacent waveguides is calculated to be below $-20\mathrm{dB}$. In contrast, the optical field distribution of a uniform waveguide array shown in Fig. 1(c) reveals a significant mode coupling. The 2D grating is fabricated by etching a series of circular holes into the silicon surface, forming a symmetric square pattern. To optimize the radiation efficiency of the grating, we conduct simulations and adjust the diameter of the central circular hole under fabrication constraints, choosing a diameter of 480 nm, a periodicity of 700 nm, and an etching depth of 70 nm. Under these conditions, the radiation efficiency of the 2D grating remains around 30% across 1500 to 1600 nm wavelength range, ensuring effective beam emission, as illustrated by the red line in Fig. 1(d). The blue line in Fig. 1(d), on the other hand, depicts the dispersion characteristics of the grating, indicating that different operating wavelengths result in varying diffraction angles. Light transmitted through the waveguide array of each OPA unit enters the 2D grating in a plane-wave-like manner, subsequently being radiated into free space to form four parallel beams at different azimuths. Each beam is independently manipulated to scan distinct regions, resulting in four discrete sub-FOVs, as presented in Fig. 1(e). These sub-FOVs are seamlessly synthesized to achieve panoramic perception. Owing to the orthogonal polarization states of beams from adjacent ports and reciprocal propagation directions from opposing ports, the grating ensures low-crosstalk transmission and emission, providing exceptional stability and reliability for the multi-beam OPA system.

For each OPA unit, lateral steering is accomplished by tuning the phase difference, while wavelength adjustment harnesses the diffraction properties of the 2D grating to enable an additional dimension of steering. The overall lateral steering range of the proposed OPA is determined by the synthesized sub-FOVs, and Figs. 2(a)–2(d) qualitatively illustrate the simulated far-field distributions of a single OPA unit under four different phase differences. Increasing the phase difference between adjacent waveguides allows for an increase in steering angle, and a noticeable side lobe appears next to the main lobe when the phase difference reaches 180°, limiting the achievable steering range. It can be observed that the longitudinal angle of the far-field spot shifts as the steering angle changes during lateral steering, resulting in a curved trajectory. This effect is likely due to the variation in incident angle of the beam relative to the grating vector at different lateral deflection angles. In Figs. 2(e) and 2(f), the far-field distributions with four OPA units operating simultaneously are presented. Four identical and discrete far-field patterns can be observed, and four sub-FOVs steer different regions in parallel. Some interference occurs when four units operate simultaneously, possibly caused by mismatched sizes between the waveguide array and the 2D grating, leading to additional diffraction that slightly increases under such operating conditions. With the coordination of four sub-FOVs (each with 130°), the lateral 360° beam steering is attainable. On the other hand, the longitudinal far-field distributions of a single OPA unit are given in Figs. 2(g) and 2(h). At a fixed phase difference, the longitudinal angle is adjustable by tuning the operating wavelength. Considering the finite width of the far-field spot, we define the angle at the maximum light intensity as the exact steering angle, leading to further quantitative analysis of the far-field distributions, as illustrated in Fig. 2(i). A single OPA unit has a 130° steering range and a 10° full width at half-maximum (FWHM) for the 0° far-field spot. Although half-wavelength spacing theoretically allows for a 180° aliasing-free FOV, the actual steering range is constrained by the far-field envelope of the individual antenna. The wide width of the far-field spot arises from the small number of waveguide channels; it however does not affect the steering characteristics. Figure 2(j) shows the normalized curve of light intensity for longitudinal steering.

Figure 3(a) displays a simplified schematic diagram of the steering range testing for the fabricated device. The packaged chip under testing is positioned at the center of a circular rail, while a near-infrared lens and a camera are mounted on the rail and can be rotated to capture the spots emitted by the OPA chip at any lateral angle. Figure 3(b) shows a micrograph of the device, which is fabricated on a 220 nm thick SOI wafer with a 2 μm buried layer. TiN micro-heaters are deposited above waveguide channels for phase control, and the entire device is covered with a silicon oxide cladding. To facilitate independent control of each OPA unit, their input ports are discretely configured. The external laser output is split and coupled into each port, allowing each unit to be activated either simultaneously or sequentially for beam steering. Heater electrodes are wire-bonded to a printed circuit board (PCB) for applying voltages to conduct large-scale phase adjustments. The image of the packaged chip is shown in Fig. 3(c). To compensate for additional phase shifts caused by fabrication errors and thermal crosstalk, an initial phase calibration is performed using a genetic algorithm, where a Gaussian function is employed as the target beam intensity distribution curve. The algorithm iteratively optimizes the voltage distribution, allowing the far-field beam to progressively converge toward the desired target distribution. For a single OPA unit, we experimentally obtain a lateral steering range of 120°. Figure 3(d) shows the synthetized images of far-field spots at different angles, exhibiting an arc-shaped scanning trajectory consistent with theoretical simulations. This indicates a 30° overlapping region between adjacent sub-FOVs. As a proof-of-concept demonstration, each OPA unit comprises only eight channels. During the testing, a focusing lens is employed to compress the spot, achieving an excellent angular resolution of $0.62°\times 0.19°$. This performance can be further enhanced by expanding the channel array. By tuning the wavelength from 1500 to 1600 nm, a longitudinal steering range of 15.75° can be achieved, and the corresponding synthetized far-field pattern images are shown in Fig. 3(e). Sequential activation of each OPA unit enables continuous lateral 360° steering, with each unit covering a 90° FOV, as illustrated in Fig. 3(f). By utilizing the 2D grating as a single antenna, the proposed OPA generates four independent sub-FOVs with overlapping edges, facilitating 360° steering in the lateral direction, while the orthogonal vertical steering dimension remains severely limited by factors such as the tunable wavelength range of the light source, the diffraction efficiency of the grating, and the wavelength tuning efficiency, like traditional OPAs.

Next, we characterize the beamforming loss of the fabricated OPA by measuring the output beam power using a free-space power meter. The experimental results for the total beam power within the FOV of a single OPA unit are shown in Fig. 4(a). Under an input power of 0 dBm, the output beam power remains approximately $-15\mathrm{dBm}$ when steering from $-45°$ to 45°, with fluctuations less than 3 dB, corresponding to an estimated beamforming loss of 15 dB. However, the power reduction is observed at larger steering angles. We further measure the beam power at 0° across various operational wavelengths, as presented in Fig. 4(b). The results reveal that beamforming losses remain consistent within the 1500–1600 nm range. Additionally, the beam power across the lateral 360° FOV is evaluated, as shown in Fig. 4(c). The power degradation at the edges of individual sub-FOVs can be mitigated through multi-FOV synthetization, resulting in a relatively uniform power distribution.

We continue to perform a proof-of-concept demonstration of parallel multi-target ranging and velocimetry. Specifically, the frequency-modulated DKS microcomb is fed into the proposed OPA chip, where it is diffracted by the 2D grating at varying longitudinal angles within each of the four sub-FOVs, and each spectral channel could acquire both distance and velocity information simultaneously. The testing system is shown in Fig. 5(a), and the distance information is mapped from the frequency variation of the echo signals. Based on Kerr-nonlinear four-photon interactions, a DKS microcomb is generated by pumping a ${\mathrm{Si}}_3{\mathrm{N}}_4$ micro-ring resonator (MRR) and stabilized with the auxiliary laser-assisted thermal balance scheme [28,29]. Figure 5(b) presents the measured optical spectrum with a 100 GHz line spacing. Utilizing a wavelength-selective switch (WSS), 10 comb lines between 1540 and 1570 nm are extracted as multiwavelength carriers for subsequent testing. An arbitrary waveform generator (AWG) is employed to generate an FMCW signal with a bandwidth of 10 GHz and a modulation period of 200 μs. The signal is loaded onto the individual comb line, as shown in Fig. 5(c). After optical amplification, a power splitter separates 90% power into the signal path and the residual power into the coherent receiver as a local reference. Then, the signal is launched into free space by the proposed OPA to probe reflectors mounted on a motorized slide rail. To be noted, the emission of these 10 comb lines is achieved through synchronized phase adjustment using a shared phase shifter. Consequently, the emission states might not be optimal, as the phases are not tailored to any specific wavelength while the comb lines have different initial phase profiles. This issue can be addressed through a specialized design, such as incorporating a Bragg grating before the shared phase shifters to manage the dispersion of different wavelengths and align their initial phases [30]. Given the spatial separation and different orientations of the diffracted comb lines, we employ a single fiber collimator to sequentially receive the reflected signals from each comb line, which are individually mixed with the reference signal in the coherent receiver. The resulting beat frequency, proportional to the distance of the target object, can be displayed on the oscilloscope screen through performing the fast Fourier transform (FFT) on the time-domain waveform. The target distance is determined by $d=\mathrm{\Delta }fc/2k$, where $\mathrm{\Delta }f$ is the beat frequency, $c$ is the speed of light, and $k$ is the frequency sweep rate. Velocity measurement is also executed by manipulating the reflector in reciprocating motion on the rail. As the radial velocity of the target causes a Doppler frequency shift in the echo signal, two frequency differences $\mathrm{\Delta }{f}_1$ and $\mathrm{\Delta }{f}_2$ can be resolved by Fourier analysis. Based on these two frequency differences, both distance and velocity information can be confirmed.

To validate the capability of high-precision ranging, we first select a single comb line for measurements. The motorized target is set to move along the rail and away from the collimator in steps of 10 cm, with the distance between them being measured. Figure 5(d) depicts that all ranging results have a precision of 1 cm. For velocity characterization, we record the beat signals of moving targets at four distinct speeds through continuous measurements. These signals, shown in Fig. 5(e), showcase that the signal-to-noise ratios (SNRs) are approximately 17 dB, and corresponding velocimetry errors are less than 0.5 cm/s. The exceptional detection performance of a single comb line ensures the feasibility of parallel detection. Based on the intrinsic multi-beam characteristics of the proposed OPA and diffraction effects of the grating, the selected 10 comb lines could disperse longitudinally within four sub-FOVs, enabling the simultaneous parallel detection of 40 objects. In experiments, the target objects are located approximately 1 m away from the chip. The measured SNRs and ranging errors are displayed in Fig. 5(f). For simplicity, the ranging results of 20 spectral channels are presented as representatives. To mitigate the impact of surface irregularities and disturbances in the optical fiber link on the test results, multiple measurements are conducted for each channel. Consequently, the results from all spectral channels exhibit high consistency, with measurement imprecision below 1 cm. Following this, by adjusting the voltage distribution applied to the phase shifters to induce beam deflection from the OPA, the 10 longitudinally separated comb lines can be steered laterally to detect another 40 targets. With the aid of a microcomb, one-dimensional scanning is sufficient for panoramic target detection. It should be noted that the selectable number of comb lines is constrained by the bandwidth of the grating antenna. By designing a broadband one, the parallelism can be further enhanced. This innovative system offers a novel approach to the development of high-performance LiDAR systems, significantly enhancing detection efficiency and system versatility.

A comprehensive comparison with existing OPA solutions has been given in Table 1. The proposed OPA demonstrates significant advantages in lateral steering range and optical power uniformity. In this work, we utilize a 2D grating as the emitting antenna and leverage its excellent symmetry to share it across four independent OPA units. This design constructs different oriented sub-FOVs, effectively expanding the overall lateral steering range, and achieves relatively uniform output power distribution through multi-FOV synthetization to compensate for the degradation in edges. Based on the diffraction characteristics of the 2D grating, the proposed OPA typically emits beams upward, forming a cone-shaped 360° FOV by combining the laser beams from four OPA units. Unlike widely used mechanical or MEMS-based vehicular LiDAR systems that symmetrically emit beams toward the horizontal direction to detect objects primarily in the front, side, and back, the proposed top-facing OPA is particularly more suitable for applications that necessitate monitoring and tracking targets in the air or above.Table 1.

Performance Comparison of State-of-the-Art OPAs^b,c

The 2D grating, serving as the antenna in the proposed OPA, is fabricated by etching circular holes into silicon waveguides. Variations in the dimensions of these holes, such as their diameter and etching depth, significantly affect the grating’s diffraction efficiency, which in turn, influences the OPA’s output power and effective detection range. Given the critical role of the 2D grating as an antenna, optimizing its design to accommodate fabrication tolerances is essential.

In addition, we utilize the fabricated OPA system combined with the microcomb to demonstrate high-precision parallel detection at a distance of 1 m. This short-range detection system shows great potential in applications such as smart homes for multi-device interaction, and facial recognition systems. To achieve high-precision long-range detection, it is essential to address the low output power of each comb line and the limited sensitivity of receivers, which require further research on materials with higher damage thresholds, highly sensitive receiving systems, and advanced signal processing technologies.

Aligning and receiving each beam with a collimator is complex and not conducive to the practical application of multi-beam OPA systems. Future research will focus on addressing this issue by achieving heterogeneous integration of OPA chips and detectors to enable integrated receivers. Additionally, by utilizing advanced signal processing techniques, signals of different wavelengths will be pre-encoded at the transmitter to prevent crosstalk during multi-channel reception. Through these integrated methods and advanced signal processing technologies, the reception and isolation of multi-beam signals will be realized, eliminating the need for separate alignment with discrete collimators.

We have proposed and experimentally demonstrated a silicon-based multi-beam OPA. A specific design with multiple OPA units sharing a 2D grating as a single emitter is introduced, and the lateral 360° FOV is experimentally observed. Furthermore, we combine the fabricated OPA with a DKS microcomb to demonstrate the massively parallel coherent LiDAR multi-target distance and velocity measurements, validating ranging errors of less than 1 cm, velocimetry errors of less than 0.5 cm/s, and signal-to-noise ratios of around 17 dB.