Lens-based integrated 2D beam-steering device with defocusing approach and broadband pulse operation for Lidar application

We propose an integrated two-dimensional beam-steering device based on an on-chip silicon-nitride switch/emitter structure and off-chip lens for light detection and ranging (Lidar) application at 1550 nm. In this device, light is guided by a 1 × 16 switch to one grating emitter in a 4 × 4 grating-emitter array. The beam from the grating emitter is collimated and steered by a fixed lens. By changing the grating emitter that emits light, different beam-steering angle can be achieved. A divergence angle of 0.06° and a field of view of 2.07° × 4.12° in the far field are achieved. The device has O(log2N) power consumption for N emitters, allows digital control and achieves 18 dB background suppression. Blind-zone elimination and broadband operation are also achieved in our lens-based beam-steering device. Therefore, it is suitable for broadband solid-state Lidar application. © 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement


Introduction
Light detection and ranging (Lidar) technology has attracted intense interest for autonomous driving, sensing, wind detection, etc. [1,2] In Lidar systems, beam-steering devices are the key components that steer the light beam and scan the target. Conventional beam-steering devices such as mechanical steering mirrors suffer from limited steering speed and are vulnerable to vibrational perturbation. Recently, many compact beam-steering technologies have been investigated, including the optical phased array (OPA) [3][4][5], microelectromechanical system (MEMS) [6][7][8], and liquid crystal [9,10]. Among these technologies, OPA has become the leading technology. Integrated OPA-based Lidar has been reported with large-scale two-dimensional (2D) emitters [5,11,12], high side-mode suppression [13], low divergence angle [14,15], and wide field of view (FOV) [16][17][18][19]. Chip-scale OPA-based transceivers have also been reported [20][21][22]. Meanwhile, a new method of beam-steering device based on integrated planar lens [23,24] and liquid lens [25] has also been proposed. Lens-based beam-steering devices have been demonstrated that they can be applied to Lidar with low power consumption and easy steering control [26]. Among the many reported works, wavelength-tuning-assisted beam steering is usually chosen to simplify the device design and control complexity. Recently, a new design concept based on on-chip switch/emitter structure and off-chip lens has been reported by D. Inoue [26] and Y.-C. Chang [27]. This design allows O(log 2 N) control power consumption and potentially very good signal-to-background suppression. Low control-power consumption and simple control complexity also enable 2D beam steering using a single wavelength. However, in current works, some problems remain to be solved for real Lidar application, e.g., beam-steering discontinuity (existence of blind zone) and wavelength constraint due to the use of resonator components. Therefore, developing a design that eliminates the blind zone and enables broadband femtosecond-pulse operation while maintaining the advantages of low control-power consumption, easy steering, good background suppression, and good power handling is very meaningful.
In the present paper, we propose and demonstrate an integrated 2D beam-steering device based on on-chip switch/emitter structure and off-chip lens for Lidar application at 1550 nm. The device is manufactured on a silicon-nitride platform. It allows 2D beam steering without the need for wavelength tuning. It has a beam divergence of 0.06°, an FOV of 2.07°× 4.12°, background suppression of 18 dB, and O(log 2 N) control-power consumption (N is the number of emitters). The blind zone is eliminated by defocusing the emitter plane, which is accompanied by an increase in beam divergence. Broadband femtosecond-pulse operation is also supported owing to the wavelength-insensitive design. We believe that this present work may pave the way for solid-state and high-performance Lidar applications such as autonomous driving.

Principle of beam steering
The basic principle of our scheme is based on the combination of an on-chip switch/emitter structure and an off-chip lens. As shown in Fig. 1, an emitter array is placed on the focal plane (FP) of a lens (denoted as "1 st FP" and "device lens" in Fig. 1). The light beams emitted from the different emitters are parallel to one another, Thus, for an ideal lens, they will virtually intersect at one point (denoted as "S") on the FP ("2 nd FP" in Fig. 1) of the device lens at the other side. If this "S" point is considered as a virtual light source, the light beams from different emitters become collimated light beams that point at different directions at the "S" point, which means beam-steering can be achieved by selecting the light emission from different emitters. According to the Fourier optics theory, the beam pattern near the "S" point, i.e., E out1 (x, y), is the Fourier transform (FT) of the beam pattern on the emitter array, which is expressed as [28] where E in (x 0 , y 0 ) is the beam pattern in the emitter array, λ is the light wavelength, f is the focal length of the device lens and F represents the Fourier transform operation. Then, a second lens ("FT lens") is added as a Fourier transform lens for far-field (FF) measurement. Similarly, the beam pattern in the FP of the FT lens ("3 rd FP" in Fig. 1), i.e., E out2 (x e , y e ), or the FF pattern is given by where the lenses of both device and FT have the same focal length f for simplicity. Obviously, E out2 (x e , y e ) equals to the central symmetry of input beam pattern E in (x 0 , y 0 ) with a constant complex coefficient. Therefore, the FF beam intensity pattern is fully determined by the beam pattern in the emitter-array plane. Meanwhile, beam divergence angle ∆θ is approximately expressed as follows [26]: where w is the near-field beam diameter. And beam-steering angle θ is approximately given by [26] θ = tan −1 l f (4) where l is the length of the emitter array. These equations indicate that the beam divergence and beam-steering angle can be engineered by properly choosing the beam diameter of the emitter, emitter spacing, and focal length of the lens. The number of FF resolved points is equal to the number of emitters. Simulation of the beam propagation is performed in Zemax to confirm the aforementioned analysis. Figure 2(a) shows that two identical plano-convex lenses (Thorlabs LA1116) with a focal length of 1.04 cm at 1550 nm are utilized to form a coaxial system as proposed and shown in Fig. 1. A 4 × 4 1550nm-light point-source array is placed on the FP of the device lens. The light-source array has pitch value of 250 and 125 µm along the x and y directions respectively, and the total size is 0.75×0.375 mm 2 . Each single-spot source emits light with a divergence angle of 10°. The insets in Fig. 2(a) show a close look of the light array and FF pattern generated by the FT lens. The simulated beam patterns on the FP of the two lenses are shown in Figs. 2(b) -2(d). Figure 2(b) shows the beam pattern of the point-source array. Figure 2(c) shows the beam pattern at the "S" point with an expanded diameter. Figure 2(d) shows the beam pattern on the FP of the FT lens in the angle domain, which is centrally symmetric with the input beam pattern, as proven in Eq. (2). The transfer between the space and angle domains is calculated using Eq. (3). An aberration can be observed in Fig. 2(d) because the lenses are not ideal, which can be improved using a multi-lens system.

Device design and characterization
An integrated beam-steering device is proposed, as shown in Fig. 3. The device is designed on a silicon-nitride (Si 3 N 4 ) platform. The light is coupled into a chip via a standard grating coupler and then sent to a 1 × 16 switch, which is made up of cascaded 1 × 2 Mach-Zehnder interferometer (MZI) switches with thermal phase tuning. The heating resistors are formed using a titanium thin film deposited on silica cladding, and the wiring lines and bonding pads are made of aluminum, as shown in the inset in Fig. 3. The 16 output channels of the switch are connected to 16 grating emitters (4 × 4 emitter array). Each time, the light is guided to one output channel of the switch and is emitted to free space via a grating emitter. At the top of the chip, a glass lens ("device lens") is installed whose FP overlaps with the emitter plane. The lens collimates and steers the light beam. This device is fabricated on a 3 × 16-mm 2 chip, as shown in Fig. 4. The scanning electron microscope (SEM) images of the emitter array and single grating emitter are also shown in Fig. 4. Here, a 4 × 4 array is fabricated, and the spacing between the emitters is 250 µm along the x direction and 125 µm along the y direction. The emitter is approximately 10 × 8-µm 2 large with a pitch of 1.2 µm. The number of resolved points in the FF is equal to the number of emitters in the chip, i.e., 16 points.
The grating emitter is designed to have a beam size of approximately 8 × 6 µm 2 , as shown in Fig. 5(a). In the experiment, the near-field beam profile is measured to has high background suppression of more than 20 dB and a beam diameter of 10 × 10 µm 2 , as shown in Figs. 5(b) and 5(c). The beam-size discrepancy between the simulation and experiment is due to the fabrication error and the limited resolution of the InGaAs camera (Xenics Bobcat-320) with a value of approximately 3-µm. The beam divergence (θ div ) of the grating emitter is individually characterized by measuring the change in the beam diameter (∆d) when the emitter plane is shifted by ∆f (∆f >> 10 µm) away from the focal plane of the microscope, expressed by θ div = ∆d/ ∆f. The summary of the beam divergence with respect to different wavelengths is shown in Fig. 5(d).
The average beam-divergence angle is 10.5°in the y direction and 9.3°in x the direction with less than 2°deviation from 1540 to 1560 nm.
The 1 × 16 switch is realized using a binary-tree structure with cascaded 1 × 2 switches. Each 1 × 2 switch has an MZI structure with a 1-mm-long thermally tuned phase shifter. The switch speed is measured to be 27.1 µs, as shown in Fig. 6(a). Meanwhile, the phase shift with respect to different applied electrical power is shown in Fig. 6(b), and the corresponding phase-control efficiency is 88.8 mW/π. The efficiency can be further improved using novel designs, e.g., a The grating emitter is designed to have a beam size of approximately 8 × 6 μm 2 , as shown in Fig. 5(a). In the experiment, the near-field beam profile is measured to has high background suppression of more than 20 dB and a beam diameter of 10 × 10 μm 2 , as shown in Figs. 5(b) and 5(c). The beam-size discrepancy between the simulation and experiment is due to the fabrication error and the limited resolution of the InGaAs camera (Xenics Bobcat-320) with a value of approximately 3-μm. The beam divergence (θ div ) of the grating emitter is individually characterized by measuring the change in the beam diameter (∆d) when the emitter plane is shifted by ∆f (∆f >> 10 μm) away from the focal plane of the microscope, expressed by θ div = ∆d/∆f. The summary of the beam divergence with respect to different wavelengths is shown in Fig. 5(d). The average beam-divergence angle is 10.5° in the y direction and 9.3° in x the direction with less than 2° deviation from 1540 to 1560 nm.  recycling structure [29]. For the control-power consumption, we can see that only the switches in the optical path from the input to the desired output emitter are operated. Thus, the control-power consumption is O(log 2 N) for the N emitters. With a pre-biased voltage compensating for the fabrication error in the Ti heater, digital signals are used to control the switch, which allows a much simpler control compared with the analog signal. The insertion loss of each 1 × 2 switch is estimated to be 0.5 dB.

Setup and results
The experimental setup is shown in Fig. 7(a). A 1550-nm continuous wave light is coupled into the chip through a grating coupler. The light is then guided by the 1 × 16 switch and emitted to space by the grating emitter. Figure 1 shows that two plano-convex lenses (Thorlabs LA1116) are utilized in this setup. The device lens is set to steer the beam, and the FT lens is used to perform the Fourier transform for the FF measurement. A microscope with InGaAs camera is focused on the FP of the FT lens. A photograph of the experimental setup is shown in Fig. 7(b). The two lenses have the same focal length of 1.04 cm at 1550 nm. Therefore, the device and FT lenses form a 4f system with magnification equal to one. The single-spot beam emission of the device is first characterized and the measured FF patterns are shown in Fig. 8. We can clearly observe that all patterns have very clean background, which means that the leakage light from other non-working emitters is negligible.
Meanwhile, the FF beam-pattern outline of each emission spot is extracted along the θ and ϕ axes, as shown in Figs. 9(a) and 9(b), respectively. An average background suppression of 18 dB is calculated from the gray images captured by the infrared camera. In principle, the background suppression ratio in our device is only determined by the extinction ratio of the 1 × 2 switch, which is expected to exceed 20 dB using optimized fabrication. As shown in Fig. 9, an average full width at half maximum beam divergence of 0.06°in the x and y directions are obtained from all channels, which are close to the theoretical value of 0.056°(tan −1 (10 µm/1.04 cm)). The sidelobes near the main lobe are caused by the near-field emission pattern of the single grating as shown in Fig. 5(a), and are thus a part of the FF pattern. These side lobes can be eliminated using a nonuniform grating structure. The other side lobes that appear far away from the main lobe, whose average intensity is below −18 dB compared with that of the main lobe, are caused by the leakage light from other emitters. When all the 1 × 2 switches are tuned to output as 50:50 beam splitters, all emitters emit light. The near-field and far-field patterns of the spot array are shown in Figs. 10(a) and 10(b), respectively. The beam-steering range is measured to be 2.07°× 4.12°in the angle domain, which is consistent with the simulation results shown in Fig. 2. The spots on the corner in the FF pattern in Fig. 10(b) are slightly distorted because of the non-ideal lens system. Moreover, an FF pattern with a shape of "N" is obtained, as shown in Fig. 10(c), as an example to indicate the programmable ability of all output emitters.

Blind-zone elimination
In contrast to OPAs that continuously steer the beam, our device, and other lens-based beamsteering works discretely steer the beam. As we have demonstrated both in the simulation and experiment, the FF beam pattern of our device lens is the same as that on the emitter plane. Therefore, the gap zone between each emitter becomes a blind zone in the FF pattern. In the current design of emitter array, the distance between each emitter can be reduced, but not eliminated. Therefore, we propose a defocusing method to eliminate the blind zone. The principle is shown in Fig. 11. The emitter plane is slightly shifted away from the FP (1 st FP) of the device lens. When light is emitted from the emitters, its beam size is expanded during propagation because of beam divergence (∼10°for our emitters) and can become sufficiently large to cover the gap between two adjacent beams when it reaches the 1 st FP. Using this method, the blind zone in the FF pattern can be eliminated. Required shifted distance ∆f is given by ∆f = l/θ div , where l is the adjacent emitter distance and θ div is the beam divergence. For our device, ∆f = 716 µm at l = 125 µm and θ div = 10°. It is worth mentioning that the blind-zone elimination is accompanied by an increase in the beam divergence, which is inevitable because the beam divergence should be equal to the minimum steering angle between two adjacent beams to eliminate the blind zone. If we want to maintain the maximum measurement distance for the Lidar, we can choose a larger focal length because the beam divergence, as expressed by Eq. (3), depends on both near-field beam size w and focal length f. Accordingly, more emitters should then be used to guarantee the required total beam-steering angle, which presents a tradeoff among total beam-steering angle, beam divergence, and number of resolved points in the far field.
Simulation of this defocusing method has been performed using Zemax, and the results are shown in Fig. 12 where the FF patterns without and with the defocusing process are shown in Figs. 12(a) and 12(b), respectively. It is noted that beam aberration exists because of the non-ideal lens system, and the aberration can be suppressed using an ideal lens system, as shown in Fig. 12(c). The experimental results are shown in Fig. 13, which agree well with the simulation. Figure 13(b) shows that the beam diameter increases, and the vertical beam spots are connected. Meanwhile, interference ring pattern appears, which is caused by the limited aperture of the microscopic system [30]. The same ring pattern also appears when we use the same microscopic system to directly measure the beam from an end-cut single-mode fiber, as shown in the insets in Fig. 13.

Broadband pulse operation
Our device is designed to work in a broad wavelength range. Grating emission angle is well known to be dependent on the emission wavelength because of the phase-matching condition. In our device, when the beam (with a divergence angle of ∼10°) from the grating emitter propagates through the device lens above the chip, the beam is collimated, and its propagation direction is steered. Owing to the lens property, the new beam propagation direction after the lens is only dependent on the position of the beam source, i.e., the position of the grating emitter, and independent of the incident angle to the lens, i.e., the emission wavelength. We choose three different wavelengths of 1540, 1550, and 1560 nm to perform the simulation, and their corresponding emission angles are 10°, 9°, and 8°. The simulated FF patterns are shown in Figs. 14(a)-14(c), respectively. A coma aberration appears because of the non-ideal lens system. The aberration disappears when an ideal lens system is used, as shown in Figs. 14(d)-14(e). It can be seen that for the three wavelengths, the beam patterns remain unchanged, indicating the wavelength insensitivity of the device.
Experimentally, seven different wavelengths from 1550 to 1580 nm are tested for broadband operation, as shown in Fig. 15. It can be observed that the beam spot-positions remain unchanged for all wavelengths, which indicating that the beam direction is independent of the wavelength. To further confirm this result, the cross sections in the ϕ direction are extracted from Fig. 15, as shown in Fig. 16(a). We also add a horizontal axis of pixels from the InGaAs camera. A zoom-in view of the beam profiles at a steering angle of −1.03°is shown in Fig. 16(b). The dots represent the pixel data, and the curves are drawn using spline interpolation. It can be seen that the beam spots only deviate 1∼2 pixels corresponding to an angle deviation within 0.03°. Owing to the bandwidth of the grating coupler and emitters, the emission efficiency at a shorter wavelength is lower. Thus, the spot brightness is lower at a shorter wavelength, as shown in Fig. 15.  For Lidar with a TOF (time-of-flight) technology, the time and distance resolutions are determined by the pulse width. Thus, a femtosecond pulse can support very high measurement accuracy up to the micrometer level [31]. Moreover, support for the femtosecond-pulse operation means that the device can directly enable integrated Lidar with an on-chip soliton comb-based Lidar technology, which has been proposed in [32] and [33]. These two studies show that soliton-optical frequency combs generated by microresonators can be used for precision ranging and tracking of fast-moving objects. Therefore, combining the integrated microresonator-based soliton-comb generation with the on-chip beam steering device will be very promising for highprecision and ultrafast Lidar application. In our work, the beam-steering device is demonstrated to support broadband femtosecond-pulse operation. In the experiment, a 37-MHz femtosecondpulse source is injected into the device. The beam patterns of the emitter plane and far field are shown in Figs. 17(a) and 17(b), respectively. A clear beam pattern can be observed, and the aberration in the FF pattern is due to the non-ideal lens system. As shown in Fig. 17(c), the waveform of the pulse train is obtained by a 10-GHz photodetector and a 2-GHz oscilloscope. The insets show the zoom-in view of a single pulse from the input and output. Limited by the electronic bandwidth, the difference between the input and output pulses can hardly be observed. The output spectrum is filtered because of the limited bandwidth of the grating coupler and grating emitter, as shown in Fig. 17(d). The input femtosecond pulse has a bandwidth of 44 nm centered at 1560 nm. After the input grating coupler, the signal is estimated to have a bandwidth of 23 nm and a pulse width of 575 fs. The pulse broadening in the device is caused by the dispersion of the Si 3 N 4 waveguide and spectral filtering of the grating emitter. The waveguide dispersion is −720 ps/(nm·km), simulated using Lumerical Mode. Thus, an ∼5-mm-long waveguide introduces a group delay dispersion of −3.6 fs/nm. The signal bandwidth after the grating emitter is 21 nm, which corresponds to pulse broadening of 75.6 fs. The pulse broadening due to the 2-nm bandwidth reduction is ∼55 fs. Hence, the pulse width after the grating emitter is estimated to be 705.6 fs. The output signal from the emitter is collected using a closely placed single-mode fiber tip. The coupling efficiency from the emitter to the fiber tip is very low. Thus, the collected optical power (−25 dBm) is too low for a standard autocorrelation measurement, and the actual pulse width cannot be obtained. The pulse-energy values coupled into and out of the chip are estimated to be 12 and 1.2 pJ, respectively.

Discussion
As a new beam-steering method, the lens-based beam-steering device for Lidar application is featured with reconfigured design of the beam divergence and steering angle by choosing a lens with a proper focal length. The low control-power consumption and low control complexity enable 2D beam steering at a single wavelength. The on-chip switch/emitter structure allows very good background suppression and strictly no grating lobes.
In this section, we present the comparison of our work with the other lens-based integrated works in Table 1. They are devices that use MZI switch and on-chip planar Luneburg lens [24], devices that use ring resonator switch and off-chip glass lens [26], and those that use ring emitter and metalens [27]. It can be seen that our work is the first demonstration of blind-zone elimination using defocusing method and broadband femtosecond-pulse operation. NA: not available * In principle, these designs also support blind-zone elimination by defocusing; however, no demonstration has been provided.

Conclusion
In conclusion, a lens-based integrated 2D beam-steering device is demonstrated at 1550 nm. We have achieved 4 × 4-point beam steering. The device has O(log 2 N) power consumption for N emitters, a 27.1-µs switch speed, digital control, 0.06°beam divergence, and 18-dB background suppression. We have also demonstrated blind-zone elimination using defocusing method and broadband femtosecond-pulse operation in this work. It is believed that this work may pave the way for a novel lens-based solid-state beam-steering technology for practical Lidar applications.

Disclosures
The authors declare no conflicts of interest.