Flexible integration of free-standing nanowires into silicon photonics

Silicon photonics has been developed successfully with a top-down fabrication technique to enable large-scale photonic integrated circuits with high reproducibility, but is limited intrinsically by the material capability for active or nonlinear applications. On the other hand, free-standing nanowires synthesized via a bottom-up growth present great material diversity and structural uniformity, but precisely assembling free-standing nanowires for on-demand photonic functionality remains a great challenge. Here we report hybrid integration of free-standing nanowires into silicon photonics with high flexibility by coupling free-standing nanowires onto target silicon waveguides that are simultaneously used for precise positioning. Coupling efficiency between a free-standing nanowire and a silicon waveguide is up to ~97% in the telecommunication band. A hybrid nonlinear-free-standing nanowires–silicon waveguides Mach–Zehnder interferometer and a racetrack resonator for significantly enhanced optical modulation are experimentally demonstrated, as well as hybrid active-free-standing nanowires–silicon waveguides circuits for light generation. These results suggest an alternative approach to flexible multifunctional on-chip nanophotonic devices.


Supplementary Note 1. Morphology characterization of CdS freestanding nanowires (FNWs)
The morphology of CdS FNWs was examined by scanning electron microscopy (SEM), as depicted in Supplementary Figs. 1a and 1b. Supplementary Fig. 1a shows typical as-grown CdS FNWs on a silicon substrate. Supplementary Fig. 1b shows a single 690-nm-diameter CdS FNW transferred onto a clean silicon wafer, with highly uniform geometrical shape and smooth surface.

Supplementary Note 2. Micromanipulation of FNWs for integration with silicon waveguides
CdS FNWs were first dispersed in ethanol, and then deposited onto a glass slide. After the FNWs were dried in the open air, they can be manipulated by tapered fibre probes mounted on 3-dimension moving stages under an optical microscope 2 . Supplementary Fig. 2a gives an SEM image of a typical tapered fibre probe drawn from a single mode fibre (Corning SMF-28), with tip size down to 50 nm. To pick up a CdS FNW, first we used a fibre probe to insert beneath one end of the FNW and lift it up slowly, while the rest part of the FNW kept attached on the glass slide due to the strong friction force between the FNW and the slide (Supplementary Figs. 2c and 2d).
Then we used another fibre probe to lift up the other end of the FNW with the same approach, and finally suspended the FNW across the two fibre probe in the air ( Supplementary Fig. 2e).
Generally, longer and thicker FNWs are easier to be manipulated. However, using the nanoscale 3 fibre tip shown in Supplementary Fig. 2a  After replacing the glass slide by a silicon chip, we roughly aligned a target silicon waveguide (SW) circuit on the chip to the suspended CdS FNW under the optical microscope for subsequent FNW assembling. Supplementary Fig. 3 shows the procedure of bridging two SWs with a CdS FNW in a side-by-side coupling scheme. First, the FNW was transferred to a place close to the SWs and dropped down onto the SiO 2 substrate (Supplementary Figs. 3a and 3b). Then we used a fibre probe to press on the FNW to make it totally attached on the substrate ( Supplementary Fig.   3c). Finally, the FNW was pushed carefully to contact with the SW ends ( Supplementary Figs. 3d and 3e). Being totally attached on a clean substrate, FNWs are usually stable enough for 4 afterward top-down processing, which suggests that electrically active FNWs can also be used for this FNW-SW integration by adding electrical leads.
For the side-coupling scheme, the nanowire was supported by the SiO 2 substrate instead of the silicon core, as shown in Fig. 1 and Fig. 2 in the main text. In this case, when assembling the nanowire-waveguide coupling devices, we placed a nanowire on the SiO 2 substrate and then pushed it towards a target silicon waveguide ( Supplementary Fig. 3) until it was stopped by the silicon core. We double-checked the side-coupling structure by using a top-viewed SEM image depicted in the lower panel of Fig. 1b. From this figure, one sees the nanowire and silicon waveguide simultaneously, which means that the nanowire is at the side of the silicon core. In contrast, for the vertical coupling case, a part of the silicon waveguide can not be seen because the nanowire on the top is larger than the silicon core, as shown in Figs. 3b-3c and 4a.  CdS FNW used in our experiment, the EI difference with the 320-nm-width SW decreases with increasing wavelength, while the case for the 280-nm-width SW shows an opposite behavior.
These results can qualitatively explain the wavelength-dependent behavior of coupling efficiency in Fig. 1d. In contrast, EI of the 300-nm-width SW is much closer to that of the 860-nm-diameter FNW, resulting in higher average coupling efficiency in Fig. 1d.

Supplementary Note 4. Simulation of the near-field optical coupling between a CdS FNW and an SW
We used Lumerical FDTD to simulate the near-field optical coupling between an SW and a CdS FNW. Supplementary Fig. 5a shows the schematic illustration of the side-by-side coupling scheme with five dashed cross-section planes for investigation (Planes 1-5). The coupling length is 2.9 m Using the 3D-FDTD simulation we obtained the FNW-SW coupling efficiencies with the side-by-side coupling scheme. Considering the nanowire's surface RMS roughness (<0.5 nm) and 8 the side-wall roughness peak amplitude of silicon waveguides 7 (<5 nm), the gap between the FNW and the SW was set as 5 nm. Supplementary Fig. 6 gives the calculated coupling efficiencies between an 860-nm-diameter FNW and 340-nm-height SWs with different widths (280-320 nm).
As the SW width increases, we can see a red-shift of the efficiency maxima similar to Fig. 1d.
Besides, the efficiencies for both coupling directions are very close around the maxima. We noted that there are small discrepancies of the efficiency dependence on SW width (shift by about 10 nm) between the simulation and the experiment, which could be attributed to errors produced in the SW-width measurement and the difference between the actual CdS refractive index and literature value 6 at telecom wavelengths.

Supplementary Figure 6. Calculated coupling efficiencies between a CdS FNW and SWs.
The diameter of the CdS FNW is 860 nm and the widths of the 340-nm-height SWs range from 280 nm to 320 nm. Solid lines: coupling from an SW to a FNW. Dashed lines: coupling from a FNW to an SW.
We checked the dependence of the coupling efficiency on the gap by using an FDTD simulation.

Supplementary Figure 7. Coupling efficiencies between a CdS FNW and an SW with different gaps.
Calculated efficiencies for the two coupling directions, i.e., from an SW to a CdS FNW (a) and from a CdS FNW to an SW (b) with different gaps (5-15 nm). The width of the SW is 290 nm and the diameter of the CdS FNW is 860 nm.
We also calculated the coupling efficiencies between a CdS FNW and a SW in a vertical coupling scheme regarding different centre misalignments. As shown in Supplementary Fig. 8, First, we measured the output of the SW circuit without the CdS FNW, and denoted it as OP1 in Supplementary Fig. 9e (black triangles). Secondly, we put a graphite flake with high optical absorption on the SW to block the light propagation in the straight SW ( Supplementary Fig. 9b), and obtained a much lower (-30 dB lower) output (denoted as OP2, blue circles in Supplementary   Fig. 9e). Then, we bridged the open arm of the SW circuit by a CdS FNW (Supplementary Fig. 9c), and measured the output (denoted as OP3, red squares in Supplementary Fig. 9e), which goes 11 back to approximately the same level as the first case, indicating a high coupling efficiency between the CdS FNW and the SWs. For reference, Supplementary Fig. 9d gives an SEM image of an integrated FNW-SW MZI circuit with the bridging CdS FNW and the graphite flake. Since the input probing light was equally split into the two arms by the Y-branch, by comparing OP1 and OP3, we obtained normalized transmission of the CdS-FNW-bridged SW arm (with 2 coupling events) as OP3 -OP1, and FNW-SW coupling efficiency for single coupling event as (OP3 -OP1)/2.
In principle, this approach is valid only when the CdS FNW and the SW share a same propagation loss. When they have different losses, the actual efficiency becomes (OP3 -loss FNW -OP1 + loss SW )/2, which means for (OP3 -OP1)/2 there is an error of (loss SW -loss FNW )/2. However, as the loss coefficient of the SWs was 2-5 dB/cm [Ref. 8], the 26-m-length straight SW only introduced a loss of 0.005-0.013 dB. Therefore, considering the largest loss of the SW, this error becomes (0.013 dB -loss FNW )/2, and is only 0.007 dB in maximum provided loss FNW is negligible.
On the other hand, by zooming in the spectra of the measured coupling efficiencies shown in Fig.   1d, we obtained the maximum coupling efficiency as -0.13 dB at around 1,585 nm. Considering the possible error of 0.007 dB, the actual maximum coupling efficiency should be no less than -0.14 dB (97%), which is very close to the -0.12 dB obtained from Supplementary Fig. 6. From the measured transmission spectrum of the MZI (red line in Supplementary Fig. 10c), we selected the wavelength of the signal light (from a tunable laser, Santec TSL-710) to be 1,574.8 nm (black arrow in Supplementary Fig. 10c), slightly shift from a valley of the oscillating transmission for higher modulation depth. Similarly, in optical modulation with the hybrid FNW-SW racetrack resonator shown in Fig. 3, the wavelength of the signal light was centred 13 around a resonance of the racetrack resonator, i.e., =1,606.2 nm. Thermal nonlinear refraction effect was believed to play a minor role here, otherwise the signs of the thermal nonlinear coefficients of CdS 11 and silicon 12 would produce opposite modulation according to the signal light position in Supplementary Fig. 10c. For an MZI, it is important to realize 50/50 split ratios at both the input and output couplers to achieve large extinction ratios. As a supplementary investigation to the experiment, numerical calculations ( Supplementary Fig. 11) for the transmissions of the couplers were also performed with the same structural parameters as in the experiment, including the bended coupling SW formed by the connected ends of two S-bends. We can see the coupling efficiencies from 1,550 nm to 1,620 nm are close to 50% and the largest split ratio is only 57/43. In addition, from the measured transmission spectrum of the MZI (Fig. 2b) high transmissions (0.46 dB in average) at the constructive interference positions were also observed as well as the large extinction ratios.

Supplementary
These results suggest that the split ratios at both the input and output couplers were very close to 50/50. When the 300-nm-diameter waveguide and 860-nm-diameter nanowire were illuminated by a Gaussian beam with a spot size of 6 m respectively, the field intensities interacting with them were almost identical as the illuminations along their width direction were nearly uniform and the illuminated lengths were also the same. Therefore the core sizes of the waveguide and nanowire is less critical here for nonlinearity comparison.

Supplementary Note 7. Reproducibility of the integrated FNW-SW device fabrication
Before the micromanipulation with the FNWs, we used numerical simulation to determine a certain range of the structural parameters for fabrication. 20 couplers were fabricated for coupling efficiency measurement, among which there were 4 showing over 80% coupling efficiencies. For the MZIs, due to further structural optimization, half of the 20 fabricated devices showed large extinction ratio over 10 dB. As to hybrid resonators with vertical coupling scheme, the assembling was a bit more challenging because the FNW was to be suspended across two