Because of the high degree of integration of electrical. Design, Fabrication, Structural and Optical Characterization of thin Si 3 N 4 Waveguides

Submitted to IEEE- Journal of Lightwave Technology 1 Design, Fabrication, Structural and Optical Characterization of thin Si 3 N 4 Waveguides Nicola Daldosso, Mirko Melchiorri, Francesco Riboli, Manuel Girardini, Georg Pucker, Michele Crivellari, Pierluigi Bellutti, Alberto Lui, and Lorenzo Pavesi Abstract LPCVD (low pressure chemical vapor deposition) thin film Si 3 N 4 waveguides have been fabricated on Si substrate within a CMOS (complementary metal-oxidesemiconductor) fabrication pilot-line. Rib, channel, and striploaded waveguides have been designed, fabricated and characterized in order to workout the best structure. Number and optical confinement factors of guided optical modes have been simulated taking into account sidewall effects caused by the etching processes, which have been studied by Scanning Electron Microscopy. Optical guided modes have been observed with a mode analyzer and compared with simulation expectation to confirm the process parameters. Propagation loss measurements at 780 nm have been performed by using both the cut-back technique and measuring the drop of intensity of the top scattered light along the length of the waveguide. Loss coefficients of about 0.1 db/cm have been obtained for channel waveguides. These data are very promising in view of the development of Si integrated photonics. Index Terms waveguides, silicon nitride, optical mode analysis, optical propagation losses, CMOS-compatible technology, silicon-integrated photonics. I. INTRODUCTION Because of the high degree of integration of electrical devices (Moore s law) electrical interconnects are reaching their fundamentals bottlenecks in term of speed, packaging, and power dissipation [1], [2]. In spite of the possibility to overtake the bottleneck for several years by improving design and materials performances without changing any present architecture paradigm, optical interconnects becomes more and more appealing not only for computerto-computer (which is yet quite established) but also for onboard and intra-chip connections. Silicon microphotonics is predicted to play a significant role in this field [3]-[5]. Silicon microphotonics is the technology that implements all the different photonics components by using standard Manuscript received August 7, 2003. This work was supported by PAT through the PROFILL project. N. Daldosso, M. Melchiorri, F. Riboli, M. Girardini, and L. Pavesi are with INFM and University of Trento, Department of Physics, via Sommarive 14, 38050 Povo Trento, Italy (e-mail: daldosso@ science.unitn.it). G. Pucker, M. Crivellari, P. Bellutti, and A. Lui are with ITC-IRST, Microsystem Division, via Sommarive 18, 38050 Povo Trento, Italy. silicon processing materials and tools, as those used in modern microelectronics. Till now, various passive optical components have been developed for WDM-based photonic networks [6]; however, most of these are not compatible with silicon technology. Hybrid technology, where several guided-wave components of different material are assembled to operate as an integrated system, makes the system fabrication difficulty and costly. Silicon microphotonic might reduce the cost of production of the components and the system. The first component ubiquitous in silicon microphotonic is the optical waveguide, which is used to channel light signal in the system. In order to look for silicon compatible materials, one of the requirements is the availability of thin dielectric films with large refractive index difference ( n = n core - n cladding, where n core is the refractive index of the core material and n cladding is the refractive index of the cladding material). In fact, the device size scales down with increasing n. Unfortunately, it is quite difficult to achieve low propagation losses for large n: scattering losses increase as ( n) 2, and therefore high index core waveguides are very sensitive to interface roughness. A second requirement on the materials is the compatibility with the conventional silicon technology. The materials have to withstand the process conditions (purity, temperature, compatibility, stability, functionality) and should be processable with the tools already common in the microelectronics industry. A natural choice is to look for some of the commonly used dielectrics in microelectronics. Stoichiometric silicon nitride (Si 3 N 4 ) for the core material and silicon oxide (SiO 2 ) for the cladding material are good candidates for optical waveguides, especially in the visible range of light. In fact, 2-D waveguides with Si 3 N 4 core surrounded by SiO 2 cladding layers represent one of the best solution to have large refractive index difference ( n 0.55), low scattering losses, no absorption losses and compatibility with Si technology. Various low-losses optical waveguides have been investigated [7]-[14], and some applications in various optical devices have been already demonstrated [15], [16]. It was shown that Si 3 N 4 waveguides present remarkable losses due to the vibrational overtones of Si-H and N-H bonds, thus preventing competitive waveguides for the third telecom window [9], [13]. However, no systematic study concerning Si 3 N 4 waveguides has been done for the visible. Indeed for on-board optical interconnects, the choice of visible light seem to be quite suitable thanks to the

Submitted to IEEE- Journal of Lightwave Technology 2 availability of high performance and low cost III-V semiconductor based VCSEL (vertical cavity surface emitting laser) transmitters and Si based photo-receivers. In this paper, we perform a first step to the realization of an integrated silicon microphotonics system by developing low-losses Si 3 N 4 /SiO 2 waveguides within a CMOS fabrication pilot-line. The developed process is compatible with standard CMOS processing, which allows a direct transfer of the technology. We investigated the optical properties of thin film Si 3 N 4 waveguides deposited on Si substrate as a function of waveguide geometry and fabrication technology. Three kinds of waveguides were designed, fabricated and characterized: rib, channel, and strip-loaded. II. WAVEGUIDE FABRICATION Among the various deposition techniques we have chosen the LPCVD (low pressure chemical vapor deposition) technique, which is widely available in CMOS lines and which allows the deposition of stoichiometric Si 3 N 4 layers (n = 2.010 at 780 nm) with negligible absorption losses; however, this results in high tensile stress of the Si 3 N 4 layer and the maximum layer thickness is therefore limited to 250 nm. Because of the absence of absorption losses in Si 3 N 4 waveguides in the visible (under the selected process conditions), propagation losses are essentially related to scattering losses due to the interfaces and to leakage losses towards the Si substrate. To minimize irregularities caused by the etching process employed to fabricate the rectangular waveguides, both wet (chemical etching) and different dry etching processes (plasma etching) have been experienced. However, the choice of relatively large rectangular structures (about 10 µm in width, limited by available masks) largely reduces scattering losses due to irregularity of lateral edges of the mesa structures. Leakage losses towards the Si substrate have been avoided by growing a sufficiently thick base layer of SiO 2 (2 µm). The thickness of the Si 3 N 4 layer was chosen to be 200 nm, this value is a compromise between avoiding too thick films resulting in mechanical rupture of the Si 3 N 4 layer and minimum film thickness necessary for wave-guiding. The process sequences are standard: initially we deposit planar waveguides and then by lithography and etching we define the two-dimensional (2-D) rectangular waveguides. The planar waveguides consist in a layered structure formed by a stoichiometric 200 nm thick Si 3 N 4 layer deposited by LPCVD on 4 Si wafers previously covered with a 2 µm thick SiO 2 BPSG (Boron Phosphor Silicate Glass) cladding layer. Starting from these planar waveguides, a series of 2-D rectangular waveguides have been produced. The three different geometries (rib, channel, and strip-loaded) are sketched in Fig. 1. The photolithographic mask used for waveguide definition consisted of approximately 500 lines, each one 4 cm long and 18 µm wide. To obtain the strip-loaded waveguides the Si 3 N 4 layer was covered by a 520 nm thick layer of TEOS (SiO 2 precursor tetra-ethyl-ortho-silicate) deposited by LPCVD. Fig. 1. Schematic view of the rib, strip-loaded and channel 2-D rectangular waveguides deposited by LPCVD on Si substrate; the Si 3 N 4 core layer is 200 nm thick on 2µm of SiO 2 BPSG (Boron Phosphor Silicate Glass) used as bottom cladding. Top cladding is the air for the rib waveguides, and 520 nm thick layer of TEOS (SiO 2 precursor tetra-ethyl-ortho-silicate) for both strip-loaded and channel waveguides. Two different plasma etchers were tested to define the waveguides. For the rib waveguides we used: - oxide plasma etcher, which attacks also Si 3 N 4 but cannot detect the Si 3 N 4 /SiO 2 interface. Therefore the etching was done with fixed time. Actually, the over-etching was quite long and, beside the 200 nm of Si 3 N 4, also about 200 nm of SiO 2 were removed. - silicon nitride dedicated etcher, which is able to detect the Si 3 N 4 /SiO 2 interface, thus ensuring an etch stop close to the interface. For the channel waveguides, the structures are very similar to the rib-waveguides: the rib was defined either with the silicon nitride plasma etcher (i.e. sample G3) or with the oxide plasma etcher (i.e. sample G4) and then covered with a 520 nm layer of TEOS to obtain the top cladding. For the strip-loaded waveguides, both chemical (wet) and plasma (dry) etching were tested to define the loading SiO 2 strip over the Si 3 N 4 layer. Wet etching has been done with BHF 7:1 (buffered HF). The Si 3 N 4 layer is very slowly attacked by BHF and the etching process stops exactly at the SiO 2 /Si 3 N 4 interface (i.e. sample G5). Plasma oxide etching was also attempted using a fixed time (i.e. sample G6). The selectivity between SiO 2 /Si 3 N 4 is low. Optical interference and SEM analysis show that besides removing the SiO 2 layer also 30±10 nm of Si 3 N 4 were removed during the plasma etching process. Fig. 2. SEM images of three 2-D rectangular waveguides (a) and details of the edge of one channel (b) for sample G4 (channel waveguide obtained by oxide plasma etching).

Submitted to IEEE- Journal of Lightwave Technology 3 Scanning Electron Microscopy (SEM) has been employed to check the deposition and etching processes by investigating geometry, sidewall as well as layer roughness of the different waveguides structures. Fig. 2 shows two SEM images of the sample G4 reported as example: three channels (a) and details of the etched edge of one channel (b). III. DESIGN AND SIMULATION The geometrical parameters of the waveguides have been designed by considering the limitations of the process (Si 3 N 4 maximum layer thickness, available resolution in the lithography). Design has been done by using a photonics CAD tool for the simulation: the mode solver software FIMMWAVE 3.0 (Photon Design Inc.), which is a fully vectorial mode finder for waveguide structures. All the simulations have been done for a signal wavelength of 780 nm, a refractive index of 2.010 for Si 3 N 4 and 1.454 for SiO 2 [17]. Slab planar waveguides have been considered to understand the influence of core and cladding thickness, and of the symmetric and asymmetric geometry on the optical confinement of the guided modes. The thickness of the bottom cladding (SiO 2 ) has been varied in the simulation to obtain the minimum value needed to isolate the guided optical modes from the silicon substrate. This will avoid leakage losses towards the Si substrate because of the high refractive index of Si. For Si 3 N 4 core thickness of 200 nm it has been found that 2 µm thick SiO 2 cladding is sufficient. Symmetric (SiO 2 /Si 3 N 4 /SiO 2 ) and asymmetric (air/si 3 N 4 /SiO 2 ) slab waveguides show single mode optical propagation at 780 nm in agreement with the theoretical cutoff prediction. For the symmetric waveguide the filling factors (defined as the fraction of mode power flux in the core guiding region) are about 72% for TE (n eff =1.756) polarization and 58% for TM (n eff =1.653) polarization, where n eff is the effective mode index; while they are about 75% for TE (n eff =1.724) polarization and 51% for TM (n eff =1.564) polarization for the asymmetric waveguide. In 2-D rectangular waveguides the number of guided optical modes is large as expected for the large channel width chosen (10 µm). Both the number of guided modes and their confinement within the core depend on the geometry of the waveguide (rib, channel and strip-loaded). The first guided modes show TE polarization with a filling factor of about 70-75%, the others show mostly TM polarization with a filling factor of about 50%. In Fig. 3, the fundamental guided modes (labeled TE 0 ) of rib, channel and strip-loaded geometries are reported. The filling factor is about 75% for rib waveguides and about 72% for both channel and strip-loaded waveguides. In details, the mode analysis has shown that the rib waveguide supports about 25 guided modes: the first 15 modes are TE modes with effective refractive index n eff from 1.723 (TE 0 ) to 1.622 (TE 14 ) with the same filling factor, about 75%. Highest guided modes are TM, with n eff from 1.563 (TM 0 ) to 1.526 (TM 8 ) and a filling factor about 51%. The channel waveguide present about 40 guided modes: TE 0 mode (n eff =1.755) has a filling factor about 72%, while TM 0 mode (n eff =1.653) about 58%. In the strip-loaded waveguide, the simulation results in 10 TE guided modes with n eff =1.755 for the TE 0 mode (filling factor about 72%) and n eff =1.726 for the TE 8 mode (filling factor about 57%). Fig. 4. Intensity plot profiles of three selected optical modes (TE 0, TE 1 and TE 7 ) of the channel waveguide simulated (bottom panels) and measured by a beam analyzer (top panels). To demonstrate the good control on the process we achieved, we report in Fig. 4 three selected guided optical mode profiles of a channel waveguide measured with a beam profile analyzer together with the simulated intensity profiles computed with the nominal parameters. The agreement between the simulated and experimental results is remarkably good. Fig. 3. Simulated intensity profile of the fundamental guided mode (TE 0 ) of rib, channel and strip-loaded Si 3 N 4 waveguides; filling factor and refractive effective index are 75% and 1.723 for the rib waveguide, and 72% and 1.755 for both channel and strip-loaded waveguides. Fig. 5. Simulated intensity profiles of the TE 8 mode of the channel and strip-loaded waveguide, which show the same filling factor (about 72%). The first 7 TE polarized optical modes are well confined within the waveguide without relevant differences between

Submitted to IEEE- Journal of Lightwave Technology 4 the three geometries. However, the higher optical modes show some differences: in the strip-loaded waveguide the guided modes higher than the 7 th are not well confined under the stripe and extend through the core (Fig. 5). This affects the filling factor of higher optical modes: channel and strip-loaded filling factors are both about 72% for TE 0, whereas the TE 8 guided mode changes from about 72% (channel) to 60% (strip-loaded). This not good confinement explains the differences in the propagation loss measurements of such waveguides, which will be discussed in the next section. wall steepness because of the large width. A possible cause of the large losses can be the different etch depth achieved in the two etching processes. In fact, a non accurate etch depth in the plasma process (oxide plasma etcher) results in waveguide structures with very different guiding properties. This is show in Fig. 7, where the filling factor of the first four TE polarized optical modes as a function of the etching depth is shown. It can be seen that for etching depth less than the top cladding thickness (negative values) the filling factor noticeably decreases and that this reduction is larger for higher guided modes. Fig. 6. Simulated intensity profiles of the TE 0 and TE 8 mode of the striploaded waveguide produced by plasma and chemical etching. A further simulation has been done to verify the influence of the different etching processes: plasma etching results in quite steep lateral edges (about 80 ), whereas chemical etching produces 45 lateral edges. To simulate these structural differences the waveguide profiles have been designed according to the SEM images. In Fig. 6, the simulated intensity profile of the guided modes is shown for the two etching processes in the strip-loaded waveguide (sample G5 and G6). It is worth noting that the field distribution is almost unaffected by the steepness of the lateral edge. The filling factor remains practically the same for the first guided modes (about 72% for the TE 0, 71% for TE 4 ). This is quite reasonable because of the width of the rectangular waveguides: 10 µm is quite large compared to 0.5 µm of modified lateral edge. However, it is worth noting that the filling factor of the wet-etched structure is slightly larger than the dry-etched one (from 60% to 58% for TE 8 ). This difference becomes more and more relevant with decreasing the width of the rectangular waveguide, showing an improvement of the confinement of the guided modes. As it will be presented in the next section, quite large differences in the propagation losses of strip-loaded waveguides have been observed between those obtained by chemical and plasma etching. These cannot be attributed to the influence of the different etching process on the strip Fig. 7. Filling factor of the first four TE guided modes as a function of etch depth of the dry-etched (oxide plasma etcher) strip-loaded waveguide. Etch depth zero corresponds to well done etching, negative or positive values correspond to less or more etching with respect to the core thickness. IV. LOSS MEASUREMENTS Propagation loss measurements have been performed by using both the cut-back and top-view technique. In the former, waveguides have been cleaved at different lengths and the intensity of the transmitted light has been measured. The top-view technique is based on the assumption that the scattered light intensity observed from the top of the waveguide is directly proportional to the light intensity propagating in the waveguide. This assumption is quite reasonable in absence of specific irregularities on the channel top surface. Hence the decrease of the scattered light as a function of the position on the waveguide is representative of the attenuation of the optical mode propagating in the waveguide. The decrease of the scattered light can be easily detected by looking at the surface of the waveguide with a CCD camera (top-view). Average of multiple measurements on several waveguides are reported. For the top-view measurements the longest waveguides (about 35 mm) have been used. The input-output facets of the waveguides have been prepared by cleavage and compared with carefully lapped and polished facets of the same waveguide. SEM analysis and transmittance measurements do not show lower quality for the cleaved facets with respect to the polished ones. In

Submitted to IEEE- Journal of Lightwave Technology 5 view of the easy of preparation by simply cleavage we used cleaved facets. Waveguides have been measured by coupling-in visible light from a laser diode (780nm about 10 mw) and coupling-out the light with microscope objectives. The experimental setup is shown in Fig. 8. five different lengths (from 10 to more than 35 mm) for each sample. Fig. 9 shows an example of the transmittance measured by the cut-back technique for the strip-loaded sample G6. The plotted values are the average of significant values and the maximum intensity over all channels, respectively. Data (reported in natural logarithmic scale) show linear behavior. From the fit results a propagation loss coefficient of about 1.5±0.2 db/cm. These losses are independent on the coupling losses. It is worth noting that the same losses are obtained by fitting either the averaged values or the maximum values. The propagation loss coefficients of the different samples are reported in Table I. Fig. 8. The experimental setup for the measurements of propagation losses. In order to efficiently inject light within the thin waveguide (0.2 µm thick core), 40x objectives have been mounted on a computer controlled nano-positioning system (Melles-Griott nanomax-ts). Care has been taken to minimize injection losses due to modal mismatch, reflections and edge defects or irregularities. The success of good injection and coupling conditions is established by both top and front view. Top-view observation is performed by means of an optical microscope (STEMI-2000C) and a CCD camera (COHU monochrome) mounted on top of the measuring system. The front view is made possible by using a microscope objective (40x) matched to a zoom (Navitar 12x) mounted on a high performance CCD camera (Dalstar SMD-1M15) computer controlled by the LBA-500 Spiricon beam analyzer software. A prism beam splitter allows to divide the transmitted light in two parts: one is directed to the CCD camera, the other is directed to a calibrated photodiode, which yields intensity measurement of the waveguide transmitted light. Table I. Waveguides samples and propagation losses. The loss data have been obtained by measuring three series of ten waveguides for each sample length: typically Fig. 9. Transmittance measurements as a function of sample length (cutback technique) of a strip-loaded waveguide (sample G6) at 780 nm. The propagation loss coefficient α is 1.5±0.2 db/cm. Fig. 10. Scattered light intensity (top-view measurements) along the sample length of a channel waveguide (sample G3) at 780 nm. The propagation loss coefficient α is 0.1±0.05 db/cm. Propagation loss coefficients have been obtained also by the top-view technique. Fig. 10 reports the data for channel waveguide (sample G3). Scattered intensity is given in natural logarithmic scale. A linear fit to the data yields a propagation loss coefficient of 0.1±0.05 db/cm. Propagation loss coefficients very similar to those obtained by the cut-back technique have been obtained also for rib and strip-loaded structures, as reported in Table I. Values

Submitted to IEEE- Journal of Lightwave Technology 6 obtained by the two different techniques are in good agreement. This is quite important in establishing a good practice: the cut-back technique appears to be more correct, but it is sample and time consuming because of the needs of reproducible coupling conditions, good waveguide facets and long measurement times. The top-view method, once validated against the cut-back method, is very simple and rapid; the only requirement is to have relatively long samples in order to observe significant decrease in the scattered intensity of low-loss waveguides. From Table I, significant differences between the various samples can be observed: channel waveguides have low propagation losses, whereas rib and strip-loaded waveguides present one order of magnitude larger losses. The quite large loss value (about 2 db/cm) of rib waveguides (sample G1) is probably caused by their high sensitivity to storing and measuring condition. In fact, in rib waveguides the guiding layer is directly exposed to the ambient thus enhancing scattering losses due to irregularities and/or defects on the top surface. The propagation losses of strip-loaded waveguides (sample G5 and G6) are comparable to the corresponding planar waveguides, which show losses of 1.4±0.2 and 2.0±0.5 db/cm for symmetric and asymmetric structures respectively. The highest order guided modes of striploaded waveguides are not completely confined under the strip indeed as discussed above (see Fig. 5). Channel waveguides (samples G3 and G4) present propagation losses of about 0.1-0.2 db/cm. No significant variation can be attributed to the different plasma etching processes used. They cause a variation in the slope of the lateral edges, which are not critical because of the good confinement of the optical modes and the large width of the channel. On the contrary, the etching depth can play a key role: differences are observed between the propagation losses of samples G5 (chemical etching, losses of about 2.5 db/cm) and G6 (plasma etching, losses of about 1.5 db/cm). The strip-loaded waveguides (sample G6) defined by plasma etching result in a structure with a large confinement of the modes under the loaded area. This is caused by the removal of about 30nm of Si 3 N 4 (as shown by SEM) during the oxide plasma etching process, which is not sensitive to the Si 3 N 4 /SiO 2 interface. Hence the effective refractive index of the lateral cladding is lower for sample G6 than for sample G5, which causes in turn a better optical confinement (as shown in Fig. 7). The propagation losses values in our waveguides are within the state of the art of silicon nitrides waveguides: 0.1 db/cm for TE 0 mode at 632.8 nm has been reported in one of the earliest work on silicon nitride wave-guiding film [7] and in more recent papers [13], [14]; whereas higher values, about 0.3 and 0.6 db/cm, have been reported at 632.8 nm on silicon nitride planar [11] and ridge-type waveguides [12], respectively. It is worth noting that the strength of our results relies in the careful control and CMOS compatibility of the deposition process, and in the reduced thickness of the silicon nitride core layer. Moreover, the use of LPCVD technique results in the absence of hydrogen in the silicon nitride layer (as shown by IR absorption measurements), which avoids the absorption losses due to the vibrational overtones of Si-H and N-H bonds, and which allows its use as core material for waveguides in the near IR region. V. CONCLUSION In this paper we have presented our first step towards a Si integrated microphotonics by developing low-losses waveguides within a CMOS fabrication line. All the used processes are standard for microelectronics industry, which allows a good transferability of the reported results. We have designed, produced and tested waveguides of different geometries and found that the best choice in terms of propagation losses is the channel waveguides, which have shown losses of 0.1-0.2 db/cm. For multimode waveguides in the visible a quite large channel width can be used which relax the constraints on the steepness of the channel edge and allows the use of wet (chemical) as well as dry (plasma) etching method. We have also validate a simple technique to measure optical losses with respect to the more sample and time consuming technique of the cut-back method. REFERENCES [1] D. P. Seraphin and D. E. Barr, Interconnect and packaging technology in the 90 s, in Proc. SPIE vol. 1390, 1990, pp. 39-54. [2] R. Haveman, Scaling and integration challenges for Cu/low k dielectrics, in Proc. Workshop Processing for ULSI: Transistors to Interconnects, Austin, TX, Apr. 22, 1999. [3] L. Pavesi, Will Silicon be the photonics material of the third millennium?, Journal of Physics Condensed Matter Topical Review, vol. 15, pp. R1169-1196, Apr. 2003. [4] E. Degoli (ed), Optoelectronic interconnects for integrated circuits Achievements 1998-2001, EC (June 2001). [5] Hei Wong, Recent developments in silicon optoelectronic devices, Microelectronics Rel., vol. 42, pp. 317-326, 2002. [6] Y. Hibino, Passive optical devices for photonic networks, IEICE Trans. Commun., vol. E83-B, pp. 2178-2190, Oct. 2000. [7] W. Stutius and W. Streifer, Silicon nitride films on silicon for optical waveguides, Appl. Optics, vol. 16, 3218-3222, Dec. 1977. [8] S. Siriam, W. D. Partlow, and C. S. Liu, Low-loss optical waveguides using plasma-deposited silicon nitride, Appl. Optics, vol. 22, pp. 3664-3665, Dec. 1983. [9] C. H. Henry, R. F. Kazarinov, H. J. Lee, K. J. Orlowsky, and L. E. Katz, Low loss Si 3 N 4 -SiO 2 optical waveguides on Si, Appl. Optics, vol. 26, pp. 2621-2624, July 1987. [10] P. May, S. Basu, G. L.-T. Chiu, and G. Arjavalingam, Modal dispersion and attenuation measurements of silicon nitride and silicon oxynitride waveguides using a streak camera, IEEE J. Lightwave Technol., vol. 8, pp. 235-238, Feb. 1990. [11] G. N. De Brabander, J. T. Boyd, and H. E. Jackson, Single polarization optical waveguide on silicon, IEEE J. Quantum Electron., vol. 27, pp. 575-579, March 1991. [12] M. Ohtani and M. Hanabusa, Silicon nitride ridge-type optical waveguides fabricated on oxidized silicon by laser direct writing, Appl. Optics, vol. 31, pp. 5830-5832, Sept. 1992. [13] T. Inukai and K. Ono, Optical characteristics of amorphous silicon nitride thin films prepared by electron-cyclotron-resonance plasmachemical-vapor-deposition, Japan. J. Appl. Phys., vol. 33 (5A), pp. 2593-2598, May 1994. [14] D. A. P. Bulla, B. V. Borges, M. A. Romero, N. I. Morimoto, L. G. Neto, and A. L. Cortes, Design and fabrication of SiO 2 /Si 3 N 4 CVD

Submitted to IEEE- Journal of Lightwave Technology 7 optical waveguides, in SBMO/IEEE MTT-S IMOC 99 Proceedings, 1999, pp. 454-457.M. [15] R. M. de Ridder, K. Worhoff, A. Driessen, P. V. Lambeck, and H. Albers, Silicon oxynitride waveguiding structures for application in optical communication, Special Issue on Silicon-Based Optoelectronics, IEEE J. Selected Topics Quantum Electron., vol. 4, pp. 930-937, 1998. [16] C. Netti, M. D. B. Charlton, G. J. Parker, and J. J. Baumberg, Visible photonic band gap engineering in silicon nitride waveguides, Appl. Phys. Lett., vol. 76, pp. 991-993, Feb. 2000. [17] E. D. Palik (ed) Handbook of Optical Constants of Solids (New York: Academic Press), 1985.