The 5th International Symposium on Advanced Science and Technology of Silicon Materials (JSPS Si Symposium), Nov. 10-14, 2008, Kona, Hawaii, USA
Si photonics platform and its fabrication Tai Tsuchizawa*, Koji Yamada, Toshifumi Watanabe, Hiroshi Fukuda, Hidetaka Nishi, Hiroyuki Shinojima and Seiichi Itabashi NTT Microsystem Integration Laboratories, NTT Corporation 3-1 Morinosato Wakamiya, Atsugi-shi Kanagawa 243-0198, Japan e-mail:
[email protected] Abstract Silicon photonic wire waveguides, whose core size is as small as around 500 nm, are fabricated by using technologies developed in the semiconductor industry. For practical application, however, we must overcome technical difficulties associated with geometrical error and surface roughness at the nanometer level. In this paper, we describe fabrication technologies that satisfy these severe fabrication accuracies. We also show some practical applications of silicon photonic wire waveguides and clarify the present status of their fabrication technologies. 1. Introduction The use of telecommunications and information processing technologies is growing rapidly throughout society. One of the crucial obstacles to thepopularization of these technologies is the poor integration performance of optical devices. The integration density of optical devices is significantly lower than that of electronic devices. Consequently, there is a strong demand for an optical waveguiding system with highly integrated optical circuits. A silicon wire waveguide based on a silicon-on-insulator (SOI) structure is a promising platform for highly integrted, ultra-small optical circuits or devices. A Si wire waveguide is a single-mode channel waveguide consisting of a silicon core and silicon dioxide (SiO2) cladding [1–3]. Because of the strong light confinement, the core size at 1.5-Pm wavelength is designed to be less than 1 μm, and a sharp bending radius of only a few micrometers is possible. These features should enable us to miniaturize optical circuits and integrate them. Furthermore, we can use existing silicon fabrication technology and SOI substrates for Si electrical circuits. This will make waveguide pattern formation easy and will be advantageous for mass production, resulting in economical optical devices. Although Si wire waveguides have these interesting features, the fabricated devices are very sensitive to size error because of the very small core. Moreover, scattering losses are far more sensitive to the sidewall roughness of the waveguide’s core because of the large refractive index contrast [4]. Therefore, accurate nanometer-scale fabrication is required. In this paper, we describe fabrication technologies that satisfy these severe fabrication accuracies. We also show some practical applications of silicon wire waveguides and clarify the present status of their fabrication technologies. 2. Silicon wire waveguides 2.1. Structure of Si wire waveguide The cross-sectional structure of a Si wire waveguide is shown in Fig. 1. The Si wire waveguide is generally the channel type with a rectangular core as shown in Fig. 1(a). The rib-type Si wire waveguide with the Si slab layer at both sides of the core, as shown in Fig. 1(b), is also used when the p-i-n carrier injection structure is made in the
Fig. 1 Structure of silicon photonic wire waveguide. (a) Channel waveguide; (b) rib waveguide.
Fig. 2 Typical fabrication process for silicon photonic wire waveguides.
waveguide. The thickness and width of the core are roughly in the ranges of 200~300 nm and 300~600 nm . A buried SiO2 layer of the SOI wafer serves as the under-cladding. The SiO2 layer should be thick enough as it optically isolates Si waveguides from the Si substrate and reduces losses due to substrate leakage. As the over-cladding, a SiO2 or a polymer material that is optically equivalent to SiO2 is used. Moreover, a spot-size converter is indispensable for connection with external components such as optical fibers because the core size of the Si wire waveguide is very small. Highly effective spot size converters have already been developed [5,6]. 2.2 Fabrication 1) Fabrication process The typical fabrication process for the Si wire waveguide is shown in Fig. 2. First, the resist patterns that will become the Si core are formed on SOI wafer by e-beam (EB) lithography or deep ultraviolet (DUV) lithography, which can provide 100-nm-level patterns. Common lithographcial technologies are optimized to delineate in a straight line for the electric circuit patterns and pattern edge roughness is not considered a serious problem. However, for optical waveguide patterns, curved lines are necessary and, in addition, sub-nanometer edge roughness is needed for the reduction of propagation loss. The data generation for the EB shot or the photo mask is the key point for the waveguide pattern formation. Next, the Si core is formed with the dry etching based on the patterned resist using a low-pressure plasma, such as electron cyclotron resonance (ECR) plasma or inductive coupled plasma (ICP). The hard mask transferred from the resist pattern is used as the etching mask in some cases. Optimizing the etching conditions such as gas and pressure for each plasma equipment is very important for making a smooth sidewalls. Finally, the wafer is coated with another layer of polymer or SiO2 for the over-cladding. On formation of the over-cladding, a low temperature process is needed in order to avoid distortion of the Si core shape and to not destroy electronic devices integrated monolithically with the waveguide. The plasma enhanced chemical vapor deposition (PECVD) method is commonly used for the over-cladding film. When the electronic device structure with a PIN diode is needed, processes such as ion implantation and annealing are performed before the formation of the over-cladding. 2) Reduction of sidewall roughness A Si wire waveguide with high index contrast is not only very sensitive to size error but also to scattering at rough areas on the sidewalls of the core. Therefore, the fabrication of a very smooth core is necessitated. The roughness on the sidewalls of the waveguide core is mainly caused by imperfections in the resist pattern. The plasma etching can also add to the roughness, depending on the etching condition. Therefore, lithography and etching are the important processes for making a smooth Si pattern. We have made patterns with smooth sidewalls using EB lithography and ECR plasma etching [2,7]. The waveguide patterns were delineated on the resist with a variable-shaped EB writer with an acceleration voltage of 100 kV. The EB writer has higher throughput than point-beam machines but is not strong at making smooth patterns, especially curved lines. Since we wanted to use it for optical waveguides, we had to improve the writing method. To form smooth patterns with uniform widths, we optimized the shot size to generate an EB shot without fluctuation and developed a multiple-exposure method in which the EB shot is shifted incrementally. Moreover, we reduced proximity effects in the EB data. These improvements significantly reduced the roughness of the resist patterns. Figure 3 shows SEM images of a ring without and with EB data optimization. We can see that a smooth pattern is made㩷even in a sharp bend with a radius of 1 Pm. To transfer the smooth resist pattern to the Si faithfully without adding roughness, we employed ECR plasma etching with a SF6-CF4 gas mixture and SiO2 hard mask. The etching is performed at low gas pressures of 0.1 Pa or less and the low-energy ions irradiate a wafer using a low RF power of less than 10 W. This is highly advantageous for making smooth sidewalls because there is hardly any redeposition of reaction products and little
Fig. 3 SEM images of ring resonators without (a) and with (b) EB data optimizations.
Fig. 4 SEM images of the silicon wire waveguiding system. (a) Core of silicon wire waveguide; (b) silicon taper for spot-size converter.
damage during etching. For smooth sidewall etching, it is important not only to select the proper plasma conditions but also to select the proper material for the etching mask. We found that a SiO2 hard mask gave good results. Figure 4 shows SEM images of Si wires and a taper tip obtained after we had made the improvements to the fabrication process. The 400-nm-wide waveguide part and the 80-nm-wide tip for a spot-size converter are both fabricated with vertical and smooth sidewalls. The roughness rms is estimated to be less than 2 nm from the TEM image shown in Fig. 5. Recent propagation loss values measured for waveguides with a core size of 480 nm × 200 nm are shown in Fig. 6. We obtained a loss of 1.2 dB/cm from the slope of the fitting line. This loss is small enough to make practical devices. 3) Spot-size converter To solve the problem of large coupling loss caused by the large mode field size difference, we developed a spot-size converter made with inorganic materials, whose structure is shown in Fig. 7 [8]. The converter has a Si adiabatic taper that gradually becomes thinner toward the end and a second low-index waveguide that covers the taper. The principle of the converter is that the low-index waveguide core efficiently captures the light that leaks from the taper.㩷Typically, the Si taper should be 300-Pm long, and the tip of the taper should be less than 100 nm. The low-index waveguide has about a 3-Pm-square core and about 3% index contrast. The core is made of SiOx, whose refractive index is 1.515, and the cladding is SiO2, whose refractive index is 1.47. The SiOx and SiO2 films were deposited by ECR plasma-enhanced CVD, which can deposit high-quality films at low temperatures of less than 200 qC [9]. Figure 8 show the refractive index and the deposition rate with changing O2 flow rate. The refractive index of SiOx film can be controlled by adjusting the flow rate ratio of O2 and SiH4. Figure 9 shows SEM images of a 3-Pm-square SiON core for a spot-size converter. The tapered tip of the Si is accurately embedded in the SiON core. Figure 10 shows the transmission spectrum of a fabricated waveguide with spot-size converters. One can see that the applicable bandwidth is very large, over 250 nm. There is absorption for O-H bonds in the 1400-nm-wavelength region, but it is weak and easily eliminated by annealing. Moreover, no significant Fabry-Perot ripples were found in the transmission spectrum, indicating the absence of fatal reflections at the spot-size converter. The coupling loss per connection was derived as 0.5 dB when optical fibers with a mode field diameter of 4.3 μm were connected to the SiOx waveguide ends.
Fig. 5 FIB-TEM image of the Si/SiO2 interface at the sidewall. The roughness root mean square of the sidewall was estimated to be 1~2 nm.
Fig. 6 Propagation loss of a typical silicon photonic wire waveguide.
Refractive index
Fig. 7 Schematic of the spot-size converter for Si wire waveguides
2 IORZUDWHVFFP
Deposition rate (Pm/min)
6L+VFFP
Fig. 8 Refractive index and deposition rate for SiOx using the ECR-CVD method.
Embedded Si taper 3Pm SiOx
Fig. 9 SEM images of a low-index waveguide for spot-size converter.
Fig. 10 Transmission spectrum of a typical silicon photonic wire waveguide with spot-size converters.
3. Functional devices 3.1. Passive devices Using the fundamental fabrication technologies, we have developed some passive devices and active devices [2,7,8,10]. First, we show the results for cascaded branches based on multi mode interference (MMI) as the most basic passive device [7]. The MMI branch consists of a simple rectangle 2.6 x 1.8 Pm2 with three waveguide ports. It can therefore be fabricated very easily. MMI branches work in a wide wavelength region, and their sensitivity to geometrical errors are essentially low. Applying this branch, we have developed cascaded branches. Figure 11(a) shows a photograph of a device cascading eight MMI branches. Figure 11(b) shows the relation between the transmittance and branching order. The transmittance decreases linearly with the number of MMI branches, and the measured values lie almost exactly on the fitted line. This indicates that each branch has the same characteristics, in other words, that all branches were fabricated identically. The slope of the fitting line is almost 3.5 dB, which means the excess loss for one branch is 0.5 dB. The cascaded branch has spot-size converters; therefore, 10-Gbps signal shows a clear eye pattern in this higher-order branch. The MMI device has already been used in applications such as modulators with a Mach-Zehnder (MZ) interferometer or for on-chip optical clock signal distribution. 3.2 Active devices Active functions, such as optical switching and modulation can be achieved by exploiting carrier-induced effects. Here, we show a Si- variable optical attenuator (VOA) based on a p-i-n carrier injection structure [7,11]. Figure 12(a) shows the cross-sectional view of the carrier injection structure, which consists of a rib-type silicon wire waveguide with the p-i-n structure. The waveguide has a 600 x 200-nm core and 100-nm-thick slab. Figure 12(b) shows a fabricated waveguide with a carrier injection structure, in which the pair of contact electrodes can be seen. Figure 12(c) shows transmission spectra for various injected currents. As the injected current increases, guided light is absorbed by the injected carriers. We can see attenuation over 30 dB. The wavelength dependence of the attenuation was very flat in the measured 40–nm bandwidth. Figure 12(d) shows the temporal response of optical output when a 15-ns electrical input was applied. The response is about 2 ns, which is almost 100 times faster than a VOA with a conventional rib waveguide. We believe our silicon wire waveguide with the p-i-n structure has a great potential for use in active functional devices. 2.6um
1.8um
IN
150 um
(a) OUT 8 7 6 5 4 3 2 1
Fig. 11 (a) Cascaded MMI branches and (b) their transmission characteristics and eye pattern at 6-th port.
Fig. 12 Fast optical attenuator using a silicon photonic wire waveguide with a p-i-n structure. (a) Schematic of the PIN structure. (b) Fabricated waveguide. (c) Transmission characteristics of the carrier-injection waveguide. (d) Temporal response of optical output when a 15-ns electrical input was applied.
4. Summary We confirmed that our precise Si microfabrication technologies enable us to make Si wire waveguides with excellent characteristics and that some of the Si-wire-based photonics devices are already at a level suitable for practical application. More stable fabrication with nanometer accuracy will hereafter be required to satisfy the severe demand for practical device performance. Moreover, it will be necessary to adjust the fabrication process for the optical device to that for the electronic device which has complexities and many restrictions. This is because the integration with electric circuit is the competitive feature for Si photonics. It is clear that these advanced fabrication technologies and processes can not be achieved by only introducing the semiconductor manufacturing equipment on the market. We think new developments in manufacturing technology suitable will be needed. ACKNOWLEDGMENT This work was partly supported by the SCOPE program of the Ministry of Internal Affairs and Communications, Japan. References [1] L.C. Kimerling: Appl. Surf. Science 159-160 (2000) 8. [2] T. Tsuchizawa, K. Yamada, H. Fukuda, T. Watanabe, J. Takahashi, M. Takahashi, T. Shoji, E. Tamechika, S. Itabashi and H. Morita: IEEE J. Select. Topics Quant. Electron. 11 (2005) 232. [3] P. Dumon, W. Bogaerts, V. Wiaux, J. Wouters, S. Beckx, J. V. Campenhout, D. Taillaert, B. Luyssaert, P. Bienstman, D. V. Thourhout, and R.Baets: Photon. Technol. Lett. 16 (2004) 1328. [4] K. K. Lee, D.R. Lim, and L.C. Kimerling, Opt Lett, 26 (2001) 1888. [5] T. Shoji, T. Tsuchizawa, T. Watanabe, K. Yamada, and H. Morita: Electron.Lett. 38 (2002) 1669. [6] D. Taillaert, W. Bogaerts, P. Bienstman,T. Ktauss, P. Van Daele, I. Moerman, S. Verstuyfi, De Mesel, R.Baets: IEEE J. Quantum Electron. 38 (2002) 949. [7] T. Watanabe, K. Yamada, T. Tsuchizawa, H. Fukuda, H. Shinojima, and S. Itabashi: Proc. SPIE 6775, 2007, pp. 67750K-1 [8] T. Tsuchizawa, K. Yamada, H. Fukuda, T. Watanabe, S. Uchiyama, and S. Itabashi: Jpn, J. Appl. Phys. 45-8B (2006) 6658. [9] S. Matsuo, and M. Kiuchi, Jpn. J. Appl. Phys., 22 (1983) L210. [10]H. Fukuda, K. Yamada, T. Tsuchizawa, T. Watanabe, H. Shinojima, and S. Itabashi, Opt. Express 16, (2008) 4872. [11] K. Yamada, T. Tsuchizawa, T. Watanabe, H. Fukuda, H. Shinojima, and S. Itabashi, IEEE Intern. Conf. Group IV Photonics, Tokyo, 2007, pp. 116.