1. Introduction

Optical microring resonators have attracted a growing interest as they can be used for ultrafast optical modulation, for add-drop filtering and for resonant enhancement of the circulating light [1]. Moreover, they are compact and can be embedded in high-density optical integrated circuits. Optical modulators, add-drop filters in wavelength division multiplexed systems and sensor elements have been demonstrated recently [2, 3]. Electrical control of microring’s resonant wavelength gives increased functionality in many of these applications. Different tuning methods have been demonstrated employing thermo-optic, electro-optic, and carrier injection effect. Among them, electro-optic tuning [2, 4, 5] allows direct electrical control and ultrafast modulation. Lithium niobate (LiNbO₃) is a material of choice as it is readily available in large high-quality single-crystal wafers and possesses good electro-optical and nonlinear properties [6] and a large transparency range (0.39–5 µm). Furthermore, the material is already used in a series of photonic applications.

Low-loss planar and channel waveguides in LiNbO₃ can be fabricated using titanium-indiffusion and proton exchange techniques [7]. However, the resulting refractive index contrast between the waveguide and the substrate (typically |Δn_o,e|<0.1) is insufficient to suppress the bending losses in strongly curved waveguides. Due to this restriction, the radii of low-loss microring resonators fabricated solely with these techniques cannot be smaller than about 1 millimeter [8, 9]. A hybrid technique, combining titanium-indiffusion with ridge structures in order to enhance horizontal optical confinement, has just recently moved the low-radius limit to 100 µm [5, 10]. On the other hand, ion implantation [11] has recently regained new attention. A high dose He⁺ implantation can be used for the slicing of single-crystal LiNbO₃ thin films from bulk crystals [12]. Ion slicing combined with wafer bonding, that yields LiNbO₃ thin films embedded in a low-index cladding, has been significantly improved in our laboratory and electrically tunable microrings have also been demonstrated [4]. Depending on the choice of the substrate material the waveguides produced with this technique can have very high refractive index contrast with respect to the substrate (Δn_o of up to 0.65). This allows much smaller microrings with radii down to ~10 µm showing no significant bending losses. Yet another more recently studied technique for the production of planar waveguides in LiNbO₃ is the implantation with medium-mass ions (O, F, Mg) [13]. Fluorine implantation with ion energies of about 20 MeV yielded waveguides with a step-like index profile and largely preserved nonlinear optical properties [14]. The obtained wide optical barrier with ample index contrast provided a strong vertical light confinement as required for bent waveguides. The waveguides were several-µm thick and thus multimode at the telecom wavelength.

In this paper we present the use of fluorine implantation for the production of effectively singe-mode planar waveguides in LiNbO₃. Furthermore, we demonstrate the first low-loss microring resonators and optical channel waveguides in fluorine-implanted LiNbO₃, fabricated by lithographic masking and dry etching techniques.

2. Waveguide fabrication

In order to fabricate channel optical waveguides and microring resonators in LiNbO₃, we first characterised planar optical waveguides obtained by implantation with fluorine ions. We studied the refractive index, width and depth of the optical barrier as a function of the implantation energy and fluence. Based on these data we then determined the waveguide dimensions and implantation parameters for low-loss, effectively single-mode waveguides in LiNbO₃. Finally we structured the channel waveguides and microrings using the lithography masking and dry etching technique. In the following we describe waveguide design considerations and the fabrication steps.

2.1 Design considerations

Implantation of fluorine ions into LiNbO₃ causes material amorphisation beneath the crystal surface upon reaching the threshold fluence of approximately 10¹⁴ ions/cm² [14]. The resulting buried amorphous layer has lower refractive index and thus constitutes an optical barrier for the above-lying wave-guiding layer. Such waveguides are determined by three main parameters: the refractive index contrast between the wave-guiding layer and the optical barrier, the barrier position, and the barrier thickness. We observed that both ordinary n_o and extraordinary n_e refractive indices remain largely unperturbed in the wave-guiding region and fall sharply to a common value of 2.1±0.01 at λ=633 nm in the amorphised barrier, which is in good agreement with previous studies [14]. Assuming the same relative drop of the refractive indices at the infrared wavelengths, we estimated n=2.03±0.01 in the amorphous layer at λ=1.5 µm, yielding the index contrast of Δn_o=0.17 and Δn_e=0.11. The amorphisation spreads outwards in both directions with increasing ion fluence ϕ, giving a barrier thickness d_B of several micrometers (see Fig. 1(a)). The origin of amorphisation coincides with the position of maximal electronic stopping power, d_el, as simulated with SRIM-2003 software (Stopping and Range of Ions in Matter [15]). The simulated d_el as a function of the fluorine ion energy E is shown in Fig. 1(b). The dependence of d_el on the ion energy E can be approximated by a second-degree polynomial d_el=a ₀+a ₁·E+a ₂· E², with a ₀=-1.08 µm, a ₁=0.165 µm/MeV and a ₂=4.3·10^-3 µm/MeV². For the purposes of waveguide design, the waveguide thickness for given implantation parameters E and ϕ can thus be estimated with the help of d_B and d_el, as shown in Fig. 1(b).

 

Fig. 1. (a). Depth position of the high-energy (closer to surface, open circles) and low-energy (closed circles) crystalline-amorphous boundaries as a function of implantation fluence. Measurements were performed on samples irradiated with 22 MeV F⁴⁺ ions (d_el=4.7 µm): our measurements (open circles), data from [13] (full circles). The barrier width d_B for fluence ϕ=2.5·10¹⁴ ions/cm² is shown. The lines connecting the points are guides to the eye. (b). Simulated depth position of the maximal electronic stopping power d_el in F-implanted LiNbO₃ as a function of ion energy. The line is the polynomial approximation discussed in the text. d_B and the estimated waveguide thickness for ϕ=2.5·10¹⁴ ions/cm² and E=14.5 MeV are shown.

Download Full Size | PDF

Propagation of optical modes in planar and channel waveguides on top of implanted Z-cut LiNbO₃ was simulated with commercial software package OlympIOs [16]. We calculated the bending and tunnelling losses (combined together), electric field distribution and effective indices of TE modes at λ=1.55 µm. Absorption and scattering losses were not considered at this point (although they represent the majority of the overall losses, see Section 3), as they could not be modelled adequately. With given refractive indices in the wave-guiding and barrier layer, the calculated maximum planar waveguide thickness for single-mode propagation is 1.05±0.05 µm. However, waveguides with a thickness of up to 1.4 µm can still be used as single-mode for preliminary experiments since the second optical mode suffers roughly an order of magnitude higher tunnelling losses. For the case of channel waveguides, Fig. 2 shows the calculated bending and tunnelling losses for the fundamental TE mode in a curved waveguide with a cross-section of 1.35×3.2 µm² (height×average width, as used in our experiments) as a function of the centre ring radius R. The bending losses decrease exponentially with increasing radius and are smaller than 2 dB/cm for waveguides with R>40 µm. Along with the increasing losses the effective mode index decreases with respect to the one in straight waveguides. This hampers the evanescent optical coupling between the bus waveguide and the microring resonator. In our experiments we fabricated microrings with R=80 µm in order to preserve low bending losses (~0.2 dB/cm) and moderate optical coupling.

 

Fig. 2. Calculated combined bending and tunnelling losses (left scale) and effective mode index N_eff at ring’s outer rim (right scale) as a function of the waveguide bend radius. N_eff of the straight waveguide with the same cross-section is 2.159. Calculation is for the first TE optical mode and wavelength 1.55 µm. Waveguide cross-section dimensions as in Fig. 3.

Download Full Size | PDF

A simulated electric field profile of the first TE optical mode for this case is shown in Fig. 3. A drift of the mode towards the outer rim is clearly visible, however, the spread of the electric field into the implanted optical barrier is much smaller than the barrier thickness.

2.2 Planar waveguides

For the requested single-mode waveguide thickness of 1-1.4 µm, the implantation parameters E and ϕ were chosen according to guidelines from Fig. 1. Implantations with E≤14 MeV resulted in a brittle crystal surface, which prevented the subsequent fabrication of channel (ridge) optical waveguides. The implantation fluence was deliberately kept as small as possible in order to introduce minimal damage to the wave-guiding layer. On the other hand, tunnelling losses at λ=1.5 µm limited the minimal acceptable fluence to ϕ=2·10¹⁴ ions/cm².

 

Fig. 3. Simulation of light propagation in the waveguide of interest. (a) Vertical index profile of the waveguide structure (at X=0). (b) Simulated electric field profile of the first TE optical mode at λ=1.55 µm in the waveguide with 80 µm bend radius. Optical barrier width is 2 µm. Waveguide cross-section is trapezoidal with waveguide height 1.35 um, ridge height 1.2 µm, base width 3.7 µm and top width 2.7 µm.

Download Full Size | PDF

Finally, implantations with E=14.5 MeV and ϕ=2.5·10¹⁴ ions/cm² yielded waveguides with the lowest measured propagation losses, while retaining their thickness within requested constraints. The thickness of these planar waveguides, as measured with dark mode spectroscopy and assuming a step-like index profile, was 1.35±0.06 µm. This is approximately 0.2 µm more than predicted above (see Fig. 1(b)) because at small waveguide thicknesses the assumption of a step-like index profile gets less accurate. Although these waveguides support both TE₀ and TE₁ optical modes, they are effectively single-mode, since the second optical mode suffers roughly an order of magnitude higher tunnelling losses.

Implantations of fluorine ions F³⁺ into Z-cut LiNbO₃ wafers have been performed at room temperature. The samples were tilted at 8° relative to normal incidence to avoid channelling effects. The ion current density was kept at ~200 nA/cm² (which corresponds to a particle current of 4.2·10¹¹ ions/(s·cm²)), to minimise unwanted charging and heating. Nevertheless, the implantation times for the required fluences were only about 10 min. In addition to relatively thick optical barriers produced by the F³⁺-implantation, the short implantation time is of great advantage compared to implantations with light ions (He⁺) requiring one to two orders of magnitude higher implantation fluences and hence longer implantation times.

After the implantation, the waveguides exhibit high propagation losses, mainly due to irradiation-induced crystal defects. Thermal annealing efficiently improves the waveguide transmission [14]. The implanted samples were therefore annealed by heating them to T=300°C in air for one hour (dwell time) and then slowly cooled down. This greatly reduced the propagation losses, from several tens dB/cm prior to annealing down to ~8 dB/cm afterwards, as measured at λ=1.5 µm in our channel waveguides (see Section 3).

2.3 Channel waveguides and microrings

Channel (ridge-type) waveguides were fabricated from the annealed planar waveguides by means of photolithographic patterning and subsequent Ar⁺ sputtering. In the past, we have already successfully used this approach in Cr:LiSrAlF₆ [17], β-BaB₂O₄ [18] and other optical crystals. Photolithography was performed with an image-reversal photoresist (Clariant AZ5214). A laser beam lithography apparatus, home built in our laboratory, was used for the photoresist exposure. For the illumination we used a violet laser beam (λ=430 nm) that was focused directly onto the photoresist film with a microscope objective (focal length f=4 mm, numerical aperture NA=0.75). The desired patterns were written in the photoresist by lateral translation of the sample (for straight waveguides) or 2D acousto-optical deflection of the laser beam (for optical microrings). The main advantages of the laser lithography are its better resolution and flexibility as compared to the standard photolithographic technique utilizing UV lamps and mask aligners. If needed, different patterns can be written on each new sample, thus permitting an iterative approach towards optimised waveguide structures. Additionally, we are able to expose the photoresist with high lateral precision also in small samples, where spun edge bead prevents mask contacting.

The lateral evanescent wave coupling between a bus waveguide and a microring resonator requires a sub-micrometer wide gap. In order to accomplish this, we used a two-step laser lithography procedure. In the first step, straight bus waveguides were written and developed in the photoresist. In the second step, a new photoresist layer was spun on the sample surface. Microrings were then written in close proximity of the already-existing bus waveguides and subsequently developed. In this way, exposures (arising from Gaussian wings of the laser beam profile) from the ring and the bus waveguide were not accumulated in the same photoresist layer in the gap region. As a result, photoresist was not cross-linked in the gap. By careful positioning of the rings to the waveguides, coupling gaps as narrow as 200 nm could be obtained.

 

Fig. 4. Scanning electron micrograph of a microring resonator and a bus waveguide, structured in LiNbO₃. (a). The whole ring and the bus waveguide. Ring radius is 80 µm, ridge height is 1.2 µm. (b). Enlarged coupling region, the gap size is ~0.2 µm.

Download Full Size | PDF

The photoresist patterns were transferred to the planar waveguide surface by Ar⁺ sputtering (Oxford Plasmalab 80, RF power 200 W) and the ridge-type waveguides with a height of up to 1.2 µm were finally fabricated (Fig. 4). LiNbO₃ is highly resistant to Ar⁺ sputtering. The measured sputtering rates were ~1.3 nm/min for both virgin and implanted crystals. Waveguide facets were finally polished using standard polishing techniques. The dimensions and surface quality of the structures were then investigated with a profilometer and a scanning electron microscope (SEM). The bus waveguide cross-section is trapezoidal with a sidewall slope angle of about 65° and slightly etched upper edges (Fig. 5). The ring has steeper and smoother sidewalls, as can be qualitatively compared in Fig. 4.

 

Fig. 5. SEM image of the fabricated ridge structure. The waveguide cross-section is trapezoidal with the base width 3.7 µm, top width 2.7 µm and ridge height 1.2 µm. The amorphised barrier layer beneath the ridge is visible.

Download Full Size | PDF

3. Optical characterisation of microrings

In order to optimise the performance of channel waveguides and microring resonator devices, their propagation losses at operational wavelengths should be as low as possible. An estimation of the propagation losses in straight channel waveguides (without adjacent microrings) was performed by measuring their light transmission around λ=1.5 µm. A pigtailed tuneable diode laser (Santec TSL-220, spectral range 1.53–1.61 µm, spectral width 1 MHz) was used as the input beam source. The laser beam was end-fire coupled into the waveguides by a microscope objective with a 50× magnification and NA=0.8. The transmitted light was collected with a 100× microscope objective and projected onto an InGaAs photodiode. On average, a total of 9% of the TE polarised and 2.1% of the TM polarised incident light was emitted from the exit facet of the 5.6-mm long waveguide. We determined the propagation losses of TE polarised light by measuring its wavelength-dependent transmission in the channel waveguides with polished facets that acted as low-finesse Fabry-Perot resonators. As described in [19], the upper limit of the propagation losses can be determined solely from the contrast K between the maximum I _max and the minimum I _min of the measured transmitted power oscillations. Here K is defined by K=(I _max-I _min)/(I _max+I _min) and is independent of the coupling efficiency. With the measured contrast K=0.10 and assuming a facet Fresnel reflectivity of 14% we estimated the propagation losses to be 7.8±0.5 dB/cm at λ=1.5 µm. Comparing this value with the directly measured total transmission suggests that about one third of the laser power was effectively coupled into the first guided optical mode. According to our calculations (see Section 2), the tunnelling losses for the first guided mode are only 0.1 dB/cm. Obviously, absorption and scattering are the main sources of the total propagation loss.

The optical response at the through ports of the coupled ring resonators around λ=1.5 µm was probed by scanning the input wavelength over a range from 1.54 µm to 1.58 µm. Resonances were not observed with TM polarised light, presumably due to smaller index contrast and higher propagation losses. With TE polarised light the characteristic responses of microresonators could be observed. The highest modulation depth was obtained with a microring slightly touching the bus waveguide (the length of the joined region was 6±1 µm). Fig. 6 shows the measured transmission spectrum around λ=1.56 µm. The measured free spectral range of the microresonator was Δλ_FSR=2.0±0.01 nm. The spectral width of the resonances (full-width at half-minimum of the dip) was δλ_FWHM=0.5±0.05 nm and the modulation depth at the strongest resonances was 14 dB. The resonator finesse was F=Δλ_FSR/δλ_FWHM=4±0.4 and the corresponding quality factor Q=λ/δλ_FWHM=3100±300. The measured free spectral range agrees well with the calculated

 

Fig. 6. Measured TE wave transmission spectrum of the bus waveguide coupled to a microring resonator. Small oscillations seen at the regions outside of resonances stem from Fabry-Perot resonances of the 5.6 mm long waveguide.

Download Full Size | PDF

The calculated group effective index N_g=N_eff -λ(∂N_eff/∂λ)=2.3 in Eq. (1) includes both the material dispersion and the modal dispersion of the effective mode index N_eff. The modal dispersion becomes particularly relevant in single-mode waveguides. A detailed description of the calculation of the resonator’s free spectral range can be found in [20].

4. Discussion

In Table 1 we review some of the important properties of already demonstrated microring resonators in LiNbO₃ and compare them with our results. Our estimation of the maximum achievable free spectral range for a given fabrication technique (allowing 2 dB/cm bending losses) is included. As it can be seen, fluorine implantation enables fabrication of microrings with a Δλ_FSR of 4 nm or even greater if bending losses higher than 2 dB/cm are acceptable.

Table 1. Properties of demonstrated microring resonators in LiNbO₃

View Table

For our envisaged application of these microring resonators as electro-optical light intensity modulators, we will implement tuning electrodes. Below we estimate the microring resonance wavelength tunability δλ/δV for the optimized electrode configuration. The shift of the resonant wavelength can be derived by differentiating the resonance condition (2πN_eff(λ)L)λ=2πm for a fixed integer value m:

The resonance shift depends on the change of the effective index and is independent of the resonator’s perimeter L=2πR. Because of the crystal’s natural birefringence and the ring’s axial symmetry, the δN_eff for TE waves depends mostly on the vertical component of the applied electric field E_z and can be modelled to a good approximation by considering only electro-optical coefficient r₁₃. We simulated the electro-optically induced change of the effective index δN_eff=(∂N_eff/∂V)δV with the OlympIOs software. The electric potential in the waveguide cross-section upon applying a voltage between the electrodes is shown in Fig. 7. One of the electrodes must be on top of the ring waveguide, to provide the necessary E_z component of the electric field. A 0.7-µm thick buffer layer of amorphous silicon dioxide (SiO₂) is used to prevent the absorption losses due to chromium electrode. The calculated effective index change is δN_eff=(5±2)·10^-4 for a 100 V potential between the electrodes. The estimated tunability is therefore δλ/δV=3.4±1.4 pm/V or, expressed in frequency, δν/δV=0.42±0.17 GHz/V. The inaccuracy is due to the uncertain value of the dielectric constant of the amorphous LiNbO₃ in the optical barrier (values between the ones from SiO₂ and bulk LiNbO₃ were used).

 

Fig. 7. Simulated electric potential in the waveguide cross-section upon the application of a voltage V=100 V between the electrodes. Equipotential contours separate a 5 V potential drop. About 55% of the potential drop occurs in the SiO₂ buffer layer.

Download Full Size | PDF

It is interesting to compare the electro-optical response of the presented microring resonator with an equivalent Mach-Zehnder (MZ) modulator. We compared the microring’s modulation sensitivity, expressed with an equivalent half-wave voltage V^eq_π, with the sensitivity of a MZ modulator [21]:

where V ⁰ _π is the voltage that produces a π phase shift in one round trip in the resonator and is also the half-wave voltage of a one-arm driven MZ modulator with an interaction length equal to the microring’s perimeter (L _MZ=2πR). The enhancement of the modulation intensity by the optical resonance in the microring is given by the factor of 2|dT/dθ|_max. The measured light transmission as a function of the phase shift θ after one round trip is shown in Fig. 8. The maximum slope on the left side of the resonance dip is |dT/dθ|_max=1.7±0.2. This gives the equivalent half-wave voltage V^eq_π=0.29(1±0.04) V ⁰ _π. The asymmetry in the transmission function can be attributed to the presence of higher spatial (horizontal) TE modes in the microring. These, due to phase mismatch, resonate at slightly different wavelengths and hence distort the transmission function.

 

Fig. 8. (red) Measured light transmission as a function of the roundtrip phase θ. The line connecting the points is a guide to the eye. (blue) Transmission curve (T=cos²(φ/2)) of an equivalent Mach-Zehnder modulator, for phase φ=θ - π. Points of maximum transmission slopes |dT/dθ|_max are marked with arrows.

Download Full Size | PDF

5. Conclusions

We have demonstrated channel waveguides and microring resonators in fluorine-implanted lithium niobate. A low-fluence fluorine implantation (2.5·10¹⁴ ions/cm²) into Z-cut LiNbO₃ generates a 2-µm wide amorphous optical barrier beneath the crystal surface with a refractive index contrast of Δn_o=0.17 at λ=1.5 µm. This, combined with lithographic patterning and Ar⁺ sputtering, yields channel (ridge-type) optical waveguides in LiNbO₃ with strong optical confinement. The fabricated channel waveguides have propagation losses lower than 8 dB/cm for TE waves.

Microring resonators with 80-µm radius that can operate as wavelength filters at 1.5 µm telecom wavelength have been realised. They have a transmission extinction ratio of 14 dB, a free spectral range Δλ_FSR=2.0 nm, a finesse F=4, a quality factor Q=3100, and a phase sensitivity of |dT/dθ|_max=1.7. An estimation of the electro-optic response in these LiNbO₃ microresonators shows that their resonance wavelength could be tuned by δλ/δV=3.4 pm/V. We believe that the microresonators’ Q factor can be further increased by optimisation of the surface patterning process.