Integrated microwave acousto-optic frequency shifter on thin-film lithium niobate

Electrically driven acousto-optic devices that provide beam deflection and optical frequency shifting have broad applications from pulse synthesis to heterodyne detection. Commercially available acousto-optic modulators are based on bulk materials and consume Watts of radio frequency power. Here, we demonstrate an integrated 3-GHz acousto-optic frequency shifter on thin-film lithium niobate, featuring a carrier suppression over 30 dB. Further, we demonstrate a gigahertz-spaced optical frequency comb featuring more than 200 lines over a 0.6-THz optical bandwidth by recirculating the light in an active frequency shifting loop. Our integrated acousto-optic platform leads to the development of on-chip optical routing, isolation, and microwave signal processing.

Here, we demonstrate an integrated gigahertz AOFSs on thin-film LN, which leverage the large piezoelectric and photoelastic coefficients of LN, as well as its low microwave and optical propagation loss. Benefiting from the fact that both optical and acoustic indices (phase velocities) of LN are greater (smaller) than those of the underlying silicon dioxide, we construct fully supported AOFSs, thereby providing improved robustness and a more straightforward fabrication process than suspended acousto-optic devices. Specifically, at the telecommunication wavelength of 1.5 µm, we demonstrate an optical frequency shift of 3 GHz with carrier suppression over 30 dB and the opposite sideband suppression >40 dB (below the noise floor). Furthermore, as an application of our device, we demonstrate a 3-GHz optical frequency comb with more than 200 lines over a 0.6-THz (5 nm) optical bandwidth by recirculating light in the active frequency shifting loop with our AOFS and an erbium-doped fiber amplifier (EDFA).

Device design and fabrication
Our integrated AOFSs are fabricated using an 800-nm-thick X-cut LN thin film on a 2-µm-thick silicon dioxide layer on a silicon substrate ( Fig. 1(a)). The phase matching condition between the light and acoustic wave ( Fig. 1(a) inset) determines the optimum incident (Bragg) angle θ B of the input light beam, given by The optical wavenumber k = 2πn eff /λ, n eff is the optical mode index, and λ is the optical wavelength in vacuum. The acoustic wavenumber K = 2π/Λ, where Λ is the acoustic wavelength in the LN thin film. For the demonstrated 3-GHz AOFS in this work, we use Λ = 1.2 µm, λ = 1.5 µm, and n eff = 2.0 yielding sin θ B = 0.3 (θ B ∼ 18 degrees). The maximum shifting frequency of ∼9 GHz for λ = 1.5 µm on the LN thin film occurs when the shifted light is fully reflected by the acoustic grating (in this case, θ B = π/2 or sin θ B = 1). For our device, the light is coupled onto the chip by a lensed fiber and guided by a rib optical waveguide that is defined by two etched grooves on the LN thin film. Following the waveguide, the input light bends and adiabatically broadens towards the central acousto-optic region at the Bragg angle ( Fig. 1(b)). The rib waveguides terminate near the acousto-optic region, and the input light beam then propagates as a two-dimensional Gaussian beam in the LN thin film. The end width of the input waveguides is 18 µm such that the Rayleigh range of the light beam is much larger than the acoustic wave width of 100 µm. A waveguide collects the expanded light beam deflected by the acoustic waves, then tapers to a on-chip routing waveguide. The 40-µm width of the collecting waveguide is experimentally selected to provide the highest collection efficiency. With two optical input waveguides and two output waveguides, our device is configured for either anti-Stokes frequency shifts of the input light at Port A or Stokes frequency shifts of the input light at Port B ( Fig. 1(b)).
The thin-film acoustic wave propagates along the y direction and is electrically generated by an interdigital transducer (IDT) via piezoelectricity. The IDT consists of cross-finger electrodes that are made of a 140-nm-thick aluminum layer deposited by thermal evaporation and patterned by the lift-off process ( Fig. 1(c)). The pitch of the IDT electrodes is 580 nm, which is equal to the half acoustic wavelength (in the IDT region). To avoid any acoustic reflection by the chip edge or other devices, etching grooves are fabricated at the far ends to deflect acoustic waves. Otherwise, standing acoustic waves formed by any reflection would result in undesired carrier and sideband light at the output.

Acousto-optic interactions
To characterize the microwave-to-acoustic transduction by the IDT, we measure the microwave reflection and transmission scattering (S) parameter spectra of an IDT pair ( Fig. 2(a)). Multiple acoustic modes are observed in the transmission spectra (labeled I-IV in Fig. 2(a)), consistent with the acoustic modes found in the numerical simulations ( Fig. 2(b)). They are referred to as the Rayleigh wave (Mode I), Love wave (Mode II) of LN thin film, and higher-order Love waves (Modes III and IV) that partially propagating in the silicon dioxide underlying layer. We employ the Rayleigh wave (Mode I) for our AOFS due to its highest transduction efficiency. Twenty-five pairs of 100-µm-width electrodes are used in each IDT to match the external impedance of 50 Ω, as indicated by the over 10 dB dip in the S 11 spectrum ( Fig. 2(a)). In the case of the highest transmission of S 21 = -10 dB at 2.9 GHz, we can infer a -5 dB (31%) microwave-to-acoustic transduction efficiency for a single IDT. The efficiency is mainly limited by the symmetric IDT design (maximum 50% in one propagating direction) as well as by the mass loading and ohmic loss of aluminum. The fast variations of the S 21 spectrum is due to weak acoustic reflections between the IDT test pair. The optical refractive index variations due to the acoustic wave ( Fig. 2(c)) are calculated from a simulated strain profile and the photoelastic coefficients of LN [33]. The refractive index variation in the z direction (∆n z ) is slightly larger than that in other directions, and thus the optical TM mode is deflected more efficiency than the TE mode. We perform a three-dimensional finite-difference time-domain (FDTD) method simulation to investigate the spatial mode profile of the light that is deflected by the acoustic grating (Fig. 6). The FDTD simulation features actual device dimensions. Instead of a traveling acoustic wave, the acoustic grating in the FDTD simulation is represented by a steady refractive index profile. This approximation significantly reduces the computational expense without affecting the spatial mode profile of light. Absent the acoustic wave, the input light beam continues propagation, and no deflection is observed ( Fig. 2(d)). With the acoustic wave, the input light beam is partially deflected (Fig. 2(e)). The deflected and the transmitted light beams are twice the Bragg angle apart in the far field. The width of the deflected beam is approximately given by where W is the width of the acoustic grating. For our device, with acoustic width of W = 100 µm and sin θ B = 0.3, the width of the deflected beam is 60 µm larger than that of the input beam.

Experimental characterization of the LN AOFS
We experimentally characterize our LN AOFS in both the Stokes (Figs. 3(a) and 3(b)) and anti-Stokes (Figs. 3(c) and 3(d)) configurations. We measure the optical spectra of the deflected output light by heterodyne detection using a second laser, which is red-detuned by a few gigahertz from the input laser. A schematic of the experimental setup is shown in Fig.8. The wavelength of the input light is 1597 nm, and the IDT is driven at 2.89 GHz with a microwave power of 15 dBm. For the Stokes configuration, the deflected light is red-shifted by 2.89 GHz and exhibits a 31-dB carrier suppression ( Fig. 3(b)). The carrier suppression of the deflected light is defined by the ratio of the optical powers at the shifted and carrier frequencies. For the anti-Stokes configuration, the deflected light is blue-shifted, and a 33-dB carrier suppression is observed (Fig. 3(d)). The presence of the carrier frequency light at the deflection output port is most likely caused by the optical scattering due to the on-chip structures, as it exists in absence of the acoustic wave. The on-to-off extinction ratio at the deflection port is also measured over to be 30 dB. Any other undesired optical sideband is not observed above the noise floor in the heterodyne measurements, thus the sideband suppression rate exceeds 40 dB. We further characterize the efficiencies and bandwidths of our AOFS in the anti-Stokes configuration (Fig. 4). An insertion loss of -15 dB is determined by fiber-to-fiber optical transmission measurements through the transmitted light port. This loss is mainly due to the optical mode mismatch between the lensed fiber and waveguides and could be reduced by employing an on-chip coupler [34]. We define the on-chip frequency shifting efficiency as the ratio of optical powers between the deflected light and the transmitted light (when microwave input is off). The measured efficiency linearly raises with microwave power and reaches 3.5% at a 30-dBm microwave power applied to the IDT pads ( Fig. 4(a)). Further increase of the microwave power can damage the IDT electrodes. We then measured the frequency-shifted optical power with varied microwave frequency (Fig. 4(b)). We perform the homodyne detection of the deflected light (Fig. 9). The 3-dB microwave bandwidth of our AOFS is 70 MHz for the Rayleigh mode (Mode I in Fig. 4(c)), which is determined by the IDT design. Although other acoustic modes have different frequencies, they feature a similar acoustic wavelength as defined by the IDT electrode pitch, and thus deflect the light at the same Bragg angle. The optical bandwidth can be estimated as Λ W λ [35], which in our case gives a theoretically predicted bandwidth of 19 nm. The measured optical bandwidth of our AOFS is 14 nm, which is close to predicted value. The discrepancy could be attributed to the mode couplings in the wide waveguides.

Acousto-optic frequency comb generation
An active acousto-optic frequency shifting loop, also known as frequency-shifted feedback laser [36] can be used for optical frequency combs and pulse generation [37][38][39]. Moreover, this approach is of interest for various applications in optical real-time Fourier transformations [40], photonic microwave channeling [41], and optical frequency domain ranging [42]. For these applications, a broad frequency comb with gigahertz frequency spaced lines could improve bandwidth or ranging resolution. Here, we generate an optical frequency comb with a 2.9 GHz line spacing using an active acousto-optic frequency shifting loop (Fig. 5). The loop contains our AOFS and an erbium-doped fiber amplifier (EDFA) to compensate the optical loss (Fig. 4 Inset). The loop is seeded with a laser at 1541 nm and couples light in and out by a 2×2 90:10 fiber coupler. The Bragg angle of the AOFS is chosen to be 17.45 degree to match the operating optical wavelength of our EDFA. The generated frequency comb features over 200 comb lines spanning 0.6 THz (5 nm) optical bandwidth.
The comb bandwidth is mainly limited by the gain saturation of the EDFA, in which the gain is reduced after the light has taken hundreds of round trips. The demonstrated gigahertz acousto-optic frequency comb could find applications in dual-comb spectroscopy [43], wavelengthdivision-multiplexed communication, and on-chip short-pulse generation.

Discussion and outlook
We demonstrate an integrated gigahertz AOFS on a thin-film-LN-on-oxide substrate. The shift frequency could be tuned from hundreds of MHz to a few GHz by adjusting the Bragg angle. Higher shift frequencies are possible for shorter optical wavelengths [44] or using materials with higher acoustic speed, such as aluminum nitride. The microwave bandwidth could be extended by using multiple IDTs or chirped IDTs [45]. The optical wavelength of our AOFSs can be designed to any wavelength in the transparency window of LN, which is from visible to mid-infrared. For mid-infrared AOFSs, LN on sapphire substrates could be utilized to avoid optical absorption by the underlying oxide layer. Although the efficiency of our AOFS is currently only 3.5%, it could be further improved by employing unidirectional IDTs [46] that could yield a 3 dB higher microwave-to-acoustic transduction efficiency than symmetric IDTs. Acoustic resonators [47] could be employed to reduce the input microwave power for acousto-optic intensity modulators, in which the input light could be deflected by a resonant standing acoustic mode. A wider acoustic wave could also improve the deflection efficiency at the same total acoustic power [4]. However, as discussed in Eq. 2, widening the acoustic wave also broadening the deflected beam, which could be challenging for collecting the light back to a single mode waveguide, especially for large Bragg angles θ B (high shift frequencies). On-chip lens [48] may be employed to better collect the broadened deflected light into a waveguide. Moreover, our integrated acousto-optic platform, in which light is deflected by acoustic waves, would lead to the development of on-chip optical routers, isolators, and scanners [35], microwave spectral analyzers [49], and carrier-envelope phase stabilizer [50].

Appendix A: FDTD Simulation
To investigate the Bragg diffraction by the acoustic mode (grating), we perform a three-dimensional (3D) finite-difference time-domain (FDTD) method simulation using Lumerical FDTD Solutions.
We simulate the acousto-optic interaction of our device using its actual dimensions. The configuration of the simulation is illustrated in Fig. 6. The acoustic grating is simulated by a 3D refractive index profile. The steady refractive index profile used here significantly reduces the computational expense without affecting the spatial profiles. The input waveguide is formed by two etched groves and is excited by the fundamental TM mode. The outer boundaries of the simulation region are set to the perfectly matched layers for the optical light. The optical profile monitor at 0.4 µm from the top surface of the 0.8 µm lithium niobate layer is shown in Figs. 2(d) and 2(e). The simulated far-field pattern after the acoustic grating (Fig. 7) shows the transmitted light at the input Bragg angle (sin θ = −0.3) of the acoustic grating and the deflected light at the opposite Bragg angle (sin θ = 0.3).

Appendix B: Experimental setups
The schematic diagrams of the heterodyne and the homodyne detection setups are shown in Figs. 8 and 9, respectively. The heterodyne measurement uses two lasers and the Stokes and anti-Stokes frequency shifts are distinguished relative to the input carrier light. Here, the reference Laser 2 is red detuned from the input Laser 1.
The homodyne measurement uses the light from the same input laser in detection. This avoids frequency drifting between two lasers and reduces the noise floor on the spectrum analyzer by using narrow bandwidth, which increases the detection time.