Titanium dioxide (${{\rm TiO}_2}$) has recently evolved as a promising candidate for photonic integrated circuits (PICs) by possessing high linear and nonlinear indices, a large bandgap (${\gt}{3}\;{\rm eV}$), a uniquely negative thermo-optic coefficient, and a large transparency window from visible to mid infrared wavelengths [1–18]. Besides, the possible compatibility with the mature silicon complementary metal oxide semiconductor (CMOS) fabrication process also endorses ${{\rm TiO}_2}$–based PICs in terms of scalable manufacturing [19]. Among different PIC devices, the microring resonator (MRR) is a fundamental building block for many applications that enhances the local optical field intensity in a small area. ${{\rm TiO}_2}$ MRRs have also been widely studied and play a vital role in applications like thermal optical filters [15] or Raman sensing [18].

Several methods have been demonstrated to pattern ${{\rm TiO}_2}$ waveguides, including ion implantation [1], wet etching [2], dry etching [3–13], and lift-off [14,15]. Among them, the top-down method, in which ${{\rm TiO}_2}$ film is first deposited and then selectively etched with a mask, has normally been used. However, different from selective etching of silicon, silica, and silicon nitride, which have been intensively optimized and easily accessible, selective etching of ${{\rm TiO}_2}$ remains a challenge, especially for fabricating sub-µm-wide waveguides with dry etching like reactive-ion etching (RIE). This may be attributed to the fact that most of the Ti-contained etch products have a high boiling point and are not easily volatile during the ion etching [20], and that ${\rm Ti}/{{\rm TiO}_2}$ is a hard material itself that acts as the hard mask for etching other metals [21,22].These not only challenge the mask selection for ${{\rm TiO}_2}$ etching, but also produce rough sidewalls with residues. Whereas thick UV sensitive polymer resists could perform as the etch mask, the fabricated ${{\rm TiO}_2}$ waveguide exhibited a high loss (5.5 dB/cm [3]) or possessed a very thin ${{\rm TiO}_2}$ core of 140 nm despite a quite low loss of 0.68 dB/cm [4]. Meanwhile, the thick UV resists are not suitable for fabricating sub-µm structures, especially for the small gaps between an MRR and the bus waveguide. Sub-µm-wide ${{\rm TiO}_2}$ waveguides could be fabricated with a metal mask, but the waveguide losses were usually high, which inevitably resulted in ${{\rm TiO}_2}$ MRRs with moderate intrinsic Qs (e.g.,  ${2.2} \times {{10}^4}$ at the wavelength of 635 nm [6], ${8.9} \times {{10}^4}$ at 1550 nm [10], ${7.35} \times {{10}^4}$ at 1550 nm [13]). In order to avoid ${{\rm TiO}_2}$ etching, lift-off processes have been developed for high-Q ${{\rm TiO}_2}$ MRRs (${2} \times {{10}^5}$ [14], ${1.55} \times {{10}^5}$ [15]). However, limited by the weak confinement of ${{\rm TiO}_2}$ waveguides fabricated with the lift-off process, MRRs usually had large footprints, e.g., with a radius of 150 µm [14] and 200 µm [15].

Here, we demonstrate a new bottom-up method for fabricating ${{\rm TiO}_2}$ waveguides and MRRs. In this method, ${{\rm TiO}_2}$ waveguides are formed by depositing ${{\rm TiO}_2}$ on pattern platforms with a T-shaped cross-section profile, which is fabricated by deep etchings to the thin top layer of silicon nitride (${{\rm Si}_3}{{\rm N}_4}$) and the thick bottom layer of silica (${{\rm SiO}_2}$) and an undercut etching with a hydrofluoric (HF) acid solution. Thus, the direct dry etching to the ${{\rm TiO}_2}$ material is avoided and all the involved etching processes are indeed easily accessible in today’s laboratories or foundries. Measurements to the fabricated ${{\rm TiO}_2}$ MRRs show intrinsic Qs of ${\sim}{1.1} \times {{10}^5}$, comparable to that from the lift-off processes, and MRRs with sub-µm waveguide widths or a small radius of 10 µm exhibit close Qs. These results not only verify the high feasibilities of our bottom-up method for fabricating high-${Q}$ ${{\rm TiO}_2}$ MRRs, but also suggest a new approach for waveguide fabrication in other materials of great importance but few off-the-shelf fabrication processes.

The proposed bottom-up fabrication flow is illustrated in Fig. 1. Starting with a Si wafer with a 2.4-µm oxidized ${{\rm SiO}_2}$ layer, a thin (50 nm) layer of ${{\rm Si}_3}{{\rm N}_4}$ is first deposited with a low-pressure chemical-vapor-deposition (LPCVD) method [Fig. 1(a)]. Then an 80-nm chromium (Cr) layer is deposited by electron beam evaporation [Fig. 1(b)]. Next, the waveguide patterns are defined by e-beam lithography, first on ZEP resist and then on Cr by dry etching [Fig. 1(c)].The Cr etching is completed in a RIE machine with a gas mixture of chlorine (${{\rm Cl}_2}$, 65 sccm) and oxygen (${{\rm O}_2}$, 15 sccm) at a substrate temperature of 50°C, a pressure of 15mTor, and a platen power of 15 W. With Cr as the hard mask, the thin ${{\rm Si}_3}{{\rm N}_4}$ layer is first dry-etched with a gas mixture of carbon tetrafluoride (${{\rm CF}_4}$, 20 sccm) and hydrogen (${{\rm H}_2}$, 10 sccm) at a substrate temperature of 0°C, a pressure of 3 mT, and a platen power of 5 W [Fig. 1(d)]. Then a deep etching is carried out to etch through the 2.4-µm-thick ${{\rm SiO}_2}$ layer with the Si substrate being a little etched in [Fig. 1(e)]. A gas mixture of octafluorocyclobutane (${{\rm C}_4}{{\rm F}_8}$, 40 sccm) and oxygen (${{\rm O}_2}$, 5 sccm) at a substrate temperature of 55°C, a pressure of 6 mTor, and a platen power of 180 W is used for the deep etching. After stripping the residual Cr in Etchant 18 solution, 5% HF solution is used to etch ${{\rm SiO}_2}$ for 5 min with ${\sim}{130}\;{\rm nm}$ cut on both sides. This leaves pattern platforms with a T-shaped cross section [Fig. 1(f)]. Then ${{\rm TiO}_2}$ is deposited by ion-beam sputtering (IBS), as described in our previous work [13]. One can deposit as thick ${{\rm TiO}_2}$ as needed, as long as the ${{\rm SiO}_2}/{\rm Si}$ etching is deep enough. Here, a 350-nm-thick ${{\rm TiO}_2}$ layer is deposited. Finally, the ${{\rm TiO}_2}$ waveguides are polished by the 5% HF solution for 3 min with 250-nm-thick ${{\rm TiO}_2}$ left.

 

Fig. 1. Bottom-up fabrication flow for ${\rm TiO}_{2}$ waveguides. (a) ${{\rm Si}_3}{{\rm N}_4}$ deposition on the oxidized Si wafer. (b) Cr deposition. (c) E-beam lithography and Cr etching. (d) ${{\rm Si}_3}{{\rm N}_4}$ etching. (e) ${\rm SiO}_2$ deep etching. (f) ${{\rm SiO}_2}$ wet etching. (g) ${{\rm TiO}_2}$ deposition. (h) ${{\rm TiO}_2}$ polishing with HF.

Download Full Size | PDF

Figure 2(a) shows the scanning electron microscopy (SEM) image of a ${{\rm TiO}_2}$ waveguide after the ${{\rm TiO}_2}$ deposition but prior to the HF polishing. From the SEM image, rough sidewalls are found. This is because the incident particle beam sputtered out from the Ti target to the sample has an angle far beyond the optimal angle with the sidewall surfaces of the ${{\rm Si}_3}{{\rm N}_4}/{{\rm SiO}_2}$ pattern platform and the ${{\rm TiO}_2}$ core and, hence, yields ${{\rm TiO}_2}$ with very low density [23]or even porosity grown on the sidewalls. Tremendous scattering loss would be introduced by such sidewalls, which, however, can be mitigated by the HF etching for 3 min, as shown by the SEM image in Fig. 2(b). Meanwhile, since there is more/wider low-density ${{\rm TiO}_2}$ on the sidewalls towards the bottom and the HF etching is faster for such low-density ${{\rm TiO}_2}$ than the denser ${{\rm TiO}_2}$ on the top surface, the ${{\rm TiO}_2}$ core is shaped to be inversely trapezoidal by the HF polishing. After polishing, the waveguides are narrowed down and gaps between the bus waveguide and the MRR are inevitably large, i.e., ${\gt}260\,\,{\rm nm}$. Racetrack structures are used to guarantee sufficient light coupling into the resonators, as shown by the SEM image in Fig. 2(c). Radius and straight length of the MRRs are denoted as $R$ and $L$, respectively, and $L$ is 20 µm for all our fabricated MRRs. Figure 2(d) shows the SEM image of the cross section of the straight coupling region of a fabricated ${{\rm TiO}_2}$ MRR. In the figure, the T-shaped ${{\rm Si}_3}{{\rm N}_4}/{{\rm SiO}_2}$ pattern platform and the inversely trapezoidal ${{\rm TiO}_2}$ waveguide core can be more clearly recognized. The top and bottom widths of the waveguide core are denoted as ${w_1}$ and ${w_2}$, respectively. Moreover, the gap in the bus waveguide is defined as the distance between the upper borders of the two inversely trapezoidal ${{\rm TiO}_2}$ waveguide cores.

 

Fig. 2. SEM images of the ${{\rm TiO}_2}$ waveguide fabricated by the bottom-up method (a) before and (b) after the HF polishing. (c) SEM images of the racetrack MRR (a) and (d) cross-section of the straight coupling region.

Download Full Size | PDF

We have fabricated ${{\rm TiO}_2}$ MRRs with two different waveguide widths, i.e., ${w_1} = {1370}\;{\rm nm}$ (${w_2} = {1180}\;{\rm nm}$) and ${w_1} = {870}\;{\rm nm}$ (${w_2} = {680}\;{\rm nm}$). The narrower one is a single-mode waveguide while the wider one can support higher-order modes. Here, we only focus on the transverse-electric (TE) polarization. Figures 3(a) and 3(b) show the simulated electric field intensity distributions of the fundamental mode (${{\rm TE}_0}$) profiles for the wider and narrower waveguides, respectively, at the wavelength of 1550 nm. The effective refractive index is calculated to be 1.745 and the effective mode area is ${0.{51\,\,\unicode{x00B5}{\rm m}}^2}$ for the wider ${{\rm TiO}_2}$ waveguide at 1550 nm, while they are 1.597 and ${0.{34\,\,\unicode{x00B5}{\rm m}}^2}$ for the narrower one. Meanwhile, power confinement ratios in the ${{\rm TiO}_2}$ core are calculated to be 63.3% and 60.4% for the wider and the narrower waveguides, respectively. While the narrower waveguide may be preferred for, e.g., nonlinear applications due to a smaller effective area [24], the wider one may exhibit a lower loss since the mode apparently overlaps less with the sidewalls, which couldn’t be perfectly smooth under current fabrication techniques.

 

Fig. 3. Electric field density distribution of the mode profile for the ${{\rm TiO}_2}$ waveguides with the width (a) ${w_1} = {1370}\;{\rm nm}$ and (b) ${w_1} = {870}\;{\rm nm}$. Here, the wavelength is 1550 nm.

Download Full Size | PDF

Figure 4(a) shows the measured and normalized (to a straight ${{\rm TiO}_2}$ waveguide) transmission spectra of the fabricated ${{\rm TiO}_2}$ MRRs with ${w_1} = {1370}\;{\rm nm}$, $R = {130}\;{\unicode{x00B5}{\rm m}}$ and gaps of 310 nm (red) and 360 nm (blue). The spectra show that the extinction ratios (ERs) increase with the increasing wavelength and the decreasing gap, both of which indicate a stronger light coupling to the MRR. Thus, it can be deduced that the MRRs are under-coupling. By using the same method as that in Ref. [14], we have used the Lorentzian curve to fit the spectrum around each resonance and extracted the loaded ${Q}$(${{ Q}_{\textit{load}}}$) and intrinsic ${Q}$(${{ Q}_{{\rm int}}}$) and calculated the bend waveguide loss at each resonant wavelength. Figure 4(b) shows the extracted ${{ Q}_{{\rm int}}}$ (filled circles) and loss (filled triangular) of the two MRRs at each resonant wavelength. ${{ Q}_{{\rm int}}}$ of both the two ${{\rm TiO}_2}$ MRRs distributes around ${1.1} \times {{10}^5}$ (dashed blue line) in the whole investigated wavelength range from 1520 nm to 1620 nm. The waveguide loss can be attributed to scattering and absorption. We use a simple formula to fit the waveguide loss α as a function of the wavelength, as shown by

 

Fig. 4. (a) Measured transmission spectra of the fabricated ${{\rm TiO}_2}$ MRRs with ${w_1} = {1370}\;{\rm nm}$, $R = {130}\;\unicode{x00B5}{\rm m}$, and different gaps. (b) The extracted intrinsic ${Q}$ and waveguide loss of the MRRs at each resonant wavelength. The solid blue line shows the fitting to the loss as a function of the wavelength, and the dashed blue line indicates ${{Q}_{{\rm int}}} = {1.1}\;{\times }\;{{10}^5}$.

Download Full Size | PDF

 

Fig. 5. (a) Measured transmission spectrum and calculated coupling efficiency of a fabricated ${{\rm TiO}_2}$ MRR with ${w_1} = {870}\;{\rm nm}$, $R = {130}\;{\unicode{x00B5}{\rm m}}$, and gap = 460 nm. (b) The extracted intrinsic ${Q}$ and waveguide loss at each resonant wavelength. The solid blue line is the fitting to the waveguide loss as a function of the wavelength.

Download Full Size | PDF

The solid blue line in Fig. 5(a) shows the measured and normalized transmission spectrum of a fabricated ${{\rm TiO}_2}$ MRR with ${w_1} = {870}\;{\rm nm},\;R = {130}\;{\unicode{x00B5}{\rm m}}$, and gap = 460 nm. We have also calculated the coupling efficiency between the bus waveguide and the MRR, as shown by the solid red line in Fig. 5(a). The coupling efficiency increases from 0.08 to 0.25 as the wavelength increases from 1520 nm to 1620 nm. Around 1555 nm, the resonator is around the critical coupling, i.e., the coupling efficiency equals the single-round trip loss of the MRR (assuming no loss in the coupler [26]), showing the largest ERs more than 25 dB. The MRR is under-coupling at wavelengths shorter than 1555 nm and over- coupling at wavelengths longer than 1555 nm. With this, the ${{ Q}_{{\rm int}}}$ of the MRR and the waveguide loss at each resonant wavelength are extracted and shown by the green dots and the red dots in Fig. 5(b), respectively. The ${{ Q}_{{\rm int}}}$ around 1550 nm is about ${0.9} \times {{10}^5}$. We also use Eq. (1) to fit the loss as a function of wavelength, as shown by the solid blue line in Fig. 5(b), and $a$ and $b$ values are fitted to be ${24}\;{\times}\;{{10}^{12}}\;{{\rm nm}^4}\;{\rm dB}/{\rm cm}$ and 1.22 dB/cm, respectively. Thus, the total loss at 1550 nm is 5.4 dB/cm with scattering and absorption accounting for 77% and 23%, respectively. Partially due to less light being confined in the ${{\rm TiO}_2}$ core, the absorption loss of the narrower waveguide is slightly smaller than that of the wider one. However, the contribution of the scattering to the total loss in the narrower waveguide is apparently larger since more field overlaps the sidewalls.

 

Fig. 6. Measured transmission spectrum of a fabricated ${{\rm TiO}_2}$ MRR with ${w_1} = {1370}\;{\rm nm}$, $R = {10}\;{\unicode{x00B5}{\rm m}}$, and gap = 360 nm. The inset shows the close-up spectrum and the Fano-like fitting around the resonance at 1565.411 nm.

Download Full Size | PDF

Thanks to the large refractive index of ${{\rm TiO}_2}$, a ${{\rm TiO}_2}$ MRR may confine light in a very compact dimension, which can considerably strengthen the light-matter interaction or boost the nonlinear effect. We have also used the proposed bottom-up method to fabricate ${{\rm TiO}_2}$ MRRs with very compact footprints. Figure 6 shows the measured and normalized spectrum of a ${{\rm TiO}_2}$ MRR with ${w_1} = {1370}\;{\rm nm}$, $R = {10}\;{\unicode{x00B5}{\rm m}}$, and gap = 360 nm. The free-spectral range (FSR) is measured to be 9.8 nm. The inset shows the close-up spectrum (blue dots) around the resonance at 1565.411 nm. Meanwhile, a Fano-like fitting (red line) is used to fit the measured spectrum by taking into account the influence of the Fabry–Perot (F-P) cavity formed between the two facets of the bus waveguide. From the fitting, one can extract a linewidth of 25 pm resulting in a ${{Q}_{{\rm int}}}$ of ${1.006} \times {{10}^5}$ and a bend waveguide loss of ${\sim}{4.2}\;{\rm dB/cm}$, which are the same as that extracted from a pure Lorentzian fitting due to the very weak influence of the F-P cavity and at the same level as that of the large MRRs (i.e., $R = {130}\;{\unicode{x00B5}{\rm m}}$). This clearly shows the feasibility of our proposed bottom-up fabrication method for ${{\rm TiO}_2}$ MRRs with very compact footprints.

In summary, by first fabricating pattern platforms with a T-shaped cross section and then depositing ${{\rm TiO}_2}$ to form the waveguide, we have demonstrated a new bottom-up method for fabricating ${{\rm TiO}_2}$ waveguides without direct dry etching to the ${{\rm TiO}_2}$. With this method, ${{\rm TiO}_2}$ MRRs are fabricated and intrinsic ${Q}$ values of ${\sim}{1.1} \times {{10}^5}$ are achieved, comparable to the state-of-the-art results [14,15]. The fabricated ${{\rm TiO}_2}$ MRR with a bending radius of 10 µm can sustain a ${Q}$ value of ${\sim}{{10}^5}$, showing the strong confinement of the ${{\rm TiO}_2}$ waveguide and the feasibility of the proposed bottom-up method for fabricating very compact ${{\rm TiO}_2}$ MRRs. Measurements and analysis of an MRR with the ${{\rm TiO}_2}$ single-mode waveguide suggest that the scattering loss is dominant while the absorption loss is still not negligible. The scattering loss may be further lowered down by optimizing the metal mask etching. Meanwhile, more optimization to the ${{\rm TiO}_2}$ deposition, e.g., finer adjustment to the ${{\rm O}_2}$ flow during the reactive sputtering, is needed to reduce the absorption loss [4]. Apart from being used for the fabrication of the ${{\rm TiO}_2}$ waveguides and MRRs with high Qs and compact footprints, our method can also be seen as a simple way to validate the feasibility of waveguides with new materials, especially the ones whose etching is complex to optimize or contaminates the existing fabrication line.