We fabricate a Mach-Zehnder interferometer-based optical isolator using a silicon-wire waveguide with magneto-optic garnet cladding using direct bonding techniques. Using Si-wire waveguides, the size of the device is greatly reduced from that of our previous device. We investigate surface-activated direct bonding with nitrogen plasma treatment, which shows better bonding results than oxygen plasma treatment. A large magneto-optic phase shift of 0.8π and an optical isolation of 18 dB are obtained at a wavelength of 1322 nm.
©2012 Optical Society of America
An optical isolator, ensuring stable operation of semiconductor lasers by protecting them from unwanted back reflections, is important for high-speed photonic network systems. Only a crystallized magneto-optic material such as Yttrium Iron Garnet (YIG) can exhibit efficient magneto-optic effects with low optical absorption. Typically, crystallized YIGs must be epitaxially grown on a garnet substrate. Although several waveguide isolators with magneto-optic garnets have been fabricated and experimentally examined [1–4], none has yet been realized in commercially available devices due to a lack of compatibility with other components. Semiconductor waveguide isolators based on nonreciprocal loss shift in ferromagnetic metals such as Fe or Co are compatible with lasers and can easily be fabricated by depositing the ferromagnetic metal [5, 6]. However, because the large absorption loss of the metal must be compensated by optical amplification, the isolator requires a power supply.
Recent demand for high-speed, low-cost integrated photonic devices is focused on silicon photonics. A nonreciprocal phase shift (NPS), which is a magneto-optic effect for polarized guiding light propagating in a layered structure, can be utilized to achieve an optical isolation in a Mach-Zehnder interferometer (MZI) or a ring resonator. Recently, waveguide isolators on Si platforms have been fabricated, with three approaches employed in order to integrate the magneto-optic garnet: deposition , adhesive bonding , and direct bonding [9, 10]. Table 1 summarizes the structure, fabrication method, and characteristics of these devices. Ce:YIG is not crystalized by depositing on Si, due to lattice mismatch. Bi et al. deposited a polycrystalline Ce:YIG on Si using a YIG buffer layer . Although this would be a revolutionary development for waveguide isolators, it requires post-crystallization by rapid thermal annealing at 850 °C, and such high temperature is undesirable for the purpose of subsequent integration with other devices.
We therefore discuss the NPSs of waveguide isolators with an interlayer. Figure 1 shows the calculated NPS for the 4-layered structure Ce:YIG/interlayer/Si/SiO2, with infinite width in the plane so as to eliminate any effect from lateral confinement. In order to compare the structures in Table 1, the calculated wavelength and Faraday rotation coefficient of Ce:YIG are set to 1550 nm and –4500 deg/cm, respectively. An enhancement of NPS due to nano-scale air gap, as reported in Ref , cannot be obtained from our calculation. In the case of adhesive bonding using benzocyclobutene (BCB), with refractive index n = 1.5 , the NPS is greatly reduced—e.g., with a 100 nm-thick BCB layer, the NPS is reduced to only 10% of its value when there is zero gap. In the case of a YIG buffer layer with a refractive index of 2.2 and Faraday rotation coefficient of 100 deg/cm , the NPS is 75% of what it is in the case of zero gap structure. Consequently, direct bonding, which achieves zero gap, is important for efficient NPS—which in turn results in high optical isolation and low insertion loss due to adequate interference. Otherwise a long interaction length for obtaining sufficient phase shift is required. The NPS values in Table 1 were estimated from the nonreciprocal wavelength shift (Δλ) and the free-spectral range (FSR) of the corresponding measured results. When the NPS is π, the forward and backward spectra are completely inverted; thus, the insertion loss is minimized and the isolation is maximized. So far, large NPSs up to π between forward and backward directions have been achieved only through direct bonding approaches. In addition, a low temperature process that utilizes surface activated bonding enables us to integrate the waveguide isolator with other electronic and photonic devices.
In our previous work, the properties of a 4-mm MZI optical isolator with large bends due to shallow etched rib waveguides were reported . These properties are listed in Table 1 above. The MZI arm was much longer than that required for isolator operation, due to the large bends. This also created difficulty when aligning the garnet chip so as to cover the proper length of the MZI arms. Actually, the previous device lacked the design accuracy of operating wavelength due to misalignment of the garnet chip. In this study, we’ve therefore fabricated an MZI isolator with Si-wire waveguide by improving our surface activated direct bonding technique. The device size is reduced to 1.5 mm, so it is fully covered by a garnet chip. We achieve a large NPS of 0.8π and an optical isolation of 18 dB.
The MZI isolator is composed of 3 dB couplers with multimode interference (MMI) and nonreciprocal and reciprocal phase shifters, as shown in Fig. 2 . The nonreciprocal phase shifter provides a ± π/2 phase difference between the forward and backward directions, in a push-pull manner. This is cancelled for the forward direction, with a reciprocal phase difference of + π/2 provided by an optical path difference, but is added for the backward direction. The forward light propagating from an input port exhibits constructive interference, and is transmitted to an output port. The backward light coming back from the output port exhibits destructive interference, and is radiated to side ports of the input side.
The NPS is a direction-dependent propagation constant, given by a magneto-optic effect induced by an external magnetic field that is applied transversely. The propagation constant difference is obtained numerically by solving an eigenvalue equation derived from anisotropic dielectric permittivity , or by solving a mode profile using perturbation theory . Although the former was used in plotting Fig. 1, so that the dependence of NPS on vertical confinement could be compared, the latter is suitable for the accurate design of Si-wire waveguides with strong lateral confinement. The NPS is induced in transverse magnetic (TM) mode for this structure.
Silicon photonics are useful for the production of low-cost optical transmitters in an optical network unit. The operating wavelength in this paper is set to 1270 nm, for intended use in up-stream transmission for 10G-EPON. The nonreciprocal phase difference is generated in the optical path of MZI arms where magnetic fields are applied in anti-parallel directions. Figure 3 shows the calculated NPS of a Si-wire waveguide, with an upper cladding of Ce:YIG having the Faraday rotation coefficient of –7900 deg/cm. The side cladding regions are assumed to be air. Although the NPS is maximized for a height of 200 nm, a height of 300 nm was chosen so as to reduce the junction loss at the boundary between the air and Ce:YIG upper cladding regions. The waveguide width is set to 300 nm for the single-mode operation. The length of the nonreciprocal phase shifter for ± π/2 is 752 μm. The reciprocal phase shifter is designed to be much longer than one that provides π/2, so that the NPS can be observed as a wavelength shift of a resonant spectrum. However, this does not increase the device size since the reciprocal phase shifter is located at the vertical part of the MZI arm, which separates two arms by 500 μm to apply the external magnetic field in anti-parallel directions. The expected FSR of the resonant spectrum is 30 nm. The length and width of the MMI coupler, which is designed by an eigen-mode expansion method, are 28.6 μm and 4.9 μm, respectively.
3. Surface-activated bonding
The surface-activated direct bonding employed in this study enables us to integrate different materials at low temperatures without the need for any adhesive. A stress induced by the difference in thermal expansion of two materials may crack the wafer. The key to achieving strong bonding is to maintain a vacuum as the process is carried out. Figure 4(a) shows the experimental setup of our apparatus for direct bonding. Two samples are exposed to RF plasma for surface activation, after which they are pressed at an elevated temperature in a closed chamber. The silicon-on-insulator (SOI), which is commercially available at SOITEC, is cleaned with an RCA cleaning process. A 0.5-μm-thick Ce:YIG single crystalline layer is grown on a (Ca, Mg, Zr)-substituted gadolinium gallium garnet (SGGG) substrate by an RF sputtering deposition. The garnet wafer is cut into 1.5 × 1.5 mm2 chips. The backside of the garnet is chamfered, as shown in Fig. 4(b), in order to apply the bonding pressure to the center of the chip and avoid stress concentration at the edge of the contact side. The fifteen garnet chips arrayed on a holder are bonded with a 20 × 20 mm2 SOI sample.
In this work, we modified our previous direct bonding technique in order to improve the bonding strength and success rate. In particular, we used N2 plasma exposure instead of the O2 plasma used previously. The surface roughness of SOI and Ce:YIG just after plasma exposure was measured with an atomic force microscope. For O2 plasma exposure, the surface roughness of SOI and Ce:YIG is reduced at the exposure time of 30 s . On the other hand, for N2 plasma exposure, the surface roughness increases gradually with the exposure time, as shown in Fig. 5 . The gas flow, process pressure, and RF power are 100 sccm, 120 Pa, and 500 W, respectively. After the bonding with annealing at 200 °C, the best results, strength, and success rate were obtained for an N2 plasma exposure of 10 s. Figure 4(c) shows a photograph of samples of garnet chips bonded on SOI. The average shearing-die strength with the N2 plasma is 1.5 times larger than with the O2 plasma. Consequently, yield of bonding with N2 plasma is much improved compared with that with O2 plasma.
4. Fabrication and characterization
The Si-wire waveguides were fabricated on an SOI wafer with a 300 nm-thick Si top layer and a 3.0 μm-thick SiO2 layer. The waveguide pattern was formed with electron-beam lithography and a reactive ion etching. The 1.5 × 1.5 mm2 garnet chips were bonded on the Si waveguides so as to cover the whole of the MZI patterns using the surface activated bonding under the above-mentioned conditions. Figure 6 shows a microscopic image of the fabricated MZI optical isolator and a photograph of the isolator array.
We characterized the MZI isolator by measuring the transmission spectra and its wavelength shift due to the NPS. An external magnetic field was applied to the MZI arms in anti-parallel directions using a small permanent magnet with three reversed poles—i.e., S–N–S or N–S–N. The magnet can provide enough of an external magnetic field to saturate the magnetization of Ce:YIG in the plane without contact to the sample. Two lensed fibers were aligned to the input and output ends of the waveguide. Light from a tunable laser diode operating at a wavelength from 1260 nm to 1340 nm, is polarized to excite the TM mode. The optical transmissions for forward and backward propagation were measured by exchanging the connection of the input- and output-lensed fibers. Figure 7 shows the measured transmission spectra. The FSR agrees well with the value estimated from the optical path difference in the reciprocal phase shifter. Since the transmission spectra without external magnetic field were almost the same for both directions, as shown by the dotted lines, the reciprocity of the measurement setup was confirmed. As shown by the solid lines, when a permanent magnet with S–N–S poles was aligned on the MZI, the spectra exhibited wavelength shifts with opposite signs, depending on the propagation direction. The maximum isolation ratio was 18 dB, at a wavelength of 1322 nm.
Figure 8 shows the nonreciprocal wavelength shift (Δλ) and FSR extracted from the peak and bottom wavelengths of measured resonant spectra. The dashed lines are linearly approximated functions of measured data. The NPS is obtained by 2πΔλ/FSR, as described in Table 1. From the approximated functions of FSR and Δλ, the measured NPS is obtained as shown by the solid line in Fig. 8. The wavelength dependence of NPS with respect to material and structure dispersions is calculated as shown in Fig. 9 . Qualitatively, the wavelength dependence of the measured NPS agrees well with the calculated dependence. The fact that the measured NPS does not reach π can be attributed to either a dimensional error of the fabricated Si-wire waveguide or to an insufficient Faraday rotation due to the degraded quality of Ce:YIG layer.
Finally, we discuss the insertion loss of the isolator. The transmission spectra shown by the green and yellow lines in Fig. 10 are the insertion losses of straight waveguides “with” a Ce:YIG cladding—i.e. adjacent to the isolator—and “without” a Ce:YIG cladding—i.e., fabricated outside of the bonded garnet chip on the SOI sample,—respectively. The coupling loss between the lensed fiber and the waveguide is estimated to be ~12 dB/facet. The insertion loss of the straight waveguide “without” Ce:YIG cladding includes the fiber coupling loss of 24 dB for both input and output facets, along with the propagation loss of 5 dB/cm. An additional loss of ~6 dB “with” Ce:YIG cladding is due to the junction loss at the cladding boundary and the optical absorption of Ce:YIG. We estimate the absorption loss at ~3 dB from the material absorption of 42 dB/mm and the optical field distribution of 4.6% in the Ce:YIG. The absorption loss can be reduced to be one third by using annealing process . The insertion loss of the MZI isolator, in addition to the above losses, includes an excess loss of ~10 dB due to a design or fabrication error of MMI couplers. Even though the total insertion loss of ~45 dB is quite high, most of them can be reduced using state-of-the-art technologies. For example, the fiber coupling loss can be reduced to less than 1 dB by integrating spot-size converters, and the excess loss of the MMI coupler should be negligible when design and fabrication techniques are improved.
We fabricated an MZI-based optical isolator with Si-wire waveguide using surface-activated direct bonding. Small garnet chips were bonded onto Si waveguides by using N2 plasma treatment and devising the appropriate chamfering method. The device size of 1.5 mm was a reduction from our previous device size of 4.0 mm. Since the garnet chip covered the entire MZI, design accuracy was improved. We obtained an optical isolation of 18 dB and an NPS of 0.8π at a wavelength of 1322 nm. We fabricated an array of isolators by aligning Ce:YIG chips on the proper positions of Si circuits, as shown in Fig. 6(b). This illustrates that direct bonding technology can be applied for the purpose of integrating multiple isolators with other photonic devices. Lasers with a III-V semiconductor active layer were also demonstrated on Si using bonding approach . Our isolator operating for TM mode can be integrated monolithically with such lasers via a TE-TM mode converter on a Si platform.
This work was partly supported by a Grant-in-Aid for Scientific Research (A) funded by the Japan Society for the Promotion of Science, and by a Grant for Practical Application of University R&D Results under the Matching Fund Method funded by the New Energy and Industrial Technology Development Organization (NEDO).
References and links
1. K. Ando, T. Okoshi, and N. Koshizuka, “Waveguide magneto-optical isolator fabricated by laser annealing,” Appl. Phys. Lett. 53(1), 4–6 (1988). [CrossRef]
2. T. Shintaku, “Integrated optical isolator based on efficient nonreciprocal radiation monde conversion,” Appl. Phys. Lett. 73(14), 1946–1948 (1998). [CrossRef]
3. J. Fujita, M. Levy, R. M. Osgood Jr, L. Wilkens, and H. Dötsch, “Waveguide optical isolator based on Mach-Zehnder interferometer,” Appl. Phys. Lett. 76(16), 2158–2160 (2000). [CrossRef]
5. H. Shimizu and Y. Nakano, “Fabrication and characterization of an InGaAsP/InP active waveguide optical isolator with 14.7 dB/mm TE mode nonreciprocal attenuation,” J. Lightwave Technol. 24(1), 38–43 (2006). [CrossRef]
6. W. V. Parys, B. Moeyrsoon, D. V. Thourhout, R. Baets, M. Vanwolleghem, B. Dagens, J. Decobert, O. L. Gouezigou, D. Make, R. Vanheertum, and L. Large, “Transverse magnetic mode nonreciprocal propagation in an amplifying AlGaInAs/InP optical waveguide isolator,” Appl. Phys. Lett. 88(7), 071115 (2006).
7. L. Bi, J. Hu, P. Jiang, D. H. Kim, G. F. Dionne, L. C. Kimerling, and C. A. Ross, “On-chip optical isolation in monolithically integrated non-reciprocal optical resonators,” Nat. Photonics 5(12), 758–762 (2011). [CrossRef]
8. S. Ghosh, S. Keyvavinia, W. Van Roy, T. Mizumoto, G. Roelkens, and R. Baets, “Ce:YIG/Silicon-on-Insulator waveguide optical isolator realized by adhesive bonding,” Opt. Express 20(2), 1839–1848 (2012). [CrossRef] [PubMed]
9. M.-C. Tien, T. Mizumoto, P. Pintus, H. Kromer, and J. E. Bowers, “Silicon ring isolators with bonded nonreciprocal magneto-optic garnets,” Opt. Express 19(12), 11740–11745 (2011). [CrossRef] [PubMed]
10. Y. Shoji, T. Mizumoto, H. Yokoi, I.-W. Hsieh, and R. M. Osgood Jr., “Magneto-optical isolator with silicon waveguides fabricated by direct bonding,” Appl. Phys. Lett. 92(7), 071117 (2008). [CrossRef]
12. T. Mizumoto and Y. Naito, “Nonreciprocal propagation characteristics of YIG thin film,” IEEE Trans. Microw. Theory Tech. 30(6), 922–925 (1982). [CrossRef]
13. Y. Okamura, H. Inuzuka, T. Kikuchi, and S. Yamamoto, “Nonreciprocal propagation in magnetooptic YIG rib waveguides,” J. Lightwave Technol. 4(7), 711–714 (1986). [CrossRef]
14. R. Takei, K. Yoshida, and T. Mizumoto, “Effects of wafer precleaning and plasma irradiation to wafer surfaces on plasma-assisted surface-activated direct bonding,” Jpn. J. Appl. Phys. 49(8), 086204 (2010). [CrossRef]
15. T. Shintaku and T. Uno, “Preparation of Ce-substituted Yttrium iron garnet films for magneto-optic waveguide devices,” Jpn. J. Appl. Phys. 35(Part 1, No. 9A9A), 4689–4691 (1996). [CrossRef]
16. A. W. Fang, H. Park, O. Cohen, R. Jones, M. J. Paniccia, and J. E. Bowers, “Electrically pumped hybrid AlGaInAs-silicon evanescent laser,” Opt. Express 14(20), 9203–9210 (2006). [CrossRef] [PubMed]