Light detection and ranging (LIDAR) provides relatively long-range and high-precision 3D imaging, essential for future autonomous vehicles [1–4]. The optical phased array (OPA) is a key component of LIDAR, and OPAs have been recently proposed as a promising solid state technology to achieve full integration due to its low material cost, high integration, and compatibility with CMOS technology [5–9]. OPA has been explored on various photonic platforms, and silicon nitride (Si3N4) has recently become of interest for waveguides in integrated OPAs due to allowing for higher input power, larger transparency, and lower phase variation [10–14]. Compared with Si3N4, lithium niobate (LN) has not only the same advantages as Si3N4, but also excellent electro-optic (EO) Pockels material with a large coefficient [15–19]. With the development of thin-film LN technology, the LN-on-insulator (LNOI) platform shows great potential for OPAs.

The OPA technique keeps moving towards high performances, such as a wide field of view (FOV), aliasing-free, high scanning speed, low power consumption, and small size. Unlike radio frequency (RF) phased arrays, emitting antennas cannot be spaced by a half-wavelength, and uniformly spaced antenna arrays lead to typically unwanted grating lobes and sidelobes when the emitter spacing is larger than half-wavelength. Well-designed aperiodic arrays have been proposed as a promising approach to achieve a wide steering range and high resolution simultaneously at the cost of decreased main lobe efficiencies and signal-to-noise ratios (SNRs) [20–24]. Recently, a new beam steering technology known as lens assisted beam steering (LABS), based on an on-chip switch/emitter structure and off-chip lens, has attracted much attention [25–31]. As the array size increases, the complexity and power consumption of the driver circuit to achieve accurate phase control become prominent for traditional OPA. This design allows potentially good SNR suppression and scaling of element counts while maintaining low power consumption. Nevertheless, additional external lenses are required, making the entire system unfriendly and bulky.

Fortunately, metasurface devices, featuring ultrathin, ultralight, and flat architectures, can be engineered to provide beam engineering functionalities with high flexibility [32–36]. They are used to incorporate with an active medium as beam steering devices, such as in systems of transparent conducting oxide materials, multiple quantum wells, micro-electro-mechanical systems (MEMS), liquid crystals, or phase-change materials [37–41]. However, few studies directly integrate metasurfaces into OPAs for beam steering with higher performances. For example, the expansion of the scanning FOV was reported by incorporating metasurfaces and a silicon photonic circuit in the spatial light path [42,43].

In this paper, a metasurface-based port-selected OPA (POPA) is designed and implemented on the LNOI platform to enable a large FOV. The POPA uses aperiodic waveguide arrays, and metasurfaces demonstrate ±41.04° lateral beam scanning with aliasing free and 7.06° longitudinal beam scanning via wavelength tuning by 95 nm. A typical LABS structure requires a fixed lens, and it is still a challenge to integrate a lens with a large numerical aperture (NA) for wide FOV angle steering. Here, in contrast, our POPA relies on waveguide arrays to act as an in-plane Fourier transform, which replaces the function of the bulky lens. To further broaden the FOV, we integrate metasurfaces to realize twice the steering range by polarization multiplexing. Our approach provides a new proof-of-concept metasurface empowered OPA with high beam steering efficiency.

The device is schematically shown in Fig. 1(a), and includes mainly three parts: waveguide arrays, grating radiation antenna, and metasurface. The proposed metasurface-based POPA design for beam steering in two dimensions is shown in Fig. 1(a), and steering the beam in the θy dimension and θx dimension can be achieved by switching input waveguides and tuning the wavelength. The metasurface is integrated on top of POPA to enlarge the FOV. Here, the LNOI platform is employed, which contains a 600 nm thick layer of LN film, and a buried layer of SiO2 with thickness of 2 μm on the Si substrate. Ridge waveguides are designed 600 nm wide, with a 300 nm slab thickness and 300 nm ridge height, and the non-uniform space range of the waveguide array is from 0.4 μm to 0.8 μm. Figures 1(b) and 1(c) show the top and cross-sectional views of the non-uniform waveguide array we designed. The wavefront interferes in the LN slab and emits into space through the grating, so the cross talk between the antennas is negligible, and those features lead to simple manufacturing and small fabrication-induced phase errors.

We adopt an aperiodic waveguide array with non-uniform distribution of coupling coefficients [see Fig. 1(e)], i.e., ci=c0(N−12)2−i2+N−12+i, where N is the total waveguide number, indices i range from −(N−1)/2+1 to (N−1)/2, and c0 is a scaling factor introduced for experimental reasons. This engineered waveguide array is designed analogous to the matrix elements of the angular momentum Jx-operator, which turns out to be an elaborated definition of the discrete fractional Fourier transform. Consequently, the designed waveguide arrays should be able to transform a point source into a plane wave, i.e., the in-plane Fourier transform [44,45]. The coupling coefficient between two LN waveguides with respect to the waveguide gap can be simulated by using 3D FDTD Solution, as shown in Fig. 1(d). As such, we are able to obtain the waveguide gap distributions in the aperiodic waveguide array [see Fig. 1(e)]. The optical field evolution in this waveguide structure is described by the coupled-mode theory (CMT) −i∂∂zφi=βiφi+ci−1,iφi−1+ci,i+1φi+1,(1)where φi and βi denote the optical field and propagation constant of the ith waveguide, respectively, and ci,j+1 is the coupling coefficient between waveguides i and i+1. Figure 1(f) shows the CMT calculated light propagation patterns. The input optical field has the form (0.47, 1, 0.47), and it is shown that the Ey field propagates to a uniform intensity distribution around the center waveguides. The input pattern (0.47, 1, 0.47) will give rise to a flatter intensity distribution and phase front at the output by inverse design. Importantly, the wavefront exhibits flat features, and the phase-tilt increases as the input port gradually moves away from the central waveguide [see Fig. 1(e), bottom]. It is evident that our designed aperiodic waveguide array can transform a single waveguide input into a nearly uniform intensity distribution with a flat wavefront, which gives rise to the point-selected phase difference.

In simulations, a 21-waveguide array is chosen in modeling through a 3D FDTD Solution. The input is a TE mode defined by placing the sources at the fifth, sixth, and seventh waveguides with an amplitude ratio of 0.47:1:0.47 to satisfy required input amplitudes. As shown in Fig. 2(a), the simulated electrical field distribution for light propagation in the waveguide array and emitting antenna region validates that the radiated waves have a uniform amplitude distribution in the emitting antenna region. To confirm the phase distribution, we simulated the Ey profile of the emitting antenna region at z=0.4μm, and it is clear that the tilted phase is consistent with the phase we designed [Fig. 2(b)]. In addition, Fig. 2(c) shows the Ey profile of the yz plane at x=105μm, which indicates the tilted phase from waveguide emission to space.

In experiments, to further improve the resolution of beam steering, we implemented an OPA device employing 41 waveguides, and the propagation length from the input to the radiating antenna was L2=200μm. The SEM images of the coupling grating, waveguide array structure, and emitter antenna are shown in Fig. 3(a). The performance of the OPA was measured using a Fourier-space imaging system with an objective lens of NA=0.75 and an infrared sensor camera (XenicsXeva-1.7-320) (see Appendix A for details). From Fig. 3(b), the far-field spot size of port -2 is 2.74°×3.76° without sidelobes at the wavelength of 1550 nm measured by a Fourier-space imaging system. The POPA far field and line profile by switching from port -6 to port 6 are shown in Fig. 3(c). The optical intensity is concentrated mainly in the main lobe without any sidelobes in the range of ±21.27°, which indicates that the power is concentrated on the main lobe and the POPA, which gives rise to aliasing-free beam steering. In fact, we fabricated another sample with 21 waveguides the same as the theoretical simulation (not shown here). It is found that the FWHM of this 41-waveguide OPA decreases from 5.07° to 2.74° in the θy dimension as compared with the 21-waveguide one, revealing the improved resolution. In addition, when we sweep the wavelength from 1470 nm to 1565 nm, we achieve a 7.06° scanning range (see details in Appendix B).

To expand the scanning angle, a metasurface-based POPA was subsequently implemented on the LNOI platform, as schematically shown in Fig. 4(a), where a well-designed metasurface is integrated on top of the OPA grating with a buffer layer. The metasurface is an array of subwavelength antennas with spatially varying geometric parameters to locally modulate the optical response, so as to flexibly manipulate the phase, polarization, and propagation of light. Here, a polymer (SU-8) layer was spin-coated to fill the waveguide gaps and cover the grating, and to act as a buffer layer to protect the structure of POPA. Then, amorphous Si nanopillars as metasurface components are distributed on the SU-8, where the unit cell is schematically shown in Fig. 4(b). The rectangular nanopillar units are designed with sizes of 370nm×220nm at a height of 1 μm and arranged in a square lattice with a period of 800 nm. Light travels in the waveguide with TE mode corresponding to linear polarization (LP), which can decompose into left-circular polarization (LCP) and right-circular polarization (RCP). In the experiment, we design the metasurface to deflect the LP light into RCP and LCP with two different output directions. The deflection is obtained by using a geometric phase that results from the polarization conversion of radiation light [32,33]. The metasurfaces will not only convert the LP into LCP and RCP, but also add a spatial phase term based on their orientation angles. The simulated polarization conversion rate (PCR) is as high as ∼99.9% for the unit structure to guarantee high efficiency at the incident angle of 0° [see Fig. 4(c) and Appendix C]. We chose a horizontal phase gradient to deflect the radiation light from POPA into the desired angle. According to the generalized Snell law of refraction [32], sin(θt)nt−sin(θi)ni=±λ02πdϕdx,(2)where θi is the radiation angle of POPA;θt is the angle of refraction; ni=1.57 is the refractive index of SU-8; nt=1.0 is the refractive index of air; λ0=1550nm; dϕdx introduces a phase gradient to increase the angles of refraction for RCP and LCP, respectively. So, the radiation angle writes as θt=arcsin(sin(θi)nint±λ02πntdϕdx).(3)

Because of the change of the ambient medium of POPA (from air to SU-8), the angle of radiation into air decreases from 21.27° to 19.13°. As the meta-induced deflection angle introduced by metasurfaces is a constant within ±15° of the incident angle, the linear phase gradient we designed introduces deflection of about 20° to meet the angle expansion and the steering of the whole FOV [Fig. 4(d)]. According to theoretical calculation, if the in-plane incident angle range can meet ±25°, the FOV of a metasurface empowered OPA would possibly reach 180°, which is an extremely high value in OPA scanning range. We also theoretically calculated the diffraction efficiency of the metasurface by 3D FDTD Solution (∼77.37%), and the working efficiency of the final device in the experiment was about 1.7%.

In experiments, we fabricated the metasurface by using electron-beam lithography (EBL, Elionix ELS-F125) and dry etching, as the SEM images of samples show in Figs. 5(a) and 5(b) (see Appendix D for the detailed fabrication process). According to the above calculation, the scanning angle range is beyond ±40°, which is out of the NA=0.75 of the objective lens; we then slightly shifted the objective lens downside to observe the complete steering angle in a negative axis. Figure 5 illustrates the images from the near-infrared camera together with the measured data. One may notice that the far field is dual-beam intensity distribution on the lateral axis coupled into the 14 input ports. The maximum angle beam of ±41.04° was measured in the negative axis, while that in the positive axis is only 34.2° due to the limitation of NA [see Fig. 5(d)]. By rotating the polarizer and quarter-wave plate, the RCP and LCP beams will be separately observed, as shown in the Figs. 5(e) and 5(f), respectively. Here, we chose the output data from port -6 to analyze the performance of metasurface-based POPA; the peak-to-sidelobe ratios of RCP and LCP are −10.31dB and −12.06dB, respectively. In fact, the only sidelobe is composed of transmitted light not modulated by metasurfaces, so its vertical coordinate is always in the center of RCP and LCP, and the modulation efficiency will also be affected by the phase gradient. Note that there is some background noise in the far field and line profile, i.e., noise between +10° and +20° in Figs. 5(e) and 5(f). It is due to the noise from the grating couplers captured by the objective lens in experiments, since the distance between the input grating coupler and the radiating antenna is less than 0.5 mm in our device. Undoubtedly, the whole performance can be improved by further optimization.

We have proposed a metasurface-based OPA on the LNOI platform with independent port-selection, which definitely shows the capability of free-space beam steering across ±41.04°×7.06° FOV, aliasing-free. In our approach, a non-periodic Jx-waveguide array is employed to achieve a flat phasefront sweeping in an aliasing-free manner. By imparting a metasurface design, the lateral scanning range of the aliasing-free POPA beam is doubled (up to ±41.04°) with background suppression of >10dB. The dual-beam scanning will not only ensure a wide FOV, but also improve the frame rate. In addition, we could employ a tunable liquid crystal polarizer to filter out one branch to return to single beam scanning to avoid the problem of cross talk without a decrease in the sweeping angle range. Note that though the current device works in a static scheme, it is possible to realize a dynamic scanning of POPA by using Mach–Zehnder interferometer (MZI) switches. In this regard, the excellent EO modulation and low-loss property of the LN waveguide promise high potential advantage for high-speed OPA technology. This work provides a successful demonstration by incorporating the metasurface design into OPA, which can open a new degree of freedom to achieve aliasing-free beam steering and FOV enlargement.

Acknowledgment. T. Li is grateful for the support from the Dengfeng Project B of Nanjing University.