Direct growth of monolayer MoS₂ on nanostructured silicon waveguides

1 Introduction

Silicon is an indirect semiconductor with a centrosymmetric crystalline structure, which is not ideal for detection, electro-optic modulation, or light emission [1–4]. Therefore, CMOS-based optoelectronic chips, which utilize optical waveguides, have remained elusive. It is challenging to integrate the well-established direct semiconductors such as III-V semiconductors due to the epitaxial mismatch with silicon. On the other hand, semiconducting Van der Waals materials can be grown on a large set of non-epitaxial substrates, yielding, e.g., field-effect transistors and other electro-optic devices [5–8]. Therefore, direct integration of these 2D materials into silicon nanostructures in a wafer scalable process may add nanoscopic light sources and detectors into the toolbox of silicon photonics, thus helping bridge the gap between the excellent electronic properties of silicon and its established waveguide capabilities.

Transition metal dichalcogenides (TMDs) have the general formula MX ₂, where M represents the transition metal, and X represents a chalcogen [9–11]. Most TMDs can be exfoliated into monolayers because individual layers are bonded by the weak van der Waals force [9]. TMDs are semiconductors [9] with a bandgap in the near-infrared or long-wavelength part of the visible spectrum. Monolayer TMDs are direct bandgap semiconductors [12] with remarkable optical and electronic properties. They exhibit strong photoluminescence (PL) [12–16] driven by excitons. The TMDs having a single or odd number of layers are highly nonlinear [17–19] due to the lack of inversion symmetry. Planar heterojunctions of TMDs are also suitable as light sources and detectors [20, 21].

The integration of TMDs with photonic structures predominantly relies on transfer-based methods [22–31]. However, transfer processes of TMD on plane substrates has multiple disadvantages and challenges due to lack of scalability, contamination, doping, strain, wrinkles, air bubble and structural damages [32–36]. Moreover scalability and cost-effectiveness is important in large scale photonic applications. Intensive research has been conducted to reduce these effects. Some methods are able to solve some of these challenges, for example: thermal release tape (TRT) method was shown to have resulted in wrinkle free transfer to a plane substrate [35]; still the effects of doping was observed in that case. In general these improved methods don’t resolve all the limitations of the transfer process simultaneously. Most of these methods are not yet tested in case of nanostructures. Structured surfaces pose additional challenges in ensuring homogenous and conformal contact of the 2D material with the device [27], and the strain fields, damages, and folds are likely to be even more pronounced compared to the case of plane target substrate. There are many differences between various techniques for integrating TMD on nanostructures in the present context, which are summarized in Table 1. From the table, we can see that the direct CVD growth of TMD on a silicon photonic platform has been under-explored.

Previous works [39–41] have demonstrated the use of van der Waals epitaxy for the integration of 2D materials with different structures. In that context, the 2D material acts as the starting material. Then bulk Si or GaAs is grown on it by classic epitaxy, albeit not at the quality required for many optoelectronic applications. Nevertheless, this suggests that van der Waals materials can be integrated with these materials because their binding properties facilitate crystalline growth without epitaxially compatible substrates.

In contrast to van der Waals epitaxy, we fabricated the silicon on insulator (SOI) waveguides in the first step. Then we grew TMD monolayer crystals directly on the nanostructure using chemical vapor deposition (CVD). CVD is a well-established method to synthesize crystalline TMD monolayers [37, 42], [43], [44], [45] or even heterostructures [46] with a material quality comparable to that of exfoliated samples [45]. Since the CVD process of TMDs is based on non-directional chemisorption, the crystals grow conformally over the nanostructure following the bends and corners of the geometry as shown in Figure 1. Both silicon [47] and SiO₂ [48] will not undergo deformation at the temperature and pressure required for the CVD growth of MoS₂. Therefore this approach is assumed to be compatible with established CMOS processes for silicon nanostructures. Although the growth temperature for hBN is much higher compared to TMDs (≈1000 °C) [49, 50], recent work demonstrated the direct growth of hBN on photonic chips [51]. However, hBN is a wide-gap dielectric, and electro-optical functionalization is thus difficult [52].

Figure 1:

Concept schematic of the MoS₂ crystal (green) grown over a Si waveguide structure (pink) via the CVD process.

2 Results and discussion

The nanostructured sample consisted of silicon ridge waveguides tapering to grating couplers at both ends [53]. The waveguides were fabricated with a multistep electron-beam lithography technique on 220  nm silicon-on-insulator substrate, later diced into 10 × 10  mm chips. The waveguides and grating couplers were designed to operate at a wavelength of 1320 nm. The waveguides were still functional after the high temperature (730 °C) CVD growth process of the MoS₂ crystals (Supplementary 4). The nonlinear optical responses of MoS₂, such as second harmonic generation [18], Kerr effect [22], and four wave mixing [30] are not limited to short wavelengths, making it an attractive material for integration with NIR silicon photonic platforms. The photoluminescence emission around 680 nm, and the resonant enhanced second harmonic response at 405 nm are discussed below to analyze the quality of the MoS₂ grown on the waveguide structure. Other TMDs such as MoSe₂ which has enhanced optical responses towards longer wavelengths can also be grown on nanostructures using the CVD process nearly at the same temperature. In this study we resorted to an in-depth analysis of the growth and characterization of the MoS₂ crystals on the nanostructured substrate.

MoS₂ monolayers were grown on the waveguide chips using a well-established CVD technique [45]. The TMDs in their characteristic angular shape had lateral dimensions ranging from 2 to 10 μm, as seen in the optical microscope image (Figure 2a). The crystals grew on the SiO₂ substrate, the waveguides, the grating couplers, and even the fabrication labels. In general, we observed more MoS₂ crystals on the grating couplers (pink regions in Figure 2a) compared to the waveguides themselves (black lines in Figure 2a). We attribute this to the couplers providing more surface roughness due to the inscribed shallow gratings and thus providing more nucleation sites for the initial growth of the TMDs. Although we do not have direct control over the nucleation density and the position of the crystals, we hypothesize that creating similar grooves would enable relatively higher nucleation density in those regions, suggesting a pathway for exquisite control over the growth positions, and which will be the subject of future work.

Figure 2:

Morphological characterization of MoS₂ crystals on SOI waveguides. (a) Optical microscope image of CVD-grown MoS₂ on waveguides. The blue areas are the SiO₂ substrate. The pink, wider parts at the top consist of Si grating couplers and the black narrower parts below are the Si waveguides. MoS₂-crystals appear as greenish. The waveguides are 220 nm higher than the substrate, therefore the top of the waveguides is slightly out of focus. (b) Scanning electron microscopic (SEM) image of MoS₂ crystals grown seamlessly over edges and materials boundaries. (inset) The focused ion beam (FIB) cross section of the waveguide along the dotted line in the SEM image to highlight the nanoscale bending radius of the edges. (c) Phase image from atomic force microscopic (AFM) measurement of a region marked by the box in (a). (d) AFM height map corresponding to the AFM phase image. The grains in the height map are due to the surface roughness (RMS value 2 nm) of the substrate. (e) AFM height profiles across the lines marked in the height map in (d).

We did not observe any apparent interruption in the individual crystals at a material interface between Si and SiO₂, edges or corners in the optical microscope image (Figure 2a). The continuity of the crystals is more evident in the scanning electron microscopy (SEM) image in Figure 2b. The crystals wrapped over a “TM” shaped nanostructure (top left of the SEM image), where the termination of the Si-nanostructure is complex. The inset of Figure 2b shows the SEM image of the focused ion beam milled (FIB) cross-section across the grating coupler with MoS₂ grown on it. This reveals the nanoscale convex bends at the top (radius of curvature <30 nm) and concave bends at the bottom (radius of curvature <25 nm). The crystal grew over these bends, along the 220 nm high sidewalls, and across the Si-SiO₂ interface at the bottom. Here we see that the crystal can grow conformally across structures of low aspect ratio (labels; 1: 0.05) as well as those of higher aspect ratio (waveguides; 1: 0.73). Although it should be possible to grow on any nanostructure irrespective of the aspect ratio, additional process optimization might be required to obtain large area single crystals.

To prove the monolayer nature of the as-grown MoS₂, we performed multiple AFM measurements. The phase image in Figure 2c clearly shows the outline of the crystal. The phase of the crystal is the same on the SiO₂ substrate as well as on the Si waveguide. This suggests that the properties of the MoS₂ crystals are unaffected by the growth substrate and not deteriorated by the nanostructure. The height map related to the phase image is shown in Figure 2d. The surface roughness of the waveguide structure is high (RMS = 2 nm) compared to the thickness of monolayer MoS₂ due to the waveguide fabrication process. Therefore it is difficult to observe the TMD-crystal clearly in topography mode. However, we can see the edges of the crystals clearly after filtering. The height profiles shown in Figure 2e, shows an average thickness of 1 nm for the monolayer crystals. The Raman spectra measurements also confirms the layer number (Supplementary 3).

This shows that monolayer TMD-crystals can be grown on nanostructures defying the epitaxial matching required for common 3D semiconductors. Particularly, the growth around a bend with radius less than 30 nm is well beyond prior observations of CVD growth on fiber core with a bending radius above 10 μm [37].

The utilization of TMD-based semiconducting devices in silicon nanostructures not only depends on the TMDs morphology but also on the conservation of its electro-optical properties. We investigated those by mapping photoluminescence (PL) [12], [13], [14, 16] and second harmonic generation (SHG) [17–19]. We choose the combination because SHG is sensitive to structural changes of the monolayer crystals, whereas PL indicates the local electronic properties, dominated, e.g., by doping and defects. Both methods are thus complementary and yield a good overview of the monolayer quality.

Figure 3a is a PL map of the waveguide structure shown in the optical microscope image in Figure 2a. The TMD crystals covering the SiO₂ substrate have strong PL. However, the signal is quenched on the Si grating couplers and the Si waveguides. PL spectra could be collected only from MoS₂ located on SiO₂. Figure 3b is the room-temperature PL spectrum collected from the positions marked in the PL map. A double Gaussian fitted to the spectrum revealed strong contributions from the A exciton at 1.82 eV and weaker contributions from the B exciton at 2.03 eV. These are consistent with literature values for monolayer MoS₂ [12], [13], [14, 16]. A more detailed analysis is provided in Figure 5 of the Supplementary information. Similarly, we map the SHG signal (Figure 3c) after excitation with an 810 nm pulsed laser. A strong SHG response shows that the observed crystals are indeed monolayers. The SHG emission from the grating couplers is slightly suppressed but not as much as the PL emission. Figure 6c, in the Supplementary information provide the power dependency plot of the SHG signal. At this point we are unable to analyze the quality of vertically grown parts of the crystals due to the limitations in our setup. This analysis shows that, although MoS₂ can be grown with good morphologic properties directly on Si nanostructures, it is not yet suitable for TMD-based semiconducting devices. We hypothesize that the strong PL quenching is related to charge transfer from the substrate to the TMD, as previously observed for 2D material growth on silicon wafer [54–56] and carrier loss at the TMD-Si interface [20, 57]. Also there is a possibility of un-intentional n-doping of silicon during the lithography assisted fabrication. These mechanisms would not only quench the A- and B-excitons but should also have a detrimental effect on the C-excitons. Besides these factors, silicon has significant absorptive loss at 405 nm [31]. These explain the observed loss of SHG-efficiency at the 810 nm pump wavelength.

Figure 3:

Optical characterization of MoS₂ on waveguide structure. (a) Photoluminescence (PL) map of MoS₂ monolayers on silicon waveguide structure, the spots marked by the star denotes the position of spectra measurement. (b) PL spectra from the MoS₂ crystal at the positions marked in (a); the experimental data is fitted with a double Gaussian fit marked by the dashed red curve. (c) Map of second harmonic generation of the MoS₂ coated waveguide structure.

To test this hypothesis and implement a way towards MoS₂-growth with high optoelectronic quality on Si nanostructures, we created a second batch of samples coated with a passivation layer [56]. This layer is for avoiding direct contact between the 2D material and silicon. It is inspired by mechanically exfoliated hBN layers used in a previous study [55]. The mechanical exfoliation of hBN is not scalable, and CVD growth of hBN requires exceedingly high temperatures (≈1000 °C) [50, 51]. Therefore, we coated the second set of nominally identical samples with a protective dielectric layer using atomic layer deposition (ALD) [58–63] before the CVD process.

ALD is widely used in the semiconductor industry, and it creates conformal thin films, i.e., it coats steep side walls the same way as planar surfaces [58]. It is an established technique to coat nanostructures with high-quality optical coatings [61, 62]. We deposited 5 nm films of Al₂O₃ on the Si structures. We chose Al₂O₃ as it is one of the widely used substrates for the CVD growth of TMDs [64–70]. This thickness, which is larger than the exciton radius provides good electronic separation. But the layer is also thin enough to avoid considerable changes in the optical properties of the waveguide modes, and the overlap of electromagnetic field with the TMD crystals.

We performed finite element mode simulations of the modes supported by the waveguides to ensure that the additional deposited layer has no considerable effect on the waveguide modes. The results are summarized in Figure 4, where we compare the intensity of the TE- and TM- like guided mode for the case of an uncoated (Figure 4a) and coated (Figure 4 (b–d)) waveguide considering different coating layer combinations as indicated by the respective sketches in the top. The second, and the third rows correspond to the intensity of the TE- and TM- like mode, respectively. The variation in the effective refractive index (n _eff) of the TE and TM modes is less than 2% in each case, indicating that the modal properties are comparable. The percentages in (c), (d) refer to the fraction of total modal power in the MoS₂, showing that the impact of the Al₂O₃ coating is minimal in terms of coupling power from the Si waveguide to the MoS₂. We also calculated the effective refractive indices of TE and TM modes for varying thickness of Al₂O₃. This plot is provided in the Supplementary Figure 7. Note that the fraction of total modal power in the MoS₂ is comparable with and without the Al₂O₃ coating (0.03%).

Figure 4:

Simulation of waveguide modes. (a) Left: schematic of the simulated structure for the bare silicon waveguide. Middle: TE-like mode intensity. Right: TM-like mode intensity. Also labelled are the effective index n _eff and the dominant electric field direction. The wavelength is 1.32 μm. The same calculations considering 5 nm of Al₂O₃ coating (orange) only are shown in (b). The same calculations considering 5 nm of Al₂O₃ coating and 1 nm of MoS₂ (red) are shown in (c). The same calculations considering 1 nm of MoS₂ are shown in (d). The percentages in (c), (d) refer to the fraction of total modal power in the MoS₂. Note the comparable modal profile in each case. The refractive indices (n) used in the model are as follows. SiO₂: n = 1.45; Si: n = 3.5; MoS₂: n = 4.2; air: n = 1, Al₂O₃: n = 1.67.

Figure 5:

Optical characterization of MoS₂ on waveguide structure with 5 nm Al₂O₃. (a) Optical microscope image of the sample with a separation layer of 5 nm Al₂O₃. (b) Photoluminescence (PL) map of MoS₂ monolayers on silicon waveguide structure, the spots marked by the blue and white stars denote the positions of spectra measurement. (c) PL spectra from the MoS₂ crystal at the positions marked in Figure 5b; the red curve is the double Gaussian fit. (d) Map of second harmonic generation of the MoS₂ coated waveguide structure. (e) Raman spectra measured at 532 nm excitation, from two regions; the blue (red) colored spectra at the position labelled by the blue (red) star shown in the Raman map in Figure 5f. The peak positions of the spectra are the same for the MoS₂ on top of the Si waveguide as well as on SiO₂. The signal intensity is quantified as photon counts per second (C. P. S.). (f) The map of the E2g peak in the Raman spectra.

We grew MoS₂ on the ALD-coated nanostructure using the CVD process as before. The optical microscope image (Figure 5a) shows triangular TMD crystals. The crystals emit photoluminescence, as shown in Figure 5b. In contrast to the uncoated sample, the PL is equally strong on the silicon waveguides and SiO₂. This shows that the Al₂O₃ barrier prevents charge transfer and recombination at the Si-TMD interface. Figure 5c is the PL spectrum from the regions marked by the stars in the PL map. The Gaussian fitting reveals prominent A excitons at 1.81 eV with an FWHM of 51 nm and weaker B excitons at 1.99 eV with an FWHM of 37 nm as expected from MoS₂ monolayer crystals of good quality.

We measured the Raman map and Raman spectra to understand the crystal quality further. Raman spectrum is not only sensitive to the number of layers [71, 72] but also to strain [73–77]. The Raman spectra in Figure 5e were measured from the positions marked in Figure 5f. The spectra showed the characteristic peaks of MoS₂ around 385.2  cm⁻¹ (E2g) and 403.5  cm⁻¹ (A1g), corresponding to the in-plane vibration of Mo and S atoms and the out-of-plane vibration of the S atoms, respectively [64, 66, 67]. The separation between the two peaks is 18.3  cm⁻¹, indicating that they are monolayer MoS₂ crystals. From multiple measurements from different positions on the crystals (Supplementary 10), we conclude that the strain is homogenous over the raised Si waveguides and the lower-lying SiO₂ substrate. No additional strain is present at the edges as opposed to the case of mechanical transfer.

The SHG map in Figure 5d also proves that the Al₂O₃ layer is helpful to retain the optical properties of the MoS₂. SHG is observable from the entirety of the MoS₂-crystals. This observation is different from the SHG quenching observed for the uncoated samples. Instead, we find a roughly fourfold SH enhancement on the couplers. While the nature of this enhancement is unclear, we speculate the possibility of local field enhancement on the high-refractive-index Si structures. SH observed from the multilayer crystals at the very right and left of the images suggests that those sections might be AA-stacked multilayers. The Raman spectra (Supplementary 10) also reveals the presence of multilayers on some positions on the waveguide.

This particular study focuses on the process compatibility, and advantages of transfer-free growth of TMDs on silicon based photonic structure. We conclude that our CVD growth process of TMD is compatible with the SOI platform to obtain good quality TMD integrated with the photonic chip in a scalable manner. Future work will focus on a comprehensive study of the optical coupling between optimized MoS₂/silicon hybrid waveguides across multiple wavelength regions.

3 Conclusion and outlook

We successfully demonstrated the direct CVD growth of conformal monolayer 2D materials on nanostructured silicon on insulator waveguides. We find that the waveguide is functional after the CVD growth process at the intended wavelength (Supplementary 4). The crystals grew around nanometer-sized edges and corners without detectable levels of induced strain. The monolayer crystals exhibit good morphologic properties. Although the interaction of MoS₂ with silicon reduces its optical properties, a 5 nm coating of Al₂O₃ is sufficient to form an electronic barrier between the TMD monolayers and the silicon nanostructures shielding the TMD from the detrimental effect from (doped) silicon. The barrier layer does not influence the optical properties of the waveguides significantly, and it creates an environment to leverage the full potential of the TMD excitons in a scalable manner for integrated optical architectures. We believe that the ramifications may be profound: integrating TMDs directly with silicon waveguides may provide for nanoscale detectors and light sources. This would not only complete the toolbox of integrated wafer-scale components available to silicon integrated architectures, but also allow for the integration of entirely new types of devices such as room-temperature polaritonic lasers [78], valleytronic devices, or single-photon light sources [29].

4 Methods