Silicon carbide (SiC) polytypes, including 3C-SiC and 4H-SiC, feature wide bandgap energies [1], reasonable second- and third-order optical nonlinearities [2,3], and readily accessible color centers [4–6]. These materials are compatible with standard silicon complementary metal–oxide–semiconductor (CMOS) fabrication processes. Thus, integrated SiC-on-insulator (SiCoI) platforms have been attracting interests for linear, nonlinear, and quantum photonic circuit applications [3–11].

On-chip nonlinear and quantum photonic components, including light sources [3–6,10–12], have been demonstrated on SiCoI platforms. Single-photon emissions from color centers have been demonstrated in an integrated 4H-SiCoI platform [4,5] and in suspended 3C-SiC resonators [6]. These color centers require a pump laser in a shorter wavelength and emit photons in the near-infrared (NIR) wavelengths. Parametric downconversion photon-pair sources based on a second-order optic nonlinearity have been demonstrated on an integrated 3C-SiCoI platform [10,11], converting the pump light in 780 nm into photon-pairs in the C- and L-bands, while single-photon detectors (SPDs) using niobate nitride nanowires working in cryogenic temperatures have been demonstrated on various heterogeneously integrated material platforms [12,13].

Beside the on-chip nonlinear and quantum light sources and SPDs, on-chip pump-rejection filters are one essential building-block component for integrated nonlinear and quantum photonic circuits, but they remain relatively less explored in the literature. A reasonable photon-counting rate of 1 MHz of the generated photons is about 120 dB weaker than an on-chip pump power of 1 mW. The excess on-chip pump photons must be removed for the on-chip SPDs to detect single-photon signals. By cascading multiple stages of optical microring-based filters [14] or Mach–Zehnder interferometers (MZIs) [15], researchers can attain wavelength-selective isolations of a few tens of dBs. However, these wavelength-agile cascaded filters limit the single and photon-pair sources to narrow spectral bands. These filters further impose accurate device fabrication or careful active wavelength alignments among all the individual filter stages. The use of active thermal and electric controls for aligning the filter wavelengths consumes energy and computing resources. Long gratings [16], directional coupler (DC)-based filters adopting integrated gratings [17] and tapered waveguides [18] can offer a reasonable extinction ratio and a relatively wide bandwidth (BW). However, these designs still demand cascading of carefully designed filter stages, which tends to increase the insertion loss (IL) and requires precision device fabrication processes.

Indirect-bandgap semiconductor materials are natural long-pass filters (LPFs) or absorbers for short wavelengths because they provide intrinsic material absorption for photons exceeding their bandgap energy without an efficient radiative recombination while the intrinsic material absorption of these indirect-bandgap semiconductor materials for the light within the transparent window is negligible for a centimeter-scale photonic chip. Among readily accessible conventional semiconductor materials, amorphous silicon (α-Si) is a promising candidate. The α-Si can be readily deposited on substrates using chemical vapor deposition (CVD) furnaces in a CMOS foundry. Although the bandgap energy of α-Si can be as large as 1.7 eV [19] depending on the deposition recipe, with proper adjustments to the deposition parameters, the CVD α-Si can absorb pump light in the visible/NIR wavelengths while remaining largely transparent to light in the O-, C-, and L-bands. The α-Si features a higher structure disorder than the crystalline silicon (c-Si), which enables a larger density of states [20]. The absorption coefficient of α-Si is about an order of magnitude larger than that of the c-Si in the visible wavelengths [21]. It is known that a deposition temperature exceeding 680^°C can partially crystalize the α-Si into polysilicon and weaken the absorption capability [21]. Due to its better absorption and simple deposition process on a wafer at a low cost, α-Si has been widely adopted in solar cells [19]. The α-Si is also deposited on a silicon nitride photonic platform to serve as an on-chip microheater because the absorbed light is converted into heat [22].

Figure 1 schematically illustrates our proposed on-chip pump-rejection LPF integrated on a CMOS-compatible 3C-SiCoI photonic platform. Different from the integrated nonlinear photonic devices leveraging the α-Si as a waveguide core [23,24], we utilize a thin-film α-Si layer deposited on the top surface and on the sidewalls of the SiC waveguides to reject the pump light in the visible/NIR wavelengths by absorption along the propagation direction while the generated signal and idler photons in the longer NIR wavelengths remain propagating without significant absorption or mode perturbation. We target such a design to offer a pump-rejection LPF over a total waveguide length of shorter than 1 mm to reject the excess pump light in 780 nm wavelengths with a significant extinction ratio (exceeding 120 dB).

 

Fig. 1. Schematics of the proposed α-Si-based on-chip LPF for an integrated SiCoI nonlinear and quantum photonic platform. The footprint can be compact if the LPF is designed properly in bending shapes.

Download Full Size | PDF

For the numerical simulation, we adopted the refractive index of our deposited α-Si as 4.300 + i0.071 and 3.800 for 780 and 1550 nm, respectively, based on our material characterizations. Because the refractive index of α-Si is significantly larger than that of SiC (∼2.56) at 1550 nm, we adopted a few 10-nm-thin thickness t of the α-Si layer sandwiched between the SiC and an upper SiO₂ cladding layer to keep the waveguide modes in the 1550 nm wavelengths (or the signal and idler wavelengths) essentially in the SiC core while minimizing the IL due to the back-reflection upon an impedance mismatch between the normal SiC waveguide with no α-Si layer and the pump-rejection LPF waveguide. Meanwhile, the spatial overlap of the pump light (780 nm) waveguide mode with the α-Si thin-film absorbing layer varies with t. We adopted an eigenmode expansion (EME) solver to simulate the transmissions of the transverse electric (TE)- and transverse magnetic (TM)-modes in the 780 nm (absorbing) and 1550 nm (transparent) wavelengths.

Figure 2(a) schematically illustrates the cross-sectional structure of the pump-rejection LPF. We adopted a 3C-SiC waveguide with a width, a SiC film thickness and a slab thickness of 850, 460, and 100 nm, respectively. We adopted a waveguide sidewall slope of 80^° based on the device fabrication results. These design dimensions are adopted based on requirements of other applications and can be varied. We assume the α-Si layer covering the waveguide top surface and sidewalls uniformly. We adopted a 30-nm-thick aluminum oxide (Al₂O₃) bonding layer underneath the SiC film. The waveguide structure is surrounded by SiO₂ cladding. Figures 2(b) and 2(c) show the EME simulation results of the transmissions assuming a waveguide length of 1 mm. We observe a small but increasing IL for the 1550 nm due to the waveguide mode impedance mismatch when entering the filter regions. When the α-Si layer is thick enough, multi-mode interference (MMI) may occur and reduce the IL for 1550 nm with careful designs. For 780 nm, the absorption is determined by the spatial overlap between the mode field and the α-Si. The modal overlap does not monotonically increase with the α-Si layer thickness t due to complex MMI behaviors. Figures 2(d)–2(g) show the simulated mode-field amplitude distributions for TE- and TM-modes in 780 and 1550 nm within the first 10 µm in the propagation direction inside the filters with α-Si layer thicknesses of 40 nm. We observe that the 780 nm waveguide modes exhibit a higher-order mode interference, which results in a different spatial overlap with periodic coupling of the 780 nm light into the α-Si thin-film guided mode, as shown in the insets of Figs. 2(d) and 2(g). Based on the simulation results, we adopted a thin-film α-Si layer thickness of 40 nm in the device fabrication to attain a significantly large absorption for 780 nm (pump) wavelengths while maintaining a negligible IL below 1 dB for 1550 nm (signal and idler) wavelengths.

 

Fig. 2. (a) Schematics of the cross section of the simulated structure. The horizontal dashed line indicates the x–y plane where the mode-field amplitude distributions in (d)–(g) are simulated. (b) and (c) Simulated transmission of the TE- and TM-polarized modes at 1550 and 780 nm wavelengths after a filter length of 1 mm, respectively. The overlap between the α-Si and the 780 nm fields is also shown in (c). (d)–(g) Simulated mode-field amplitude distributions for the TE- and TM-polarized modes at 780 and 1550 nm wavelengths within the first 10 µm in the propagation direction inside the filters with the α-Si layer thickness of 40 nm. Insets: cross-sectional field amplitude distributions within the filter waveguides.

Download Full Size | PDF

We adopted the commercial 4^″ epitaxial 3C-SiC-on-Si wafer from the NOVASiC. The thickness of SiC is 1.5 µm with a chemical mechanical polishing. We adopted a 4^″ Si substrate with a thermal SiO₂ layer of 3 µm as the carrier wafer in the wafer-to-wafer bonding. We first use a standard 120^°C H₂SO₄ (98%):H₂O₂ (30%) = 10:1 solution to clean both wafers. We transfer the two wafers into an Oxford atomic layer deposition (ALD) equipment to deposit a layer of alumina (Al₂O₃) with a thickness of 15 nm simultaneously, which gives a total layer thickness of 30 nm after the bonding step. The films deposited by ALD feature a high -OH surface density which is favorable for molecular bonding [25]. After the deposition, we put the two wafers into contact and press them to initiate a pre-bonding. Then, we transfer the wafer pair into a Karl Suss SB6 bonder. We anneal the pre-bonded wafers at 300^°C in a vacuum for 3 h to attain a permanent bonding. After the bonding steps, we use a Si grinder to remove the bulk of the Si substrate on the SiC side and remove the rest of the Si substrate with a tetramethylazanium hydroxide solution heated to 80^°C. The remaining SiC film after the substrate removal is less than 50% of the original wafer. We dice the bonded wafer into dies with a size of 1.2 × 1.2 cm.

The exposed SiC surface corresponds to the first layer of SiC epitaxially grown on a Si substrate, which has a rather poor crystal quality due to the lattice mismatch between the SiC and Si [26]. We adopted a deep reactive ion etching (DRIE) recipe with SF₆ and O₂ to thin down the film to our target thickness of 460 nm. We deposit a plasma-enhanced CVD (PECVD) SiO₂ layer of 500 nm thickness as the hard mask. We use electron-beam (e-beam) lithography and DRIE to pattern the SiO₂ hard mask. We then use the same SF₆/O₂ DRIE process to pattern the SiC. The SiC-to-SiO₂ selectivity is 1.45. We remove the remaining SiO₂ hard mask using buffered oxide etchant and transfer the sample to an CVD furnace to deposit the α-Si layer with a 40 nm thickness. We adopted a SiH₄ gas flow of 45 sccm with the furnace pressure kept at 600 mTorr. The deposition temperature was kept at 550^°C to avoid the α-Si being partially crystallized. We have characterized the bandgap energy of our deposited film to be 1.25 eV based on a fitting from a built-in ellipsometer model for α-Si. This step is not compatible with our previously demonstrated 3C-SiCoI platform which has a glass substrate that would melt under this temperature [27]. After that, we use an i-line photolithography process to cover the desired filter region with a photoresist and to completely remove the α-Si layer elsewhere by wet etching (Freckle etch solution). We deposit a PECVD SiO₂ upper-cladding layer of 1000 nm thickness to protect the whole sample. Finally, we dice the samples into columns for butt-coupling.

Figure 3(a) shows the microscope image of the fabricated filters with different filtering lengths. Figure 3(b) shows the scanning electron microscopy (SEM) image of the waveguide-to-filter transition region. We design the transition in a tilted manner intended to gradually modulate the impedance mismatch between the two regions.

 

Fig. 3. (a) Microscope image of the filters with a length of 0–300 µm in a step of 100 µm. (b) SEM picture showing the waveguide-to-filter transition region.

Download Full Size | PDF

Details of the experimental setup are described in Supplement 1. Figures 4(a) and 4(b) show the normalized transmission spectra in the O-, C-, and L-bands in the TE- and TM-polarizations from the fabricated filters with different lengths. We do not observe a systematic increase in IL with the filter length increasing to 100 µm steps, suggesting the passive waveguide loss in the transparent window is negligible in this length scale. However, we observe a systematic difference between the reference waveguide with no α-Si and the three filters. Figures 4(c) and 4(d) show the normalized transmission spectra for both polarizations in the 780 nm wavelengths. As we adopted the filters in straight waveguides, the background caused by the direct scattering of the input light is high at ∼−40 dB. We observe that, even for the shortest filter with a length of 100 µm, the transmitted light is at the background level of the scattering light. This is further confirmed by the near-field images of the output butt-couplers captured by a charge-coupled-device camera from the reference waveguide and from the 100-µm-length filter, respectively shown in insets in Figs. 4(i) and 4(j). The exposure times are 1 and 200 ms for insets in Figs. 4(i) and 4(j), respectively, corresponding to a 23 dB enhancement. The transmitted light from the butt-coupler after the absorption is at the background level. Figure 4(e) shows the wavelength-averaged transmissions over three wavelength windows (1550/1310/780 nm) for TM and TE polarizations in various filter lengths. Compared with the reference waveguide (0 µm filter length), we extract ILs of 0.8 ± 1.0 and 1.9 ± 0.5 dB for the TE- and TM-polarized waveguide modes in the O-band while we obtain 1.3 ± 1.2 and 1.5 ± 0.9 dB for the TE- and TM-polarized waveguide modes in the C- and L-bands. For the transmissions at 780 nm wavelengths, we estimate a lower-bound IL of 230 ± 8 and 248 ± 12 dB/mm for the TE- and TM-polarized waveguide modes based on the 100-µm-length filter, limited by the scattering of the input light. The stray light issue can be potentially addressed by placing the input and output edge couplers at widely separated positions in orthogonal orientations on the chip. It would be helpful to adopt a bright field layout in which the majority area of SiC is removed to reduce the light scattering in the SiC film surrounding the waveguide structures. Table 1 summarizes the working principles and the key performance of our devices and of a few related work [14–18].

 

Fig. 4. Lens-to-lens normalized transmission spectra of the TE- and TM-polarizations from different filter lengths for (a) and (b) O-band, C-band, and L-band and (c) and (d) the 780 nm wavelengths. Insets: (i) and (j) show the near-field images of the transmitted TM-polarized modes after the reference waveguide and the shortest filter with a length of 100 µm, respectively. (e)Wavelength-averaged transmissions over various wavelength windows for the TE and TM polarizations and the filter lengths.

Download Full Size | PDF

Table 1. Comparison of Rejection Filters With This Work

View Table

Figure 5 shows the measured spectra using an optical spectrum analyzer (OSA) with and without the 780 nm light coupling into the 100-µm-length filter. We do not observe any significant optical powers resulting from the radiative recombination above or below the bandgap wavelength of ∼1.0 µm up to 1550 nm. The spectra shown from ∼800 to 1550 nm is at the noise floor of the OSA, which indicates no significant re-emission of light above the −90 dB level upon the 780 nm light being absorbed. This ensures that the on-chip filters do not act as secondary light sources, which is critical toward nonlinear and quantum photonic circuits applications.

 

Fig. 5. Output spectra captured by an OSA with and without the 780 nm pumping into the filter with the length of 100 µm, showing no re-emission above the −90 dB level compared with the absorbed pump power of 1 mW.

Download Full Size | PDF

In summary, we have designed, fabricated, and experimentally tested on-chip pump-rejection LPF filters based on the absorption of the α-Si absorbing layer on an integrated CMOS-compatible 3C-SiCoI photonic platform. We have extracted ILs at ∼1 dB for the wavelengths in the O-, C- and L-bands while demonstrated an estimated a lower-bound pump-rejection exceeding 230 dB/mm for the 780 nm wavelengths. We anticipate that these simple on-chip passive pump-rejection LPFs can become a building-block component for future monolithic SiC-based nonlinear and quantum photonic chips. We believe that this structure is also applicable to other integrated nonlinear and quantum photonic platforms, including silicon nitride, lithium niobate, and aluminum nitride.