High-performance racetrack resonator in silicon nitride-thin film lithium niobate hybrid platform

In this paper, we propose an electro-optic modulator design in a hybrid Si3N4-X-cut LiNbO3. The modulator is based on a modified racetrack resonator and performs at both DC and heightened frequencies. Here the driving electrodes are defined along the straight section of the racetrack. This is done to maximize modulation and minimize modulation-cancelation that occurs in a conventional X-cut LiNbO3-based resonator due to the directional change of the electric field in the micro-ring. The single bus racetrack resonator is formed in a hybrid Si3N4-LiNbO3 platform, to guide the optical mode. The fabricated device is characterized and has a measured tunability and intrinsic quality factor (Q) of 2.9 pm/V and 1.3 × 105, respectively. In addition, the proposed racetrack device exhibits enhanced electro-optic conversion efficiency at modulation frequencies that match with the racetrack’s optical free spectral range (FSR). For example, at the modulation frequency of 25 GHz, which corresponds to the fabricated device’s optical FSR frequency, a ∼10 dB increase in electro-optic conversion efficiency is demonstrated. With the enhancement, our measured device demonstrates a conversion efficiency comparable to non-resonant thin-film LiNbO3 devices that possess RF electrodes that are 10 times longer in length. © 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement


Introduction
The micro-resonator has become a key building block in photonic integrated circuits (PICs), with applications spanning from microwave photonics to quantum computing, and high sensitivity sensors [1,2]. It offers various functionalities such as optical switching, filtering, high-speed modulation, wavelength division multiplexing (WDM) and frequency comb generation on a chip-scale [3]. The crystal ion sliced (CIS) thin-film LiNbO 3 -based standalone and hybrid photonic platforms (heterogeneous integration of LiNbO 3 with other materials) have drawn significant attention within the research community because of its tight optical mode confinement resulting low voltage operation, low drive power, and compact footprint. [4][5][6][7][8][9][10][11][12][13][14][15][16]. In addition, silicon-based micro-rings have been widely investigated [17][18][19][20][21]. However, due to the intrinsic properties of silicon, the application of these devices has been limited by its nonlinear response, temperature sensitivity and high insertion loss [22][23][24]. The Si 3 N 4 -LiNbO 3 hybrid platform has recently been investigated with great potential due to its superior optical properties and the versatile functionality [6][7][8]24]. This hybrid material system combines the strong linear electro-optic (EO) effect in LiNbO 3 with the ultra-low propagation loss, low second-order nonlinearity, small thermo-optic coefficient, high power handling capability, and wide optical transparency window of Si 3 N 4 [24,25], to achieve low optical insertion loss, low nonlinearity due to plasma dispersion, ultra-fast modulation and low voltage operation compared to alternative methods. Furthermore, a strip-loaded Si 3 N 4 waveguide on thin-film LiNbO 3 allows for an etch free LiNbO 3 modulator fabrication. Lastly, the highly confined optical mode together with closely placed RF electrodes significantly reduces the driving voltage required for electro-optic tuning. In general, the electrodes are placed along an optical micro-ring resonator [7,18,26,27], however, in an X-cut LiNbO 3 , the EO effect in r 33 crystal orientation dominates. As a result, such a ring structure would fail to take full advantage of r 33 due to the change in crystal direction along the optical waveguide path in the resonator [7]. Thus, to attain optimal modulation, the resonator structure needs to be modified to maximize the use of r 33 and provide a strong modal overlap between the RF signal and the fundamental transverse electric (TE) optical mode. Recently, modified micro-ring electrode design has been demonstrated in standalone X-cut LiNbO 3 [24,28] and hybrid X-cut ChG-LiNbO 3 platform [23]. However, hybrid Si 3 N 4 -LiNbO 3 platform remains unexplored. Furthermore, high-frequency operation and RF modulation enhancement using FSR coupling [29,30] of micro-ring on this hybrid platform have yet to be studied.
In this paper, we demonstrate a PECVD Si 3 N 4 -X-cut LiNbO 3 -based modified hybrid racetrack resonator where the modulating electrodes are placed on both sides of the long arm of the racetrack so that only the r 33 Pockels coefficient is utilized for resonant wavelength tuning/modulation. The modified structure efficiently utilizes the maximum EO coefficient and thereby improves the tunability by a factor of 1.6 over that of conventional micro-ring resonators [7]. A high intrinsic Q-factor of 1.3 × 10 5 and racetrack loss of 0.2 dB/cm are calculated for this modified structure. The demonstrated device shows a low V π L, high modulation efficiency, and high linearity. The device is characterized electrically up to 32 GHz RF frequencies, thus demonstrating high RF modulation efficiency. An enhanced modulation efficiency of about 10 dB is observed when the modulated optical sidebands are coupled to the adjacent free spectral range (FSR) modes [29,30]. This paper is organized by first presenting the design and fabrication of the modified structure, followed by the racetrack resonator transmission and DC characterization. Subsequently, high-frequency modulation and modulation efficiency are compared with a micro-ring resonator and the effect of FSR coupled-mode on the racetrack is discussed.

Device design and fabrication
A schematic of the proposed hybrid racetrack resonator is shown in Fig. 1(a). The hybrid waveguide core is formed by depositing a 200 nm plasma-enhanced chemical vapor deposition (PECVD) Si 3 N 4 strip on top of the 300 nm X-cut thin-film LiNbO 3 overlaid with a 2 µm SiO 2 bottom cladding layer. The thickness and width of the Si 3 N 4 strip are 200 nm and 1.2 µm, respectively. The dimensions are chosen to only support the fundamental TE mode that offers a bending loss of less than 0.01 dB/90°bend, 300 µm radii are used for all bends in this device. The mode confinement factor in the LiNbO 3 is ∼ 65% for this configuration [7]. The inset in Fig. 1(a) shows the field distribution of the fundamental TE mode in the straight waveguide section. The modulated racetrack consists of a 1.5 mm long (L Probe ) straight section, three half circles, and two S-shaped bends; with the same bending radius of 300 µm. After etching the Si 3 N 4 the parallel plate driving electrodes are patterned on the top of the exposed X-cut LiNbO 3 film. The racetrack straight arm is centered within the electrodes' 6 µm gap. The racetrack is designed such that the device conserves the optical quality factor (bus to racetrack coupling coefficient, absorption loss) and maintains maximum electro-optic interaction (see Appendix). An optical microscope image of the fabricated device is shown in Fig. 1(b). Figure 2(a) shows the cross-section schematic of the complete fabrication steps. The fabrication process involves three main critical steps: (i) thin-film Si 3 N 4 (200 nm) deposition by PECVD, (ii) optical waveguide and resonator patterning by e-beam lithography (EBL) and dry etching, and (iii) electrode definition by direct laser lithography. The fabrication steps are performed on the thin-film X-cut CIS LiNbO 3 on insulator (LNOI), which is procured from NanoLN. The 3" LNOI wafer consists of a 300 nm thick layer of the X-cut LiNbO 3 layer, bonded to a 500 µm thick quartz handle wafer via a 2 µm PECVD SiO 2 intermediate layer. The hybrid material platform is realized by depositing a thin-film Si 3 N 4 on the top of the LNOI substrate. A 200 nm of PECVD Si 3 N 4 is deposited using the following parameters: SiH 4 (10 sccm)/ NH 3 (6 sccm)/N 2 (300 sccm), chamber pressure of 100 Pa, RF power of 50 W, and temperature of 300°C. The refractive index is 1.943 at 1550 nm, measured by an ellipsometer and the deposition rate is ∼42 nm/min. To define the optical waveguide, an 80 nm thick chromium (Cr) layer is sputtered onto the insulating substrate. The Cr layer acts as a conducting layer to mitigate charging phenomenon during EBL. A soft-mask of the optical waveguide and resonators are defined by EBL writing on AR-N 7520.18 negative tone resist, the patterns are aligned in such a way that light propagates along the crystal Y-axis, this ensures maximum access to the r 33 electro-optic coefficient. The soft-mask patterns are then transferred into the Cr hard-mask with Cl [(45 sccm)/O 2 (2 sccm)] based inductively coupled plasma (ICP) dry etching and subsequently into the Si 3 N 4 thin-film using an Fl based ICP. The following parameters are used for pattern transfer into Si 3 N 4 : SF 6 (5 sccm) / C 4 F 8 (5 sccm)/ Ar (90 sccm), chamber pressure of 1.5 Pa, RF plasma power of 500 W and DC bias power of 150 W. The etch rate of Si 3 N 4 is ∼2.6 nm/sec. The remaining EBL resist on the sample was removed by oxygen plasma cleaning followed by piranha wet cleaning [H 2 SO 4 (75%): H 2 O 2 (25%)] at 80°C for 2 minutes. Finally, the remaining Cr is stripped in a chemically selective wet etch and followed by a 3-minutes piranha and oxygen plasma clean. This cleaning process strips the residual resist and insoluble particles that are stuck to the waveguide sidewalls. Finally, Coplanar Stripline (CPS) electrodes are defined on the top surface of the LiNbO 3 layer. A bilayer lift-off process is performed to transfer the electrode mask onto the soft-mask to obtain a clean solvent lift-off. The lift-off resists LOR-5A (MicroChemicals) is coated first, followed by the positive tone AZ 1512 (MicroChemicals) coating. A direct laser writer (Heidelberg MLA 100) is used to perform lithography, which maintains high-precision alignment of the electrode placement with the optical structure. The photoresists are developed simultaneously with AZ 300 MIF developer (MicroChemicals) for 2 minutes. An oxygen plasma is performed to strip residual resist and to promote metal adhesion of the subsequently deposited metal layers. Electron beam evaporation is used to put down 20 nm of Ti followed by 350 nm of Au. Metal lift-off is performed in NMP solvent (1-methyl-2-pyrrolidinone) at 80°C in the ultrasonic bath for 15 minutes. Devices are then diced, and the facets are polished at ∼15°top to bottom angle to ensure low fiber-chip coupling loss. The fabricated device with integrated electrodes is shown in Figs. 2(b) and 2(c).

Racetrack optical transmission and DC characterization
The transmission spectrum of the modified racetrack resonator is extracted using a tunable laser (Keysight 81608 A) swept at wavelengths around 1550 nm. The experimental setup of the measurement is shown in Fig. 2(d). The input and output optical light is coupled to the device under test (DUT) using a ∼ 2 µm spot size, TE-polarized tapered lensed fiber. The output light is detected by an InGaAs photodetector (DET08CFC) and directed to an oscilloscope via a trans-impedance amplifier (TIA). The measured spectrum of the modified hybrid racetrack resonator around 1550 nm is shown in Fig. 3(a). The spectrum exhibits periodic dips with uniform spacing between the adjacent resonating dips. The loaded quality factor of ∼ 6.5×10 4 is extracted at 1550.108 nm by fitting the central transmission dip with a Lorentzian model [31] as depicted in Fig. 3(b). The corresponsing intrinsic quality factor is ∼ 1.3 x10 5 [31]. The measured linewidth and free spectral range (FSR) are δλ FWHM = 0.0242 nm and ∆λ FSR = 0.207 nm, respectively. The propagation loss per unit length in the modified racetrack is calculated as ∼ 0.2 dB/cm [31]. The total optical insertion loss for this device is ∼13 dB, mainly attributed to the mode mismatch between the optical waveguide and input fiber. The tunability of the modified racetrack device is measured by applying a varied DC voltage to the driving electrodes and measuring the corresponding resonance wavelength shift. The applied DC voltage modifies the optical path length due to the change of the refractive index, causing a shift of the resonance wavelength. The DC voltage is applied to the racetrack's integrated electrodes using a DC source and a DC needle probe. The wavelength tuning spectra at different voltages are shown in Fig. 3(c). The resonance wavelength shifts by ∼ 72.75 pm from the resonance wavelength 1550.108 nm as the voltage are swept from 0 V to 25 V, demonstrating a wavelength tunability (δλ/δV) of ∼ 2.9 pm/V. The achieved tunability is 1.6 times better than that of our previous micro-ring based resonator [7]. The racetrack resonator also exhibits good linear dependence of the wavelength shift on the applied voltage as shown in Fig. 3(d). The DC modulation efficiency (V π L) of the racetrack resonator is also calculated using the following equation, V π L= (0.5*L Probe *FSR)/(δλ/δV) [32]. The calculated V π L of the modified racetrack resonator is 5.1 V·cm.

High-frequency characterization
To characterize the high-frequency response of the micro-ring and modified racetrack resonator configurations, we examine the generated peak sidebands that appear under the application of a modulating RF signals. The sidebands are offset from the input optical carrier by an equivalent to the applied modulation frequency. The setup used for sideband measurement is shown in Fig. 4(a). An RF electrical signal is fed from a signal generator (N5183A MXG) to the device electrode using a standard 50 Ω impedance ground-signal (GS) configuration probe (GGB industries) with a 67 GHz operational bandwidth. The optical signal is coupled into the DUT directly from the tunable laser via a TE polarized lensed fiber. The modulated optical signal is directly fed into a Yokogawa AQ5319 Optical Spectrum Analyzer (OSA) and the intensity of the observed modulation sidebands is recorded for each applied RF frequency. The modulation spectra are normalized to the measured power of the carrier at the off-resonant frequency of the resonator. The optical sideband normalization accounts for probe insertion loss and feed loss. Normalized modulation spectra for the modified racetrack and micro-ring are shown in Figs. 4(b) and 4(c), respectively. The normalized sidebands are plotted at 2 GHz increments for each device. The broad peak located at the center of the spectra corresponds to the laser wavelength, which lies in the slope of the resonance curve. The colored peaks on both sides of the center peaks are the modulation sidebands. The laser carrier wavelength is chosen such that the frequency detuning (optical frequency detuning of the laser from the mirroring resonance) is equal to ∼ 0.58 times of the half width at half max (HWHM) bandwidth of the resonator, which is considered as optimal detuning for maximum gain [33]. In the racetrack modulator, the sideband rolls-off as the RF modulation frequency (f m ) remains within FWHM < f m < FSR. The sidebands increase around adjacent resonances and approach a local maximum at ∼25 GHz: the FSR of the racetrack. This is due to both the sideband and optical carrier frequencies being resonant in the cavity simultaneously. The generated sidebands are transmitted through the ring, which increases the modulation efficiency [29,32] and the sideband power is increased by 9.5 dB as f m sweeps from 8 GHz to 25 GHz. The sideband power of the micro-ring resonator is also examined up to 32 GHz as shown in Fig. 4(c). Here the micro-ring is surrounded by the electrode (L probe = ∼1.5 mm) except the bus to the ring coupling region. In the micro-ring resonator, the sideband power rolls-off as frequencies increases upto 32 GHz, modulation enhancement is not observed since the modulation frequency remains below the FSR frequency of the ring (∼70 GHz). Here, the generated sidebands are not resonant in the ring, therefore the modulation efficiency decreases. At the modulation frequency of 8 GHz, the modified racetrack modulation efficiency is ∼10.32 dB higher than that of the micro-ring resonator, as shown in Fig. 5(a). These results indicate that for the same electrode length, the modified racetrack resonator structure improves the DC tunability, modulation efficiency, and energy consumption significantly. Besides, the modulation efficiency is comparable to the long thin film LiNbO 3 phase modulator [16]. The CPS transmission characteristics of the racetrack are measured using a vector network analyzer (VNA) up to 67 GHz. The VNA was calibrated using Short-Open-Load-Thru (SOLT) standards using calibration substrate 8 (CS8) from GGB. Ground-Signal probes were used for both the signal launch and terminated. The high-speed performance of the modulator devices depends on impedance matching of the CPS to the source impedance [34] and RF and optical index mismatch [34]. In Fig. 5(b), the electrical transmission (S21) shows a smooth trend up to 67 GHz and the loss is close to 0 dB due to the short electrical length, the trend also matches with the simulated S21 (see Appendix). The electrical return loss (S11) is below ∼12 dB upto 67 GHz indicating that the impedance is likely near 50 . From the S-parameter measurement [35], the characteristic impedance and the RF group index (n RF ) of the integrated racetrack electrodes have been estimated to be ∼ 58 Ω and 2.09, respectively. From the  simulated optical group index (n opt ), the modal effective index mismatch for this type of electrode is ∼0.08. The modulation efficiency can be further improved by optimizing the device electrode design (e.g. electrode gap, height, and RF cavity) [30,34,36]. Extended operational frequency is achievable by modifying the optical path length, such that the resonator FSR matches with the operating frequency [30]. The optical loss can be further improved by introducing a glass cladding layer and using low-pressure chemical vapor deposition (LPCVD) Si 3 N 4 instead of PECVD Si 3 N 4 , reducing the hydrogen content in the Si 3 N 4 [37][38][39]. Future work will require an investigation of an efficient edge coupling into and out of the optical waveguide by introducing either a mode transition structure using inverse taper or a grating coupler [40,41].

Conclusion
In summary, a tunable modified racetrack resonator is demonstrated using a Si 3 N 4 -LiNbO 3 hybrid platform that exhibits a high intrinsic Q factor of 1.3× 10 5 along with high resonance tunability of 2.9 pm/V. The modified racetrack design in both optical and electrical domains offers maximum utilization of the electro-optic coefficient of LiNbO 3 to the resonant structure, thereby achieving enhanced modulation efficiency compared to a micro-ring resonator. Our modified racetrack resonator also exhibits higher modulation efficiency (∼10 dB) when coupled to the adjacent FSR resonance modes. To the best of our knowledge, this is the first demonstration of the FSR coupled-mode resonance-based high-frequency modulation in hybrid Si 3 N 4 strip loaded thin-film LiNbO 3 platform. This unique device also exhibits excellent linearity, low voltage operation, banded modulation (can operate at n th adjacent resonance mode), low resonator optical loss, and compact optical modulation.

Appendix: Simulation results of the racetrack modulator
The single TE mode optical waveguide was designed in Lumerical Mode solver software. The optimal device dimensions are chosen to be maximizing the mode confinement in the LiNbO 3 and minimizing optical bending loss. The detail results were shown in our previous work [7]. The racetrack structure on the hybrid Si 3 N 4 -LiNbO 3 platform is simulated using 2.5-D variational finite difference time domain simulations (varFDTD) from Lumerical solutions. The transmission spectrum of the simulated microresonator is shown in Fig. 6(a). The resonance peak near 1550.115 nm is fitted with a Lorentzian as shown in Fig. 6(b). The simulated linewidth and the free spectral range are δλ FWHM = 0.021 nm and ∆λ FSR = 0.208 nm, respectively. The simulated result agrees well with the measured data. The linewidth variation might occur due to the additional losses in the fabricated waveguide. The loaded Q of the simulated micro-ring is ∼7.4×10 4 , which is higher than our measured loaded Q of 6.5×10 4 due to the additional losses in the waveguide. Another most critical factor for this type of modulator design is the effect of the electrode gap on the micro-ring quality factor. To maximize the modulation efficiency, a small electrode gap is desired. However, closely placed metal electrodes induce an additional absorption loss in the optical waveguide. As a result, the closely placed metals reduce the quality factor of the micro-ring. The metal-induced absorption loss for different electrode gaps is calculated in Lumerical MODE solutions. From Fig. 7(a), the absorption loss is significantly higher below 5.5 µm electrode gap, while the loss is nearly unchangeable above 5.5 µm electrode gap. The effect of the electrode gap on the intrinsic quality factor has experimentally measured. Figure 7(b) shows intrinsic quality factor ascending with increasing electrode gap, and it remained constant when the electrode gap reached to 6 µm. Thus, a 6 µm electrode gap is chosen for designing the CPS electrode and maintaining an acceptable metal absorption loss. The electric field distribution of our designed electrode is simulated with ANSYS High Frequency Structure Simulation (HFSS). The electric field map and the field profile of the device in contact with the probe are shown at Figs. 8(a) and 8(b). The electrical field along the optical waveguide is extracted. The simulated field illustrated about ∼9.8 × 10 4 V/m. The cross-sectional view of the simulated optical TE mode profile and the electric field directions (arrows) between the electrodes is shown in Fig. 9(a). Finally, the simulated S-parameters of the designed CPS is shown in Fig. 9(b).