Development of the SC-RTA Process for Fabrication of Sol-Gel Based Silica-on-Silicon Integrated Optic Components

Transcription

1 Journal of Sol-Gel Science and Technology 13, (1998) c 1998 Kluwer Academic Publishers. Manufactured in The Netherlands. Development of the SC-RTA Process for Fabrication of Sol-Gel Based Silica-on-Silicon Integrated Optic Components R.R.A. SYMS, A.S. HOLMES, W. HUANG, V.M. SCHNEIDER AND M. GREEN Optical and Semiconductor Devices Section, Dept. of Electrical and Electronic Engineering, Imperial College, Exhibition Road, London SW7 2BT, UK Abstract. The SC-RTA process for fabricating silica-on-silicon PLCs from sol-gel glass is described. A wide range of glasses has been deposited, process temperatures have been reduced, and components fabricated by reactive ion etching, reflow and burial of channel guides have shown steadily decreasing loss. Propagation losses are 0.2 db/cm at λ = µm in a high n system. Passive components demonstrated include tree-structured power splitters and thermo-optic switches. Keywords: sol-gel glass, silica-on-silicon, integrated optics, planar lightwave circuit 1. Introduction Several processes have been developed for depositing the thick silicate glass layers used in silica-on-silicon integrated optics, the most promising system for lowcost planar lightwave circuits (PLCs). The most advanced are flame hydrolysis (FHD) [1], chemical vapor (CVD) [2] and plasma-enhanced chemical vapor deposition (PECVD) [3]. The search for suitable glasses has paralleled work on silica-based fibers. In these, loss is dominated by Rayleigh scattering at short wavelengths, O H absorption at near-infrared wavelengths and absorption from the vibrational bands of the dopants at long wavelengths. Scattering is minimized by avoiding dopants prone to crystallization (particularly, TiO 2 ); OH contamination by dehydration, and IR absorption by adopting the dopant whose band is furthest in the infrared, GeO 2. O H absorption occurs in PLCs, but can be removed by annealing. Other losses include coupling loss, substrate loss, scattering from core walls and bend loss. Low-loss coupling to single-mode fibers is only obtained with large (6 8 µm square) and weakly confining ( n/n = %) cores. Minimization of substrate loss requires a thick buffer layer (20 µm inlow nsystems). Sidewall scattering is reduced by reflowing the core; this requires the core glass melting point to be lowered, e.g., by addition of P 2 O 5. However, the P 2 O 5 concentration is limited, and further temperature reduction requires addition of B 2 O 3. Bend loss is reduced using smaller, more strongly confining cores. TiO 2, GeO 2 and As 2 O 3 all allow a large index change. The best composition is unclear, but GeO 2 -B 2 O 3 - P 2 O 5 -SiO 2 gives excellent results in FHD [1]. The sol-gel process is an alternative method of depositing glassy films, based on the hydrolysis and polycondensation of alkoxides. Three stages are involved: formation of a particulate suspension (a sol), processing to form a solid network permeated by liquid (a gel), and drying to form a porous glass. For silica, the overall reaction is Si(OR) 4 + 2H 2 O SiO 2 + 4ROH, where R denotes an alkyl group. This reaction proceeds in steps. The alkoxide is first hydrolyzed, so that several OR groups are replaced by OH. Mixing the alkoxide with a quarter of the stoichiometric ratio of water and refluxing with a catalyst (e.g., HCl) will replace on average one group; repeating the process replaces two. These reactions may be slowed by dilution with solvent. When some groups have been hydrolyzed, polymerization can take place. The byproduct is water, which causes further hydrolysis and

2 510 Syms et al. Figure 1. SC-RTA process for channel guide silica-on-silicon PLC fabrication. polymerization. This reaction can also be slowed by dilution; however, when the sol is coated onto a substrate, the solvent is driven off, causing gelation. The gel is then baked to form a glass. Sol-gel films have been used in antireflection coatings [4]. Planar waveguides have also been formed from sol-gel glass on SiO 2 substrates [5]. Porous material has been used as a host for sensing dyes [6], optically nonlinear organic [7] and semiconductor [8] materials, and rare-earth [9, 10] dopants. Channel guides have also been fabricated from planar layers by CO 2 laser irradiation [11] and by cross-linking an embedded photopolymer [12]. In sol-gel films, shrinkage often causes cracking when the thickness exceeds a few µm, especially on Si substrates. Cracking is avoided in thick (>30 µm) films using repetitive spin-coating and rapid thermal annealing (SC-RTA) [13]. These films have been formed into channel guide PLCs by the conventional topographic method. Figure 1 shows the process. A thick buffer of doped silica is first deposited, followed by a thinner layer of higher index material to act as the guide. These films are built up as multilayers. At each iteration, the wafer is spin-coated, heated by a rapid thermal annealer, and cooled. The bilayer is then consolidated. The cores are etched and reflowed; further spin-coating is used to bury them, and the cladding is consolidated. This paper describes the evolution of the process, and demonstrates the performance obtained using optimized materials. 2. Sol Preparation, Film Deposition and Glass Properties Work has mainly been performed on proprietary sols, including silica-titania, and phosphosilicate, borosilicate, borophosphosilicate and germanophosphosilicate glass (PSG, BSG, BPSG and GPSG). The number of components has slowly been increased to optimize different glass properties. Alkoxide precursors are used as sources of SiO 2, TiO 2 and GeO 2. Suitable Si alkoxides include silicon ethoxide (TEOS), silicon methoxide and methyl triethoxysilane; in our work, we have used TEOS. Ti alkoxides include titanium ethoxide, isopropoxide (TPOT) and butoxide; we have used TPOT. For GeO 2, we have used germanium ethoxide (TEOG) [14]. Trialkyl borates are suitable precursors for B 2 O 3, while P 2 O 5 can be derived from trialkyl phosphites and phosphates. However, we have used the alternative of oxide solutions in alcohol [15]. Sols are prepared using a two-step acid-catalyzed hydrolysis [6]. TEOS is first dissolved in alcohol. An HCl solution is added to achieve a water : alkoxide molar ratio of unity (R = 1), and the alkoxide is refluxed at 70 C for 2 hours. A similar mixture containing the primary dopant precursor is then added, together with sufficient HCl to take the water : alkoxide ratio to R = 2, and the solution refluxed again. An alcohol mixture containing the secondary dopant (if required) is then added. Since these solutions are often highly reactive, no further catalyst is used. For example, in a recent process for GPSG the primary dopant was P 2 O 5 in IPA, and the secondary dopant a TEOG solution. The latter was so reactive that the sol required dehydration using a molecular sieve before its addition [16]. Films are deposited by spin-coating for 40 s, and heated for 10 s in O 2. The single-layer thickness t f is determined by the spin speed ω and the sol viscosity, typically following the variation t f k(1 v a )ω γ, where k is a constant, v a is the volume fraction of solvent and γ 0.5. Multilayers are built up from films with thicknesses of 2500 to 5000 Å. Below this range, the number of layers is excessive, while above it, beading of the wafer rim occurs. Good thickness

3 Development of the SC-RTA Process 511 Figure 2. Cracking patterns in silica-titania (LHS) and germanosilicate glass on (100) Si. uniformity is obtained, but there is a slow radial thickness variation and a faster angular variation due to striations [13]. Even thin ( 2 µm) films on Si tend to crack, due to tensile stress. For (100) wafers, the cracks are uniformly distributed and oriented at 45 to the intersection of {111}planes with the surface, as in the LHS of Fig. 2. Since the stress generally becomes compressive as the anneal temperature rises, it can be made compressive during annealing itself by careful control of the RTA temperature [17]. The stress at room temperature (σ f ) can be found from the wafer curvature, and the stress during annealing (σ i ) estimated from the mismatch in expansion coefficient between glass and substrate. Figure 3 shows the variation of σ f and σ i with anneal temperature T A for a silica-titania glass [17]. σ f is tensile below a critical temperature T AC 1000 C. Above this, it is compressive. However, σ i is still tensile until T AC 1075 C. Near T AC, there is a dramatic rise in the thickness to failure. The thickness is then limited by other factors. Silica-titania films are limited to 15 µm Figure 3. Variation of film stress, intrinsic stress and thickness with anneal temperature, for silica-titania.

4 512 Syms et al. Figure 4. Variation of RTA and reflow temperature with composition, for the BPSG system. Figure 5. Variation of refractive index with composition for silica glass containing different dopants. by the wafer distortion and lattice damage caused by cycling to 1100 C, which interferes with the formation of fiber alignment grooves by anisotropic etching. P 2 O 5 and B 2 O 3 both lower the anneal temperature, allowing thicker (>30 µm) films to be made [18 20]. Figure 4 shows the anneal temperature for the BPSG system; a reduction of more than 300 C can be achieved by heavy doping. One explanation is that both dopants increase the expansion coefficient of SiO 2, reducing the thermal stress. Dopant concentrations are limited, however. Heavily-doped PSG is hygroscopic, and the maximum practical P 2 O 5 content is 10 mol% [20]. Films containing more than 10 mol% B 2 O 3 often appear milky after spin-coating, possibly due to the formation of polycrystalline B 2 O 3. Films containing a high level of B 2 O 3 and P 2 O 5 can also become opaque at high temperatures. The optimum compositional range is the clear area in Fig. 4. Additional annealing causes a gradual decrease in thickness and porosity, coupled with changes in stress [17]. Further processing is therefore required for consistent properties. Consolidation in a tube furnace

5 Development of the SC-RTA Process 513 eliminates residual porosity, and forces the thickness and stress to stable values. For waveguides, appropriate refractive index differences must be established between core and cladding. Figure 5 shows the variation of index with concentration for different dopants. TiO 2 and P 2 O 3 increase the index, while B 2 O 3 decreases it. GeO 2 also raises the index. Unfortunately, we have been unable to form thick films of pure germanosilicate glass, since the rapid reaction of TEOG causes uneven gelation and cracking. In contrast to silica-titania, the cracks are sinusoidal, and follow the radial striations as on the RHS of Fig. 2. This cracking mode can be controlled using P 2 O 5 [16]. Channel guide cores are fabricated by reactive ion etching in a Plasma Technology RIE 80 using CHF 3, Ar and O 2, with a Cr mask. Silica-titania cores can be etched with almost vertical sidewalls [13]; however, some undercut occurs with PSG [18]. Generally, modest RF power (140 W) and pressure (50 mtorr) are used, but high power and low pressure are needed to avoid micromasking in glasses forming involatile byproducts (e.g., GeO 2 -SiO 2 ) [16]. Compositions are chosen so that cores can be reflowed without melting the buffer. Because TiO 2 has little effect on the viscosity of SiO 2, the reflow temperature of silica-titania is very high (1260 C), and core and cladding melt together. Furthermore, silica-titania crystallizes at high temperatures. However, P 2 O 5 and B 2 O 3 lower the melt temperature. Figure 4 shows data for the BPSG system, for a 5 minute reflow in O 2. Temperatures of 950 C can be achieved by heavy doping [20]. 3. Passive Channel Waveguides and Devices For low fiber-device-fiber insertion loss at near-ir wavelengths, all sources of loss must be systematically reduced. Low scatter requires avoidance of crystallization, and the ability to control melt temperature. Low absorption requires elimination of OH contamination. Low substrate loss requires a thick buffer (in turn needing low anneal temperatures) and a high core-cladding index difference n, so that the guided mode is well confined and spaced away from the substrate. High n is also needed for low bend loss. Low coupling loss actually requires low n; however, waveguide tapers may overcome this conflict. Low substrate damage is a prerequisite for V-groove etching. Channel guides have been constructed from (1) silica-titania [13], (2) PSG [18], (3) BPSG [19, 20] and (4) GeO 2 -B 2 O 3 -P 2 O 5 -SiO 2 (GBPSG) [16]. In each case, a core width of 7 µm was used; however, the layer thicknesses differ from system to system, typically being the largest practical. In the last two systems, different compositions have been used for all three glass layers; most recently, PSG has been used for the buffer, GPSG for the core, and BPSG for the cladding. Steadily improving optical performance has been obtained. Losses are highest in silica-titania, with a significant differential between the TE and TM modes. The main loss is substrate absorption, due to the restriction on buffer thickness caused by the high RTA temperature. However, other losses are significant. Figure 6 shows a spectral loss variation for 3.5 cm length of Figure 6. Spectral loss variation of guides formed in silica-titania, PSG and GBPSG.

6 514 Syms et al. Figure 7. Process temperatures of guides formed in silica-titania, PSG and GBPSG. Figure 8. Variation of insertion loss with transition length for back-to-back S-bends with low and medium n cores formed in BPSG, and high n cores formed in GBPSG. silica-titania guide [16]. The increase in loss at long wavelengths is due to substrate absorption, while the rise at short wavelengths is caused by scattering following from incomplete reflow. There is also significant O H absorption near λ = 1.39 µm. In PSG, substrate and scattering losses are reduced due to the increased layer thickness allowed by a reduced RTA temperature, and due to improved reflow. Figure 6 shows a further advantage: there is no O H absorption. One explanation is that excess water is retained in volatile phosphorus compounds, and driven off during the RTA step. These characteristics are retained in BPSG, but substrate and scattering losses are reduced further because of temperature reductions. Polarization-independent losses of 0.2 db/cm at µm were obtained using this system [19]. However, the usefulness of both PSG and BPSG is limited by the small n achievable (due to restrictions on the doping level) and by deformation of the core during cladding. For higher performance, B 2 O 3 and P 2 O 5 must be relegated to controlling process temperatures, and a further dopant introduced to control the index. GeO 2 is the optimum candidate, since it avoids the problems associated with TiO 2.

7 Development of the SC-RTA Process 515 Figure 9. Variation of insertion loss with number of stages for tree-structured splitters with low n cores formed in BPSG. Figure 10. BPSG. Transmission and switching characteristics of a thermo-optic Mach-Zehnder interferometer with medium n cores formed in Figure 6 shows the low, spectrally flat and polarizationindependent performance obtained with GBPSG [16]. Figure 7 compares the temperatures used in the fabrication of silica-titania, BPSG and GBPSG PLCs, showing the reduction in temperature accompanying this improvement in performance. Low process temperatures also allow fabrication of integrated fiber alignment grooves by anisotropic etching. Figure 7 shows a packaged PLC formed in PSG with an 8-core singlemode ribbon fiber pigtail at each end. Using passive alignment, coupling losses were 1.3 db/facet at λ = µm. The achievable n has great influence on bend losses. Figure 8 compares the loss of back-to-back sinusoidal S-bends in low and medium n BPSG systems and a high n GBPSG system. Shorter transitions, with smaller bend radii, can be achieved in the high n system. Limited work has been performed on passive components. Figure 9 shows the insertion loss of Y-junction based tree-structured power splitters. Although these were formed in low n BPSG, excess losses below 4 db were obtained for 1 4 splitters. Similarly, Fig. 10 shows the performance of a Y-junction based Mach-Zehnder interferometer in medium n BPSG.

8 516 Syms et al. Switching is obtained by the thermo-optic effect, and a thick cladding is required to isolate the guided mode from the Ti heater electrode. A TE mode insertion loss of 3.5 db was achieved for a length of 3.4 cm. Switching times of 2 4 ms were achieved using 0.5 W drive power [20]. 4. Conclusions Sol-gel glasses used to form silica-on-silicon PLCs by SC-RTA have been compared. Each material has shown significant improvement over its predecessor, and the main losses have been identified and eliminated. GBPSG allows low, polarization-independent and spectrally flat propagation loss to be combined with low bend loss. Improving performance has been accompanied by reduced process temperature. The most significant problem remaining is low process yield, caused by glass flakes originating from the wafer rim. Automated equipment is now needed to make SC-RTA competitive with FHD, CVD and PECVD. References 1. M. Kawachi, IEE Proc. (Optoelectronics) 143, 257 (1996). 2. Y.P. Li and C.H. Henry, IEE Proc. (Optoelectronics) 143, 263 (1996). 3. S. Valette et al., Sensors and Actuators A21, 1087 (1990). 4. B.E. Yoldas and T.W. O Keefe, Appl. Opt. 18, 3133 (1979). 5. P.P. Herrmann and D. Wildmann, IEEE J. Quant. Elect. QE-19, 1735 (1983). 6. A. Martin and M. Green, Proc. SPIE 1328, 352 (1990). 7. J-L.R. Noguès and M.V. Moreshead, J. Non-Cryst. Solids 121, 136 (1990). 8. M.A. Fardad et al., IEE Proc. (Optoelectronics) 143, 298 (1996). 9. D. Barbier, R. Almeida, and X. Orignac, 8th World Ceramics Cong. (Florence, paper SVII-1 LO4, 1994). 10. X. Orignac and R.M. Almeida, Silica-based sol-gel optical waveguides on silicon, IEE Proc. (Optoelectronics) 143, 287 (1996). 11. M. Guglielmi, P. Colombo, L.M.D. Espositi, G.C. Righini, and S. Pelli, Proc. SPIE 1513, 44 (1990). 12. C.-Y. Li, J. Chisham, M. Andrews, S.I. Najafi, J.D. Mackenzie, and N. Peyghambarian, Elect. Lett. 31, 271 (1995). 13. A.S. Holmes, R.R.A. Syms, Ming Li, and M. Green, Appl. Opt. 32, 4916 (1993). 14. T.N.M. Bernards, M.J. van Bommel, E.W.J.L. Oomen, and A.H. Boonstra, J. Non-Cryst. Solids 147, 13 (1992). 15. P.N. Kumta and M.A. Sriram, J. Mat. Sci. A28, 1097 (1993). 16. R.R.A. Syms, V.M. Schneider, W. Huang, and M.M. Ahmad, Elect. Lett. 33, (1997). 17. R.R.A. Syms and A.S. Holmes, J. Non-Cryst. Solids 170, 223 (1994). 18. A.S. Holmes and R.R.A. Syms, 8th World Ceramics Cong. (Florence, paper SVII-1 LO9, 1994). 19. R.R.A. Syms, V.M. Schneider, W. Huang, and A.S. Holmes, Elect. Lett. 31, 1833 (1995). 20. R.R.A. Syms, W. Huang, and V.M. Schneider, Elect. Lett. 32, 1233 (1996).