Optical phased arrays (OPAs) have gained interest in recent years as an alternative to traditional mechanical beam steering, such as MEMS micromirrors [1], because they lack inertia (which limits the ability to reach a large steering range at a high speed) and can be single chip (the steering elements may be integrated with an on-chip laser).

Phased arrays in radio were first explored more than a century ago [2], but OPAs are less than a decade old [3–20]. A phased array consists of several coherent emitters. By aligning the emitters’ phases, the emitted light interferes constructively in the far field at certain angles. For uniform emitter spacing greater than about half a wavelength, aliasing occurs: strong constructive interference is observed at more than one far-field angle. Aliased lobes are called grating lobes, and any power between them (aside from the main lobe) constitutes side lobes. Non-uniform radar arrays were first investigated in the 1960s [21]. Their aperiodic spacing suppresses grating lobes while maintaining a narrow main lobe, at the cost of higher side lobe power. Since they achieve a narrow main lobe with fewer antennas (including the circuitry behind each antenna), a sparse aperiodic array is also less expensive. Returning to optical phased arrays, the situation is different: emitting waveguides cannot be spaced by a half-wavelength since this would cause uncontrollable on-chip coupling between emitters, and the marginal cost of adding another emitter may not be as significant.

The main lobe has a full width at half-maximum (FWHM) divergence $\delta \psi$ and can be steered within a steerable range ${\psi }_s$; the resolution is then ${\psi }_s/\delta \psi$. Our device is steered in one axis using phase control and the other axis using the wavelength [3–10]. It is also possible to steer in both axes using phase control [11–17], but in the general case this requires $N\times M$ phase controls for an array of size $N\times M$; in contrast, our approach requires $N+1$ controls ($N$ phase controls and one wavelength control).

To date, OPA steering resolution has been limited. To our knowledge, the widest demonstrated steering range of any OPA was 51°; however, the beam divergence was relatively large (3.3°) [9]. The narrowest beam divergence was 0.3°; however, the steerable range was relatively small (0.9°) [12]. The highest-resolution device had a resolution of 23, achieving the best ratio of steering range to beam divergence (23° and 1°, respectively) [10]. In these cases, the steering range was limited by aliasing. Indeed, a wider steering range can be achieved by inserting a diverging lens after the emitters, but this also expands the beam itself, and therefore the resolution is unchanged.

We improve on each of these OPAs by achieving a much wider steering range, a smaller beam divergence, and therefore a higher resolution in the phased-array axis ($\psi$) and also a narrower beam in the wavelength-controlled axis ($\theta$) than ever reported. We achieve the widest steering range in $\psi$ (80°, measurement limited) through non-uniform emitter spacing to suppress lobes that otherwise limit the steering range and by using wide-angle emitters. We achieve the smallest beam divergence in $\psi$ (0.14° average, 0.11° best measured) through a large collection of phase-controlled emitters. The combination of these two results allows us to steer the beam to over 500 resolvable points in $\psi$ (calculated based on the beam size and the achieved steering range). Furthermore, we achieve the smallest beam divergence in $\theta$ ever reported for a silicon OPA (0.142°) by using a carefully controlled weak-grating coupler. The devices are fabricated in an advanced 300 mm CMOS facility.

Our device is shown schematically in Fig. 1(a). Light from a tunable laser is either coupled in through a grating coupler (i) or edge-coupled (after the chip is diced along the dashed line). The TE grating coupler is used for testing at fixed wavelength ($\sim 1.3\mathrm{\mu m}$), is less alignment-sensitive than edge coupling, and ensures that the coupled light is TE. However, it was not optimized for bandwidth, having only about 25 nm FWHM. For the full two-dimensional steering we show later, we prefer to dice off the grating coupler and edge couple. After the light is in the waveguide, it is split into 128 channels using a star coupler (ii), though we only draw five channels for clarity, each channel being phase-controlled by thermo-optic tuners (iii). The thermo-optic tuner is a strip of p doped silicon (red) running alongside each rib waveguide (gray), contacted by tungsten vias (yellow) and metal wires (blue; the common ground wire was omitted for clarity). The waveguide array is condensed to the desired emitter positions and emitted from gratings (iv). The emitted light traces out the green cone in the far field: as an example, the light is shown steered to $\theta \ne 0$, $\psi =0$ (red arrow) or to $\theta \ne 0$, $\psi \ne 0$ (blue arrow). Fig. 1(b) is an optical microscope image of the star coupler, and Fig. 1(c) is a scanning electron microscope image of one of the emitters.

 

Fig. 1. Device concept. (a) A simplified drawing of the device, showing only five channels for clarity (our device has 128); (i)–(iv) are described in the text. The dashed line indicates where the chip may be diced in order to edge-couple light. After passing through the star coupler and phase tuners, the emitted light traces out the green cone and is steerable using the wavelength ($\theta$ axis) and by tuning the phase of each channel ($\psi$ axis). (b) An optical microscope image of the star coupler. The input waveguide is at center left; the 128 output waveguides begin near the center and fan out toward the right of the image. The scale bar is 500 μm. (c) A tilted scanning electron microscope image of the emitter with the cladding removed. The waveguide with grating notches is in the center, and the dark lines above and below show an isolation trench between emitters. The scale bar is 2 μm. The inset is a simplified drawing of the emitter at a slightly different angle so the scanning electron microscope image can be understood. The grating etch is indicated in blue.

Download Full Size | PDF

To achieve small beam divergence in $\theta$, we need long emitters in $x$ (since the far field is the Fourier transform of the emitted light); thus, we use a shallow grating etch with weak out-coupling. The grating etch depth was 16 nm (measured by profilometry) with a uniform pitch of 490 nm and a 50% duty cycle. To achieve small beam divergence in $\psi$, we need the collection of emitters to have a large $y$ extent. In a waveguide-based device, this amounts to splitting the light into many channels (128 in our case).

To achieve large steering in $\psi$, we need both wide-angle emission from the individual emitters [the envelope in Fig. 2(a)] and minimal aliasing. The former condition implies we must use emitters that are narrow in $y$ [using the coordinate system defined in Fig. 1(a)]. We use a 0.4 μm-wide rib waveguide, the minimum practical width with our thickness of silicon (0.4 μm), and a rib etch depth (0.2 μm) with reasonable waveguide confinement.

 

Fig. 2. (a) and (b) Far-field simulation along $\psi$ for uniform and non-uniform emitter pitches, respectively. The beam is shown steered to different angles in 5° increments [two angles in (a) and ten angles in (b)]. We avoid aliasing and therefore achieve a much larger steering range using the non-uniformly spaced emitters. (c) A zoom-in of the main lobe from (a) and (b), showing it is the same in both cases.

Download Full Size | PDF

Next, we investigate the latter condition (aliasing) in more detail. An OPA with uniformly spaced emitters, with the spacing larger than about half of a wavelength, generates a main lobe and grating lobes; larger emitter spacing results in smaller grating lobe spacing. We simulate our uniformly spaced OPA in Figs. 2(a) and 2(b), where the main lobe is at $\psi =0°$ and the grating lobes are at about 10° increments. The main lobe may be steered halfway to an adjacent grating lobe [Fig. 2(a), black arrow]. If it is steered further, then an adjacent grating lobe will enter the field of view and become the main lobe; thus, the grating lobes limit the steering range. For a fixed number of emitters, decreasing the emitter spacing would increase the grating lobe spacing and thus the steering range; however, that also increases the divergence of the main lobe. Therefore, the resolution is nearly constant [10]. Even so, the pitch cannot be made too small: it must be several times the wavelength to avoid on-chip coupling between emitters.

When the pitch is several times the wavelength, the regularly spaced emitters constructively interfere at many possible output angles. However, by tailoring the emitter positions to be non-periodic, the light no longer necessarily exhibits strong constructive interference at angles outside of the main lobe. Thus we can suppress the grating lobes to break the steering limit they impose, and the beam can be steered much further [Fig. 2(b)] at the cost of increased side lobe (“background”) power. The side lobe level (the maximum side lobe power at any angle) can be small for high-channel count devices such as this: better than $-10$ or $-8.8\mathrm{dB}$ when the beam is steered to 0° or $\pm 40°$, respectively, for our device, as simulated in Fig. 2(b). The main lobe is not significantly affected [Fig. 2(c)]; thus, we can expect the same beam divergence but an increased steering range.

Non-uniform OPA emitter spacing has been demonstrated on a smaller scale in the past [10,18–20]. One device [20] employs 12 emitters on a non-uniform pitch but emits a one-dimensional steerable line instead of a two-dimensional (2D) steerable beam. Two other devices [10,18] use 32 emitters on a non-uniform pitch but only demonstrate 10° or 23° steering in the OPA axis (a resolution of about 8 or 23, respectively). The last device [19] uses an injection-locked 2D array of 64 vertical-cavity surface-emitting lasers (VCSELs), but steering to large angles is difficult because the VCSEL emission envelope is narrow, so as the steering angle increases, the beam is quickly suppressed. Steering was demonstrated up to 2.2° (equivalent to a resolution of 14).

To characterize the phased-array emission from our device, we use an infrared CCD and a pair of lenses [Fig. 3(a)]. The real image (lens 1 only; blue rays) is used to align the imaging system to the chip, and then the Fourier image (both lenses; red rays) gives the far-field measurement. The initially arbitrary phases at the emitters [Fig. 3(b)] are aligned by a gradient-search algorithm using feedback from the camera image to condense the beam to the desired angle [Figs. 3(c) and 3(d)]. By first measuring the OPA with uniformly spaced emitters, we calibrate the image by comparing the measured lobe spacing on the camera image with the known lobe spacing from simulation (see Supplement 1), and then use that calibration for our measurements of the non-uniform OPA. The beam can be condensed to any angle within the numerical aperture of the measurement setup.

 

Fig. 3. Condensing and steering the beam. (a) The test setup, indicating the light rays used to align the imaging system to the emitters (blue rays, lens 1 only), and the rays used to characterize the far-field beam (red, lenses 1 and 2). (b) Without phase control, the phase at each emitter is initially arbitrary; thus we observe a smeared-out elliptical line in the far field. It is an ellipse because we are imaging the edge of the cone shown in Fig. 1(a). (c) and (d) The far field after aligning the phase at each emitter, for the OPA with non-uniform and uniform pitches, respectively. (e) The condensed beam for the uniformly spaced OPA, showing 75 separate measurements of the beam steered to different angles across about 10°; each measurement is a different color.

Download Full Size | PDF

In Fig. 3(e), we show the beam steered to 75 angles at a fixed wavelength (1.27 μm), with adjacent points spaced by approximately the beam FWHM, spanning about 10°. These cross sections must be taken along an elliptical arc [blue dashed line in Fig. 4(a)] because we are projecting the cone from Fig. 1(a) onto a plane. From these 75 points, we measure the divergence in $\theta$ to be $0.142\pm 0.005°$ and in $\psi$ to be $0.14\pm 0.02°$, where the uncertainty value is the standard deviation over the 75 measurements (see Supplement 1).

 

Fig. 4. (a) The uniformly spaced emitter showing aliased lobes at $\pm 5°$. (b) The cross section of (a) in the OPA-steered direction (along the dashed blue line) compared to the simulation. The non-uniformly spaced OPA has the same typical beam width. (c) The cross section of (a) in the wavelength-steered direction (green). The cross section in the other direction is overlaid for comparison (blue dotted). (d) Images at three wavelengths stitched together; the image boundary is indicated by a dashed gray line. For these measurements, we used the uniformly spaced OPA so the direction of wavelength tuning is clear, but the non-uniform OPA behaves in the same manner.

Download Full Size | PDF

For these statistical measurements, we used the uniformly spaced OPA so we can ensure we are measuring the beam width accurately (since there is more than one lobe in the measurements, and we use the spacing between those lobes to calibrate the measurement), but we measure the same typical beam width for the non-uniform OPA, as expected [see Fig. 2(c)].

To our knowledge, this divergence in the wavelength-steered axis $\theta$ is the smallest from a silicon emitter to date. The previous record was a measured divergence of 0.3–0.5° [22] or a similar device with a divergence of 0.3° [23]. Our very small divergence is a result of a well-controlled shallow grating etch and can probably be further improved. As for the phased-array axis $\psi$, the smallest divergence from those 75 measurements was 0.11°, matching the expected divergence from simulation. The slightly wider-than-optimal average beam width may have been caused by how well the algorithm approaches a perfectly coalesced beam or thermal crosstalk between the channels, since such crosstalk may prohibit perfect beam collapse. We note that the beam width could be made about 40% smaller in $\psi$ by using a splitter that gives equal power in each emitter (for example, cascaded $1\times 2$ splitters), but we chose a star coupler for this device since it can have a smaller footprint for a large output channel count.

One of those 75 measurements with its cross sections is shown in Figs. 4(a)–4(c). We also demonstrate the steerability in the perpendicular (wavelength-tuned) axis in Fig. 4(d). Tuning the wavelength from 1260 to 1360 nm changes $\theta$ by 17°, close to the expected angle (see Supplement 1). Half of that tuning range, also shown in the same figure, may be considered approximately a reasonable tuning range for an on-chip tunable laser [24,25]. For this measurement, the input grating coupler was diced off and the chip was edge coupled using a lensed fiber. Otherwise, we would have to change the launch angle into the input grating coupler to be able to inject light across this entire range. We show results from the uniform OPA so that the direction of steering is obvious, but the non-uniform emitter OPA behaves the same.

Returning to the non-uniformly spaced OPA, we demonstrate (to our knowledge) record-breaking wide beam steering from an OPA in Fig. 5. The beam steered to five angles spanning 80° is shown in Fig. 5(a), and the cross sections are shown in Fig. 5(b) compared to the simulation. The steering range here is limited by the numerical aperture of the objective lens.

 

Fig. 5. (a) The beam from the non-uniformly spaced OPA, steered to five angles across 80°. (b) Elliptical cross sections of Fig. 3(e) with the simulated result overlaid. The gray area is the viewable range of our imaging setup. The worst side lobe level in these five measurements is noted on the graph.

Download Full Size | PDF

Since our infrared camera has a finite number of pixels (640 in its widest axis), we need to adjust our optics depending on whether we are capturing a narrow field of view (FOV) for measurements such as Fig. 4(a) (to be able to resolve the beam) or a wide FOV for measurements such as Fig. 5(a) (to be able to observe wide-angle steering). For the former we use an objective lens with numerical aperture (NA) of either 0.29 or 0.42, and for the latter we use a lens with an NA of 0.7. To switch between FOVs, we adjust the positions of both lenses and the camera until the far field is focused on the camera plane, then measure a uniformly spaced OPA to calibrate the image before proceeding with the other measurements. Since our camera has a limited dynamic range, while we can see the main lobe and the tallest side lobes simultaneously [a typical case is noted on the figure in Fig. 5(b)], we can only see a detailed side lobe profile if we increase the laser power, which saturates the main lobe. Supplement 1 shows typical results as we vary the laser power. We do not observe significantly better or worse side lobe levels than most other OPAs in the literature; however, our intent was not to have significantly lower side lobes, but rather to form a small beam steerable over a large range. A non-uniform sparse array inherently trades side lobe power (by spreading a grating lobe out across each interference order) for an alias-free beam.

In conclusion, we demonstrate a two-axis steerable optical phased array with extremely wide-angle steering and a small divergence. To achieve both simultaneously, we have designed and fabricated a device with an appropriate non-uniform emitter spacing to suppress the grating lobes. Our comparison device with uniform emitter spacing has grating lobes separated by about 10° near normal, limiting its steering range to that value. In contrast, we steered the non-uniform emitter OPA across an 80° range (measurement limited), allowing us to direct the beam to over 500 points on that axis. In the other axis, we have also demonstrated a record small beam divergence for a silicon photonics device (0.142°), and steered it across 17° (100 nm). The two axes taken together represent tens of thousands of resolvable points. Such high-resolution beam steering enables new applications in free-space optical communications and steered laser sensing.