1. Introduction

Silicon carbide (SiC) is a key semiconductor material for electronic and optoelectronic applications, specifically high-power [1,2] and high-temperature [3,4] applications as it provides high mechanical strength and excellent thermal conductivity. The large indirect bandgap (> 2.2 eV) and high index of refraction (~2.6) of SiC allow to develop low-loss and compact integrated optoelectronic devices over a wide wavelength range from visible to mid-infrared (IR) [4]. In addition, the possibility of achieving low-loss and high quality-factor (Q) waveguides and resonators in SiC along with its relatively high third-order nonlinearity [5–7] enable to develop low-threshold nonlinear photonic devices for applications such as optical switching [8] and frequency comb generation [9].

Crystalline SiC has recently attracted attention for quantum optics applications, as it can harbor optically-active defects (i.e., vacancies) that can be employed as solid-state ‘qubits’ [10,11] for various photonic, spintronic, and quantum information applications. Both optically-pumped [12] and electrically-pumped [13] SiC single-photon sources have been demonstrated at room temperature with high brightness and photo-stability. Among three common polytypes of crystalline SiC, namely 4H, 6H and 3C, only cubic (3C) SiC can be epitaxially grown on a silicon (Si) substrate [4,14], which makes it compatible for integration with electronic devices.

Several research groups have demonstrated integrated photonic devices (e.g., microcavities) on thin-film 3C-SiC [15–24], where the Si substrate is partially undercut to achieve photonic devices with light confinement in the SiC layer. However, such suspended devices suffer from low yield and low reliability because of their fragile mechanical structure and they are not suitable for development of large-scale integrated photonic devices. Furthermore, the reported Qs of the resonator-based devices based on this approach to-date are relatively low (far below those of resonators formed in Si or silicon nitride (Si₃N₄)). This shortcoming is partially due to the low optical quality (i.e., high-loss) of the SiC transition layer (i.e., the first few 100s of nanometers of 3C-SiC grown on a Si substrate), which is caused by the high density of the defects (because of the Si/SiC lattice mismatch [14]), and therefore, higher scattering and absorption loss in the SiC transition layer.

One solution for development of a high-quality 3C-SiC photonic material platform with mode confinement in the SiC film is to use a SiC-on-insulator (SiCOI) structure formed by integration of a thin SiC film on a relatively thick silicon oxide (SiO₂) isolation layer. Previous efforts for forming high-quality SiCOI platforms have been focused on developing thin films from highly-pure 6H/4H-SiC bulk wafers using the smart-cut technique [25,26]. The original 4H-SiC bulk wafer has an ultra-low optical loss of ~0.3 dB/cm. However, after going through all the fabrication processes, the integrated waveguide on this platform has a much higher loss of ~7 dB/cm [26]. In addition, the brittle nature of crystalline 4H-SiC as well as the incurred complex fabrication processes (including the 4H-SiC bulk wafer production, high-energy ion implantation, and heavy-duty chemical-mechanical polishing (CMP)), impose major limitations on commercialization with low-cost, high-efficiency, and high-yield. Several previous works have reported different approaches for forming SiCOI platform using 3C-SiC by utilizing either high energy ion implantation [27] or high temperature thermal oxidation [28–30] to create the SiO₂ layer in SiC. However, they have resulted in SiCOI with extensive roughness and nonuniformity and cannot yield a low-loss platform suitable for photonic applications. In particular, thermal oxidation of SiC is very slow due to the chemical inertness of SiC [31], thus making the process costly and inefficient. Despite the awareness of the high density of defects in SiC/Si transition layer and some initial efforts to reduce the defects by homoepitaxial growth of SiC after bonding [30], the low-quality transition layer has never been completely removed.

In this work, we use an optimized low-temperature hydrophilic bonding process [32,33] to transfer an 800 nm-thick layer of 3C-SiC, epitaxially grown on a Si substrate, onto a 4 µm-thick thermally-grown SiO₂ layer on a Si substrate. The process can readily yield a high-quality SiCOI platform for integrated photonic devices without the need for undercutting the substrate or adopting high-loss processes. The SiC film is flipped through the bonding process, and the lower-quality SiC transition layer will be on the top of the transferred film in the SiCOI platform, and it is removed by CMP. After polishing only 100 nm from the top of the SiC film on the SiCOI platform, we achieve a factor three reduction in the surface roughness of the 3C- SiC film, measured using atomic force microscopy (AFM). The process reported in this paper enables us to achieve microring resonators (40 µm radius) with record-high intrinsic Qs of at least 142,000. The propagation loss extracted from our microrings (40 µm radius, 800 nm thickness, and 4 nm free spectral range (FSR)) with this Q is about 2.9 dB/cm. To the best of our knowledge, these Qs are about three times higher than the highest Qs (51,200) [23] previously reported [15–24] in crystalline SiC and higher than the highest Qs (130,000) [7] achieved in amorphous SiC materials, which do not possess the unique optical and quantum properties as those of the crystalline polytypes. The propagation loss extracted from the record-high Q device in 3C-SiC and in amorphous SiC is about 8.7 dB/cm (in a microdisk resonator with 6.25 µm radius, 700 nm thickness, and 24 nm FSR) [23] and 4 dB/cm (in a microdisk resonator with 6 µm radius, 570 nm thickness, and 21.5 nm FSR) [7], respectively.

In the rest of this paper, we present the fabrication details of our SiCOI material platform (Section 2), discuss its original quality together with the critical improvements made by CMP (Section 3), and present and discuss the fabrication and characterization results of the fabricated photonic devices (Section 4). Final conclusions are made in Section 5.

2. Fabrication of SiCOI material platform

Figure 1 shows the fabrication process flow of our SiCOI platform. The general idea here is to build a material platform fully-compatible with wafer-scale processing based on transferring of a thin 3C-SiC layer from the Si substrate onto a low-refractive index SiO₂ (thickness: 4 µm) on a Si substrate. Considering the refractive indices of SiO₂ and 3C-SiC (1.44 and 2.6, respectively), this structure provides the required refractive index contrast for optical mode confinement in the SiC layer and isolation from the Si substrate. The resulting SiCOI platform is compatible with large-scale integration and can be used to form integrated SiC devices for linear, nonlinear, active, and quantum optical application.

 

Fig. 1 Fabrication process flow of the SiCOI material platform. (a) Piece #1, a prime Si wafer. (b) Epitaxial growth of 3C-SiC with top surface smoothened by CMP, leaving a 3C-SiC film with an average thickness of 800 nm (thickness variation ~100 nm). (c) Deposition of a 30 nm SiO₂ layer using ALD. (d) Piece #2, a prime Si wafer. (e) Wet oxidation to grow 4 μm of thermal SiO₂. (f) Piece #1 and piece #2 are bonded using a low-temperature hydrophilic bonding process. (g) Removal of the Si handle layer using Bosch process and KOH wet etching. Inset: A photo of the bonded SiCOI piece.

Download Full Size | PDF

The fabrication process starts with two prime Si wafers (or pieces). The first piece (piece #1) is composed of a thin layer of 3C-SiC epitaxially grown on a Si substrate (NOVASiC SA) and then is smoothened (to reduce the surface roughness) and thinned-down using CMP to achieve an average SiC layer thickness of 800 nm [Fig. 1(b)]. A 30 nm-thick SiO₂ layer is deposited on the smoothened SiC film using atomic layer deposition (ALD) to provide bonding interface with low surface roughness and high bonding strength [33] [Fig. 1(c)]. Meanwhile, a 4 µm-thick layer of SiO₂ is grown on another Si substrate piece (piece #2) through wet oxidation [Fig. 1(e)]. The two pieces (piece #1 and piece #2) are then bonded using a wafer-scale low-temperature (300 °C) hydrophilic bonding process to form one monolithic piece [Fig. 1(f)]. We use low-temperature bonding (300 °C) due to the concern that SiC and SiO₂-on-Si have different thermal expansion coefficients. Bonding or annealing the sample at higher temperatures might result in higher interfacial stress, which may crack the SiC thin film. We observe a reduction in the yield by ~20% when we increase the bonding temperature by 100 °C (to 400 °C). We have previously used a similar bonding process to form high-quality double-layer crystalline Si-on-insulator [32] and crystalline Si-on-Si₃N₄ [33] platforms. Finally, we remove the Si handle layer of the SiC piece using a combination of Bosch dry-etching process and potassium hydroxide (KOH) wet etching to quickly remove the 500 µm-thick Si substrate without damaging the SiC layer [Fig. 1(g)]. We use Bosch process to etch down the top Si handle layer to the thickness of about 2 μm and then switch to KOH wet etching to finish etching the rest of Si as SiC is stable in KOH [34]. Figure 1 inset shows a photo of the developed 2” × 2” SiCOI platform with nearly 100% yield. The colorful interference pattern on top of the sample is the result of variation in the SiC film thickness caused by the initial CMP step after SiC layer epitaxy. The same process here, without any changes, can be extended to develop wafer-scale SiCOI platforms for large-size (e.g., 4-inch or 6-inch) wafers.

To examine the actual film thickness as well as the bonding quality of our sample, we take several scanning electron microscopy (SEM) images, as shown in Fig. 2. Figure 2(a) shows the cross-section of a SiC-on-Si sample before the bonding process. We can clearly observe large defects (as deep as ~1 µm) in Si near the SiC/Si growth interface. These interfacial defects are formed at the transition layer during the SiC layer growth (epitaxy) due to the large lattice and thermal-expansion-coefficient mismatches of Si and SiC and result in a low-optical-quality deposited film at the first few 100’s of nm of the grown SiC film (i.e., transition layer) [35]. As it can be seen from the cross-sectional SEM images, the SiC layer after the initial after-growth CMP has an average thickness of 800 nm (thickness variation ~100 nm). Figure 2(b) shows the cross-section SEM image of a SiCOI sample, where the SiC layer is on the top of the buried oxide (BOX) layer (similar to a commercial silicon-on-insulator (SOI) wafer, with SiC instead of Si as the device layer). While the thickness of the SiC layer has not changed in this process, the SiC film is flipped upside down, and the SiC transition layer is now on the top of the SiCOI platform.

 

Fig. 2 (a) Cross-sectional SEM image of the original 3C-SiC-on-Si sample. The average thickness of the SiC film is 800 nm. (b) Cross-sectional SEM image of the SiCOI sample. The average thickness of the SiC film maintains as 800 nm. The thickness of the BOX layer is 4 μm. (c) Top view SEM image of the broken edges of the SiCOI sample along the SiC crystal directions.

Download Full Size | PDF

To monitor the quality of the SiC/Si transition layer, we conduct transmission electron microscopy (TEM) on a SiCOI lamella etched and transferred onto the TEM grid by focused ion beam (FIB) incorporated with a micro-manipulator and SEM. Figure 3(a) shows the zoomed-out view of the SiCOI lamella. It shows the cross-sectional structure of the SiCOI sample. The platinum (Pt) layer on top of SiC is to protect its surface from being exposed to the ion beam. The line-shape patterns shown in the region where SiC exists are from the diffraction of electrons due to the defects of the SiC lattice. It is clearly shown that the density of the diffraction pattern becomes much higher when it approaches the top of the SiCOI lamella where the transition layer sits. Figure 3(b) shows a zoomed-in view of this transition layer at the atomic scale; it clearly shows (indicated by white color) a high density of lattice dislocation as well as stacking faults in the SiC crystal. The transition layer has high-density defects, which will cause scattering and absorption of light inside the material. However, when it comes to the other side of the SiC film (i.e., the bottom of the SiC layer in the SiCOI lamella), as shown in Fig. 3(c), SiC atoms are uniformly stacked.

 

Fig. 3 (a) Zoomed-out TEM image of the SiCOI lamella. The line-shape patterns shown in the region where SiC exists are from the diffraction of electrons due to the defects of the SiC lattice. The density of defects is reducing in the direction of SiC growth (from region A to B). (b) Zoomed-in TEM image of the top surface of a SiCOI lamella (region A in (a), also known as the transition layer) where there is a high density of defects. Some of the regions with defects are indicated by white color. (c) Zoomed-in TEM image of the bottom layer of the SiCOI lamella (region B in (a)) where there is a low density of defects in SiC.

Download Full Size | PDF

3. Quality analysis and improvements

To assess the optical quality of the developed SiCOI platform and hence the possible influence of the bonding process, we fabricated different microresonator devices on this platform and characterized their intrinsic Q’s as an indication of the material platform quality. We use alumina (Al₂O₃) as hard mask patterned by electron beam lithography (EBL) with negative resist flowable oxide (FOX, Dow Corning). FOX is selected as the EBL resist due to its high resolution and good etching selectivity. We also develop an optical recipe for SiC dry etching using reactive-ion etching (RIE) with a fluorine-based plasma chemistry (40 sccm CHF₃ + 10 sccm O₂ in an Oxford End-Point RIE chamber). Our recipe etches SiC at ~24 nm/min with selectivity (Al₂O₃/SiC) of ~0.25. The developed technique for etching SiC based on alumina hard mask is both simpler than the previously reported approaches based on using metal masks such as chromium (Cr) or aluminum (Al) [20,23], and it also provides good selectivity for deep SiC etching.

Figure 4(a) shows the SEM image of the fabricated micro-donut resonators on SiCOI platform with outer radius of 5 µm, 10 µm, and 20 µm and width of 1.25 µm, 3 µm, and 4 µm, respectively, coupled to bus waveguides (width: 800 nm). Figures 4(b) and 4(c) show a close view of the resonator-waveguide coupled-region for a 20-µm radius resonator and input/output tapered-grating couplers, respectively. The grating couplers are designed to achieve maximum coupling to the fundamental quasi-transverse-electric (TE)-polarized modes (i.e., electric field in the plane of the resonator) of the input waveguide at wavelengths around 1550 nm. The finite-difference time-domain (FDTD) simulation shows our grating (800 nm thick, 950 nm pitch, 0.5 duty cycle) has a maximum efficiency of –6 dB (or 25%) at 1563 nm wavelength with a 3-dB bandwidth of ~90 nm. However, the experimentally measured coupling efficiency is lower (maximum efficiency of – 11.7 dB (or 6.8%) at 1552 nm wavelength with a 3-dB bandwidth of ~90 nm) caused by fabrication-induced imperfections (e.g., the change of the shape of gratings from sidewall etching). Our grating could be further optimized according to the exact etching profile for higher coupling efficiency. The fundamental quasi-TE resonant mode of the fabricated micro-donut resonator (with a relatively large width) is similar to that of a micro-disk resonator with the same radius while its higher-order modes are suppressed by etching the center region of the microdisk. Figure 4(d) shows the cross-sectional normal magnetic field (H_z, with z being the out-of-plane coordinate) profile of the first-order (i.e., fundamental) quasi-TE mode of a 20 µm-radius microdisk resonator, simulated using three-dimensional finite element method (3D FEM implemented in the COMSOL environment). As shown by the mode-profile, the fundamental radial mode only interacts with the outer sidewall of the resonator, which results in maintaining the high Q of the microdisk (and micro-donut) resonators. Therefore, using such a design, we can fairly compare the performance of our device with those of undercut microdisk resonators fabricated on SiC-on-Si substrates [15–24].

 

Fig. 4 (a) Top view SEM image of fabricated micro-donut resonators on SiCOI platform. Resonators have outer radius of 5 µm, 10 µm, 20 µm and width of 1.25 µm, 3 µm, 4 µm, respectively. (b) Top view SEM image of a 20 µm-radius micro-donut resonator-waveguide coupled region. The width of the bus waveguide is 800 nm. (c) Top view SEM image of input/output grating-taper coupled region. (d) Cross-sectional H_z profile of the first-order quasi-TE mode of a 20 µm-radius resonator according to the designed geometry.

Download Full Size | PDF

All resonators studied in this work are based on the same coupling scheme [Fig. 4(b)]. The bus waveguide width is 800 nm, and the gap between the waveguide and the resonator is varied between 100 nm to 400 nm to achieve near-critical coupling. We use a tunable semiconductor laser covering wavelengths around 1550 nm as the light source. Light coming out of the laser is coupled into a single-mode fiber and goes through a fiber polarization controller (to adjust the input polarization) before being coupled to the input waveguide through the input grating coupler. The light from the output waveguide is also coupled to a single-mode fiber through the output grating coupler and is detected and converted to an electrical signal using a near-infrared (NIR) photo-detector before sampling it using a National Instruments data acquisition card.

Figure 5(a) shows the transmission spectrum of the quasi-TE-polarized mode of a 20 µm-radius micro-donut resonator with waveguide-resonator gap of 100 nm. The width of the bus waveguide in this structure is 800 nm. The non-flat background is caused by the Fabry-Perot (FP) resonance between the two gratings. Figure 5(b) shows the magnified plot of the transmission of a near-critically-coupled mode of the resonator with a resonance wavelength of 1559.1 nm and FSR of ~8 nm. The dashed-curve shows a Lorentzian fitting (using a model that includes the FP response in the background) to the raw experimental data (blue dots). An intrinsic Q of ~42, 000 is extracted from the fitting curve, which is in-par with the highest reported Q on any 3C-SiC/Si platform [23]. This indicates that transferring of the SiC film from the original SiC/Si wafer onto the oxide-on-Si substrate and the bonding process have not degraded the SiC film quality or introduced any extra surface damage to the SiC film. However, the current value of Q is still orders of magnitude lower than the material-loss-limited Q of SiC. As predicted by theoretical calculation based on ab initio pseudopotential-plane-wave method and random-phase approximation, all polytypes of SiC have very small intrinsic loss at near-IR [36]. The loss of the bulk SiC is mainly limited by the crystal defects and impurities (or doping) in SiC. Also, the characterization of the Q of the micro-donut resonator with different sizes [Fig. 5(c)] showing slow increase of Q of their fundamental modes by increasing the resonator radius indicates that the Qs of the fundamental quasi-TE modes of the fabricated resonators are not limited by radiation-loss caused by the etched sidewall roughness. Therefore, we believe that the defects at the SiC/Si interface [14,35] are responsible for the large optical loss and low Q of the developed resonators, and the crystal defects near the top surface as well as the top-surface roughness of the SiC film have a major effect on the achievable Qs from 3C-SiC resonators. To have better statistics of the resonator Q with any given radius, we fabricate 5 identical devices with the same geometry; the variation of the measured Q for each resonator radius is shown with error bars in Fig. 5(c).

 

Fig. 5 (a) Transmission spectrum of a 20 µm-radius micro-donut resonator under TE polarization near 1550 nm wavelength. (b) Normalized transmission spectrum of the resonant mode at 1559.1 nm marked in (a), with experimental data and Lorentzian fitting shown in blue and red, respectively. The mode is near-critically coupled with an intrinsic Q of about 42,000. (c) The measured intrinsic Qs versus the outer radius of the micro-donut resonators fabricated on SiCOI platform.

Download Full Size | PDF

To evaluate this conclusion, we perform AFM measurement on the grown SiC at Si/SiC interface (now on the top of the SiCOI platform) to analyze the root-mean-squared (RMS) roughness (σ) of the device top surface. Figure 6(a) shows the two-dimensional (2D) AFM scan of the top surface of the original SiC/Si sample, which has gone through an initial CMP process after SiC expitaxy. We measure σ ~1.33 Å with a range from −4.538 Å to 5.152 Å, which is close to the surface roughness of ultra-low-loss Si₃N₄ as reported in [37]. However, the 2D AFM scan of the top layer of the SiCOI sample (which is the SiC/ Si interface layer) shows a roughness of σ = 7.25 Å with a range from −2.5 nm to 3.5 nm, which is 5-7 times higher than that of the top layer of the original sample, shown in Fig. 6(b). This large roughness mainly comes from lattice dislocations at the SiC/Si interfacial layer. To reduce the scattering losses from the Si/SiC interfacial layer, we use CMP (Entrepix, Inc) to reduce the roughness of the top layer of the SiCOI sample. The inset in Fig. 6 shows the SiCOI sample after polishing the top SiC layer and thinning it by 100 nm. The SiCOI can endure the CMP process, which shows the good mechanical strength of the bonded sample. Also, the smoother “rainbow” color of the surface (inset in Fig. 6) as compared to that of inset in Fig. 1 indicates the improved thickness uniformity of the sample after CMP. The 2D AFM scan after the final CMP step is shown in Fig. 6(c). The σ has decreased from 7.25 Å to 2.47 Å with range from −8.824 Å to 9.963 Å, corresponding to a factor of 3 reduction in the surface roughness. Therefore, we expect the resonator Qs to improve by at least the same factor as we not only reduce the surface roughness of the SiC, but also reduce the bulk defects by removing 100 nm of the low-quality interfacial SiC film (that has a high density of defects). Figures 6(d)-6(f) show 3D AFM scans of the top surface corresponding to 2D scans in Figs. 6(a)-6(c), respectively.

 

Fig. 6 (a) 2D AFM scan of the top surface of the original SiC/Si sample (with initial CMP after epitaxy) before the bonding process, scaled to −4.538 Å to 5.152 Å with RMS roughness (σ) of 1.33 Å. (b) 2D AFM scan of the top surface of SiCOI sample (same as the bottom surface of original SiC/Si sample) before the final CMP, scaled to −2.5 nm to 3.5 nm with σ = 7.25 Å, representing the large surface roughness of the SiC/Si transition layer (now on top of the SiCOI sample); (c) 2D AFM scan of the top surface of SiCOI sample after the final CMP, scaled to −8.824 Å to 9.963 Å with σ = 2.47 Å. (d) 3D AFM scan of the top surface of the original SiC/Si sample in (a), scaled to −8 nm to 8 nm with σ = 1.33 Å. (e) 3D AFM scan of the top surface of the SiCOI sample before the final CMP in (b), scaled to −8 nm to 8 nm with σ = 7.25 Å. (f) 3D AFM scan of the top surface of the SiCOI sample after the final CMP in (c), scaled to −8 nm to 8 nm with σ = 2.47 Å. Inset: A photo of the SiCOI sample after the final CMP.

Download Full Size | PDF

4. Demonstration of high Q microresonators

To evaluate the optical quality of the resulting SiCOI platform after CMP, we fabricate microring resonators in it, similar to the previously explained structures in the SiCOI platform using the same fabrication as explained before. The devices are fabricated in the SiC film with a thickness of 800 nm after CMP to have the same SiC thickness that used for the fabricated devices without CMP (we could still find 800 nm-thick SiC after 100 nm-thick CMP due to the initial thickness variation of the original sample). A zoomed-in SEM image [Fig. 7(b)] shows that we get sidewall etching on the two sides. The sidewall etching mainly comes from the low-anisotropic nature of the RIE etching process, which is usually less pronounced in a biased inductively coupled plasma (ICP)-RIE system. We also clearly observe that some defects appear on the sidewall. These defects could be from the original material itself, and they are exaggerated by dry etching. However, the sidewall looks smooth under the 1 µm scale. Figure 7(c) shows the cross-sectional H_z profile of the fundamental quasi-TE mode of a 40 µm-radius ring resonator simulated using 3D FEM according to the real geometry observed from Fig. 7(b). We can see from Fig. 7(c) that the optical mode is highly confined by both sidewalls, and its Qs is sensitive to sidewall roughness. Note that here we fabricate microring resonators (instead of micro-donut resonators in the previous case) to show the clear advantage of the CMP process for loss reduction.

 

Fig. 7 (a) Angled-view SEM image of fabricated microring resonators on SiCOI platform (after CMP). Resonators have outer radii of 40 µm, 60 µm, 80 µm and width of 2.5 µm. (b) Top view SEM image of a 40 µm radius microring resonator-waveguide coupled region. Bus waveguide width is 800 nm. Due to sidewall etching, the topside width of the ring is reduced to 1.7 µm (c) Cross-sectional H_z profile of the first-order quasi-TE mode of the 40 µm radius-resonator according to the real geometry measured in (b).

Download Full Size | PDF

Figure 8(a) shows the transmission spectrum of the quasi-TE-polarized mode for the 40 µm microring resonator in Fig. 7 with waveguide-resonator gap changed from 300 to 100 nm to ensure near-critical coupling. The width of the bus waveguide is 800 nm. Figures 8(b)-8(d) show the magnified plots of the transmission of the coupled waveguide-resonator device near three resonance wavelengths with FSR of ~4 nm. The dashed-curve in each case shows a Lorentzian fitting to the raw experimental data (blue dot). Intrinsic Qs of 110, 000, 97, 000, and 142, 000 are extracted from the fitting curves. To the best of our knowledge, these Qs are highest values reported to date [15–24]. Our best Q here is about three times higher than the highest Q (51,200) achieved in 3C-SiC [23]. It is also higher than the highest Q (130,000) achieved in amorphous SiC [7]. It should also be noticed that the previous record-Q devices are all microdisk resonators while our Q is demonstrated in microring resonators, which shows further advantage of our platform over the previously reported ones from the optical quality perspective. Nevertheless, it should be mentioned that, similar to Fig. 5(c), the Qs of resonators still do not scale according to the size of the ring on the SiCOI platform. To further prove the Qs of our device are not limited by our etching process, we also perform AFM measurements to analyze σ of the etched sidewalls of our microring resonators as shown in Fig. 9.

 

Fig. 8 (a) Transmission spectrum of a 40 µm microring resonator at the quasi-TE polarization near 1550 nm wavelength when the resonator-waveguide spacing is changed from 300 nm to 100 nm. For easy visualization, the spectra are shifted by a value of −15 (dB) with respect to each other along the vertical axis. (b) Normalized transmission spectrum of the cavity mode at 1550 nm marked in (a), with experimental data and Lorentzian fitting shown in blue and red, respectively. The mode is under-coupled with an intrinsic Q of about 110,000. (c) Normalized transmission spectrum of the cavity mode at 1533.8 nm wavelength marked in (a), with experimental data and Lorentzian fitting shown in blue and red, respectively. The mode is critically coupled with an intrinsic Q of about 97,000. (d) Normalized transmission spectrum of the cavity mode at 1561.3 nm wavelength marked in (a), with experimental data and Lorentzian fitting shown in blue and red, respectively. The mode is near-critically coupled with an intrinsic Q of about 142,000.

Download Full Size | PDF

 

Fig. 9 (a) Zoomed-in angled-view SEM image of the microring resonator in Fig. 7(b). (b) 2D AFM scan, and (c) 3D AFM scan of the sidewall of the microring resonator with RMS roughness σ = 2.56 nm.

Download Full Size | PDF

We use a similar concept as that in [38] to perform AFM measurements on our sidewalls. By taking advantage of the non-vertical sidewall of our waveguides after the etching process, we tilt our sample by about 60° so that sidewall angle is compensated. We are able to perform AFM measurements on the near-horizontally-tilted sidewall of the waveguides. Figure 9(a) shows the zoomed-in angled-view SEM image of the microring resonator in Fig. 7(b). Figure 9(b) shows the 2D AFM scan of part of that sidewall. We measure the RMS roughness σ = 2.56 nm, which is close to that of a typically etched Si₃N₄ sidewall [38] (corresponding to a scattering loss-limited Q in the orders of 10⁷ [37]). Figure 9(c) shows 3D AFM scan of the sidewall of the microring resonator shown in Fig. 8(b), with σ = 2.56 nm. Comparing the Qs reported here with those of the reported Si₃N₄ resonators with similar σ, we conclude that the sidewall roughness does not significantly contribute to the loss of the SiC resonator. We estimate that the material loss, mainly caused by the remaining defects in the SiC thin film, is responsible for most of the 2.9 dB/cm effective waveguide loss. Further efforts in improving the quality of the grown 3C-SiC (e.g. growing ultra-thick SiC to achieve less defects) can thus result in considerably higher Qs. Also we observe that there are holes on the SiO₂ layer after plasma etching [Fig. 9(a)]. We think that these holes correspond to the defects in the SiC layer, translated to the SiO₂ substrate during the etching process.

To directly compare the Q of the devices before and after CMP on our SiCOI platform, we fabricate micro-donut resonators [Fig. 10(a)] with 20 µm radius on our platform after extra CMP (~500 nm-thick SiC film). Figure 10(b) shows the transmission spectrum of the quasi-TE-polarized mode of the micro-donut resonator shown in Fig. 10(a). Figures 10(c) shows the magnified plots of the transmission of the coupled waveguide-resonator device near 1562.0 nm wavelength. Intrinsic Qs of 126,000 are extracted from the fitting curves. Compared to the similar resonators fabricated in the SiCOI platform without CMP (with Q of 42,000 as shown in Fig. 5), the Q of the device after CMP is 3 times higher. We also observe from Fig. 10(a) fewer holes on the substrate compared to the previous cases [Figs. 9(a) and 4(b)], which is a clear sign showing that the defect-rich transition layer is removed after thinning down the material by CMP.

 

Fig. 10 (a) Top view SEM image of a 20 µm-radius micro-donut resonator-waveguide coupled region. The thickness of the device is 500 nm. (b) Transmission spectrum of a 20 µm-radius micro-donut resonator under TE polarization near 1550 nm wavelength. (c) Normalized transmission spectrum of the resonant mode at 1562.0 nm marked in (b), with experimental data and Lorentzian fitting shown in blue and red, respectively. The mode is near-critically coupled with an intrinsic Q of about 126,000.

Download Full Size | PDF

5. Conclusion

In conclusion, we have successfully utilized a low-temperature hydrophilic bonding process to form a high-quality crystalline 3C-SiCOI platform. The proposed platform is considerably more robust as compared to those realized using existing approaches for the development of the SiC nanophotonic devices. It alleviates the need for undercutting the SiC devices on a Si substrate (resulting in devices with more mechanical strength) and provides a much better yield as compared to the previously reported devices based on suspended structures. We systematically reduce losses in 3C-SiC by “flipping” the SiC film and using CMP to reduce SiC surface roughness and partially remove the SiC film transition layer, which has high density of defects. The entire fabrication process is CMOS-compatible, inexpensive, and could be reproduced with high yield. Using the resulting 3C-SiCOI platform, we demonstrated here the record high Q of 142,000 for microring resonators on the SiCOI platform after CMP. We believe that our platform can address the shortcomings of existing SiC-based platforms to enable a novel set of on-chip devices and systems beyond what currently available in the SOI platform by taking advantage of unique linear, nonlinear, and quantum features of SiC. Our process can also be extended to build high-quality material platforms based on a wide range of epitaxially grown materials including III–V materials and diamond for both electronic and integrated optic applications.