1. Introduction

Ion-exchange technique was first used to make waveguide in glass substrate by T. Izawa and H. Nakagome in 1972 [1]. Since then, this technique has been widely used to fabricate passive [2] or active waveguide devices [3,4] because it has the advantages of simplicity, flexibility and low fabrication cost. But unfortunately, conventional ion-exchanged waveguide devices are fabricated only in bulk glasses, they are not easily compatible with other waveguide devices [5,6]. This drawback makes ion-exchanged devices difficult to realize more complex functions and will largely limit their commercial viability.

Silica-on-silicon (SOS) is a widely accepted waveguide structure; its fabrication processes can be adapted from the matured semiconductor manufacturing processes [7]. Various optical components of this structure can be integrated on one substrate to make planar lightwave circuits (PLCs). Many methods have been developed to fabricate silica-on-silicon waveguides, such as plasma-enhanced chemical vapor deposition (PECVD) [8], flame hydrolysis deposition (FHD) [7], sputtering [9], and sol-gel [10]. Reactive ion etching (RIE) is a normal technique to fabricate channel SOS waveguides. However, this expensive and time-cost technique may generate a significant roughness level on the channel sidewalls, which increases the scattering loss and polarization dependent loss (PDL). The roughness induced scattering loss is significant since these waveguides have a step-index-contrast. The roughness-induced polarization dependent loss is caused by the fact that roughness is present on the sidewalls but not on the upper and lower interfaces. In principle, ion-exchanged waveguides have a graded-index profile. Consequently, both scattering loss and the polarization dependent loss can be substantially reduced. Furthermore, ion-exchange technique is tolerant to the imperfections in the photolithography because edge roughness on masks that define the waveguide geometry renders less damaging due to the followed diffusion process. Waveguides with a propagation loss less than 0.1 dB/cm and a PDL below 0.1 dB have been realized by ion-exchange and field-assisted burying processes [3].

A combination of SOS and ion-exchange technique provides a challenging way of fabricating low-loss on-chip channel waveguides with relatively low-cost, as RIE is not necessary any more. In 2000, J. Fick et al. reported the fabrication of ion-exchanged channel waveguides in sol-gel films on silica glass substrate [11]. But the device had a large insertion loss of 16 dB at 1550 nm.

In this paper, we report the fabrication of ion-exchanged sol-gel waveguides and Y-branch power splitters in silica-on-silicon. Li⁺ ions-containing ion-exchangeable aluminosilicate glass thin film was fabricated on silicon substrate by sol-gel technique. Ag⁺-Li⁺ thermal ion-exchange was used to achieve lateral confinement. Propagation loss of the channel waveguide was as low as 0.50 dB/cm. The total fiber to waveguide and waveguide to fiber coupling loss was estimated to be 1.51 dB. Our results demonstrated that ion-exchange in SOS is a practical and low-cost method to prepare low-loss optical channel waveguides.

2. Experiment

The sol-gel solution was prepared by hydrolyzing tetraethoxysilane (TEOS) with a molar ratio of water to TEOS around 2. HCl was added as a catalyst. After reaction at 70 °C for 2 hours, 17 mol% Al(NO₃)₃ and 12 mol% LiNO₃ was added to the solution for the ion-exchangeable layer (or 5 mol% Al(NO₃)₃ only for the buffer layer). Then both solutions were stirred at 70 °C for another 2 hours. Highly mobile Li⁺ ions in the ion-exchangeable layer can be used to exchange with Ag⁺ ions outside of the glass [12]. The reason why we used Li⁺ ions instead of widely used Na⁺ ions for ion-exchange is that Na⁺ ions-containing salts will make the gelation process too fast to be controlled. The Al(NO₃)₃ used here is to avoid the cracking of Li⁺ ions-containing glass film and to reduce the glass transform temperature as well.

The process of channel waveguide fabrication is schematically illustrated in Fig. 1. Li⁺ ions-containing ion-exchangeable sol-gel glass film was deposited by a multilayer deposition technique named DC-RTA (dip-coating and rapid thermal annealing). This technique is put forward by us after SC-RTA (spin-coating and rapid thermal annealing) technique reported by A. S. Holmes et al. [13], but using dip-coating instead of spin-coating. For each cycle, the deposited sol-gel layer was consolidated at 1000 °C for 45 seconds. The thickness of a single layer is about 60~100 nm after annealing. Crack-free aluminosilicate glass films with a thickness of a few µm can be formed by repeating cycles of this DC-RTA process because the stress brought from glass consolidating process can be released by each annealing process. To fabricate waveguides, a 5 µm thick buffer layer without Li⁺ ions was firstly deposited on silicon substrate to serve as a forbidden region for ion-exchange and to avoid the propagation of light penetrating into the Si substrate. Then a Li⁺ ions-containing ion-exchangeable layer with a thickness h=11 µm was deposited on it.

 

Fig. 1. Schematic diagram of the channel waveguide fabrication. (a) Buffer layer formed by DC-RTA; (b) Li⁺ ions-containing ion-exchangeable layer formed by DC-RTA; (c) Channel windows in Al film opened by standard photolithography; (d) Ag⁺-Li⁺ ion-exchange process; (e) Mask removed and channel waveguide realized.

Download Full Size | PDF

To fabricate channel waveguide, a 150 nm Al film was thermally evaporated on the glass film and a standard photolithograph was applied to form an open window on Al film. Then Ag⁺-Li⁺ thermal ion-exchange was carried out to realize a two-dimensional confinement. The ion-exchange was carried out in 15wt% AgNO₃ and 85wt% KNO₃ eutectic salt melt at 336 °C (±1 °C) for 75 sec. This diluted AgNO₃ salt melt can help to accurately control the refractive index profile. Unlike conventional bulk glass based ion-exchanged devices that need end polishing, ion-exchanged SOS sample can be simply cleaved for butt coupling with single mode fiber.

3. Results and discussion

Table 1. m-line measurement and fitted data

View Table

The refractive indices of the planar waveguide (TE mode) before and after ion-exchange were evaluated by prism coupling and m-line measurements at 632.8 nm and 1550 nm. The results were summarized in Table 1, where n_bf and n_ie are the refractive indices of the buffer layer and the ion-exchangeable layer measured before ion-exchange, respectively. The error of refractive index measurement was below ±2×10^-4. After ion-exchange, there are two types of waveguiding modes in the planar ion-exchanged waveguide, which can be easily identified in the m-line measurement at 632.8 nm (Fig. 2(a)). Optical modes with effective refractive indices N_eff0, N_eff1, N_eff2 above n_ie are conventional modes in ion-exchanged layer. On the other hand, the combination of ion-exchangeable layer and buffer layer can support optical modes with n_bf<N_eff <n_ie. 9 such modes can be found in Fig. 2(a). From another point of view, these two types of modes can also be unified as the modes of the two layer structure waveguide with unified effective refractive index N’_effm (m=0, 1,…11). Leaky modes with N_eff<n_bf can also be recognized in the figure.

The spatial refractive index distribution of all the waveguiding modes can be deduced from the inverse Wentzel-Kramers-Brillouin (WKB) approximation [14], which is shown in Fig. 2(b). The difference between the deduced and the measured n_ie and h is due to the error induced by WKB calculation near the step-index point. However, accurately measured N_eff0, N_eff1, N_eff2 can be used to fit the refractive index profile of the ion-exchanged region because they are far from the step-index point. The surface index difference Δn and diffusion depth d were assumed to have the following expression according to the ion-diffusion process:

where D is the ion-diffusion coefficient, t is the ion-exchange time, erfc(y/d) is the complementary error function. From the fitting, we obtained Δn=0.0183±0.0004, d=5.0±0.2 µm and D=4.8 µm²/min at 632.8 nm with the ion-exchange condition mentioned above. The green dashed line in Fig. 2(a) shows the fitted surface refractive index n_surf after ion-exchange. As the refractive index depth profile follows the Ag⁺ ions concentration profile in the ion-exchanged region in the glass film, we can assume that the refractive index profile at 1550 nm is almost the same as that at 632.8 nm. The only difference is on Δn due to the dispersion. Using the analytic transform matrix method and also the N_eff0 measured at 1550 nm, Δn=0.0196 at 1550 nm was deduced.

 

Fig. 2. (a) m-line measurement of the planar waveguide at 632.8 nm after ion-exchange in the mixed melt salt at 336 °C for 75 sec; (b) Inverse WKB approximation deduced spatial refractive index profile of the planar ion-exchanged waveguide

Download Full Size | PDF

The refractive index difference between the TE and TM mode was also measured to be 4~5×10^-4 after ion-exchange. This value is almost the same as it was before ion-exchange, which indicates that birefringence do exist in the DC-RTA process, and the followed ion-exchange process does not induce additional birefringence.

Waveguides with high optical quality were prepared. After ion-exchange, no noticeable micro-cracks were observed. The immersion coupler method [15] was used to measure the propagation loss of the planar waveguide at 650 nm, 975 nm and 1550 nm before and after ion-exchange. Before ion-exchange, the propagation losses measured decreased with wavelength almost as 1/λ⁴, which indicates that the propagation losses are coming mainly from Rayleigh scattering loss. The propagation loss was 0.14 dB/cm at 1550 nm. After ion-exchange, almost no change was found from propagation loss measurement. This result indicates that the Ag⁺-Li⁺ ion-exchange process in the Li⁺ ions-containing sol-gel glass film does not induce additional scattering or absorption loss. Loss due to the reduced Ag⁰ in the common Ag⁺-Na⁺ ion-exchange is negligible here because Ag⁺ ions concentration is not high and the sol-gel process uses high purity precursors.

Channel waveguides were formed through the patterned Al mask. The ion-exchange process is basically an ion-diffusion process. Assuming that Ag⁺ and Li⁺ are equally mobile and the molar fraction of the incoming ions is much less than that of the host alkali ions in the glass, the two-dimensional (2D) ion diffusion equation could be simplified as:

where C_A is concentration distribution of the Ag⁺ ions exchanged in the glass film, w is the width of the mask window, C₀ is the concentration of Ag⁺ ions on surface of the glass film. In our experiment, we had w=3 µm and the exchange time is t=75 sec. Because the refractive index change should be proportional to the Ag+ ions concentration, the refractive index change also satisfies Eq. (2). Replacing C₀ by Δn and C_A(x, y, t) by Δn(x, y, t), we can obtain the refractive index distribution after ion-exchange. The 2D diffusion process was numerically modeled using a finite element method (FEM). Fig. 3(a) shows the contours of the mode intensity profile (red dashed lines) calculated by RSOFT BeamPROP software with the FEM simulated 2D refractive index profile. The near field pattern measured experimentally (solid lines) is shown as well. In Fig. 3(a), the surface of the glass film both in experiment and in simulation is unified as zero of Y axis (blue straight line) according to Eq. (2). It can be seen from the figure that the calculated mode intensity profile matches well with the measured one, but small difference exists, especially in vertical direction. Most likely the erfc function profile we used in the fitting may have a deviation from the actual value. The waveguide dimension (1/e² of the maximum intensity) W_H=12.1 µm in horizontal and W_V=7.5 µm in vertical were obtained from contour of the measured near field pattern.

 

Fig. 3. The characteristic of channel waveguide. (a) Contours of the numerical simulation of mode intensity profile (dashed lines) and the near field intensity distribution at 1550 nm of the waveguide (solid lines), corresponding to 0.2, 0.4, 0.6 and 0.8 times the maximum mode intensity value. The background is the measured near field intensity of the channel waveguide; (b) Loss characteristic of the channel waveguide measured by cutback method at 1550 nm.

Download Full Size | PDF

The insertion loss of the channel waveguide was measured by a tunable laser source (ANDO AQ4321D) set at 1550 nm and an optical multimeter (ANDO AD2377). Single mode fibers at 1550 nm were used to couple the light into and out of the channel waveguide. The total fiber-waveguide insertion loss of a 3 cm long waveguide at 1550 nm was 2.99 dB. The PDL of the waveguide was also measured by inserting a polarization controller before the input fiber. The measured PDL was about 0.4 dB. This PDL may be generated from the asymmetrical waveguide structure or the residue stress after multilayer deposition process.

A propagation loss of 0.50 dB/cm was measured by the cutback method (Fig. 3(b)). Comparing with the propagation loss of 0.14 dB/cm of the planar waveguide indicates that it is likely to improve the loss characterization by better optimizing our fabrication process. The coupling loss was deduced to be 0.76 dB/facet in Fig. 3(b) by extrapolating the waveguide length to zero. The coupling loss caused by mode overlapping between the waveguide and the single mode fiber of 1550 nm could be estimated by the following formula [2]:

where a is mode field diameter of the single mode fiber (10.5 µm). With the values of W_H and W_V obtained above, the coupling loss was evaluated to be about 0.30 dB. The additional 0.46 dB loss is caused by Fresnel reflection and the nonparallelism of the waveguide and the fiber in the measurement. The total Fresnel reflection loss of fiber to air and air to waveguide was calculated to be 0.31 dB. But the measured value is between 0.15~0.31 dB due to the interference enhancement effect when the two ends form an F-P cavity.

 

Fig. 4. (a) Schematic map of Y-branch power splitter; (b) Output mode pattern of the two branches.

Download Full Size | PDF

A Y-branch power splitter was also fabricated to apply this technique into a functional element of PLCs. The configuration of the Y-branch power splitter is shown in Fig. 4(a), its length is 3 cm and two output ports are separated by 250 µm. The output mode patterns of the two branches are shown in Fig. 4(b). The insertion losses of the two branches at 1550 nm are 6.40 dB and 6.65 dB, respectively. Therefore, the uniformity is 0.25 dB. This imbalance is due to the non-perfection of the fabrication process. The excess loss of a 3 cm long Y-splitter is 3.52 dB, calculated from the measured insertion loss of the two branches. Compared with the 2.99 dB insertion loss of the channel waveguide with the same length, the additional loss caused by the Y junction is 0.53 dB, which is close to the typical reported value of 0.4 dB [16]. The PDL of the splitter were also measured for the two branches. Both are around 0.50 dB, similar to that of the channel waveguide.

4. Conclusion

Low-loss channel waveguides were successfully fabricated in silica-on-silicon by a combination of ion-exchange and sol-gel thin film techniques. A 3 cm long waveguide has a total inserting loss of 2.99 dB and a PDL of 0.4 dB. A Y-branch power splitter with additional loss of 0.53 dB and uniformity of 0.25 dB was also realized by this technique. These results show that the combination of ion-exchange with sol-gel techniques makes the conventional ion-exchange method being suitable for producing silica-on-silicon devices and makes ion-exchanged waveguides more viable for application in commercial processes. We believe that this thin film ion-exchange technique can also be extended to other thin film techniques.