1. Introduction

The successful commercialization of lithium niobate on insulator (LNOI) wafer fabricated by the smart-cut technique and wafer-bonding process [1,2], has increasingly aroused a great enthusiasm in the research of high-performance photonic waveguide components, due to the preeminent electro-optic (EO) property in single crystal thin film lithium niobate (TF-LiNbO₃, TFLN), and the high refractive-index contrast between LN optical waveguide core and its surrounding cladding [3–6]. There are two common approaches for all the off-chip light-coupling schemes between optical fiber and waveguide, namely, the end coupling and grating coupling methods. It was reported that the high coupling efficiency (CE) can be realized in end-face couplers with thick cladding layer, but it is sensitive to end alignment-error, and complicated to be fabricated with end-polishing process [7,8]. On the contrary, surface grating couplers (GCs) device is ease of fabrication due to its simple structure, and can be placed at arbitrary locations on the sample. Moreover, there even exists a high alignment-tolerance in GCs for the convenient devices measurements [9–11]. Therefore, the GCs have been growing as one mainstream of the instrumental light-coupling methods in the developments of large-scale optical devices integration.

Until now, a very few literatures on conventional LNOI GCs have achieved a low insertion loss (IL). The GCs either need to be precisely designed by novel methods, or required unusual geometrical parameters, adopted the complicated fabrication process. For example, the ridge-waveguide tapered GCs employed with the curved focused non-uniform grating-stripes had been investigated, and it was additionally designed with the side-arm fins for the following lift-off process, a minimum loss of approximately −9.45 dB/coupler at 1537 nm was measured [12]. Other reported literatures on uniform LNOI GCs, which probably featured a relatively much higher efficiency, usually required to add the upper cladding layer and refractive index of matching liquid, as well as the set of highly-cost bottom reflector. Worse still, the output performance was even more sensitive to the misalignment-tolerance and actual imperfect fabrication process, and the range of 3-dB optical bandwidth (BW) was very narrow. For instance, the integrated GCs deposited with SiO₂ upper cladding layer employed as the index-matching layer and the following etching mask layer had been fabricated, and a peak efficiency of −12 dB/coupler was reported [13]. By using focused ion beam (FIB) etching, a minimum loss of −9.1 dB/coupler was measured in uniform GCs without the tapered structures [14]. In contrast to the taperless GCs with uniform gratings [9], the integrated GCs with the tapers are much preferable, and increasingly employed in the efficient mode conversion for interfacing the photonic waveguides with off-chip single mode fibers (SMFs) [10,15]. The measurement of −9.95 dB/coupler was presented in the tapered GCs with uniform gratings [16]. And a measured loss of around −8 dB/coupler was also reported [17]. A much lower loss of around −7 dB/coupler was measured at the wavelength of near 1550 nm for vertical coupling GCs with SiO₂ cladding, and the hard-etching TFLN device layer was fully etched by further optimization of geometrical structure and parameters [18]. A minimum loss of −5.5 dB/coupler was measured at 1582 nm in the y-cut integrated vertical coupling GCs with both uniform gratings and bend waveguide structure [19]. Recently, the taperless GCs based on LNOI-on-Si structure had been fabricated by the highly-cost FIB setups, and merely realized a minimum loss of −8.3 dB/coupler for transverse-electric (TE) polarization [20].

To the best of our knowledge, state-of-the-art investigations still have not paid enough attention on the LNOI integrated GCs, the high-efficiency devices with traditional structure even have not reported in previously demonstrated literatures. In this work, the ridge-waveguide GCs with tapered structures had been designed, modeled and fabricated. The proper selection of SiO₂/handle-substrate stack, as well as the optimum choice for substrate material, had been independently investigated, the device geometrical structure and parameters were fully optimized. The measured optical response exhibited a low loss of −6.3 dB/coupler at the wavelength of 1543 nm, and a wide 3-dB bandwidth of larger than 90 nm. Moreover, all the GCs structures simultaneously defined on the sample were patterned by one step of electron-beam lithography (EBL) process, followed by improved inductively coupled plasma (ICP) dry-etching and cleaning process.

2. Device design and simulation

The schematic top-view of LNOI integrated GCs layout is presented in Fig. 1(a). Seen from Fig. 1(a), it is clear to see that the GCs device consists of the straight uniform grating-stripes, the mode tapers and the photonic stripe-waveguide. In order to obtain a high CE in the tapered GCs, all the geometrical parameters had been optimized in our work. The scale size of each part in the device is marked with respective symbol, as plotted in Fig. 1(a). L_stripes is the total length of uniform gratings, and W_stripe is the nominal width of all these same grating-stripes. L_wg is the length of ridge-waveguide photonic wire, and W_wg is the width of straight waveguide. L_taper is the length of ridge-waveguide taper. The side-view ridge-waveguide GCs is showed in Fig. 1(b), the partial zoomed-in image is also presented in Fig. 1(c), and the resulting structure parameters are marked, respectively. w is the width of single grating-stripe, Λ is the period of grating-stripes, w/Λ is the filling factor. H is the total thickness of TFLN layer, and h is the etching depth of ridge-waveguide. Considering that the ridge-waveguide photonic devices had been widely investigated, and the selection of etching depth larger than the un-etched portion in ridge-waveguides is always favorable, due to the fact that much stronger light mode coupling and confinement can be realized in the deeply-etched ridge structure [7,19,21,22]. Then, the deeply etching of TFLN device layer was also adopted in our case, and only one step of EBL pattering process was implemented to simultaneously define all the structures on the sample. It is obvious to find that the LN crystallographic axis is also embedded in Fig. 1(c), and the direction of x-cut LN ridge-waveguide photonic wire is aligned to the y crystallographic axis, due to the electro-optic effect in the further modulator applications [8,23,24].

 

Fig. 1. Schematic top-view (a) and side-view (b) of the integrated ridge-waveguide LNOI GCs with the tapers. The enlarged image of partial gratings region is plotted, as marked with blue dashed lines (c).

Download Full Size | PDF

Seen from Fig. 1(b), the schematic of light coupling process from optical fiber to the GCs is also exhibited. It is easy to find that the fundamental light mode was injected into the top left fiber core, which was used as the input light source for TE polarization. In our simulated model, the coordinate axes was located at the leftmost end of the left taper, and the origin was inserted in Fig. 1(b), as schematically plotted with the arrows and a symbol of “O”, respectively. The fiber was placed above the TFLN layer, where it was optimized to be located at the position of x=+5 µm, and y=−7.76 µm. The light power distributions had also been analyzed, as marked with the white arrows, and the relation could be simply described as follows: P₀=P₁+P₂+P₃+P₄, where P₀ is the light power inputs, P₁ is the light power coupled to the left side and it can be dissipated finally, P₂ is the power radiated downward, P₃ is the power transmitted to the right side, and P₄ is the power reflected back into the incident fiber. The light power distribution in GCs was simulated by using the finite-difference-time-domain (FDTD) methods (by Lumerical, Ansys). The output optical performance was characterized, and the simulated result of electric-magnetic field distributions is also presented in the following sections.

The electric field distribution of the optical wave in the simulation region is shown in Fig. 2(a), and the center of coordinate axes was also inserted here with respect to the one in Fig. 1(b). It is easy to find that a part of light mode was effectively coupled into the GCs, and the partial light power radiated downwards was further leaked into the bottom high-index Si substrate, due to the fact that there exist one source of mode losses in GCs [19]. Moreover, the relation between CE and the thickness of SiO₂ lower cladding layer, had also been simulated and analyzed, as shown in Fig. 2(b). It is obvious to find that the curve of coupling efficiency oscillates between the minimum and the maximum value with respect to the buried oxide (BOX) thickness. According to the existing literatures [9,19,25], we inferred that the effect might be attributed to the constructive interference of light modes among the part of light power radiated upward at a particular etching-depth, the part of power scattered to the bottom substrate and then reflected back by the SiO₂/Si interface, as well as the part of power diffracted upward at the top surface of the partially etched grating-stripes.

 

Fig. 2. The calculated electric field distributions (a) and coupling efficiency as a function of the buried oxide (BOX) thickness (b) in the GCs for TE polarization.

Download Full Size | PDF

Based on the researches [6,9], we noted that the best grating coupling characteristics experience a dependence on the structural parameters, such as the selections of Λ, h, and w as well as the optical fiber position. In our simulations for the designed LNOI GCs model, the corresponding parameters tolerances on the CE product were systematically studied, respectively. The FDTD optimization algorithm was referred to the particle swarm algorithm (PSA), to result in the optimum results during the process of self-specified design-intent sweeping parameters, and a part of the scanned results are shown in Fig. 3, respectively. Seen from Fig. 3(a), we can find that there is an obvious phenomenon of the existence additional coupling loss in these GCs transmissivities operated at the wavelength of 1550 nm, and the average discrepancy is of around 0.35 dB with a 10 nm deviation of grating period. Namely, the obtained coupling efficiency characteristic is greatly sensitive to the parameter of Λ. Besides, the coupling efficiency as a function of the etching depth (h) is also simulated and calculated here, as shown in Fig. 3(b). It is clear to see that there is no obvious difference between these transmission curves for the case of a 10 nm deviation in etching depth.

 

Fig. 3. The simulated coupling efficiency (CE) as a function of the grating period (c) and etching depth (d) for TE polarization in the x-cut LNOI GCs, respectively.

Download Full Size | PDF

Considering that the output performance in GCs also depends on the choices of LNOI wafer material and structure, and then, the selection of substrate material is also an important factor. Compared with the expensive LN wafer used as the substrate, the complementary metal oxide semiconductor (CMOS) compatibility of Si wafer is more promising [4,26]. Thus, the TFLN wafer bonded on Si substrate by using the middle material of SiO₂ was customized in our case, the thickness of the bottom Si substrate was measured to be of 500 µm, H was adopted with 600 nm, and the thickness and sizes of the purchased TFLN chips was verified by the film thickness measurements (from FILMETRICS, KLA-Tencor). The BOX thickness was selected to be of 4.7 µm, which was also corresponding to one of the maximum CE dimensions, as marked with the dashed blue line in Fig. 2(b). Based on the existing researches [1,18,27], the GCs structure parameters and optical effective index for supporting fundamental polarizations had been systematically investigated. In order to fully meet the effective mode theory and diffraction conditions for realizing the single mode transmission at the wavelength of 1550 nm. Our original parameter selections for deeply-etching ridge-waveguide device had also been conducted, and the proper choice of W_wg=1 µm. The optimized parameters are listed as follows: Λ=1.16 µm, w=0.71 µm, h=550 nm, L_wg=10 µm, L_taper=200 µm, L_stripes=15 µm. Moreover, the optimal geometrical parameters and actual fabrication results for the ridge-waveguide LNOI GCs are compared, as shown in Table 1. Noted that there is a discrepancy between the simulation and experiment results, which can be attributed to the fabrication errors sensitivity in actual experiments, since there always exist a challenge for effectively obtaining the ideal rectangle optical waveguide due to the nature of hard-etching in LN material.

Table 1. Comparison between the simulated and actual measured geometrical parameters in ridge-waveguide GCs.

View Table

3. Device fabrication

Based on the optimized structure parameters, all the GC devices on the same sample were simultaneously fabricated. In our case, the sufficient x-cut LNOI sample was availably purchased (from NANOLN), the 3 inch wafer was sliced into 1.0 cm by 1.1 cm segments (by DISCO DAD3650). The fabrication procedure flow of LNOI GCs is plotted in Fig. 4. Seen from Fig. 4, the LNOI wafer was firstly cleaned by wet process and oxygen plasma treatment (by PVA TePla Plasms System). And a 800-nm-thick amorphous silicon (α-Si) hard mask layer was deposited on the TFLN device layer by using the plasma enhanced chemical vapor deposition (PECVD) systems (by Oxford PECVD system), followed by spin-coating a 400-nm-thick electron-beam resist of 6200.13 on the α-Si mask layer. Then, only one step of EBL writing was performed to define all the structures (by Vistec EBPG 5200). After that, the exposed resist was developed, and dry-etching in inductive coupled plasma (ICP) reactive ion etching (RIE) systems was implemented to transfer the patterns into α-Si mask layer (by SPTS DRIE-I). Then, the residual resist was fully removed by acetone wet-etching and oxygen plasma cleaning. And then, the improved argon ions (Ar⁺) dry-etching process in the ICP systems was conducted to transfer the patterns into the underlying TFLN device layer (by Sentech SI 500), and it was similar to the reported process [7,8,28]. Finally, the residual α-Si hard mask was precisely removed by wet-etching process of diluted potassium hydroxide, followed by the preferable sample cleaning after fabrication.

 

Fig. 4. Schematic of the ridge-waveguide GCs fabrication process flow.

Download Full Size | PDF

Moreover, during the fabrication process, the EBL defined resist was additionally post-baked after EBL process prior to the resist development process in our case, in order to increase the hardness of resist, and to properly relieve the LN charging effect induced by EBL. Thus, the etching selectivity for the following etching process was enhanced, to obtain a much smoother side-wall profile. Besides, the fabrication of α-Si film was systematically investigated, as for there is a difficulty in achieving high-quality thick α-Si film during the PECVD deposition process. The etching process for α-Si hard mask, such as the over-etching effect under the condition of too thin α-Si hard mask, was also fully considered to avoid the transferred patterns distortion, and to reduce the surface roughness. Thus, the total thickness of α-Si hard mask was properly selected, and the un-etched α-Si thickness with a few tens of nanometers was intentionally designed in case of LiF byproducts during the α-Si etching of fluorine-based gases.

Based on the existing literatures [1,28,29], the relationship between the Ar⁺ etching parameters in ICP systems and the TFLN etching results had been studied. We found that a much lower value of chamber pressure can realize a much smoother side-wall, the higher ICP powers and gas flow rate can lead to a much smaller side-wall roughness, but the slanted angle of LN side-wall decreased. We deduced that the phenomenon can be attributed to the effects of serious etched-material re-deposition and re-sputtering during LN etching, thus the α-Si mask selectivity and etched LN thickness were sacrificed. It was even reported that the LN layer etched by Ar⁺ milling at a controllable low temperature of −50 °C, which could result in a higher LN quality after fabrication [16]. That is, the temperature transferred on the sample might also play a role in the TFLN quality during etching, and high chamber temperature could reduce the α-Si mask selectivity. Meanwhile, the actual operation temperature can be cooled down to around 0 °C in our available etching systems, and there exist the poor heat-dissipation problem for a long time etching process. Considering these factors as discussed above, we even found that the actual temperature transferred on our samples was as high as 40 °C after the etching process, and the heat-dissipation problem might lead to an etching profile discrepancy between the simulations and experiments.

Considering that the output performance of GCs experiences a dependence on the fabrications of TFLN waveguides, and then the optimizations of LN-etching processes were regarded as one of the important points to obtain a high CE in GCs. In our following experiments, the Ar⁺ dry-etching process was repeatedly implemented, and it was set of multiple etching times. In the chamber setups, the temperature was of 0°C, the pressure was of 0.7 Pa, the Ar⁺ gas rate was 80 sccm. Besides, the conditions of relatively high RF power of 400 W and ICP power of 600 W were selected to enable a direct current (DC) ion beam bias voltage, in which it was sufficient for accelerating the Ar⁺ towards the LN sample, in combination with the efficiently ion-bombardment and physical etching of TFLN device layer, resulting in a much smoother profile and more precipitous waveguide side-wall. In order to greatly relieve the effects of etched LN re-deposition and sputtering, as well as other particulars deposited on the device sidewall and surface during LN etching, we firstly re-optimized the device structure with more suitable parameters that were less affecting the actual fabrication sensitivity. Moreover, the fabrication processes had also been further optimized due to the fact that there is a long time etching of LN, in our group, the multi-steps etching procedure was developed by separating with a few etching sub-cycles, and each etching process was conducted under the identical conditions. In detail, the total time for etching LN device layer was divided into 3 more shorter etching sub-periods, in order to ultimately avoid the high temperature effect gradually transferred into the LNOI chip. Besides, according to the existing reports [19,29], to remove any residues and byproducts, the cleaning methods after respective etching step of α-Si mask and LN layer were considered and optimized, respectively. And then, the standard cleaning process was even selectively adopted and improved in our experiments, especially for dealing with the water-bath heating mixed solutions of NH₄OH, H₂O₂ and H₂O (denoted as the SPM process), and the temperature was controlled to be of 60–80 °C for each cleaning of 10–15 mins.

4. Device characterization and measurements

The optical microscope characterization was implemented, and the captured image for a part of the fabricated LNOI GCs structure is presented here in Fig. 5(a). As for the whole device structure is a bit long, thus the optical micrograph of the integrated GCs with the tapered structures was measured with a proper magnification. Scanning electron microscope (SEM) measurements were also performed, the non-conductive LN sample was measured at a low voltage condition, and the obtained graphs were scanned with the SE2 mode, as shown in Fig. 5(b) and Fig. 5(c), respectively. Seen from the fabricated devices, the sample surface is clean after fabrication, and the measured w was of 0.73-µm-width, Λ was of 1.17 µm, and W_wg was of 1.03 µm, as presented before in Table 1. The measured results were in good agreement with our originally designed parameters, and there exist the effect of fabrication imperfections on the GCs device sizes, such as the tapered structures and gratings, especially for etching the fine ridge-waveguide grating-stripes. The estimated etching selectivity between the α-Si mask and TFLN layer was of around 1.4: 1, before the significant erosion of α-Si mask layer. The measurement of approximately 470-nm-thick etched LN layer had been effectively achieved by using our etching scheme, and the LN etching rate was calculated to be of 36 nm/min.

 

Fig. 5. Optical micrograph and the corresponding SEM images of a part of the fabricated ridge-waveguide GCs.

Download Full Size | PDF

For the output performance characterizations of optical properties, the schematic of testing systems employed to measure the transmission response of the fabricated GCs is presented, as shown in Fig. 6(a). In detail, the TE-polarized mode was emitted from a tunable continuous-wave (CW) laser, the laser power could be fed into the devices by the polarization maintaining fiber, and then the transmission signals were recorded by a power-meter (by Keysight 81960A). The concrete parameters of measurement setups are presented as follows: the initial laser power in the input power sensor was set to be of 10 dBm, the starting wavelength was of 1504.1 nm, and the stop wavelength was of 1620 nm, and the step of stepped sweeping size was set to be of 2 pm (by Keysight 81636B). The polarization of the input light mode was adjusted by a polarization controller. The color-picture of partial actual testing setups and the chip is also presented in Fig. 6(b). Considering that the optical fibers were fixed on the holder to precisely change its position, and the relative angle to the sample surface could be theoretical adjusted by means of the manipulators. While the incident angle (θ) is always fixed at 10° in simulations and experiments, due to fact that there is a limitation on actual tolerance of our measurement systems, as shown in Fig. 6(b). The output power from the fabricated device was collected through the standard SMF by the GCs structure after each manual adjustment for reaching its maximum throughput.

 

Fig. 6. Schematic of the optical testing systems (a) and the zoomed-in image of partial actual testing setups (b) for the transmission measurements of the GCs.

Download Full Size | PDF

The simulated and measured transmission spectrum of the GCs for TE-polarized fundamental mode is presented, as shown in Fig. 7. Seen from Fig. 7, it is clear to find that the maximum CE of −3.4 dB/coupler and −6.7 dB/coupler was simulated and measured at the telecom wavelength of 1550 nm, respectively. And the fabricated best GCs even exhibited a peak CE of −6.3 dB/coupler at 1543 nm. To the best of our knowledge, the measured efficiency is high enough so far. Moreover, multiple GCs with the same structure design could effectively operate at the wavelengths ranged from 1504.1 nm to 1620.3 nm, and it revealed that the measured performance for all the simultaneously fabricated GCs are repeatable and stable. The difference between the simulations and experimental results, which could be attributed to the LN etching error during the GCs device fabrications. Compared with the device sizes between the simulations and actual experiments, we find that the fabricated devices lateral sizes are nearly in good agreement with the simulated values, such as the parameters of Λ and w. However, compared with the designed parameter of h, it even revealed that there is an obvious difference in contrast to the designed value of expected etching depth due to the actual manufacturing imperfections, as for there is always a big challenge to effectively etch the LN material in the vertical direction, and the error of deeply-etching depth in our case lead to the imperfect fabrications of standard rectangle waveguides, and even featured with the slant waveguide side-walls. Namely, the final etching depth could be reached at a sacrifice of the device lateral sizes. For our fabricated GCs device, we can also find that the peak wavelength was shifted with a few nanometers, which might be correlated to the constant incident angle (θ) in our measurement setups. Meanwhile, it could not be denied that the high-performance was obtained in our fabricated LNOI GCs. In our opinion, this could be attributed to the customized LNOI on Si wafer with the optimized 600-nm-thick LN device layer, the optimal geometrical parameters with a relatively thick SiO₂ lower cladding layer, and the CMOS-compatible Si handle-substrate for supporting the TFLN layer. Moreover, the compact GCs structures designed with optimized gratings, and proper width selection of ridge-waveguide photonic wire, as well as the tapers designed without too narrow sections, which might lead to a much lower transmission loss and mode mismatch. Besides, the improved dry-etching and cleaning process could also play an important role, as discussed above.

 

Fig. 7. The simulated and measured transmissivity results for the ridge-waveguide GCs for TE polarization.

Download Full Size | PDF

Moreover, the 3-dB optical bandwidth (BW) for the fabricated GCs was also characterized, as marked with the dashed blue lines in Fig. 7. It is clear to find that the optical bandwidth is broad, and a 3-dB BD of more than 90 nm was calculated. Moreover, we even made a conclusion to the reported researches focused on the interest of coupling characteristics in the GCs structures, such as the taperless GCs with the highly-cost bottom reflector [9,14], the novel designed tapered GCs structure and added with the bottom reflector [10], the vertical coupling GCs employed with uniform and chirped hybrid gratings [19], as well as the FIB fabricated taperless GCs both with the bottom reflector and chirped apodized gratings [20]. In contrast to the reported complicated and highly-cost GCs structures with a high loss, our tapered GCs devices are highly-efficient and ease of fabrication, due to its simple structure with uniform gratings. Moreover, without employing these complex structures and high-cost process as discussed above, the output performance in our GCs even could be enhanced in a wide margin by further optimizing the device structure, accurately controlling the fabrication process, and improving the sensitivity of measurement setups, as well as matching the best roll angle (θ) of the injected optical fiber. During the devices manufacturing process, by ultimately improving the fabrication tolerance of structure parameters such as h, Λ, and w, and conducting systematical investigations on the sample cleaning methods, there may yield a transmission response close to the ideal model with high-performance in our further fabricated devices.

5. Conclusions

In a conclusion, the conventional ridge-waveguide GCs with straight uniformly-distributed grating-stripes and the tapers, which were based on a 600-nm-thick x-cut thin-film LNOI-on-Si platform, had been designed, fabricated and characterized. In our work, the measured optical transmission in the LNOI GCs exhibited a low IL of −6.3 dB/coupler at 1543 nm for TE mode, and a high performance of −6.7 dB/coupler at the telecommunication wavelength of 1550 nm, with a broad 3-dB optical bandwidth of more than 90 nm. Moreover, all the GCs structures patterned in the sample were ease fabrication, compatible with CMOS process, and could be simultaneously defined by one step of EBL process. The results may provide significant TFLN applications to the further researches of traditional GCs with high-efficiency, and boost the developments of photonic devices on LNOI.