1. Introduction

Potassium titanyl phosphate (KTP, KTiOPO₄) is a nonlinear optical crystal that possesses high nonlinear-optical and electro-optical coefficients, wide transparency range, thermal stability and high optical damage threshold [1]. It has therefore found widespread application in second harmonic generation, optical parametric oscillation and Q-switching. By periodic poling of its ferroelectric domains with a fine period, phase-matching can be achieved to the deep-blue [2]. KTP is also particularly attractive for generation of entangled photons in the optical telecommunication bands, as is can fulfill the extended phase-matching condition [3].

Using waveguide geometries in KTP allows for increased efficiency in frequency conversion processes and enables integration of optical components. The most common way to fabricate waveguides in KTP substrates is Rb-ion exchange. To do this a metal mask on the surface of a z-cut substrate is lithographically structured, followed by immersion of the substrate into a bath of Rb nitrate at temperatures of 300-400 °C for 30-240 min [1]. At these temperatures Rb ions diffuse in and increase the material’s refractive index, thus forming optical waveguides. A drawback of this method is that crystal inhomogeneity and spatially varying ionic conductivity can result in a poorly defined refractive index profile along the z-axis [4,5]. Also Rb-diffused waveguides can further diffuse and degrade at slightly elevated temperature, especially when they contain fractions of Ba [6]. This interdicts operation of such waveguides in KTP at higher temperature, e.g. to prevent gray tracking [7]. To overcome the problem of the mobile alkali and alkaline earth metals, liquid phase epitaxy of a KTP isomorph waveguiding thin-film with increased refractive index on top of another KTP-type substrate was introduced [4,8]. Furthermore, fs-laser writing or ion implantation were used to prepare thermally stable waveguides [9,10].

In this work we propose a new fabrication scheme that is based on ridge definition by diamond-blade dicing and ion exchange. The method makes use of the anisotropic diffusion behavior of Rb and other ions in KTP, as the diffusion constant along the z-axis is several orders of magnitude greater than that in the xy plane [1]. Contrary to the technique developed by Bierlein et al. [1], waveguides are fabricated in substrates that have their z-axis lying in the surface plane, with at least a component of the z-axis being perpendicular to the waveguide direction. In this way the fabricated side-exchanged waveguides have, after additional annealing treatment, a homogenously increased refractive index along the waveguide cross section, and hence a step-like index profile along the depth direction [4]. The new method proposed here thus leads to symmetric refractive index profiles that allow for an improved spatial overlap of interacting waves of different wavelength. Furthermore, the waveguides are less affected by elevated temperatures. Ridges fabricated by this method show very low optical losses and an excellent nonlinear performance demonstrated by type-II SHG of 1064 nm fundamental wavelength.

2. Waveguide fabrication

The ridges in this work were defined using a diamond-blade dicing saw, which is a method well known to enable optically smooth surfaces, high ridges with large aspect ratio, and near 90° flanks [9,11]. Precise dicing is a rapid fabrication method, as there is no need for e.g. photolithography requiring clean room environment or thin film deposition. Also ridges of different depth can be easily cut into a single sample. A drawback is that this method does not allow for the fabrication of curved waveguides. However, this could be overcome by fs-laser ablation [12] or reactive ion etching [13], which are methods that have been demonstrated in the KTP isomorph RTP.

An overview of the fabrication scheme is depicted in Figs. 1(a) - (d). First, grooves are cut into a commercial flux-grown KTP substrate that has its z-axis lying in the surface plane. The substrate is cleaned with acetone and isopropanol and then immersed into a nitrate salt melt to diffuse Rb and Ba through the cut flanks along the z-axis into the crystal. Through diffusion wide waveguides are created. After removal of residual salt in deionized water, narrow ridgewaveguides are formed in a second cutting step. For both cuts a 100 µm thick blade (P1 A863 SD6000 N100 BR50) and a feed rate of 0.5 µm per revolution are used on a DAD322 dicing saw of Disco Corp. The following annealing step levels the inhomogeneous ion concentration along the z-axis, leading to ridge waveguides with step-like index profile. By using ion-exchange from one side only, possible risk of ridge fracturing in the salt melt, presumably because of expansion of the solidifying salt mixture during cool-down, is minimized. Such fracturing was frequently observed when initially performing a two-side ion-exchange on already defined ridge waveguides of ~10 × 10 µm² cross section.

 

Fig. 1 Sample fabrication: a) grooves are cut with a wafer saw into KTP substrates with the z-axis lying in the surface plane; b) in a salt melt Rb/Ba ions diffuse through the flank along the z-axis forming a waveguide; c) definition of the ridge waveguide by a second cut after cool-down and sample cleaning; d) homogenization of the diffused ions by annealing treatment in air; e) confocal microscopy image of a completed ridge waveguide with dimensions of 10x10 μm².

Download Full Size | PDF

To achieve good overlap with a single-mode optical fiber the exchanged cross section needs to measure ~10 × 10 µm². By simulations with Lumerical Mode we found that the homogenous index increase over the bulk KTP index across the waveguide needs to be in the range of (1.3-3.0) ·10⁻³ to make the waveguide single-mode for both polarizations at 1064 nm. The height of the exchanged area is defined by the cutting depth of the groove through which the waveguide is exchanged, and the width is defined by the distance between the first and second cut.

To obtain the desired index increase the diffusion process and the annealing step have to be considered. During diffusion Rb⁺ ions replace K⁺ ions and move along the crystal’s z-axis via a hopping mechanism. When divalent ions, such as Ba²⁺, are diffused into KTP, they replace two K⁺ ions to fulfill charge neutrality. Thereby they introduce additional K vacancies, thus accelerating the hopping mechanism of the Rb and K ions and making the ion exchange more homogenous. For determining appropriate salt compositions, we made use of findings from [14]. They described the relation between the Rb concentration of a salt melt of rubidium, barium and potassium nitrate and the index increase at the surface of immersed KTP. We used a salt concentration of 50 mol% RbNO₃, 3 mol% Ba(NO₃)₂, 47 mol% KNO₃ that supposedly leads to an index increase at the substrate surface of 2.8·10⁻³ (3.6·10⁻³) for TM (TE) polarization. Figure 2 shows the calculated diffusion profile after successive diffusion treatment for intervals of 4 h at 330 °C, and the profile after 16 h diffusion averaged over the first 10 µm to account for the annealing step of a prepared ridge waveguide. As a result, the homogenous index increase is given by 74% of that at surface. After 16 h diffusion, ridge definition and annealing the fabricated ridge waveguides are therefore supposed to have a homogenous index increase of 2.1·10⁻³ (2.7·10⁻³) for TM (TE) polarization and be single-mode at 1064 nm. In the experiments samples were annealed at 370 °C for 16-20 h.

 

Fig. 2 Calculated normalized refractive-index profiles for different times of the ion-exchange process. The green line indicates the profile after 16 h exchange, averaged over the first 10 µm to account for ridge definition and annealing.

Download Full Size | PDF

The relatively long diffusion time was chosen to obtain a large exchange depth and hence better homogeneity as influence of crystal inhomogeneity and fabrication inaccuracies is reduced. For example, in this way the dependence of the index increase of a ridge waveguide becomes less dependent on the exact position of the second cut. Also influence of salt drops, which remain on the sample after pulling it out of the melt and lead to further diffusion during slow cool-down, is minimized. For the calculations a diffusion coefficient of D = 6.3 µm²/h was used. The value has been determined from diffusion of a stronger composition (80 mol% RbNO₃, 3 mol% Ba(NO₃)₂, 17 mol% KNO₃) and measurement of the index profile by M-line spectroscopy (Metricon 2010/M). The original composition was too weak to enable proper profile determination.

For type-II second harmonic generation (SHG) of 1064 nm wavelength in bulk KTP at room temperature, a crystal orientation of (ZX w) ~23.5° has to be used [1]. Calculations show that the preparation of the presented ridge waveguides in this widespread crystal orientation would lead to a phase-matching wavelength at ~1077 nm. To compensate this shift resulting from waveguide dispersion, the waveguides in this work were instead fabricated in substrates of type (ZX w) 37.0°, i.e. ridges form an angle of 37° to the x-axis.

Following the described fabrication scheme we were able to reproducibly fabricate ridge waveguides of 1 cm length; the smallest ridge width achieved without cracking was 1.5 μm. The facets for coupling light into and out of the ridges were prepared by two more dicing cuts perpendicular to the ridge direction. A confocal microscopy image of a completed ridge waveguide is shown in Fig. 1(e). The sidewalls are near vertical (86°) and the surfaces are smooth. By use of a white light interferometer we determined a roughness of ~5 nm (RMS) of the ridges’ sidewalls.

3. Characterization

All fabricated samples were characterized with a polarization maintaining (PM) fiber-coupled tunable laser amplifier system (Sacher Serval Plus) in a wavelength range of (1036-1070) nm, which allows for fiber-coupled power of up to 800 mW. The PM fiber was adjusted to the waveguide ridges at a polarization angle of 45° for type-II SHG and the light exiting the waveguides was imaged through a 20 × microscope objective and split via a dichroic long-pass mirror (Thorlabs DMLP900) onto two detectors for the fundamental (FH) and the second harmonic (SH) wave. To filter out residual FH power from the SH beam a KG3 filter was used. Conversion efficiencies were calculated from the detected powers behind the sample, and corrected for transmission/reflection at the objective, beam splitter and filter.

For ridge waveguides of 1 cm length we determined total insertion losses as low as 1.0 dB (1.1 dB) for TM (TE) polarization at 1060 nm wavelength. Taking into account reflection losses of two times 0.3 dB (0.4 dB) and neglecting losses due to imperfect overlap, we can estimate upper bounds for propagation losses of 0.4 dB/cm (0.3 dB/cm).

As described before, the aim of the annealing step is to level the inhomogeneous Rb/Ba concentration and hence the index increase along the waveguide cross section. To investigate the effect of annealing, the SHG performance of waveguides was measured before and after treatment at 370 °C for 20 h. We found that the SH spectrum had shifted towards shorter wavelengths, which is understandable since the waveguide becomes less dispersive as the SH mode cannot concentrate anymore in the region of elevated index increase (see Fig. 3a). The intensity redistribution of the TM₀₀ SH mode that is initially leaning towards the side, through which the waveguide was exchanged, is well visible in the inset of Fig. 3(a). With the SH mode becoming more symmetrical, the overlap between the FH and SH mode also increases leading to enhancement of the conversion efficiency.

 

Fig. 3 a) Conversion efficiency of a ridge waveguide before and after annealing for 20 h at 370 °C. The two insets show how the initially asymmetric SH mode (top) monitored from the output facet by a CCD camera becomes symmetrical after annealing (bottom). b) Measured conversion efficiency of three ridge waveguides of different width (height: 8 µm, length: 9.06 mm) and a calculated curve. The inset shows the experimental phase matching wavelength along with the simulated curve.

Download Full Size | PDF

To oppose the side-exchanged waveguides to conventional top-diffused ones, we prepared a planar waveguides by immersion of a z-cut KTP substrate in a melt of 95 mol% RbNO₃ and 5 mol% BaNO₃ for 1 h at 300 °C. The substrate was then repeatedly annealed at 225 °C for 10 h and the change for the mode indices was monitored by M-line spectroscopy at 532 nm. The results of the experimental series in Fig. 4 clearly indicate a significant change of the mode indices at a moderate temperature of 225 °C. The change of mode index for m = 0 at 532 nm by 6.7·10⁻³ after 60 h annealing translates into a change of the FH phase-matching wavelength of 87 nm, when assumed for simplicity that the mode index at the FH wavelength remains unchanged.

 

Fig. 4 Annealing of a planar waveguide in z-cut KTP. Mode indices were measured with a Metricon 2010/M prism coupler at 532 nm wavelength for TE polarization.

Download Full Size | PDF

In simulations we have found that the phase-matching wavelength depends on the height of the waveguide (i.e. depth of the groove used for the ion-exchange), the ridge width and the index increase. The narrower the ridge becomes, the further the FH modes is pushed into the substrate and the waveguide dispersion is increased (while the SH mode is less affected). Therefore, narrower waveguides result in an increased phase-matching wavelength. Experimental results for three neighboring waveguides of 8 µm height each, but different width are displayed in Fig. 3(b). The phase-matching wavelength shifts by −0.52 nm/µm over the width. In the inset the experimental findings are compared to a simulation and a fairly well agreement is found. The spectral bandwidth (FWHM) of the three shown curves are 0.69/0.64/0.74 (nm cm) and are only slightly higher than the theoretical value for bulk SHG of 0.59 (nm cm). The higher experimental value can be explained by slight variation of the width / mode indices along the waveguide. The black dotted curve in the diagram was calculated for an assumed variation of the mode indices along the waveguide. Its maximum efficiency is lowered by 16%, compared to a calculated curve with constant mode index along the waveguide.

Maximum conversion efficiency was found in a ridge waveguide with a width of 12 µm and length of 9.8 mm. We determined an internal conversion efficiency of η_exp = 3.3%/(W cm²). From mode images we calculated an effective area of SHG of 84 µm². To compare this result with the theory and to conclude on possible degradation we calculated the theoretically expected conversion efficiency using

To investigate thermal stability, we annealed a sample containing ridge waveguides for 10 h at 500 °C and found that the waveguides were still guiding, without noticeable changes of the initial refractive index profile. Insertion loss and phase-matching wavelength had not significantly changed, but conversion efficiency had more than halved. While other effects (mode overlap, waveguide losses) may thus be excluded, we assume that this degradation is due to (partial) ferroelectric domain switching and/or domain wall motion. Formation of domain-inverted thin surface layers due to ion-diffusion has been observed in undoped KTP for annealing at 946 °C, i.e. close to the Curie temperature [16]. Further, in a recent publication it was demonstrated that annealing of slightly (0.3%) Rb-doped KTP crystals above 550 °C already induces effective domain wall motion [17].

4. Summary and outlook

In conclusion, we have demonstrated fabrication of ridge waveguides in KTP by use of diamond-blade dicing and single-side Rb/Ba ion exchange. Initially ions are in-diffused from one side into a diced sidewall of the waveguide; afterwards a second cut is defined and the waveguide is further annealed. In this way step-like refractive index profiles are obtained which have two main advantages. First, they show high resistance against further annealing treatment, thus allowing long-time use at elevated temperature. Second, this leads to improved mode overlap in nonlinear frequency conversion processes when compared to typical gradient-index profiles. When compared to the latter, the higher index steps in lateral direction improve light confinement, too, and thus further improve spatial overlap of interacting modes [11].

We have demonstrated the nonlinear optical performance of our fabricated ridge waveguide by using type-II second harmonic generation of the fundamental wavelength 1064 nm, and found a conversion efficiency for 532 nm light to be very close to the theoretically expected value. The ridge waveguides presented in this work were fabricated with the z-axis lying perpendicular (ϕ = 90°) to the waveguide propagation direction. However, all geometries where 0° < ϕ < 90° are technically feasible, too, which can provide additional freedom for phase-matching, e.g. for generation of entangled photon pairs in the near infrared [18].