1. Introduction

Photonic integrated circuits (PICs) are widely employed for wavelength multiplexing in telecommunication [1] and for sensing in life-science applications [2] to name just two examples. For the realization of compact PICs, waveguides with strong light confinement are necessary for realizing elements with small bending radii in the 10 − μm range. Thus, ridge waveguides in thin-film substrates are frequently used due to their high refractive index contrast. However, for low scattering losses it is mandatory to be able to structure ridges with sub-nm surface roughness. Therefore, standard materials for PICs are thin-films of silicon, silicon oxide [3] or silicon nitride [4], since these materials are CMOS compatible and thus established semiconductor fabrication techniques, which are deployed for decades for building electronic circuits, can be used. However, for many optical applications materials showing higher functionality, e.g. second-order optical nonlinearity or the linear electro-optic effect are of great interest [5].

One promising material also known as the “silicon of photonics”, is lithium niobate. It shows a large transparency window (absorption below 1 cm⁻¹ between 300 to 4500 nm wavelength [6]) and, due to its non-centrosymmetric structure, a linear electro-optic effect, that can be used for optical modulators or tunable filters [7–10] in lithium-niobate-on-insulator (LNOI) substrates. Furthermore, lithium niobate shows a high second-order optical nonlinearity, which makes this material appealing for integrated frequency converters [11–15]. However, lithium niobate is difficult to structure with standard semiconductor manufacturing techniques like reactive ion etching, since nonvolatile LiF is generated during the etch process [16], causing serious surface roughness. That’s why more sophisticated techniques like femto-second machining with subsequent focused-ion-beam milling has been deployed, achieving in the telecom range Q-factors in microdisc resonators of up to 2.45 × 10⁶ [17], corresponding to a loss of 15 dB/m. Furthermore, ridges have been fabricated by diamond-blade dicing [18] with 120 dB/m loss as well as with precision cutting [19], achieving losses below 10 dB/m. By electron-beam lithography and reactive ion etching narrow single mode waveguides with low propagation losses of 40 dB/m have been achieved [20]. Recently Zhang et al. reported on ridge waveguides in lithium niobate with ultra-low losses of 2.7 dB/m utilizing a multipass exposure of the photoresist and an improved inductive-coupled-plasma-reactive-ion-etching process [21]. Nevertheless, outstandingly low optical losses were achieved, these waveguides are still dominated by scattering losses since the absorption-loss limit of lithium niobate is of the order of 0.1 dB/m [6].

Here we report on a method to reduce the sidewall roughness of ridge waveguides of lithium niobate by a polishing postprocess. This improves the waveguide losses by more than one order of magnitude.

2. Fabrication and roughness reduction

The microring resonator and ridge-waveguide fabrication is conducted in two steps. First, we use standard semiconductor fabrication techniques to structure ridges into a lithium-niobate-on-insulator (LNOI, from NanoLN) substrate. Second, we polish the waveguide-ridge sidewalls to minimize scattering losses caused by surface roughness.

The LNOI substrates consist of a 1-μm-thick 5-%-MgO-doped z-cut LN thin film on top of a 2-μm-thick silicon dioxide layer, providing a large refractive index step and thus strong vertical light confinement. A 100-nm-titanium layer is deposited. At the subsequent image reversal lithography process [Fig. 1(a)] this layer prevents light from penetrating into the transparent LNOI sample. Multi-pass reflections within the LNOI sample would lead to an exposure of areas behind the mask which should be avoided. At the next step, we deposit 1400 nm chromium by physical-vapor deposition (PVD). Subsequently by lift-off, we obtain the chromium mask defining the waveguides with a width of 4 μm and the microring resonators with a width of 8 μm and a radius of 100 μm. [Fig. 1(b)]. This pattern is transfered by reactive-ion etching into the top LN thin film and the buried silicon-dioxide layer by reactive-ion etching [Fig. 1(c)]. We use CHF₃ and argon as etch gases with a flow rate of 25 sccm, an RF-power of 450 W, and the pressure of the chamber is very low at 0.9 Pa. Due to the high RF-power and the low pressure, we achieve large sidewall angles of up to 70°. However, since lithium fluoride (LiF) is redeposited in the etching process, we end up with ridge waveguides having high surface roughness.

 

Fig. 1 LN ridge waveguide fabrication steps (a–c) and schematic of polishing postprocess (d).

Download Full Size | PDF

To minimize the surface roughness, we developed a chemo-mechanical polishing postprocess [Fig. 1(d)]. We use a standard-wafer-polishing machine (Logitech, PM5), which enables us to scale this process up to the wafer level. This machine consists basically of a large polishing plate with a polishing tissue and a jig, where the sample is mounted upside down. Via adjustment screws the chip can be leveled to the polishing plate and via a spring the polishing pressure can be set. For successful sidewall polishing, four points are crucial: First of all a very soft polishing tissue (WhiteFelt, Buehler) has to be used, such that the 3-μm-high waveguide ridges can be pressed into the tissue and sufficient material removal can take place at the ridges [Fig. 1(d) zoomed view]. Second, to protect the waveguide top, the residual chromium mask is essential as a shielding layer. This works well, since we use chemo-mechanical polishing suspensions with silicon dioxide grains (MasterMet), which polishes LN faster than chromium. Third, it is important to apply very low polishing pressures since at pressures higher than 10⁴ Pa, the LN thin film starts to detach from the silicon dioxide layer [Fig. 2(b)]. Finally, an appropriate cleaning procedure has to be applied to remove all the polishing grains. Rinsing with water or cleaning with solvents does not work, since the silicon dioxide particles stick to LN [Fig. 2(b)]. We clean the waveguides successfully by rinsing in KOH at 45 °C for 5 minutes and subsequent rinsing in deionized water at 45 °C with ultra-sonic assistance, getting totally rid of all particles [Fig. 2(c)].

 

Fig. 2 Colored SEM images (red: chromium; yellow: LN; gray: SiO₂). (a) overview of microring resonator and zoomed views of ridge sidewalls (b)+(c). (b) shows detach of LN-layer and polishing residues. (c) perfect, cleaned sample.

Download Full Size | PDF

The product t_pol p of polishing time and pressure is important for the material removal. We varied this parameter from 1 × 10⁴ to 2.5 × 10⁵ min Pa. Figure 3(b) shows the sidewall for a low t_pol p. Here, a waviness with a periodicity in the 100-nm range and also strong roughness can be seen on both, the LN and the silicon dioxide. The waviness is caused by imperfections in the chromium mask, which are transfered into the ridges at RIE. The roughness may stem from LiF micro masking. At a polishing-time-pressure product of 1.2 × 10⁵ min Pa the waviness is completely gone and also the roughness is reduced [Fig. 3(c)]. At 2.5 × 10⁵ min Pa no roughness can be seen on the LN surface in the SEM images anymore. The polishing slurry comprises silicon-dioxide particles. They are considerably harder than lithium niobate. Consequently, the lithium-niobate film is affected stronger by the polishing procedure than the silicon-dioxide substrate. That’s why still some roughness can be seen on the silicon-dioxide surface. However, we cannot further increase the polishing-time-pressure product t_pol p since the chromium protection layer is used up at this point. For stronger polishing, one would need a thicker chromium protection layer. In our process this is not possible due to mechanical stress, which is induced in the chromium by the PVD process, causing delamination of thicker layers.

 

Fig. 3 (a) Colored SEM image of polished waveguides and (b–d) zoomed view of the sidewall (red box of (a)) of polished samples (yellow: LN; gray: SiO₂) applying different products of polishing time and pressure t_pol p.

Download Full Size | PDF

3. Optical loss measurements

To determine quantitatively the loss reduction, we employ whispering-gallery-resonator-loss measurements for differently polished samples and at different optical wavelengths. The method works as follows: The linewidth Δν_FWHM of whispering-gallery modes with a resonance frequency ν₀ is measured, and the Q-factor Q can be calculated via Q = ν₀/Δν_FWHM. The Q-factor results from two loss mechanisms: Intrinsic losses α_i and coupling losses α_c. Intrinsic losses α_i include absorption loss and bulk and surface scattering losses. If the radius of curvature of the waveguides is sufficiently large, radiation losses can be neglected. Then 1/Q = 1/Q_i + 1/Q_c is valid, where Q_i and Q_c quantify the loss mechanisms [6]. The total loss α is linked to the Q-factor by α = 2πn_eff/(Qλ), with the effective refractive index n_eff and the wavelength λ. In the undercoupled regime, the coupling loss α_c become small with respect the intrinsic loss α_i. Then, $Q_{\text{c}}^{-1}$ can be neglected and just the bulk- and scattering losses contribute to the measured Q-factor.

For being able to adjust the coupling strength to the undercoupled regime, we use a configuration like it is shown in Fig. 4. We use a tunable laser as the pump source. The polarization direction of the light is adjusted with a fiber polarization controller and coupled into a fixed chip with a straight coupling waveguide via end-fire coupling. Below this chip we have a second chip comprising the microring resonators under test, mounted on a x, y, z-piezo-actuator stage. The microring-resonator ridge width is 8 μm, why the waveguides are multimode, and we choose a large radius of 100 μm to be sure that we have no radiation losses. For radii below 40 μm also the roughness reduction of the inner sidewall of the microring might not work well enough. By adjusting the distance between the coupling waveguide and the microring resonator, we are able to adjust the coupling strength to the undercoupled regime. This is important since we want to measure the wavelength dependence of the waveguide losses. The evanescent field reaches farther out of the waveguides for longer wavelengths, which means that the coupling distance has to be increased, compared to smaller wavelengths, to stay in the undercoupled regime. The outcoupled light is collimated on a detector. Typical measurements of the least and most polished sample can be seen in the insets of Fig. 5. A Lorentzian fit is applied to the data to extract the linewidth and thus the Q-factor of the mode.

 

Fig. 4 Optical setup for whispering-gallery-resonator loss measurement.

Download Full Size | PDF

Figure 5(a) shows the loss of different samples, measured with light at 980 nm wavelength and TE polarization. The errors for the polishing-time-pressure product are due to pressure-setpoint stability limitations of the polishing machine used. For the Q-factor measurement, uncertainties arise from the fits to the data to determine the linewidth (below 5 %). Furthermore, the resonators are multimode, resulting in a Q-factor variation up to 2.5 × 10⁵ for different modes. The latter error, however, just leads to an Q-factor underestimation.

 

Fig. 5 (a) Q-factors vs. product of polishing time and pressure t_pol × p. (b) Q-factors and corresponding losses at different wavelengths. Insets: Scans over a WGM of a high (orange) and low (green) polished sample.

Download Full Size | PDF

The Q-factor increases continuously with stronger polishing and can be improved by more than one magnitude. There is no sign of a saturation effect, which indicates that still scattering losses at the ridge sidewalls are dominant. To further clarify this, we measure with the most strongly polished sample the Q-factor at different wavelengths (Fig. 5(b)) with TE polarized light. A power-law dependence with an exponent of 2.2 can be seen. The loss values range from 1.1 to 0.04 dB/cm, which is much higher than one would expect from absorption losses (for LN in the range of 10⁻³ dB/cm [6]). These results further confirm that scattering losses at the ridge sidewalls are dominant. We determined also the loss Q-factors for TM polarized modes at 980 nm wavelength for the least and most polished sample. The Q-factors are equal to TE polarized modes. Due to the waveguides’ asymmetry one might expect a polarization dependent loss. This would be true, if the top and bottom surfaces and the sidewalls of the waveguide show the same roughness. In our case, however, losses due to the sidewall roughness are dominant. Since our waveguides are considerably wider than high, there is not much difference between the field distribution of the TE and TM modes at the waveguides sidewalls. Thus, no polarization dependent loss is measured.

To build absorption-loss-limited waveguides, one would need stronger polishing. However, for reasons mentioned above this is currently not possible with our fabrication process. Another possibility would be to improve the reactive-ion etching process and to start polishing with waveguides already showing lower initial scattering losses. Recently Zhang et al. demonstrated that this is possible. Ridge waveguides in LN with ultra-low losses of 2.7 dB/m were fabricated with a multipass exposure of the photoresist and an improved inductive-coupled-plasma-reactive-ion-etching process [21]. Improving the losses of such waveguides by one order of magnitude by our polishing process may result in absorption limited ridge waveguides.

4. Conclusion

We showed how to achieve ultra-low-loss LN ridge waveguides by a polishing postprocess. We achieved optical loss reduction by more than one order of magnitude resulting in losses as small as 4 dB/m. This waveguide type is highly appealing for building compact PICs with high functionality, due to large refractive index steps resulting in high light confinement in a material providing large electro-optic and nonlinear-optic coefficients. Our method will help to overcome the scattering-loss limitation of such waveguides making on-chip narrow-band electro-optically-tunable filters and highly-efficient frequency converters more feasible. Furthermore, this polishing process is not restricted to LN but can also be applied to other materials.