Development of low-loss TiO 2 waveguides

: TiO 2 channel waveguides were fabricated using a DC sputter deposition process, followed by photolithography and reactive ion etching. A SiO 2 cladding was deposited using evaporation. SEM, TEM and Raman measurements indicate the presence of both an amorphous and a crystalline phase. As the layer thickness increases, poly-crystalline structures start forming. Loss measurements were performed by imaging the scattered light from the top of the channel waveguides and ﬁtting an exponential decay to the intensity proﬁle. Propagation losses of 7.8 ± 0.52 dB/cm at a wavelength of 632.8 nm and 0.68 ± 0.46 dB/cm at a wavelength of 1010 nm were experimentally characterized.


Introduction
The demand for photonic integrated circuits is rapidly increasing. TiO 2 is an emerging material for integrated photonics, which has received increasing interest in the last few years [1][2][3][4][5]. Its high refractive index (i.e., 2.2-2.3 for amorphous layers and even higher when in its crystalline phases anatase or rutile [1]) permits the design of compact integrated circuits. TiO 2 is transparent over a broad wavelength range, spanning from visible to infrared wavelengths, due to its high bandgap of 3 eV [6]. Its high refractive index makes TiO 2 a promising material for waveguide enhanced Raman spectroscopy [7], experimentally showing a 50 fold higher Raman efficiency compared to Si 3 N 4 waveguides [4].
TiO 2 can be doped with, for example, rare earth ions [8] to achieve optical gain in different wavelength ranges. The low phonon energy of TiO 2 , with a maximum at 633 cm −1 for anatase [9], limits the amount of non-radiative decay. Examples of active photonic applications include Si:TiO 2 waveguides doped with Er 3+ ions [10] and rare-earth ion doped layers produced by the sol-gel method [11,12]. TiO 2 exhibits a high non-linear refractive index, approximately 3 times larger than that of silicon nitride [8]. Several non-linear processes have been recently demonstrated in TiO 2 waveguides, such as third harmonic generation [13], supercontinuum generation [2], spectral broadening [14] and four wave mixing [5].
The thermo-optic coefficient of TiO 2 is negative and ranges between −0.5×10 −4 K −1 and −2.15×10 −4 K −1 [15][16][17]. This variance is related to the variation in the material characteristics of the layer, such as crystallinity, stress and density, as a result of varying deposition methods. By combining TiO 2 with a material with a positive thermo-optic coefficient, athermal waveguides can be designed. Athermal waveguides have been demonstrated by combining TiO 2 with Si 3 N 4 [18,19] and silicon [17]. Such athermal behavior was demonstrated over a wavelength range of up to 300 nm [20].
In order for TiO 2 to find actual applications, low propagation losses need to be obtained to get high quality devices, which still remains a challenge. The lowest losses reported to date for channel waveguides fabricated using conventional fabrication techniques (i.e., deposition, lithography and reactive ion etching) are 9.7 dB/cm at 632.8 nm [21] and 4 dB/cm at 1550 nm [22]. Lower losses were obtained using a dielectric lift-off process, which resulted in 7.5 dB/cm at 632.8 nm and 1.2 dB/cm at 1550 nm [3]. The lift-off process, however, limits the freedom in the design of photonic integrated circuits.
In this work, we developed a DC reactive sputtering process to deposit low-loss TiO 2 layers. Sputter deposition allows for fast wafer-scale deposition of thin layers providing good control over their stoichiometry and morphology. Patterning is performed using UV contact lithography followed by reactive ion etching. These techniques are widely available and allow fast processing at the wafer scale. Waveguides are cladded with evaporated SiO 2 to reduce the index contrast between core and cladding, which decreases optical propagation losses. Propagation losses as low as 7.8 ± 0.52 dB/cm at 632.8 nm and 0.68 ± 0.46 dB/cm at 1010 nm were measured in the fabricated waveguides, which are comparable to the losses obtained with the lift-off process [3], but considerably lower than any previous work utilizing similar standard fabrication techniques [5,21,22].

Fabrication
A thin TiO 2 layer was deposited using DC reactive sputter deposition from a 99.999% pure 4 inch titanium target in the TCOater in the MESA+ cleanroom. The substrate consists of a 4 inch silicon wafer with an 8 µm thick layer of thermal SiO 2 . During deposition, the target to substrate distance was kept at 176 mm, the reactor pressure at 6·10 −3 mbar and the temperature of the substrate was kept at room temperature (i.e., between 20-30°C). A DC power of 500 W was applied to the titanium target. No annealing was performed after deposition, since it is known to lead to high amounts of crystallization, increased surface roughness [23,24] and a high extinction coefficient [25].
The O 2 flow has major impact on the quality of the deposited films. As the O 2 content in the reactor increases, the bias voltage between the target and the wafer increases. The increase in the bias voltage is related to an increase in the oxidation state of the target with increased O 2 flow, passing from fully metallic to fully oxidized. Since the secondary electron yield of titanium is higher than that of TiO 2 [26], an increasingly oxidized target exhibits a reduced secondary electron current and, since the applied power is kept constant, the bias voltage increases. Figure 1(a) shows the variation of bias voltage as a function of O 2 flow into the chamber. As the O 2 flow is increased, the argon flow is adjusted to keep a total flow of 40 sccm. The data is collected for an increasing O 2 flow rate. At around 4.5 sccm of O 2 flow, a jump in the bias is observed, known as the 'knee point' [27]. During deposition both the sputtered titanium and the titanium target are getting oxidized. For low O 2 flows the removal rate at the target is sufficient to keep the target from forming a full oxidized layer. However, at the knee point a threshold is reached causing a self-enforcing loop, where the target gets oxidized, resulting in a lower deposition rate. The lower deposition rate causes less gettering of oxygen at the wafer, which in turn increases the O 2 partial pressure leading to more oxidation of the target. Layers with the highest deposition rate and the lowest propagation losses are deposited with an O 2 flow just after the knee point (i.e. at higher O 2 flows). For an O 2 flow below the knee point, the deposited layers are not fully oxidized, therefore exhibiting high optical propagation losses. This point is indicated in Fig. 1(a) with the triangle. As the O 2 flow increases passed the knee point, optically guiding layers are obtained. However, the deposition rate drastically decreases due to the increased oxidation state of the target. Layers deposited at a too high O 2 flow show higher scattering losses, due to an increased crystallization of the layer [28]. Figures 1(b)-(d) show the guiding (from right to left) of the first order TE slab mode in a TiO 2 layer, deposited at 4.5, 5 and 5.5 sccm of O 2 flow. The layers deposited at higher O 2 flows show higher propagation losses. Coupling into the slab mode is performed using a Metricon 2010M prism coupling setup with a rutile prism. In order to get a stable process, the O 2 flow is kept at 4.6 sccm. In this way, accidental shift towards the metallic regime is prevented. The layer used for the channel waveguides have estimated losses of 3 dB/cm at 632 nm, based on an estimated 15 dB attenuation over 5 cm.  After deposition, 1 µm wide waveguides were defined using UV contact photolithography using a 1.7 µm thick photoresist layer (Olin Oir 907-17). Etching is performed using an Oxford Plasma Pro 100 Cobra reactive ion etching machine. During etching, the pressure is kept at 10 mTorr and the temperature is set at 10°C. An ICP power of 1500 W and a CCP power of 20 W are used. The plasma consists of SF 6 , O 2 and Ar with flowrates of 25, 6 and 5 sccm respectively. This process results in channel waveguides as shown in Fig. 2(a). The waveguides are covered by a 1 µm thick SiO 2 cladding, which is deposited by e-beam evaporation from a 99,999% pure SiO 2 target in the BAK600 in the MESA+ cleanroom. The target to substrate distance is kept at 15 cm. Evaporation was chosen in order to keep the sample at room temperature to prevent crystallization of the TiO 2 . The settings for the evaporations were a beam current of 200 mA and a base pressure of 6·10 −6 mbar.
Simulations using Lumerical MODE solutions show four TE modes and three TM modes at a wavelength of 632 nm and two TE modes and one TM mode at a wavelength of 1010 nm. Normalized mode profiles of the fundamental TE modes are shown in Figs. 2(b) and (c) for the wavelengths of 632 nm and 1010 nm respectively. These modes show confinement of the light in the waveguide core of 76.7% and 51.5% respectively.

Characterization
AFM measurements were performed on deposited TiO 2 layers of different thicknesses with a Bruker dimension fastscan atomic force microscope (AFM). They are shown in Figs. 3(a) and (b). The surface roughness of these layers is determined by calculating the root mean square of the height profile of these images. The rms roughness values are shown in Fig. 3(c) as a function of layer thickness. The surface roughness increases linearly with increasing layer thickness. The increasing grain size indicates increasing amount of crystallization [29,30]. As a result, thick waveguides exhibit higher scattering losses. mTorr and the temperature is set at 10°C. An ICP power of 1500 W and a CCP power of 20 W are used. The plasma consists of SF 6 , O 2 and Ar with flowrates of 25, 6 and 5 sccm respectively. This process results in channel waveguides as shown in Fig. 2(a). The waveguides are covered by a 1 µm thick SiO 2 cladding, which is deposited by e-beam evaporation from a 99,999% pure SiO 2 target in the BAK600 in the MESA+ cleanroom. The target to substrate distance is kept at 15 cm. Evaporation was chosen in order to keep the sample at room temperature to prevent crystallization of the TiO 2 . The settings for the evaporations were a beam current of 200 mA and a base pressure of 6·10 −6 mbar.
Simulations using Lumerical MODE solutions show four TE modes and three TM modes at a wavelength of 632 nm and two TE modes and one TM mode at a wavelength of 1010 nm. Normalized mode profiles of the fundamental TE modes are shown in Figs. 2(b) and (c) for the wavelengths of 632 nm and 1010 nm respectively. These modes show confinement of the light in the waveguide core of 76.7% and 51.5% respectively.

Characterization
AFM measurements were performed on deposited TiO 2 layers of different thicknesses with a Bruker dimension fastscan atomic force microscope (AFM). They are shown in Figs. 3(a) and (b). The surface roughness of these layers is determined by calculating the root mean square of the height profile of these images. The rms roughness values are shown in Fig. 3(c) as a function of layer thickness. The surface roughness increases linearly with increasing layer thickness. The increasing grain size indicates increasing amount of crystallization [29,30]. As a result, thick waveguides exhibit higher scattering losses. The crystallinity of the layer was determined using transmission electron microscope (TEM) imaging. The TEM-sample is prepared by thinning a cross section of the TiO 2 by using focused ion beam milling, to obtain a thin TiO 2 lamella of several tens of nanometers. To prevent damage to the TiO 2 layer during the thinning process, a protective coating of 3-4 µm of platinum is deposited on top of the TiO 2 layer. This deposition might, however, cause amorphization of the top surface of the sample. This is prevented by applying a 20 nm carbon layer before the platinum deposition. The TEM imaging is performed using a Philips CM300ST FEG TEM. Figure 4(a) shows an overview of the TEM sample, consisting of the SiO 2 substrate, TiO 2 layer and a protective coating of carbon and platinum. In this image, it is possible to observe a column-like structure appearing after the initial 50 nm of the TiO 2 layer. The same structure can be seen in the SEM image, Fig. 2(a). A zoomed-in image of the initial 50 nm of the TiO 2 layer is shown in Fig. 4(b). The corresponding Fourier transform (FT) is shown in Fig. 4(c). The almost smooth halo indicates that the first 50 nm of the deposited layer consists of predominantly amorphous TiO 2 . A zoomed-in image of the top part of the TiO 2 layer exhibiting the column-like The crystallinity of the layer was determined using transmission electron microscope (TEM) imaging. The TEM-sample is prepared by thinning a cross section of the TiO 2 by using focused ion beam milling, to obtain a thin TiO 2 lamella of several tens of nanometers. To prevent damage to the TiO 2 layer during the thinning process, a protective coating of 3-4 µm of platinum is deposited on top of the TiO 2 layer. This deposition might, however, cause amorphization of the top surface of the sample. This is prevented by applying a 20 nm carbon layer before the platinum deposition. The TEM imaging is performed using a Philips CM300ST FEG TEM. Figure 4(a) shows an overview of the TEM sample, consisting of the SiO 2 substrate, TiO 2 layer and a protective coating of carbon and platinum. In this image, it is possible to observe a column-like structure appearing after the initial 50 nm of the TiO 2 layer. The same structure can be seen in the SEM image, Fig. 2(a). A zoomed-in image of the initial 50 nm of the TiO 2  Fig. 3(d), showing some features that are hidden in the full FT due to averaging effects. layer is shown in Fig. 4(b). The corresponding Fourier transform (FT) is shown in Fig. 4(c). The almost smooth halo indicates that the first 50 nm of the deposited layer consists of predominantly amorphous TiO 2 . A zoomed-in image of the top part of the TiO 2 layer exhibiting the column-like structure, is shown in Fig. 4(d). As the thickness of the deposited layer increases, a crystalline phase becomes visible, as shown in Figs. 4(e) and (f), which correspond to the FT of the full area and the marked area of Fig. 4(d), respectively. The main features in Figs. 4(e) and 4(f) show periodicities of around 2.1 and 2.8 Å, which correspond to the Ti-O and O-O bonds in anatase, respectively [31]. A possible explanation for the structure of the deposited film could be found in the structure zone model [32]. The deposited films in this work show poly-crystalline features, shown in the TEM images, and changing structure with increasing film thickness, shown in the AFM images, which corresponds to zone T in the structure zone model.
Confocal Raman measurements were performed on a 100 nm thick TiO 2 layer on a fused silica substrate to further confirm the crystalline structure of the layers. A WITEC alpha300R/S/A instrument was utilized. Pump light with a wavelength of 785 nm and a power of 185 mW is used in combination with a 100x objective with an NA of 0.9. The spectrum is measured with an integration time of 1 s and 10 accumulations. It was verified that the power density in the laser focal point did not induce the crystallization of the layer. The Raman spectrum is shown in Fig. 5, which clearly shows the presence of the anatase crystalline phase with peaks at 151, 409, 515 and 633 cm −1 [9].  a HeNe laser is used, the measurements at 770, 859, 979 and 1010 nm are performed using a Ti-sapphire laser, the wavelength of which is measured using a spectrometer. After optimizing the alignment for maximum coupling efficiency, top view images are taken with a Point Grey monochrome camera (BFLY-U3-23S6M-C). The linearity of camera was measured to be 0.1% over 2 decades of optical power. Figure 6(a) shows a top view image of the light propagation at 632.8 nm. The corresponding exponential fit of the intensity as a function of propagation length is given in Fig. 6(b), showing propagation losses of 7.8 ± 0.52 dB/cm. The same is shown in Figs. 6(c) and 6(d) for a wavelength of 1010 nm. Losses as low as 0.68 ± 0.46 dB/cm can be observed. For the measurement at 1010 nm, the scattering point has not been taken into account for the exponential fit, since this single defect is not representative for the overall waveguide losses. The error margin of the losses is obtained by calculating the standard deviation of the fit parameters.
The wavelength dependent propagation loss is summarized in Fig. 7. The significant increase in propagation losses towards the visible wavelength range, indicate that the losses are mainly caused by scattering, which is a result of the poly-crystallinity of the layers and top and sidewall roughness. This is to be expected, since TiO 2 only starts absorbing below a wavelength of roughly 420 nm. For higher wavelengths, where the loss starts to decay, the relative error increases. Due to the short length of the spirals utilized for these measurements, the lowest value for the loss that can be measured is limited. Loss-measurements at wavelengths of 1310 nm and 1550 nm where non-successful, due to low signal-to-noise ratio. At these wavelengths, the Si substrate becomes transparent. The rough background of the chip scatters the light and causes higher amounts of background noise. Further reduction of the propagation losses could be obtained by reducing surface roughness by using a chemical mechanical polishing step, as it has been done in the Si 3 N 4 waveguide platform [33]. Furthermore the sidewall roughness might be reduced by optimizing the etching recipe and by using e-beam lithography for higher resolution of the resist. Further optimization of the deposition process can reduce the amount of crystallinity, which will reduce volume scattering losses and reduce the surface roughness.

Conclusion
TiO 2 waveguides were fabricated with DC reactive sputter deposition, followed by photolithography and reactive ion etching. A SiO 2 cladding is applied to obtain a less fragile device with lower propagation losses. Combination of TEM imaging and Raman measurements revealed the presence of anatase crystalline phase in the upper TiO 2 layer. The lower TiO 2 layer remained completely amorphous. AFM imaging shows that the surface roughness increases as the thickness of the layer increases, which corresponds to an increased crystallinity of the layer. Channel waveguide losses as low as 0.68 ± 46 dB/cm at a wavelength of 1010 nm were measured. The waveguide losses increase towards lower wavelengths, due to the increased amount of scattering losses. Further reduction of the propagation losses might be obtained by using E-beam lithography, chemical mechanical polishing and further optimization of the deposition process.