Increase Optical Bandwidth with Epi CVD

Because light has an information-carrying capacity many orders of magnitude greater than radio frequency signals, optical networking has become the most important medium for information transfer. A demand for increasing the speed and density of information transfer drives a demand for increasing the bandwidth of optical networks. Optical systems, such as planar lightwave circuits (PLCs), can carry signals with increasingly larger numbers of wavelengths that have to be managed through devices with higher degrees of functionality and sophistication.

PLCs can include both passive and active components. An example of a passive device implemented in integrated form is the arrayed waveguide grating (AWG). This structure can be viewed as a two-dimensional diffraction grating, and it has been successfully used for multiplexing/demultiplexing, add-drop multiplexers, and dynamic gain filter equalization. In addition to passive waveguide devices, PLCs may include active optical components such as switches and photodetectors. Having active components is highly desirable for optical circuits with a compact integration of devices and different but complementary functionality.

It is desirable to integrate various optical components on a single wafer. There is a need for techniques and equipment that can accomplish this task in a cost-effective, reliable and manufacturable fashion.¹

One PLC fabrication approach makes use of III-V materials such as indium phosphide 3 in. wafers. Manufacturing difficulties have, however, prevented large-scale and low-cost production of InP-based optical devices.

Another approach to PLC fabrication is to use silicon as a waveguide material in the silicon dioxide cladding. A large difference in the refractive index, Dn, between core and cladding materials — 3.5 for silicon and 1.5 for SiO₂, respectively — can allow for waveguide structures with a small bend radius and thus a reduced optical device size. On the other hand, the large Dn causes a variety of problems. For example, the Si-in-SiO₂ waveguide is very sensitive to line-edge roughness and may require edge roughness control beyond the capabilities of commercial optical systems. Also, systems with a large Dn tend to carry numerous modes simultaneously, as opposed to single-mode propagation in waveguides with a smaller Dn. These modes travel different distances inside the waveguide, a phenomenon called modal dispersion, resulting in a poor signal quality at the output. It is therefore desirable to use single-mode waveguides.

Lower-Dn waveguides can be fabricated using silicon oxynitride/ SiO₂ or doped silica. A 1 Tb/sec optical packet switch fabric based on this approach with an 80 × 80 channel AWG has been demonstrated.² SiO₂ waveguide technology, however, is mainly passive and lacks the potential of flexible active device integration.

In the following study, we found that epitaxial silicon germanium can address the disadvantages of the previously mentioned methods. Indeed, standard silicon processing equipment, such as silicon etch systems and epi CVD (chemical vapor deposition) reactors, exist to form SiGe structures in a production-worthy fashion. Due to the high quality of epitaxial films,^3,4 the propagation losses can be expected to be low. Changing germanium concentration should allow the Dn to be controllably changed to meet specifications of optical devices. SiGe waveguides can be fabricated on the same wafer with SiGe or germanium photodetectors using the same deposition and etch equipment. By doping silicon, one can form conductive electrodes that locally change the refractive index and thus create active devices.

1. Epi SiGe waveguide fabrication flow (a-c). Top-view optical micrograph of an individual Si/SiGe/Si waveguide structure (d). A pyramid above the SiGe core is formed on the top surface (c-d).

Epi SiGe waveguides

The epi SiGe waveguide fabrication flow we studied is illustrated in Figure 1. First, a blanket silicon film was deposited on the silicon substrate (Fig. 1a) to serve as the lower cladding layer. Next, a blanket SiGe film was deposited on the substrate to serve at the core layer. For the study, photolithography was used to pattern the wafer and an Applied Materials Silicon Etch DPS (decoupled plasma source) Centura system was used to etch the pattern. Etching leaves a SiGe strip on the silicon substrate (Fig. 1b ). The strip typically has a square cross section, in the 0.5-5 µm² range. The etch was followed by an ex situ wet cleaning step, after which the wafer was loaded into the Centura deposition system. A bake in hydrogen was necessary to remove the native oxide layer before depositing the next epitaxial layer. The bake can be done either in an epi deposition chamber or a dedicated EpiClean chamber,⁵ which allows for native oxide removal at a reduced thermal budget. Performing the bake in a dedicated chamber was possible because of the multi-chamber architecture of the Centura platform. After the deposition of a thick silicon layer (Fig. 1c ), the fabrication was complete. In this structure, SiGe forms the core of the waveguide, and the silicon layer forms the cladding.

The crystalline order is maintained throughout the growth of the structures from the substrates all the way up to the top surface of the upper cladding, as manifested by the clearly visible crystalline facets in Figure 1d . The optical micrograph shows that these facets form a pyramid. If necessary for device integration, the pyramids above the SiGe cores can be removed by chemical mechanical polishing (CMP).

2. Top-view optical micrograph of Si/SiGe/Si strips, showing no defect structure (right) and a structure with multiple defects (left). The defects were formed as a result of the absence of a hydrogen bake.

To achieve a high-quality crystalline growth, special care was given to cleaning the surface upon which an epitaxial layer was grown. We used an ex situ wet clean to remove organic, ionic and heavy metal contaminants and followed this with an in situ hydrogen bake. The hydrogen bake removes residual contamination. The concentration of impurities inside the layers was below detection limits of secondary ion mass spectroscopy (SIMS): <2×10¹⁷ atoms/cm³ for oxygen, <1×10¹⁷ atoms/cm³ for carbon, <5×10¹⁵ atoms/cm³ for fluorine, and <2×10¹⁶ atoms/cm³ for boron and phosphorous. The right image in Figure 2 shows a top-view optical micrograph of a defect-free waveguide structure. The structure consists of pyramids that are 80 µm apart, with a SiGe core below each pyramid. No defects can be detected either between pyramids or on the pyramid facets.

The high quality of the epitaxial growth can also be seen in Figure 3 . The cross-sectional SEM image demonstrates that there are no voids between the SiGe core and silicon cladding.

3. Cross-sectional SEM micrograph of the waveguide structure. The dotted line shows the location of the SiGe core (left). Digital camera image of the distribution of the light propagated by the SiGe core (right).

Inadequate cleaning may result in defect formation. To assess the effects of inadequate cleaning, we eliminated the hydrogen bake. The result is seen in Figure 2 (left). Each defect is formed on the surface before growth of the top silicon cladding layer (Fig. 1b). The lateral dimensions of the defects increase during the formation of the upper silicon cladding so that, when the waveguide fabrication is completed, the defect size on the top surface is several microns. Figure 2 also shows an example of a crater-like defect.

In addition to achieving high quality in silicon and SiGe deposition, it is important to utilize high-growth-rate processes in fabrication of waveguides because their total thickness is on the order of several microns. In fabricating SiGe waveguides using the epi system we have achieved growth rates of >0.15 µm/min. Multi-chamber architecture of the system^2,3 allows the tool's throughput to be further increased, resulting in production-worthiness of the SiGe waveguide fabrication process.

The right panel of Figure 3 presents direct evidence of the waveguiding effect. To obtain that image, light at a wavelength of 1.55 µm from an optical fiber was coupled to one end of the waveguide. A miniature infrared camera detected the light on the other end of the waveguide. The signal from the camera was converted to a visible light image and displayed on a monitor. It can be seen that light is propagated by the SiGe core only.

4. Marcatili’s model of the electric field distribution. The square indicates the SiGe waveguide core. The blue squares in the corners are areas where the model is not applicable.

This observation is in agreement with calculations of the electrical field in the vicinity of the SiGe core, which have been done based on the model by Marcatili.⁶ Figure 4 depicts the electrical field distribution both inside and outside the core. The solid red line presents the boundary of the SiGe core. The Marcatili model does not work in the vicinity of the waveguide corners, and these areas near the corners are shown in blue. The calculations show that the electrical field exponentially decreases outside the core of the waveguide according to the following expression: E~exp(-0.7 µm/x), where x is the distance from the center of the core.

5. Difference in the refractive index between SiGe and silicon (top), and cutback measurement results (bottom). The slope of the log (Pi/Po) vs. waveguide length gives a propagation loss value of 0.31 dB/cm (Pi is the input power of the light beam, Po is output power).

The larger the index contrast, the greater the confinement of the electromagnetic field inside the core. Figure 5 (top) shows the difference in the refractive index between Si_1-x Ge_x and silicon. The wavelength for index measurements was 0.633 µm. The Δn increases linearly with the increasing germanium atomic concentration x in the SiGe cores Δn=αx, where α=1. The germanium concentration is therefore a very convenient and well-controlled parameter to tune the optical properties of the waveguide, and tailor them to the needs of a specific optical device design.

Another important property of waveguides is propagation loss, usually measured in decibels per centimeter (10 dB/cm represents a decrease of power by an order of magnitude per 1 cm of waveguide length). We determined the waveguide propagation losses by measuring the ratio between input and output optical power through straight waveguides of different lengths. The starting wafer sample was ~2 cm long and covered with several nominally identical channel waveguides. The coupling surfaces were polished and the transmissivity through the waveguides was measured. The propagation length was then reduced by cutting the sample and repolishing the coupling surface. By averaging each value over more than 10 measurements per waveguide length, we extrapolated the propagation losses by determining the slope of the input-to-output power ratio vs. waveguide length plot (Fig. 5 , bottom). Because of the averaging procedure we could assume the coupling losses between fiber and silicon wafer to be constant.

For waveguides with the Δn between the SiGe core and the silicon cladding Δn/n=0.6%, the loss value for the wavelength of 1.55 µm is ~0.31 dB/cm. No measurable polarization dependence of the losses was detected in waveguides. Polarization dependence could be produced by strain in the SiGe core. Its absence is in agreement with the results of X-ray diffraction (XRD) analysis that showed that SiGe was relaxed.

It is interesting to note that the loss value of 0.31 dB/cm obtained in this work is about an order of magnitude lower than propagation loss values of 3 dB/cm in SiGe waveguides fabricated by other methods — ultrahigh vacuum (UHV) CVD⁷ and molecular beam epitaxy (MBE)⁸ — using a SiO₂ clad layer.⁹

We have demonstrated waveguiding properties in the infrared wavelength range in Si/SiGe/Si structures. The light in these single-mode waveguides propagates with 0.31 dB/cm signal attenuation (at a wavelength of 1.55 µm). Relaxation of SiGe results in no polarization and a high refractive index contrast between SiGe and silicon. These photonic structures have been fabricated using standard silicon semiconductor equipment, making particular use of a low-pressure CVD chamber for epitaxial deposition of the core and cladding, with a high throughput of the deposition processes.