Silicon photonics devices are fast evolving from discrete optical components to complete circuits that will eventually require low-loss waveguide interconnects and devices for high-density integration. The reliance of a silicon-on-insulator (SOI) platform for building up active optical devices limits the fabrication freedom in terms of process temperature for the formation of a multilayer photonics platform. Therefore, low temperature deposition of low-loss optical materials is essential for building high-density optical integrated circuits. There have been a variety of low temperature deposition techniques for optical waveguide materials. For example, the use of polymer waveguides is constrained to the uppermost optical layer because of its incompatibility with the CMOS thermal processes ^[1,2]. Low temperature plasma enhanced chemical vapor deposition (PECVD) materials such as silica- and nitride-based waveguides have been demonstrated. These materials can achieve low propagation losses, but their low refractive indices (∼1.5 and ∼2.1 at 1.55 μm, respectively) require mode matching to the silicon devices, adding a layer of complexity to the design and fabrication. In addition, silicon oxynitride and silicon nitride contain N-H stretching bonds from using ammonia in the deposition processes, which can cause absorption in the s- and c-band wavelength range. Although the N-H bonds can be annealed out at high temperatures close to 1000°C, such processing renders these materials unsuitable for back end of the line (BEOL) processes ^[3,4].

Hydrogenated amorphous silicon (a-Si:H) waveguides fabricated by the conventional deposition method (PECVD) have been investigated ^[5–12]. Their low refractive index difference compared with crystalline SOI waveguide devices simplifies integration, especially in multilayer photonics. The waveguide losses in a-Si:H can be investigated from the perspective of its lattice network, which affects absorption in the material. In crystalline silicon (c-Si), the tetrahedral lattice structure has a bond angle of 109.5° and a bond length of 2.35 Å (1 Å = 0.1 nm). The amorphous silicon lattice network comprises 5% and 10% deviation from the crystalline in the bond angle and bond length, respectively ^[13]. These deviations in amorphous structure do not affect the first few neighboring atoms, which is also known as the short range order (SRO) of an a-Si network. However, the amorphous disorder can induce structural stress to the intermediate range order (IRO; that is, beyond the neighboring atoms in the network), leading to bond breakage and the occurrence of dangling bonds. This has a negative impact on the optoelectronic properties of the a-Si waveguide material, which will suffer from high absorption and degradation over time. Therefore, hydrogen passivation is crucial during or after the deposition of amorphous silicon for the development of waveguide device applications.

There has been a lot of research carried out on different types of deposition and hydrogenation methods for a-Si:H material, such as sputtering, evaporation, glow discharge plasma chemical vapor deposition, PECVD, and catalytic chemical vapor deposition (Cat-CVD) ^[14–19]. PECVD has been widely reported for the deposition of a-Si:H waveguides ^[5–12], and the optical propagation losses are summarized and compared to this work in Table 1. Hot-wire chemical vapor deposition (HWCVD), also known as Cat-CVD, has been used to deposit a-Si:H, micro-crystalline silicon (μc-Si:H), and polycrystalline silicon for photovoltaic and thin film transistor applications ^[23]. Previous work performed by Takahiro et al. ^[24] demonstrated a waveguide loss of 15 dB/cm at a wavelength of 800 nm. The advantages of HWCVD are its low deposition temperature capability, high deposition rate, no plasma induced damage, and large area homogeneity ^[25–27]. The complete catalytic and thermal decomposition of the precursor gas silane (SiH4) is achieved by the hot filament dissociation of Si and four H atoms. In this work, we demonstrate low-loss a-Si:H waveguides for the telecoms wavelength band using the HWCVD process at temperatures compatible with BEOL (below 400°C). Raman spectroscopy was used to analyze the a-Si:H peaks and the amorphous network characteristics due to hydrogenation, thereby correlating its material quality to the waveguide propagation loss at the 1550 nm wavelength. Optical propagation losses were measured using the cut-back method, and control samples with 2 μm wide waveguides were used to initially evaluate the optical propagation characteristics. Sub-micrometer wire waveguides were then fabricated, and a propagation loss of 0.8 dB/cm has been obtained.

The fabrication process started with thermal oxidation of a p-type Si substrate to produce a 4.6 μm thick SiO2 cladding, followed by HWCVD of a-Si:H layer on top of the oxide layer. The thick SiO2 was for optical isolation and to emulate the buried oxide (BOX) of a SOI wafer. The a-Si:H samples used in this work were prepared with the Nitor 301 HWCVD system from ECHERKON Technologies Ltd. The system used 15 tungsten filaments mounted parallel to each other with 2 cm separation. Each filament has a length of 30 cm with diameter variations from 0.125 to 0.2 mm. The different filament wire diameters provide thermal heating to the substrate temperature (Tsub) ranging from 190°C to 320°C. The reason for using tungsten filaments is their good catalytic performance, low activation energy (71±20kJ/mol at filament temperature, Tfil<1700K ^[28]) for silicon, low cost, and that they are readily available. The distance from the substrate to the filaments was maintained at 6 cm. The system used a constant Tfil of 1850°C, which is high enough to achieve the complete dissociation of silane ^[29,30]. In order to maintain the control of film thickness, the substrate is first pre-heated without introducing any gases. By using different filament diameters, it is possible to obtain different film thickness. The a-Si:H thickness is 400 nm deposited at 0.01 mbar (1 bar = 100 kPa) with a silane and hydrogen gas mix ratio of 2.4∶1. The 400 nm thickness is chosen so that the Raman laser (wavelength) penetration depth can fully interrogate the deposited a-Si:H film without interference from the background silicon dioxide layer. After the a-Si:H deposition, the film was covered with 5 nm thick PECVD SiO2 at 350°C as a passivation layer to prevent atmospheric degradation. We prepared two batches of a-Si:H waveguides. The first batch was fabricated by standard photolithography and anisotropic inductively coupled plasma (ICP) etching to produce 2 μm wide, fully etched wire waveguides, which were used as control samples to study the role of hydrogen in the structural ordering and to correlate this to the optical propagation loss. The second batch was fabricated by direct write lithography and anisotropic ICP etching to produce fully etched wire waveguides with widths of 250, 350, 550, 650, and 850 nm. For each width in batch two, waveguides with different lengths ranging from 3 to 10 mm were produced with grating couplers. Finally, the a-Si:H waveguides were clad with a 1 μm thick PECVD SiO2 layer.

There are other measurement and characterization techniques such as electron spin resonance (ESR) and the constant photocurrent method (CPM) to evaluate the defects of amorphous silicon ^[12,31]. These are the viable methods to evaluate absorption in a-Si. In this work, however, we used Raman spectroscopy to evaluate the atomic structure to understand the crystallinity and density of the material. A Renishaw Invia Raman spectrometer with a 532 nm Nd:YAG excitation laser was used. The laser has a penetration depth of 100 nm for a-Si:H material. A scanning electron microscope (SEM) and an atomic force microscope (AFM) were used to examine the cross section and surface roughness of the waveguide. Based on the analysis of the Raman spectra, the best Gaussian fit of a-Si:H network material with optimal Tsub was determined. Preliminary transmission measurements were then conducted on the a-Si:H waveguides deposited at various Tsub, with dimensions of 2 μm width and 400 nm height, using the Fabry–Perot ^[32] method to correlate with the Raman results. The light was coupled into the device by the end-fire coupling approach with a 40× magnification objective lens at the input and output of the waveguide. The output light was fed to an InGaAs photodetector. Using the optimized material, wire waveguides with a range of widths were fabricated to evaluate its optical propagation losses. The propagation losses were measured by using the cut-back method with the fiber grating couplers with a total insertion loss of 3 dB, and the optical transmission was normalized to the grating couplers. Agilent 8163B, an integrated tunable laser source, and a photodetector system were used to couple light in and out of the waveguides by gratings and TE polarized optical fibers.

Raman spectroscopy has been used for atomic structural characterization of a-Si:H ^[33,34]. The Raman spectrum of standard a-Si:H has four Raman active phonon modes, namely, a transverse acoustic (TA) peak at 135cm−1, a longitudinal acoustic (LA) peak at 330cm−1, a longitudinal optical (LO) peak at 450cm−1, and a transverse optical (TO) peak at 480cm−1, which can be used to determine the structural quality of a-Si:H ^[35]. Due to the symmetry in the face centered cubic (fcc) lattice, c-Si has a narrow sharp Raman peak from the TO phonon mode ωTO, positioned at 520cm−1, with a full width at half-maximum (FWHM) ΓTO of about 3.5cm−1 ^[35]. In amorphous silicon, the small change in the structural distortion of the SRO network induces substantial modification to the phonon density of states. Thus, there is no translation symmetry, and ωTO has a broad Raman peak positioned near 480cm−1, with ΓTO∼60cm−1 ^[36,37]. Structural distortion also induces a change in bond length and bond angle Δθb. In the Raman spectrum of the amorphous silicon network, the TO phonon (vibration) mode is attributed to the SRO, while the TA, LA, and LO phonon vibration modes are from the structural distortion in the IRO network ^[34]. Moreover, the area intensity ratio of ITA/ITO, ωTO, and ΓTO are sensitive to the change in bond angle of the a-Si network and ITA/ITO also indicates structural disorder in the IRO.

Figure 1 shows the Raman shift results from the effect of the deposited temperatures and the structural behavior of the a-Si:H. Figure 1(a) presents the Raman spectrum of a-Si:H material deposited at 230°C, in which all vibration modes are defined. Figure 1(b) provides the Raman shift of ωTO as a function of Tsub, where the dotted line is a guideline for the center of the Raman line 480cm−1. The plot shows that ωTO does not depend on the Tsub from 190°C to 320°C as compared to PECVD ^[34]. The results indicate two regimes for the HWCVD, namely, the diffusion control regime and kinetic control regime indicated in Fig. 2(a), where hydrogenation influences the a-Si:H quality during deposition. It has been shown that the concentration of hydrogen increases with a decrease of Tsub in HWCVD ^[38]. In our experiment, the change of Tsub was achieved by changing the diameter of the tungsten filaments. This could induce the additional change in hydrogen concentration, even though the flow rate of the precursors and filament temperature were maintained. It appears there is an interplay of the two mechanisms due to hydrogen and Tsub during the deposition process. The a-Si:H material deposited at 210°C, 230°C, and 250°C shows a low structural disorder of the amorphous network with ωTO appearing around 480cm−1, which indicates high-quality a-Si:H deposited in this temperature range. It also suggests that hydrogen passivation into the dangling bonds stabilizes the SRO of a-Si. For a-Si:H material deposited at 190°C and 320°C, there is an upshift in ωTO to 490cm−1 and 493cm−1, respectively, as shown in Fig. 1(b). This is due to the mixture of micro- and nano-scale voids in the amorphous silicon network ^[33,39,40]. As shown in the inset of Fig. 1(b), the deconvolution of the amorphous peak into two Gaussian functions is related to the amorphous and intermediate voids.

In the Arrhenius plot of Fig. 2(a), at Tsub 320°C of the diffusion control regime, the hydrogen concentration in a-Si:H is expected to be less for the passivation of dangling bonds than in the one deposited at Tsub 190°C, which has a reduced densified layer with micro/nano-scale voids and increased average bond length of Si-Si atoms. These a-Si:H films with voids will have an inhomogeneous strain, which translates to a broadening of the ωTO peak, as reported by Hishikawa et al. ^[34], showing a shift in ωTO is due to the change in the stretching force of Si-Si bonds in the SRO. For the presence of the Si-H and Si-H2 stretching modes, we managed to obtain broad Raman shift spectra between 1600cm−1 and 2300cm−1 (not shown) for the 190°C, 230°C, and 320°C samples. When these spectra were fitted with the Voigt function, we obtained low to medium stretching modes of 2091cm−1, 2048cm−1, and 2018cm−1 for the respective deposition temperatures of 190°C, 230°C, and 320°C. The Raman shifts of the 230°C and 320°C samples are the closest to the Si-H bond at 2000cm−1, which indicates hydrogenation of the a-Si lattice network. Our measurement agrees with Johnson et al. ^[41], Matsumura ^[42], and Smets et al. ^[43], whose a-Si:H also has a broad Raman peak, showing evidence of Si-H and Si-H2 bonds and also suggesting nano-crystalline and voids in the amorphous silicon network ^[44]. However, for further verification of Si-H and Si-Hn bonds in a-Si network, Fourier-transform infrared spectroscopy (FTIR) could provide more information of the hydrogen content in the material.

The peak position of TO also indicates the effect of stress in the film, where tensile stress downshifts ωTO and compressive stress upshifts ωTO. Ratnayake et al. ^[45] suggest that there is a linear relationship for the internal tensile and compressive stress with regard to the deposition temperature. The film deposited at 210°C shows evidence of tensile stress with a downshift to 476cm−1, while the film deposited at 320°C indicates compressive stress with an upshift to 493cm−1. The Raman spectral information for a-Si:H networks deposited at different Tsub, at different deposition rates, and with different surface roughnesses is tabulated in Table 2.

The FWHM width of the TO peak (ΓTO) is proportional to the structural angular distortion, which means the larger the ΓTO, the more structural distortion in the bond angle distribution ^[36]. Figure 1(c) shows the dependence of structural deformation on the substrate temperature Tsub. The material deposited at 210°C has an FWHM ΓTO of 67±1cm−1 compared to the pure amorphous Si, which has a ΓTO of 60cm−1. Han et al. ^[33] reported that device-quality a-Si:H has a ωTO centered at 480cm−1 with FWHM ΓTO of 70cm−1, which closely matches our experimental ωTO result at 480.5cm−1 with ΓTO at 68.9cm−1 for the a:Si-H material deposited at 230°C. Based on Beeman’s linear relationship ^[36] (ΓTO=15+6Δθb) between ΓTO and bond angle deviation Δθb, the a-Si:H material deposited at 230°C has a bond angle deviation of 8.98°, with respect to the crystalline Si bond angle. Our bond angle deviation result is within the range reported by Beeman et al. and Han et al. on a fully formed amorphous silicon network ^[33,36]. The FWHM ΓTO becomes wider when Tsub is at 190°C and 320°C (shown in Table 2), which implies changes in inter-atomic lattice spacing due to either compressive or tensile strain and thus larger bond angle fluctuations. The exact bond angle fluctuation requires future X-ray diffraction measurements and correlates to ωTO and ΓTO with a Tsub that requires more temperature points. For now, our goal is to achieve low temperature deposition of a-Si:H for silicon photonics applications.

The TA phonon mode at the Raman peak position of ∼135cm−1 is attributed to the bond bending mode in the IRO of the amorphous network, while the TO phonon at 480cm−1 is attributed to the bond stretching mode in the SRO of the amorphous network ^[38]. By taking the Raman area intensity ratio of ITA peak to ITO peak, we can study the role of hydrogen on the TA bending modes (IRO) and TO stretching modes (SRO). Figure 1(d) presents information related to the role of hydrogen on the IRO or SRO. In the hydrogen-dominated diffusion control regime, bond bending decreases [(ITA/ITO)↓] as the hydrogen concentration increases, which indicates that the effect of hydrogen on the structural disorder is more prominent in the IRO than in the SRO. This observation matches the experimental results of Gupta et al. ^[38]. In the temperature (Tsub)-governed kinetic control regime, bond stretching increases with decreasing Tsub, indicating that thermally induced structural distortion is more obvious in the SRO. These results show a balance between hydrogen diffusion and thermal deposition required to achieve device-quality a-Si:H. The experiments also show the optimized temperature of 230°C, which meets the above requirement.

The cross-section SEM images of the a-Si:H films are shown in Fig. 2(b), where the red box highlights the core thickness of 400 nm surrounded by SiO2. The films deposited at 190°C and 320°C show granular structure due to defects such as microvoids. The a-Si:H film deposited at 230°C is relatively defect free with high densification. The small particles on the facets are due to residue from the cleaving.

Initial propagation loss measurements were performed with the Fabry–Perot method on the 2 μm wide fully etched waveguides deposited at different Tsub temperatures. Figure 2(a) shows the deposition rate and the propagation loss of the a-Si:H waveguide for the different temperatures. The mass transport control (also known as diffusion control) deposition regime is observed at temperatures above 210°C, which is characterized by small changes of the deposition rate with temperature. The surface-limited or kinetic control regime occurs below 200°C, where the deposition rate is dependent on temperature changes. The small drop in the deposition rate at 320°C could also suggest that gas reactions are more prominent above this temperature, but temperatures above this point were not investigated as they are outside the temperature range of interest. The average deposition rate of a-Si:H material is 0.55 nm/s for the temperature range 210°C–320°C, while the deposition rate drops to 0.45 nm/s at Tsub of 190°C. It is noteworthy that the trend in the Raman shift matches the trend in the propagation loss measurements, where the loss of a-Si:H deposited at 190°C and 320°C is higher compared to those of a-Si:H at 210°C, 230°C, and 250°C. The waveguide deposited at 230°C gives the lowest loss of 1 dB/cm. The inset in Fig. 2(b) shows the optical intensity profile from the end facet of the waveguide. We note that the AFM root mean square (rms) surface roughness of the a-Si:H film deposited at 190°C and 320°C is 1.32 nm and 3.58 nm, respectively, while the a-Si:H deposited at 210°C, 230°C, and 250°C has an rms roughness of about 1 nm (see Table 2).

From the Raman characterization, the a-Si:H film deposited at Tsub of 230°C is a promising candidate for low-loss silicon photonics platforms. Therefore, a 400 nm thick a-Si:H film was deposited at 230°C, wire waveguides with different widths were fabricated as mentioned in Section 2, with their propagation losses assessed by the cut-back method, and coupling with optical fibers was performed via grating couplers. The ellipsometer measurement showed that refractive index n=3.695, κ=3.167×10−6, and estimated band gap energy Eg=1.624eV were measured for the bulk material at 1550 nm using the Cody–Lorentz model ^[46].

The measured propagation losses at a wavelength of 1550 nm for different widths are shown in Fig. 3(a). The cross-section image of a waveguide with a width of 350 nm is shown in the inset of Fig. 3(b). Waveguides with widths ranging from 350 to 850 nm were measured under TE polarization, while the narrow 250 nm wide waveguide was measured under TM polarization only. The propagation loss [decibels per centimeter (dB/cm)] as a function of waveguide width is plotted in Fig. 3(b). The propagation loss was decreased with increasing width. The lowest propagation loss is 0.8 dB/cm for the 650 nm wide waveguide, while the 350 nm wide waveguide has a loss of 27 dB/cm. The higher loss is due to the weaker optical confinement and mode interaction with the surrounding sidewall roughness and air voids as seen in the inset of Fig. 3(b). For an 850 nm wide waveguide, the loss was increased to 3 dB/cm, which could be attributed to the fabrication error. Since the optical confinement is well within the 650 nm wide waveguide, and as the rms surface roughness is low, we focused our analysis on how much material absorption contributes to the loss after the hydrogenation and deposition at 230°C. The extinction coefficient κ can be calculated using the measured loss as the absorption coefficient α, where α=(4πκ)/λ. Using the excitation wavelength at 1550 nm, the extinction coefficient κ is 1.85×10−6.

To evaluate the origin of the losses, we performed numerical analysis using a commercial software called MODE Solutions by Lumerical that uses a finite-difference eigenmode (FDE) solver approach to calculate the loss. We modeled all fabricated widths of the wire waveguides in the following order: (a) with absorption but without roughness and (b) with an absorption coefficient, sidewall/surface roughness, and air voids, which are artifacts from fabrication. Simulations for different waveguide widths were executed using the n and κ values from the ellipsometer. First, the extinction coefficient was added into the material to simulate the propagation loss of the waveguide due to the material absorption at the 1550 nm wavelength. The extracted extinction coefficient is 1.845×10−6, which closely matched our analytical estimate of 1.85×10−6. Under the assumption of high optical confinement, the simulated loss of 0.73 dB/cm is comparable to the experimental loss of 0.8 dB/cm for the 650 nm wide waveguide. Further simulation with only the extinction coefficient on other waveguide widths shows no significant changes in the propagation loss. AFM measurement of surface roughness is shown in Fig. 4(a). From the cross-sectional view of the waveguide [inset of Fig. 3(b)], air voids are present between the SiO2 cladding and the a-Si:H waveguide. All these roughness factors were taken into account in simulation condition (b). The sidewall of waveguide has the anisotropic roughness (7 nm correlation length) to resemble the etching result, while the AFM measured top and bottom surface roughness is 3.58 nm (worst case). The same sidewall roughness was applied to the air void, where the width of the air void is 50 nm at an angle of 30° to emulate the fabricated features shown in Fig. 3(b). All waveguides support the fundamental TE mode, except from the waveguide that has 250 nm width, as shown in Fig. 4(b). The simulated propagation losses (red squares) were plotted together with the experimental loss (black dots) in Fig. 3(b). The simulation and the experimental results show excellent agreement, which strongly suggests that roughness scattering dominates the propagation loss more than the material absorption. This means the HWCVD a-Si:H deposited at 230°C has very good hydrogen passivation in the SRO amorphous network.

From the Raman results, in addition to an a-Si:H Raman shift of 480cm−1, it is critical to achieve an FWHM of less than 70cm−1 and a bond angle deviation of 6.6°<Δθb≤10°. This is to ensure the optical quality of a-Si:H material for waveguide device applications. We have achieved those conditions at a deposition temperature of 230°C and shown that the absorption loss is lower than the scattering loss. We demonstrated, for the first time to our knowledge, a very low optical loss of 0.8 dB/cm at the 1550 nm wavelength using the HWCVD method. This low-loss waveguide offers a potential interconnect solution for high-density silicon photonics integration.

Acknowledgment. The authors thank Southampton Nano Fabrication Centre, University of Southampton. G. T. Reed is a Royal Society Wolfson Merit Award holder. He is grateful to both the Royal Society and the Wolfson Foundation. The authors acknowledge Dr. Antoine F. J. Runge for his help in the optical measurement using the Fabry–Perot method. All authors also thank Dr. David Thomson for letting us use his grating coupler transmission measurement setup.