1. Introduction

Whispering-gallery resonators (WGRs) made of non-centrosymmetric material have been shown to have great potential for extending the emission range of laser sources by frequency conversion and efficient three-wave-mixing processes such as second-harmonic generation and optical-parametric oscillation in classical and non-classical regimes [1,2]. Due to their high Q-factors up to 10⁹ [3] efficient optical frequency conversion processes with low pump powers in the mW and even in the μW range could be demonstrated [4–7]. Lithium niobate (LN, LiNbO₃) was frequently used for applications in the visible and near-infrared wavelength range, while other materials were used to operate WGRs even in the ultra-violet (beta barium borate [8], lithium tetraborate [9]) or mid-infrared region (silver gallium selenide [10]). However, these millimeter-sized bulk WGRs were fabricated on a lathe by diamond cutting and polishing, processes which are not suitable for realizing integrated laser sources. Thus, fabrication of integrated on-chip LN WGRs has become of great interest, but limitations in fabrication quality have led to orders of magnitude smaller Q-factors compared to their bulk counterparts. Therefore only electro-optic-tuning of filters [11–14], second harmonic generation (SHG) [15–18] and sum-frequency generation (SFG) [19] have been demonstrated on-chip.

On-chip high-Q WGRs, with Q-factors up to 10⁷ to 10⁸, are typically made of materials like silicon dioxide [20,21] and silicon nitride [22–24]. Transient melting of the resonators of these amorphous materials by selective heating provides almost perfect surfaces and standard semiconductor fabrication processes can be applied to yield free-standing resonators. Applications like Brillouin lasers [25], emitting light with ultra-low linewidth, and frequency combs [26] have been demonstrated using such structures.

These materials are centrosymmetric, such that they exhibit no second-order nonlinearities. However, for fabricating tunable light sources, second-order nonlinearities have some advantage SHG and OPO, can be employed. Furthermore, active stabilization and tuning of resonances is possible via the Pockels effect which is why researchers tend to induce second-order nonlinearities by applying electric fields [27] or strain in centrosymmetric materials [28] or by breaking the symmetry of waveguides [23]. An easier way to obtain frequency conversion through high second-order optical nonlinearities, however, is to use non-centrosymmetric materials like LN; gallium arsenide [29,30]; aluminum gallium arsenide [31]; gallium nitride [32]; or aluminum nitride [33]. LN has the advantage that, with the help of electric fields, stable domain structures can be imprinted allowing for quasi phase matching [34]. Therefore, LN is of great interest for realization of integrated frequency converters. There are two design concepts commonly used for realization of on-chip WGRs. One employs microdiscs on micropillars [12, 13, 15–17, 35, 36] where a tapered fiber is used to couple light into the WGRs. The other is based on on-chip microrings with a coupling waveguide placed close to it, which leads to a higher integration level and better mechanical stability [11,14]. Although lithium niobate is good for fabricating bulk frequency converters, it is not easy to fabricate on-chip WGRs with low surface roughness using this material. That is why sophisticated, serial fabrication techniques like femto-second laser writing with additional focused-ion-beam milling have been applied to achieve Q-factors in the range of 10³ to 10⁴ for ring-like WGRs and 10⁵ to 10⁶ for disc-like WGRs. Furthermore, for efficient frequency conversion it is important to adjust the coupling strength between the WGR and the coupler in a well-controlled manner since the coupling strength has a large impact on the non-linear optical processes [2].

We report here on a novel fabrication method to generate on-chip WGRs with improved quality factors up to 3.0 × 10⁶ using a flexible coupling scheme and show to our knowledge for the first time cascaded nonlinearities. These results prove the high quality of the fabricated structures and are an important step towards totally frequency-agile integrated nonlinear-optical laser light sources.

2. Fabrication and optical characterization

We use mass-production processes like standard UV lithography, reactive-ion etching, and a highly-parallel polishing process as outlined in Fig. 1(a). We employ lithium niobate-on-insulator (LNOI) wafers from NANOLN for the fabrication of our devices [Fig. 1(a)]. We used two wafer configurations. On the one hand a wafer with a 500-μm-thick z-cut lithium niobate substrate, a 2-μm-thick silicon dioxide intermediate layer and a 0.7 μm z-cut LN thin film on top to establish the polishing process. On the other hand a wafer with a 500-μm-thick z-cut lithium niobate substrate, a 1-μm-thick silicon dioxide intermediate layer and a 2 μm z-cut LN thin film for the non-linear optical frequency conversion. We deposit 100 nm of titanium via physical vapor deposition (PVD) on the substrate. This layer serves as a reflection layer enduring the image reversal lithography to prevent multi reflections in the transparent LNOI substrate, which would lead to an exposure of the dark areas. In the image reversal lithography step the photoresist AZ 5214E is used, and after structuring of the resist via UV lithography we deposit 900 nm of chromium on the sample. Via lift-off we obtain the chromium hard mask, which defines our waveguide structures on top of the titanium layer. Next we transfer this pattern into the LN thin film and the silicon dioxide layer via reactive ion etching (RIE). We use 25 sccm Argon and 25 sccm CHF₃ as etch gases. The chamber pressure is 0.9 Pa and the RF-power 450 W. Figure 1(b) shows such an etched WGR. In the zoomed view of Fig. 1(c), a high roughness and waviness of the sidewall becomes evident. This is caused by limitations in the etching process such that we developed a chemo-mechanical polishing process to remove the waviness and decrease the roughness to 4 nm as measured by atomic force microscopy (AFM); the final result is seen in Fig. 1(d). The polishing of the sidewalls of the waveguide ridges is done with a chemo-mechanical polishing procedure with a silicon dioxide suspension (MasterMet 1 and MasterMet 2 from Buehler) on a standard wafer polishing machine (Logitech PM5). We use a soft tissue so that there is mechanical material removal not only at the ridge top but also at the sidewalls. However, since there is residual chromium left on top of the ridges from the chromium mask, used at the RIE process, the LN film is protected from top and polished at the side only. That’s how we are able to selectively polish just the sidewalls of the waveguide ridges without destroying them by a strong removal from the top side. Last, we remove the remaining chromium and titanium layers via chemical wet etching and finally get microring waveguides with a diameter of 200 μm and a ridge width of 7 μm.

 

Fig. 1 a) Fabrication steps of the WGRs. b) Colored SEM image of the WGR, 200 μm in diameter and the ridge width is 7 μm, after RIE. c) Zoomed SEM view of ridge sidewall before polishing. Grey: LN substrate; Blue: 2 μm thick SiO₂ layer; Red: 0.7 μm thick Single-crystalline +z-cut LN thin film; Yellow: Cr and Ti layer. d) Sidewall after polishing and removal of Cr and Ti layer.

Download Full Size | PDF

The Q-factor was measured by scanning a pump laser at 976 nm across resonances of the WGR and determining the linewidth of the whispering-gallery modes (WGMs). A schematic of the setup can be seen in Fig. 2. To couple light into the WGR, we fabricated a second chip with straight 3-μm-wide ridge waveguides as coupling waveguides, where ordinary polarized light is fed in by end-fire coupling. We monitor the power of the pump light via a photodetector before the microscope objective. At the output of the coupling waveguide, we record the transmission of light in the infrared and visible wavelength ranges. For this, the detector for infrared light is used to measure Q-factors of the pump WGMs, and the detector for visible light is employed to record the frequency-doubled light. The WGR chip is placed below the chip with the coupling waveguide and mounted on a x-y-z-θ_x-θ_y-piezo-actuator stage to align the coupling waveguide to the WGR, enabling evanescent coupling.

 

Fig. 2 Schematic of the optical setup for Q-factor measurement and for investigation of nonlinear optical processes. The pump light is polarized perpendicular to the optical axis of the LN thin film.

Download Full Size | PDF

By changing the distance between the coupling waveguide and the WGR in the z-direction via a piezo actuator, one can adjust the coupling efficiency from undercoupling to critical coupling and also overcoupling in a well controllable manner [Fig. 3]. The coupling strength follows hereby the function K(V_piezo) = 4r /(1 + r)² with r = c₁exp(−c₂V_piezo) [2] with the voltage V_piezo applied to the piezo actuator, the constant c₁, and the constant c₂ which describes the expansion of the piezo actuator. At critical coupling we can couple more than 95 % of the pump light from the coupling waveguide into the WGR. To determine the intrinsic Q-factor of our WGRs, we select critical coupling and measure the linewidth of the WGMs at this point, since we have here the strongest coupling so that we can measure the linewidth of the WGM with the lowest error. At critical coupling, the intrinsic losses of the WGR are equal to the coupling loss, and the measured linewidth is twice the intrinsic linewidth δν = 2δν_i. The dependency of the linewidth with coupling distance is δν(V_piezo) = δν_i(1 + r) [2]. The Q-factor can be calculated as Q = ν/δν_i with the frequency of light ν and the intrinsic linewidth δν_i [39]. The Q-factors at 976 nm vary between 0.9 × 10⁶ and 3.0 × 10⁶ and we could not observe a significant influence of the light polarization on the Q-factor. We note that these values are more than two orders of magnitude higher when compared to previous work on integrated LN ring WGRs [11, 14]. Compared to integrated LN disk WGRs, the Q values are slightly higher than the previously published values ranging from 10⁵ to 2.45 × 10⁶ which were, however, measured at a wavelength of 1.5 μm [12,13,15–17,35,36]. At this longer wavelength, the Q-factors of our resonators should be even higher since at shorter wavelengths scattering losses become more prominent [17].

 

Fig. 3 Linewidth δν and coupling contrast K as a function of the voltage applied to the piezo actuator being proportional to the distance between the coupling waveguide and the WGR. The piezo actuator expands approx. 100 nm/V. Inset: Transmission at the output of the coupling waveguide as a function of the detuning of the pump-laser frequency showing a Lorentzian shape.

Download Full Size | PDF

3. Cascaded optical nonlinearities

Since we aimed for type-I phase matching we used ordinary polarization for the pump light. By keeping the temperature of the WGR constant and scanning pump wavelength from 970 to 980 nm, we recorded the WGMs in the transmission spectrum and observed the light scattered from the WGR surface with a CCD-camera. We recognize blue light emission in the WGR for most of the WGM, resulting from SHG. We measured the power of the SHG light at the end of a coupling waveguide with a detector to optimize the phase matching, by changing the temperature in the Kelvin level. Furthermore we optimized the coupling distance in terms of SHG output power. By tuning the power of the pump laser, we achieve an internal power in the coupling waveguide in the range of 5 to 24 mW. Some of the WGMs show also green light for higher power. In order to investigate this in more detail, we selected such a WGM with a resonance wavelength at 975.79 nm and inspected the scattered light using a grating spectrometer. Figure 4 shows the spectra and corresponding images of the WGR at 4, 7, and 13 mW pump power. At 4 mW pump power, the SHG of the pump wave from 976 to 488 nm can be seen. If we increase the pump power to 7 mW, a peak at 1040 nm wavelength appears, which corresponds to a wave number shift of 632 cm⁻¹ with respect to the pump light.

 

Fig. 4 a) Spectra of the scattered light from the WGR at 3, 7 and 13 mW pump power. Blue: Data measured with Bluewave spectrometer (measurement range 300 to 1100 nm). Grey: Data recorded by NIRQuest spectrometer (measurement range 900 to 2500 nm). Green inset: Spectrum measured at the coupling waveguide output. b) Microscope images of the WGR taken at 3, 7 and 13 mW pump power.

Download Full Size | PDF

This can be attributed to the well-known frequency of an ionic vibration of the lithium niobate crystal lattice [37], so that stimulated Raman scattering (SRS) is responsible for this wavelength shift. Furthermore, we observe green emission at 520 nm, which occurs due to SHG of the Raman peak. The camera picture shows that this emission is present only in the backward path of the coupling waveguide, in perfect accordance with previous work on bulk WGRs, where it was shown that SRS generates a WGM with a reverse propagation direction. The reason for this is the phase-matching condition for this process. Energy conservation with ω_p − ω_s = Ω is fulfilled with ω_p, ω_s and Ω being the angular frequencies of the pump light, Stokes light and the phonon. Furthermore, momentum conservation has to be fulfilled with k_p − k_s = K with k_p, k_s and K being the wave vectors of the pump light, Raman light and the phonon. Only for large values of K = k_p − k_s, Ω is close to the phonon resonance [38]. Furthermore, |k_p| ≈ |k_s|. Thus, the Stokes wave has to propagate in the opposite direction with respect to the pump, e.g. k_p ≈ −k_s. Increasing the pump power to 13 mW, we observe cascaded SRS from 1040 to 1114 nm and frequency-doubled emission of this Raman line at 557 nm. In addition, we detect emission at 520 nm in the forward direction, which is generated by sum frequency generation of the pump and the Raman peak at 1114 nm.

Furthermore, we recorded the power of the light converted into the visible range at the coupling waveguide output to calculate the conversion efficiency. The maximum measured conversion efficiency at 19 mW pump power is in the range of 1.9 × 10⁻⁵ which is rather low. The reason for this is that we use only one coupling waveguide for both the pump light and the frequency-doubled light, and thus we cannot couple out the frequency doubled light efficiently to the detector at the waveguide output, a well known problem [17]. However, this issue can be addressed by using a second coupling waveguide such that the coupling efficiencies for the pump light and the frequency doubled light can be optimized individually.

4. Conclusion

In conclusion, we have shown the high potential of on-chip LN WGRs for frequency conversion from the infrared to the visible wavelength range by cascaded nonlinear-optical processes. This advance was enabled by two innovations: Firstly, we were able to fabricate record-high-Q microring WGRs thanks to a special polishing post process, while emphasizing that only highly-scalable fabrication processes were used opening the perspective for low-cost mass-production of these structures. Secondly, as compared to previous work, we demonstrated a stable but tunable coupling scheme to optimise the coupling for the nonlinear optical processes. Without this additional degree of freedom, demonstration of cascaded nonlinear optical process would have been difficult to achieve.