1. Introduction

Although most of the major advances in silicon photonics have been based on single-crystalline silicon-on-insulator (SOI), the demands for polycrystalline silicon (polySi) based photonic devices are increasing owing to the unique feature of polySi as it can be easily grown on almost any substrate by standard techniques and simultaneously it has electron mobility of the order of 100 cm²/V·s, making it still capable of being used as electrically active layers. In high-speed metal-oxide-semiconductor (MOS) capacitor-based silicon optical modulators and in horizontal slot waveguides for electrical injection [1,2], polySi has been utilized as part of the waveguide material above a thin oxide layer where the fabrication of a single-crystalline silicon layer is difficult. More importantly, polySi-based photonics provides a simple solution for multi-level or three-dimensional (3-D) integration of optical networks in Si integrated circuits. Ring resonators and electro-optics modulators entirely based on the deposited polySi films have been demonstrated recently [3,4].

However, polySi-based photonic devices suffer a major limitation − a relatively large propagation loss in polySi waveguides due to light absorption and scattering at the grain boundaries is present in polySi. Because polySi’s nature depends strongly on the fabrication details [5], the losses in polySi waveguides also depend strongly on the detailed fabrication process. Historically, polySi waveguides fabricated on directly deposited polySi layers exhibit very large optical loss (>70 dB/cm) mainly because of the rough surface of the polySi layer and/or the small grain size [6,7]. The loss was reduced to ~34 dB/cm after smoothing the surface by chemical mechanical polishing (CMP) [7]. On the other hand, polySi waveguides fabricated by solid-phase crystallization (SRC) of deposited amorphous silicon (a-Si) exhibit a relatively low loss due to its smooth surface and/or large grain size [8,9]. ~11 dB/cm loss was reported for the 200-nm-thick polySi waveguides and the loss was further reduced to ~9 dB/cm by remote electron cyclotron resonance (R-ECR) plasma hydrogenation [8]. By replacing the cladding material from SiO₂ to SiON, ~6.45 dB/cm loss for the TE mode and ~7.11 dB/cm for the TM mode were reported [9], which approaches its SOI counterparts [10]. However, in all the above reports [2–4,8,9], a 1100°C/long-time annealing procedure was carried out in order to crystallize the a-Si into polySi with the maximum grain size. However, such a high thermal budget restricts the polySi layers to be fabricated before any doping procedure, thus making it difficult to be implemented in the 3-D integration and even in the MOS-type modulator fabrication because some doping procedures have to be done before the polySi waveguide fabrication. Thus, it is extremely important to reduce the thermal budget for the polySi waveguide fabrication without sacrificing its optical loss property. Moreover, when polySi is used as active layers in photonic devices, it is usually is doped. However, the doping effect on the optical loss in polySi waveguides has not been reported yet in literature.

In this work, firstly, the maximum temperature for the polySi waveguide fabrication is reduced from 1100°C to 1000°C. Secondly, two fabrication approaches are compared: in one (normal) approach, the as-deposited a-Si film is first crystallized into polySi and then this polySi film is patterned into waveguide structures by dry etch; in the other (modified) approach, the as-deposited a-Si film is first patterned into the waveguide structures by dry etch and then crystallized them into polySi. Thirdly, some polySi waveguides are uniformly doped by phosphorus with the concentration ranging from 1 × 10¹⁷ to 5 × 10¹⁸ cm⁻³ by ion implantation, and the doping effect on the propagation loss is evaluated. Finally, a standard forming gas (FG) annealing is carried out for some wafers and the influence of FG annealing on both undoped and n-doped polySi waveguides is investigated.

2. Experimental

~2-μm-thick SiO₂ layer was deposited on 8-inch, <100>-oriented Si wafers using high-density plasma (HDP) chemical vapor deposition (CVD) at 700°C as the bottom cladding layer. The as-deposited HDP-SiO₂ film has quite a smooth surface with the root mean square (RMS) surface roughness of ~0.13 nm, as measured by atomic force microscope (AFM). The scanning area in the AFM measurements is kept at 5 μm × 5 μm throughout this paper. A 220-nm-thick a-Si layer was then deposited by low-pressure CVD (LPCVD) at 540°C from SiH₄. The surface roughness of as-deposited a-Si is ~0.34 nm. In the normal approach, the a-Si film was crystallized into polySi using a two-step annealing procedure in a pure N₂ ambient: first at 575°C for 15 h and then at 1000°C for 0.5 h. The surface roughness after the first and second annealing step is ~0.38 nm and ~0.39 nm, respectively, in agreement with the previous reports that the polySi film formed by SPC has quite a smooth surface [8,9]. After a 50-nm-thick plasma-enhanced CVD (PECVD) SiO₂ deposition and photoresist coating, the waveguide structures were patterned using 248-nm deep UV lithography. The thin SiO₂ layer was first dry etched by CF₄ using the photoresist as mask, followed by photoresist stripping by O₂ + N₂, and then the polySi layer was dry etched down to the bottom SiO₂ layer using the thin SiO₂ layer as the hard mask by the gaseous mixture of Cl₂ + He-O₂. After wet etching the remaining SiO₂ and depositing a 5-nm-thick screening SiO₂, two-energy (30 and 100 keV) phosphorus implantation was carried out for some wafers, followed by rapid thermal annealing (RTA) at 1030°C for 30 s for dopant activation. The implantation dose was chosen based on the Tsuprem4 simulation to make a uniform phosphorus distribution in polySi waveguides with concentration of ~1 × 10¹⁷, ~5 × 10¹⁷ cm⁻³, ~1 × 10¹⁸ cm⁻³, and ~5 × 10¹⁸ cm⁻³, respectively. Standard forming gas (10% H₂ + 90% N₂) annealing was carried out at 420°C for 30 min for some wafers. Then, ~2-μm PECVD SiO₂ was deposited at 400°C as the upper cladding layer. To form smooth chip facets for the end-fire coupling, deep trenches were fabricated by dry etch of 4-μm-deep SiO₂ and subsequent ~120-μm-deep Si, located at the intervals between chips. The wafers were then diced along the middle of the deep trenches without touching their sidewall. Since the sidewall of the deep trenches was quite smooth hence there was no need to manually polish the chip facets after the wafer dicing.

In the modified approach, the a-Si film was first patterned with the waveguide structures using lithography and dry etched down to the SiO₂ layer using a 50-nm-thick SiO₂ layer as the hard mask. The same abovementioned dry etch recipe was used. Then, the a-Si wire waveguides were crystallized into polySi using the abovementioned two-step annealing procedure. The other processes are exactly the same as those in the normal approach. For comparison, some chips taken from the fabricated wafers were additionally annealed at 1100°C for 1 h in N₂ ambient.

The waveguide patterns have a width of 200, 300, 400, and 500 nm. All waveguides contain a 200-nm-wide, 200-μm-long inverted taper structure at both facets for coupling. The scanning electron microscopy (SEM) image of a waveguide near the tip area is shown in the inset of Fig. 1 . No differences are observed from SEM images in waveguides fabricated by the normal and modified approaches. There are 7 waveguides with lengths ranging from 0.69 to 2.68 cm for each width set. Each waveguide contains 10 identical bends with a large bend radius of 20 μm (thus the bend induced loss is negligible [10]). 1550-nm light output from a stabilized laser source is coupled from a lensed fiber in to the waveguide in one facet and coupled out to another lensed fiber in the opposite facet. The fibers and the chip under test were mounted on an XYZ micrometer piezo-stage for precision alignment to search the maximum output power (dBm). The insertion loss (dB) of waveguide is obtained after subtracting the fiber-to-fiber loss (dBm), i.e., the maximum output power measured without propagating through the waveguide. The propagation losses of TE mode were also measured for several samples using a polarization-maintaining fiber and a polarization controller. The measured loss values are quite close (the difference is within 1 dB/cm) to those measured without the polarization control, in agreement to the previous report [9]. Therefore, for simplicity, the loss data reported in this paper were all measured without polarization control. Figure 1 shows one example of the measured insertion losses as a function of waveguide length. They exhibit quite good linearity. By linearly fitting the experimental points, the propagation loss and the coupling loss are extracted for each waveguide width. The coupling loss (including two facets) is around 5.5–6.5 dB for all waveguide sets. Taking the possible variation of coupling loss among waveguides into account, the error of propagation loss determined by such a cutback method is estimated to be less than ± 1.0 dB. However, the linearity of experimental points becomes poorer and the error becomes larger if the output power becomes very small (e.g., < −50 dBm). Therefore, the propagation loss in waveguides with very large loss (e.g., >50 dB/cm) cannot be accurately determined.

 

Fig. 1 Insertion losses of polySi waveguides as a function of the waveguide length for three width sets. The linear fitting lines are also shown, from which the propagation losses and the coupling losses are extracted. The inset shows a SEM image of the waveguide near the tip area.

Download Full Size | PDF

3. Results and discussion

3.1 Effect of waveguide width

Table 1 summarizes the propagation losses at 1550 nm in the undoped polySi wire waveguides with height of 220 nm and widths of 200, 300, 400, and 500 nm, fabricated using the normal and the modified approaches. For waveguides fabricated by the normal approach (i.e., first SPC and then patterning), the propagation losses are 13.0, 11.3, and 9.8 dB/cm for width = 500, 400, and 300 nm, respectively, while the 200-nm-wide waveguides have large loss (>70 dB/cm). It is well known that the light attenuation in polySi waveguides can be attributed to two main origins: one relates to the light absorption and scattering at the grain boundaries present in the waveguides (namely the bulk loss) and the other relates to the light scattering at the core/cladding interface (namely the interface loss). Because the narrower wire waveguides contain relatively less grains in the waveguides and they confine light less tightly (i.e., the light mode spreads more widely in the surrounding SiO₂ layer), the bulk loss decreases while the interface loss increases with the waveguide width shrinking. The results show that the overall loss decreases with the waveguide width shrinking from 500 to 300 nm, indicating that the bulk loss dominates in waveguides with the width larger than 300 nm. Because the bulk loss is determined by the grain size and/or crystalline fraction of polySi, which can be improved by higher temperature annealing, we can expect that polySi waveguides fabricated at lower temperature have larger bulk loss. This is seen in the 575°C annealed polySi waveguides which have much larger losses and this loss increases more rapidly with the waveguide width increasing from 300 to 500 nm (see Table 1). For the 1100°C annealed polySi waveguides reported in [9], on the other hand, the overall loss is relatively small and it decreases (not increases) slightly with the waveguide width increasing from 300 to 700 nm [9], indicating that the bulk loss is not the dominant contributor for those waveguides.

Table 1. Propagation losses at 1550 nm in various undoped polySi wire waveguides, extracted from the cutback method. The error is estimated to be less than ± 1.0 dB/cm.

View Table

After an additional 1100°C/1 h annealing, the losses in the initial 1000°C annealed waveguides become 10.9, 10.0, and 11.1 dB/cm for widths of 300, 400, and 500 nm, respectively, very close to those reported by Liao et al. where the polySi layers were crystallized at a maximum temperature of 1100°C. We can see that the losses in the 400- and 500-nm-wide waveguides reduce slightly after the additional annealing, which can be attributed to the reduction of bulk loss due to the further crystallization of polySi. For the initial 575°C annealed polySi waveguides, as expected, the loss reduction is more significant after the additional 1100°C annealing. The 400- and 500-nm-wide waveguides have losses of 10.2 and 11.7 dB/cm, respectively, very close to those of the initial 1000°C annealed waveguides, indicating that the crystallization of polySi is almost saturated after the 1100°C annealing [9]. However, the loss in the 300-nm-wide waveguides becomes slightly larger after the additional 1100°C annealing, which should be attributed to the increase of interface loss because the higher temperature annealing may slightly roughen the polySi waveguide interface. The above results indicate that although the higher temperature annealing (here 1100°C) is beneficial for the bulk loss reduction due to the improvement in crystallinity, its contribution to the overall loss decreases with the waveguide width shrinking and it may even becomes detrimental to the overall loss as the interface loss may increase slightly after the higher temperature annealing. For the 300-nm waveguides, the 1000°C anneal is sufficient to reach the lowest overall loss.

3.2 Effect of fabrication approach

The 200-nm-wide waveguides fabricated by the normal approach have a very large loss, which can be attributed to the large interface loss. Based on simulation of waveguides with identical interface roughness, it has been known that the interface loss increases with the waveguide width decreasing, and becomes substantially large for the width narrower than 300 nm [9,11]. Because the polySi waveguides have quite smooth bottom and top interfaces, we deduce that the loss mainly originates from the relatively rough sidewall induced by the dry etch process. Due to the lack of suitable measurement tools, e.g., the 3-D AFM tool, we cannot directly measure the sidewall roughness. Alternatively, the dry etch related surface roughness is estimated using the following experiments. In one experiment, a 400-nm-thick a-Si film was crystallized into polySi using the abovementioned two-step annealing method. The surface roughness was measured to be ~0.7 nm. Then the polySi film was partially etched by ~200 nm using the same recipe as that was used for the waveguide etch. The roughness of the remaining polySi film became ~4.0 nm. The relatively rough surface of the partially dry-etched polySi film may be attributed to the non-homogeneous etch nature of polySi layer because of the presence of grain boundaries and/or different grain facets in the surface of polySi layer. We suspect that the waveguide sidewall may have the RMS roughness in the range of ~4.0 nm, close to the other groups’ estimation [8,12]. In the other experiment, a 400-nm a-Si film (the surface roughness is ~0.4 nm) was partially etched by ~300 nm using the same dry etching recipe, the surface roughness of the remaining ~100-nm a-Si layer was measured to be ~0.4 nm. After crystallization using the abovementioned two-step annealing process, the roughness of polySi was measured to be ~0.5 nm. The relatively smooth surface of the partially etched a-Si can be attributed to the homogenous etch nature of a-Si, which is also partially confirmed by the fact that very low loss a-Si waveguides can be fabricated [13]. Our experiment indicates that the partially dry-etched a-Si layer retains the smooth surface after SPC, just like the as-deposited a-Si layer. We thus suspect that the waveguides fabricated using the modified approach (i.e., first patterning the a-Si and then crystallization) may have smooth sidewall with the RMS roughness in the range of ~0.5 nm.

The losses of polySi waveguides fabricated by the modified approach are also listed in Table 1. The losses are 13.9, 16.3, 18.1, and 25.7 dB/cm for the waveguides with widths of 200, 300, 400, and 500 nm, respectively. We see that the 200-nm-wide waveguides have much lower loss than those fabricated by the normal approach, confirming that these waveguides really have a reasonable smooth sidewall. However, the wider waveguides have larger loss and the loss increases with the waveguide width more rapidly than those fabricated by the normal approach. It indicates that the waveguides fabricated by the modified approach have larger dominant bulk loss, most probably due to their smaller grain size and/or poorer crystallinity than those fabricated by the normal approach. Since the same thermal treatment is carried out in both approaches, it indicates that the crystallization of a-Si may be layout-dependent, namely, the crystallization of a-Si wire is not as effective as the crystallization of a-Si film under the same annealing condition. One possible reason may be attributed to the fact that the a-Si wire has a much larger interface-volume ratio than the a-Si film − since the initial nucleation may occur more easily at the interface and the subsequent grain growth may terminate at the interface, the grain size in the final polySi wire is smaller (thus containing more grain boundaries) than that in the final polySi film after the same thermal treatment. Another possible reason is that the grain growth rate in the a-Si wire may be smaller than that in the a-Si film under the same thermal treatment, thus leading to a smaller crystalline faction in the final polySi wire waveguides. To understand this behavior unambiguously, more experimental studies are necessary, such as refractive index measurement, x-ray diffraction (XRD), and transmission electron microscopy (TEM), etc. which is still ongoing. After an additional annealing at 1100°C for 1 h, as expected, the losses in waveguides with the width ≥ 300 nm are reduced due to further crystallization of the polySi waveguides, like those fabricated by the normal approach. However, the final losses in these waveguides (except the 200-nm-wide waveguides) after the additional 1100°C/1 h annealing are not as low as those fabricated by the normal approach, which can also be ascribed to the insufficient crystallization of polySi wire waveguides. For the 200-nm-wide waveguides, the loss keeps almost the same after the additional 1100°C annealing probably due to the contrary effects of the additional high-temperature annealing on the bulk loss and the interface loss.

Nevertheless, the modified approach can significantly improve the sidewall roughness of the polySi waveguides, which compensates the degradation of its crystallization. This approach can be used to fabricate very narrow wire waveguides (e.g., ≤200 nm) where the interface loss dominates. Furthermore, if the crystallization of a-Si wire can be improved, such as using the laser annealing technology, very low loss polySi wire waveguides may be expected.

3.3 Effect of phosphorus doping

The doped Si waveguides may have two additional loss origins: one from the free carrier absorption and the other from the absorption and/or scattering in the possible doping-related defects. The former is an intrinsic and useful effect in the active photonic devices whereas the latter should be minimized. Using the well known formula for the 1550-nm wavelength: Δα = 8.5 × 10⁻¹⁸ · ΔN + 6.0 × 10⁻¹⁸ · ΔP, where Δα (cm⁻¹), ΔN (cm⁻³), and ΔP (cm⁻³) are the absorption-coefficient, the free electron density, and the free hole density, respectively [14], it can be calculated that the 1 × 10¹⁸ cm⁻³ n-type doping causes ~36 dB/cm bulk loss purely due to the free-carrier absorption if every dopant contributes a free electron. For the SOI strip waveguides, we have experimentally measured that the undoped waveguide has a loss of ~1 dB/cm, the 1 × 10¹⁸ cm⁻³ phosphorus doped waveguide has a loss of ~24 dB/cm, and the 1 × 10¹⁹ cm⁻³ phosphorus doped waveguide has a very large loss, roughly in agreement with those calculated from the above formula taking the waveguide geometry (thus the light mode confinement factor) into account.

Figure 2 shows the additional loss induced by phosphorus doping in polySi waveguides as a function of the doping concentration. As expected, the additional loss increases with the doping concentration. The increasing rate depends on both the waveguide size and the fabrication approach, and is relatively slow for concentration up to 1 × 10¹⁸ cm⁻³, whereas the 5 × 10¹⁸ cm⁻³ doped waveguides have a very large loss due to the domination of the free carrier absorption. Comparing the 1 × 10¹⁸ cm⁻³ doped polySi and SOI waveguides, we can see that the additional loss induced by phosphorus doping in the polySi waveguides is smaller than that in the SOI waveguides, probably because some dopants in polySi waveguides are segregated at the grain boundaries so that they do not contribute free electrons. The polySi waveguides fabricated by the modified approach may have more significant phosphorus segregation effect because they have smaller grain size and/or poorer crystallinity. We can see that these waveguides have even smaller additional loss induced by the 1 × 10¹⁸ cm⁻³ phosphorus doping. The above results allow us to draw a conclusion that the phosphorus doping induced addition loss in polySi waveguides may be still mainly determined by the free carrier absorption, as the SOI counterparts. In other words, the loss induced by the possible doping-related defects, such as those segregated at the grain boundaries, is minor.

 

Fig. 2 The phosphorus-induced additional loss ( = loss of doped waveguide – loss of corresponding undoped waveguide) as a function of phosphorus concentration. The solid symbols represent those fabricated by the normal approach and the open symbols represent those fabricated by the modified approach.

Download Full Size | PDF

3.4 Effect of forming gas anneal

It has been known that hydrogen passivation of dangling bonds at the grain boundaries in polySi waveguides can reduce the propagation loss [8]. It was argued that among various passivation techniques, the standard forming gas (FG) annealing is not an effective technique to passivate the dangling bonds in polySi films [5]. However, there was a report that the polySi waveguide loss is reduced from ~14 to ~11 dB/cm after the FG annealing [12].

Figure 3 shows the FG annealing induced loss variation of our undoped and n-doped polySi waveguides. For the undoped polySi waveguides fabricated by either the normal approach or the modified approach, ~1-2 dB/cm loss reduction is obtained after the standard FG annealing, in agreement to that reported in [11]. The lowest loss is achieved to be 7.9 dB/cm in the 300-nm-wide polySi waveguides and the 200-nm-wide polySi waveguides have a loss of 12.8 dB/cm. The effectiveness of the FG annealing for the loss reduction in undoped polySi waveguides may be attributed to the fact that the FG annealing is carried out after the waveguide patterning so that the surface-to-volume ratio is larger than that in the case of poly-Si film. The dangling bonds at the polySi waveguide sidewall can be passivated by hydrogen from the FG gas ambient, unlike in the case of polySi films where the passivation depends on the diffusion of hydrogen from the surface into the bulk. Several FG annealed chips were re-measured after storing them at room-temperature for several weeks, the measured propagation losses show no substantial variation, indicating that the FG annealing induced hydrogen in polySi waveguide is quite stable at room temperature.

 

Fig. 3 The forming gas anneal-induced loss variation ( = loss of FG annealed waveguide – loss of corresponding waveguide without the FG anneal) for undoped and doped polySi waveguides. The solid symbols represent those fabricated by the normal approach and the open symbols represent those fabricated by the modified approach.

Download Full Size | PDF

However, for the phosphorus doped polySi waveguides, Fig. 3 shows that the loss increases, by ~0.1–4.9 dB/cm, after the FG annealing. The amount of loss increase depends on both the phosphorus concentration and the waveguide width: slightly larger for the higher phosphorus concentration. A possible explanation of this phenomenon is that the dangling bonds in the grain boundaries may already have combined with phosphorus because phosphorus in polySi tends to segregate at the grain boundaries [5], therefore, during the FG annealing, hydrogen cannot passivate the dangling bonds, instead, the hydrogen and phosphorus may form a cluster at the dangling sites during the FG annealing. These clusters may contribute to light absorption and/or scattering, leading to a slightly enhanced light attenuation in the FG annealed n-type doped polySi waveguides. More studies are still on-going to fully understand the physics beyond this phenomenon.

4. Conclusion

The propagation losses in various polySi wire waveguides are measured using the cutback method. Although a lower crystallization temperature leads to a higher bulk loss, its contribution to the overall loss decreases as the waveguide width decreases. For the 300-nm-wide waveguides, the 1000°C annealing is sufficient to give the lowest loss. The modified approach, i.e., patterning the a-Si into waveguides first and then crystallization, can significantly improve the sidewall roughness of final polySi waveguides but the crystallization property of polySi is degraded, leading to a relatively low loss in the 200-nm-wide waveguides but lager loss in the wider waveguides. The additional loss in polySi waveguides due to phosphorus doping is smaller than that in the corresponding SOI waveguides. This is probably due to the phosphorus segregation in polySi. The loss due to the possible doping-related defects is minor. For undoped polySi waveguides, the standard FG annealing can further reduce the loss by 1–2 dB/cm, while for the phosphorus-doped polySi waveguides, the losses increase slightly after the FG annealing.