An intra-chip coupling scheme from optical fibers to silicon strip waveguides is demonstrated. The couplers consist of silicon inverse width tapers embedded within silicon dioxide cantilevers that are deflected out-of-plane by residual stress. Deflection angles from 5 to 30 degrees are obtained and controlled by thermal annealing. Butt-coupling from tapered fibers or collimation-coupling from lensed fibers may be employed. The coupling scheme enables direct access to devices on the entire chip surface without dicing or cleaving the chip. Coupling efficiencies of 1.6 dB per connection for TE polarization and 2 dB per connection for TM polarization are achieved. The coupling efficiency shows little wavelength-dependence, with less than 1.6 dB fluctuation over the wavelength range of 1500 nm to 1560 nm.
©2009 Optical Society of America
Silicon photonics is a promising candidate for large-scale integrated optics . It is compatible with well-developed and cost-efficient CMOS technology and can be integrated with electronic devices monolithically. Single-mode silicon strip waveguides surrounded by SiO2 cladding have submicron cross-sections of typically 450 nm by 250 nm due to the high contrast between the refractive indices of Si (nSi=3.48) and SiO2 (nSiO2=1.44). Highly confined optical modes allow for densely integrated waveguides and small radius of curvature waveguide bends. The high confinement, however, also produces challenges when attempting to couple light between silicon strip waveguides and optical fibers. Mode conversion between single-mode fibers with mode field diameters (MFD) of 10 μm and silicon strip waveguides with MFDs of 0.5 μm can result in coupling losses of more than 20 dB.
Several coupling schemes have been proposed to bridge the mode size mismatch between silicon strip waveguides and optical fibers. MFDs of optical fibers can be reduced by exploiting tapered fibers  or lensed fibers [3,4]. Tapered fibers may be fabricated by etching silica optical fibers with concentrated hydrofluoric (HF) acid. Constant evaporation during the etching process creates a concentration gradient beneath the liquid surface. MFDs of tapered fibers can be reduced to 0.8 μm by controlling the etch time . Lensed fibers employ endface microlenses realized by laser-shaping  or mechanical polishing . MFDs of lensed fibers are limited by the light wavelength in air. At 1.55 μm wavelength, commercially available lensed fibers have minimum attainable MFDs of ~2.5 μm. Improved focusing can be obtained through the use of high-refractive-index coatings deposited on the microlenses where MFDs can be curtailed to 1.6 μm . Alternatively, MFDs of silicon strip waveguides can be increased by means of waveguide tapers [6–9] or inverse width tapers [10–13]. Silicon waveguide tapers expand optical modes by adiabatically broadening the waveguides. To maintain the mode shape at the taper terminals, both the waveguide width and the height are increased simultaneously. Inverse width tapers, on the other hand, reduce silicon waveguides to sharp tips, where the fundamental waveguide modes are cut-off. Evanescent modes are delocalized from the waveguides thereby producing larger MFDs. Inverse width tapers can be defined in the same lithography process as the silicon waveguides and can have coupling losses as low as 0.5 dB per connection .
One disadvantage of waveguide tapers or inverse width tapers is that they have to be terminated on cleaved or diced edges for coupling to fibers. To couple light out-of-plane, alternative schemes may be employed. One scheme exploits surface diffraction gratings and width tapers [15–17]. Grating couplers have relatively low coupling efficiency and narrow bandwidth. For example, 7 dB coupling loss and 60 nm 3dB-bandwidth were reported in , and 2.5 dB coupling loss and 20 nm 3dB-bandwidth were reported in . The performance of grating couplers can be improved by employing double grating structures  or bottom reflectors [19,20]. With bottom mirrors, 1 dB coupling losses and 60 nm bandwidth have been theoretically predicted . Another scheme employs prisms [21, 22] or tapered-curved fibers for evanescent coupling [23–26]. For prism couplers, the input beam is deflected by total internal reflection, where a small low-index gap between prism and waveguide is maintained. Light tunnels through the gap and is coupled to the underlying waveguides. For tapered-curved fiber couplers, optical fiber is tapered down at intermediate points to diameters of 0.8~1.5 μm and bent to radii of curvature of ~50 μm. The tapered-curved portion of the fiber piggybacks onto a waveguide on the chip surface with gaps of ~100 nm in-between. Coupling efficiencies of 98% have been achieved for such evanescent coupling techniques .
In this paper, fiber-to-chip couplers that consist of stress-engineered SiO2 cantilevers, at the center of which are conventional silicon inverse width tapers, are demonstrated. The freestanding ends of the SiO2 cantilevers are deflected upward by stress, enabling out-of-plane intra-chip coupling to tapered or lensed fibers. Control of the SiO2 cantilever deflection through the use of thermal annealing is demonstrated. The mode conversion from fiber to cantilever and the bending loss are also examined. Transmission measurements are taken to determine the coupling efficiency, scattering loss, and wavelength dependence.
The paper is organized as follows. Section two discusses the coupler design, the fiber-to-coupler mode conversion, and the cantilever bending loss. Section three describes the fabrication process and the control of the cantilever profile. Section four describes the experimental characterization. Section five gives concluding remarks.
The couplers consist of SiO2 cantilevers and Si inverse width tapers, as shown in Fig. 1. The Si inverse width tapers are embedded in the center of the SiO2 cantilevers which are comprised of buried oxide (BOX) and plasma-enhanced chemical vapor deposition (PECVD) SiO2. The SiO2 cantilevers are released from the substrate at the terminations of the Si inverse width tapers. The PECVD SiO2 layer functions as an optical cladding for the Si waveguides and as a stress layer to deflect the free-standing SiO2 cantilevers. The stress in the BOX-PECVD SiO2 bilayer deflects the SiO2 cantilevers out-of-plane. Tapered or lensed optical fibers are used to couple to the SiO2 cantilevers. Coupling of light from the Si strip waveguides to optical fibers consists of two mode conversions, as shown in Fig. 2. The first mode conversion is from the Si waveguide mode to the delocalized mode at the end of the Si inverse tapers. The second mode conversion is from the delocalized mode to the optical fiber mode. The distance between the endface of the SiO2 cantilevers and the silicon inverse tapes is 1 to 2 μm. The Si inverse tapers are not exposed to air on the endfaces because the etching processes required to release the SiO2 cantilevers also attack the Si inverse tapers.
The SiO2 cantilevers are designed sufficiently long such that the released ends are deflected enough above the chip surface to allow for out-of-plane coupling to tapered or lensed fibers. Preliminary experiments showed that for the fabrication processes and the measurement setup described in Sections 3 and 4, the cantilever length should be greater than 40 μm. The width of the SiO2 cantilevers is designed to accommodate the Si waveguide mode delocalized by the Si inverse tapers. A Si inverse width taper buried in SiO2 cladding is simulated with beam propagation method (BPM). The waveguide end of the taper is 450 nm in width and 250 nm in height, and the reduced end of the taper is 100 nm in width. The inverse taper is 40 μm long and has a quadratic taper profile. Contour maps of the simulated electric fields at the tip of the Si inverse taper are shown in Fig. 3. It can be deduced from Fig. 3 that the delocalized TE mode has an MFD of ~1.5 μm, and the delocalized TM mode has an MFD of ~0.8 μm. The delocalized TE and TM modes are well confined within a 4 μm wide by 2.1μm high cross section, which is denoted in Fig. 3 by white dashed lines. In the following fabrication process, SiO2 cantilevers with nominal length of 40 μm and width of 4 μm are fabricated. The estimated coupling efficiency, for a tapered fiber MFD of 1.5 μm and a cantilever endface to taper tip distance of 2 μm, is -0.35 dB (TE) and -0.74 dB (TM). The bending loss of the SiO2 cantilevers is worth consideration since dielectric waveguides cannot guide light around bends without losing power . It is shown in Section 3 that the maximum deflection of the 40 μm long cantilevers is 12 μm, which corresponds to a radius of curvature of 67 μm. The bending loss calculated using effective index method (EIM) and the weak guidance approximation  can be safely ignored for this radius of curvature. When the deflection is smaller than 12 μm, the bending loss will be even smaller.
Test structures for optical characterization of the SiO2 cantilever couplers were fabricated on a silicon-on-insulator (SOI) wafer by electron-beam and ion-beam lithography processes. First, silicon strip waveguides were fabricated. An SOI wafer with 250 nm Si and 1 μm BOX was cleaned and dehydrated. HSQ in MIBK (Dow Corning XR-1541) was spun-coated at 3000 rpm for 45 sec. The residual solvent in the HSQ film was driven out by soft-bake on hotplates at 120 °C for 2 min and then at 220 °C for 2 min. Si strip waveguides were defined using an electron-beam lithography (Leica EBPG-5000) tool at 50 kV. Two types of silicon strip waveguides were fabricated, one with inverse width tapers at the input and output ends and the other without inverse tapers at either end. After exposure the resist film was developed in 0.26 N TMAH solution. The resist patterns were transferred to the 250 nm thick silicon top layer by inductively-coupled plasma etching with HBr-chemistry. The HBr etching automatically stops at the BOX surface with Si/SiO2 etching selectivity of more than 20. Figure 4 shows scanning electron micrographs of silicon waveguides after the HBr etching. Second, SiO2 cantilevers were fabricated with the process flow shown in Fig. 5. A 1.1 μm SiO2 thin film was deposited by PECVD with SiH4-N2O chemistry at 200 °C. A 150 nm titanium-nickel mask was then evaporated on top of the PECVD SiO2. The patterns for the SiO2 cantilevers were written directly on the Ti/Ni mask by focused Ga+ ion beam (FIB) milling at 30 kV with a milling depth of 250 nm. The metal mask patterns were transferred to the SiO2 layer by reactive ion etching using SF6 chemistry. The etch recipe was tuned to etch SiO2 anisotropically and to etch silicon isotropically with large undercut in order to fully release the cantilevers from the substrate. Finally the Ti/Ni mask was removed with HNO3 and HCl solutions.
Figure 6 shows scanning electron micrographs of the SiO2 cantilevers after removal of the Ti/Ni mask. The final cross-sectional dimensions of the SiO2 cantilevers were 3.85 μm in width and 2.1 μm in height, as shown in Fig. 6(a). The released ends of the SiO2 cantilevers were deflected by 2.6 μm from their original positions. The cantilevers have a maximum tilt angle of 4.7° at the end, which is within the 6° travel of the pitch angle of our 6-axis fiber positioning stages. To provide mechanical support 5 μm long and 1.2 μm wide SiO2 struts are used as anchors, as shown in Fig. 6(b). The struts will deform slightly under stress. By increasing the length of the struts the cantilever deflection can be increased for a given cantilever length or stress.
To verify the stress source that deflects the SiO2 cantilevers, similar cantilevers were fabricated on two different SiO2 films. First, SiO2 cantilevers were fabricated on a blank area of the same chip that was used to fabricate the actual test waveguides. These cantilevers consist of the same BOX and PECVD SiO2 layers, but do not have Si waveguides in the center. Second, a bare SOI wafer with 250 nm Si and 1 μm BOX was cleaned and the Si top layer was completely removed by TMAH wet etching. SiO2 cantilevers that consist of only BOX were fabricated with the same method as that described in Fig. 5. The BOX-PECVD SiO2 cantilevers that contain no Si core had the same bending profile as that of the cantilevers with Si cores. The cantilevers that consist of only BOX did not deflect. It can be concluded that the stress in the BOX-PECVD SiO2 bilayer accounts for the deflection of cantilevers.
The cantilever profile is expected to change if the PECVD SiO2 deposition migrates to other PECVD platforms. The cantilever or strut length may be increased to produce an increase in deflection. Alternatively, post-fabrication modification of cantilever profiles may be utilized. Active controls of cantilever profiles have been reported with electrostatic  and piezoelectric actuators . As optical couplers, the cantilevers in this paper can be controlled passively and statically by means of thermal annealing . It was reported  that the residual stress in the PECVD SiO2 can be controlled by rapid thermal annealing (RTA). When being annealed at higher temperature than its deposition temperature, non-stoichiometric PECVD SiO2 will release impurities and undergo network reconstruction, which will change the stress in the PECVD SiO2 film. This process will cease within 15 sec for a given temperature.
To explore thermal control, the chip was annealed with an RTA in N2 ambient for 1 min at 400 °C, 600 °C, and 800 °C sequentially. After annealing at each temperature, the cantilever profile was measured by scanning electron microscopy. The measurement results showed that the residual stress in the BOX-PECVD SiO2 bilayer increases with increasing annealing temperature, as shown in Fig. 7. For as-fabricated SiO2 cantilevers the tilt angle was only 4.7° and the deflection at the end of the cantilevers was only 2.6 μm. In this case, only tapered fibers can be used for butt-coupling. After annealing at 800 °C, the tilt angle increased to 30°. The larger tilt angle allows lensed fibers with conical angle greater than 60° to be used for light coupling.
An important issue regarding practical applications of the cantilevers is uniformity and repeatability of the SiO2 cantilevers. Since the deflection is mainly caused by the stress in the BOX-PECVD SiO2 structure, it is expected that cantilever profiles will be the same in regions where the stress is uniform. After release from the silicon substrate, the SiO2 cantilevers have not undergone noticeable change after 3 months in laboratory ambient.
The experimental setup for optical characterization is shown in Fig. 8(a). An infrared continuous-wave laser source was first connected to a polarization controller which outputs linearly polarized TE or TM light with cross-polarization rejection ratio of more than -17 dB. Two tapered optical fibers with tip diameters of ~1.5 μm were fabricated and mounted on 6-axis positioning stages. Light in the output fiber was collected by a photodetector and measured by a power meter. The tapered fibers were tilted to match the tilt angle of the SiO2 cantilevers, and then butt-coupled to the input and output cantilever couplers of a waveguide, as shown in Fig. 8(b). One visible microscope was mounted at an oblique angle of ~30° for aligning tapered fibers to the cantilever couplers. A second visible-infrared microscope was mounted on top of the chip vertically to further assist fiber alignment and to monitor light injection in waveguides, as shown in Fig. 8(c).
Transmission measurements were taken on silicon waveguides of various lengths. The insertion loss measured at 1.55 μm wavelength is plotted versus waveguide length in Fig. 9(a) and 9(b) for TE and TM polarization respectively. The waveguide length includes the 40 μm input and 40 μm output cantilever couplers. Each waveguide was measured four times independently. Upper and lower bounds, due primarily to fiber alignment errors, are denoted by vertical bars. The discrete data points are fitted by linear regression. Coupling loss per pair of cantilever couplers is extracted from the Y-interception of the regression line, and the propagation loss of the Si waveguides per unit length is extracted from the slope of the regression line. As shown in Figs. 9(a) and 9(b), the propagation loss of the Si waveguides is 9.5 dB/cm for TE polarization and 16.5 dB/cm for TM polarization. The fiber-to-waveguide coupling loss is 1.6 dB per connection for TE polarization and 2 dB per connection for TM polarization. The insertion loss of the waveguides with inverse tapers at both ends was improved over that of the waveguides without tapers at either end by 16 dB for TE polarization and 7.8 dB for TM polarization given the same waveguide length.
It is worth noting that the abrupt decrease of insertion loss of waveguides without tapers is caused by the formation of air-backed SiO2 microfibers. When the waveguide length is close to the total length of the input and output cantilever couplers, all or most of the Si waveguide is in the air-backed SiO2 cantilevers and the cantilevers function as a microfiber. Since there is no Si inverse taper for mode conversion, most of the light power will propagate in the SiO2 microfibers rather than in the Si waveguides.
The spectral response of Si waveguides with inverse tapers at both ends is shown in Fig. 10. When the Si waveguides are very short, Fabry-Perot cavity effects becomes significant. For the case of an 80 μm long waveguide, the structure consists of two back-to-back silicon inverse width tapers. Reflections occur everywhere along the tapers so that the two tapers form a distributed Fabry-Perot cavity. The measured free spectral range is consistent with finite difference time domain simulations. As the Si waveguide length increases, the Fabry-Perot cavity effect diminishes. For a 750 μm long Si waveguide, the insertion loss changes by less than 1.6 dB over the wavelength range of 1500 nm to 1560 nm, illustrating little dependence of the coupling efficiency on wavelength.
An intra-chip cantilever coupler using stress-engineered SiO2 cantilevers for fiber-to-chip coupling to Si photonic circuits is proposed, fabricated, and characterized. PECVD SiO2 is deposited on silicon photonic circuits to serve as both an optical cladding and as a stress layer. SiO2 cantilevers with embedded Si strip waveguide terminations were fabricated and released. The residual stress in the BOX-PECVD SiO2 bilayer deflects the cantilevers out-of-plane and enables butt-coupling to tapered optical fibers. The tilt angle of as-fabricated cantilevers is 4.7° and the deflection at the end of the cantilevers is only 2.6 μm. Only tapered fiber can be used for butt-coupling in this case. By employing thermal annealing at 800 °C, the tilt angle is increased to 30°, allowing coupling by lensed fibers with large conical angles. The coupling method demonstrated in this work enables direct access to devices on an entire chip surface without dicing or cleaving. Coupling losses of 1.6 dB per connection for TE polarization and 2 dB per connection for TM polarization are achieved. The coupling efficiency shows little wavelength-dependence, with less than 1.6 dB fluctuation over the wavelength range of 1500 nm to 1560 nm.
This work was supported by NSF Grant ECCS-0725657.
References and links
1. H. Yamada, T. Chu, S. Ishida, and Y. Arakawa, “Si photonic wire waveguide devices,” IEEE J. Sel. Top. Quantum Electron. 12, 1371–1379 (2006). [CrossRef]
2. T. Alder, A. Stohr, R. Heinzelmann, and D. Jager, “High-efficiency fiber-to-chip coupling using low-loss tapered single-mode fiber,” IEEE Photon. Technol. Lett. 12, 1013–1015 (2000). [CrossRef]
4. G. Milton, Y. Gharbia, and J. Katupitiya, “Mechanical fabrication of precision microlenses on optical fiber endfaces,” Opt. Eng. 44, 123402 (2005). [CrossRef]
5. N. Kawasaki, M. Umetsu, H. Yoda, H. Tsuchiya, K. Shiraishi, K. Watanabe, and S. Shikano, “A novel lensed fiber with a focused spot diameter as small as the wavelength,” in Proceedings of IEEE Conference on Optical Fiber Communication and the National Fiber Optic Engineers Conference (Institude of Electrical and Electronics Engineers, New York, 2007), paper OWI2. [CrossRef] [PubMed]
6. D. X. Dai, S. L. He, and H.-K. Tsang, “Bilevel mode converter between a silicon nanowire waveguide and a larger waveguide,” J. Lightwave Technol. 24, 2428–2433 (2006). [CrossRef]
7. J. K. Doylend and A. P. Knights, “Design and simulation of an integrated fiber-to-chip coupler for silicon-on-insulator waveguides,” IEEE J. Sel. Top. Quantum Electron. 12, 1363–1370 (2006). [CrossRef]
8. V. Nguyen, T. Montalbo, C. Manolatou, A. Agarwal, C.-Y. Hong, J. Yasaitis, L. C. Kimerling, and J. Michel, “Silicon-based highly-efficient fiber-to-waveguide coupler for high index contrast systems,” Appl. Phys. Lett. 88, 081112 (2006). [CrossRef]
9. K. Shiraishi, H. Yoda, A. Ohshima, H. Ikedo, and C. S. Tsai, “A silicon-based spot-size converter between single mode fibers and Si-wire waveguides using cascaded tapers,” Appl. Phys. Lett. 91, 141120 (2007). [CrossRef]
10. T. Shoji, T. Tsuchizawa, T. Watanabe, K. Yamada, and H. Morita, “Low loss mode size converter from 0.3μm square Si wire waveguides to singlemode fibres,” Electron. Lett. 38, 1669–1670 (2002). [CrossRef]
12. K. K. Lee, D. R. Lim, D. Pan, C. Hoepfner, W.-Y. Oh, K. Wada, L. C. Kimerling, K. P. Yap, and M. T. Doan, “Mode transformer for miniaturized optical circuits,” Opt. Lett. 30, 498–500 (2005). [CrossRef] [PubMed]
13. K. Yamada, T. Tsuchizawa, T. Watanabe, J. Takahashi, E. Tamechika, M. Takahashi, S. Uchiyama, H. Fukuda, T. Shoji, S. Itabashi, and H. Morita, “Microphotonics devices based on silicon wire waveguiding system,” IEICE Trans. Electron. 87, 351–358 (2004).
14. S. McNab, N. Moll, and Y. Vlasov, “Ultra-low loss photonic integrated circuit with membrane-type photonic crystal waveguides,” Opt. Express. 11, 2927–2939 (2004). [CrossRef]
15. D. Taillaert, W. Bogaerts, P. Bienstman, T. F. Krauss, P. Van Daele, I. Moerman, S. Verstuyft, K. De Mesel, and R. Baets, “An out-of-plane grating coupler for efficient butt-coupling between compact planar waveguides and single-mode fibers,” IEEE J. Quantum Electron. 38, 949–955 (2002). [CrossRef]
16. G. Roelkens, P. Dumon, W. Bogaerts, D. Van Thourhout, and R. Baets, “Efficient silicon-on-insulator fiber coupler fabricated using 248-nm-deep UV lithography,” IEEE Photon. Technol. Lett. 17, 2613–2615 (2005). [CrossRef]
17. L. Vivien, D. Pascal, S. Lardenois, D. Marris-Morini, E. Cassan, F. Grillot, S. Laval, J.-M. Fedeli, and L. El Melhaoui, “Light injection in SOI microwaveguides using high-efficiency grating couplers,” J. Lightwave Technol. 24, 3810–3815 (2006). [CrossRef]
18. G. Masanovic, G. Reed, W. Headley, B. Timotijevic, V. Passaro, R. Atta, G. Ensell, and A. Evans, “A high efficiency input/output coupler for small silicon photonic devices,” Opt. Express 13, 7374–7379 (2005). [CrossRef] [PubMed]
20. P. Cheben, S. Janz, D. X. Xu, B. Lamontagne, A. Delage, and S. Tanev, “A broad-band waveguide grating coupler with a subwavelength grating mirror,” IEEE Photon. Technol. Lett. 18, 13–15 (2006). [CrossRef]
21. R. Ulrich, “Theory of the prism-film coupler by plane-wave analysis,” J. Opt. Soc. Am. 60, 1337–1350 (1970). [CrossRef]
23. I. K. Hwang, S. K. Kim, J. K. Yang, S. H. Kim, S. H. Lee, and Y. H. Lee, “Curved-microfiber photon coupling for photonic crystal light emitter,” Appl. Phys. Lett. 87, 131107 (2005). [CrossRef]
24. C. Grillet, C. Smith, D. Freeman, S. Madden, B. Luther-Davies, E. Magi, D. J. Moss, and B. Eggleton, “Efficient coupling to chalcogenide glass photonic crystal waveguides via silica optical fiber nanowires,” Opt. Express 14, 1070–1078 (2006). [CrossRef] [PubMed]
25. C. Grillet, C. Monat, C. Smith, B. Eggleton, D. Moss, S. Frederick, D. Dalacu, P. Poole, J. Lapointe, G. Aers, and R. Williams, “Nanowire coupling to photonic crystal nanocavities for single photon sources,” Opt. Express 15, 1267–1276 (2007). [CrossRef] [PubMed]
26. C. P. Michael, M. Borselli, T. J. Johnson, C. Chrystal, and O. Painter, “An optical fiber-taper probe for wafer-scale microphotonic device characterization,” Opt. Express 16, 4745–4752 (2007). [CrossRef]
27. D. Marcuse, Light Transmission Optics, (Van Nostrand Reinhold, 1982), Chap. 9.
28. E. Ollier, “Optical MEMS devices based on moving waveguides,” IEEE J. Sel. Top. Quantum Electron. 8, 155–162 (2002). [CrossRef]
29. J. S. Pulskamp, A. Wickenden, R. Polcawich, B. Piekarski, M. Dubey, and G. Smith, “Mitigation of residual film stress deformation in multilayer microelectromechanical systems cantilever devices,” J. Vac. Sci. Technol. B 21, 2482–2486 (2003). [CrossRef]
30. J. W. Noh, R. Anderson, S. H. Kim, J. Cardenas, and G. P. Nordin, “In-plane photonic transduction of silicon-on-insulator microcantilevers,” Opt. Express 16, 12114–12123 (2008). [CrossRef] [PubMed]
31. R. Charavel, B. Olbrechts, and J. P. Raskin, “Stress release of PECVD oxide by RTA,” Proc. SPIE 5116, 596–606 (1993). [CrossRef]