1. Introduction

Silicon photonics is the subject of intense research and with the tremendous progress in fabrication techniques a rich variety of silicon-based photonic devices have been demonstrated over the past decade. This success is partly due to the optical characteristics of the material, which allows for strong optical waveguiding properties in the near infrared, and partly due to the compatibility of the manufacturing process with standard semiconductor growth technologies. It is thus not surprising that silicon as an optical material has attracted considerable interest with the enticing prospect to combine both electrical and optical functionalities on a single chip with nanometer scale precision whilst still allowing for mass fabrication and reduced manufacturing costs.

All-optical signal processing is often envisioned as the ultimate solution to increase the speed of data transfer. Indeed, a wide range of functionalities has been demonstrated in a silicon photonic platform ranging from signal filtering and amplification to wavelength conversion [1–3] making silicon-based nanostructures ideal candidates for all-optical signal processing. However, the realization of optical signal processing generally requires the use of highly nonlinear materials and, unfortunately, silicon inherently suffers from two major drawbacks limiting its use as a nonlinear material: (i) two-photon absorption (TPA) which dominates over the Kerr nonlinearity at wavelengths below 2 microns and in particular at the telecom wavelengths and (ii) the absence of the second-order nonlinearity due to the intrinsic centrosymmetric nature of the material.

In order to overcome the large nonlinear losses arising from TPA, several alternatives have been proposed including e.g. implanting hydrogen atoms into the silicon material [4] or embedding silicon nanocrystals into a silica matrix [5]. Another promising alternative to prevent large TPA and simultaneously enhance the nonlinearity of a silicon based device is to use slot waveguides where the field is tightly confined in a low refractive index material sandwiched in between two silicon ridges of higher refractive index. When filled with a highly nonlinear material the slot waveguide nonlinearity can reach extremely high levels [6] with significantly reduced TPA [7] allowing e.g. to perform all-optical switching at much faster speeds [5]. In this context, it was recently shown that atomic layer deposition (ALD) is particularly adapted for filling silicon slot waveguides with inorganic nonlinear materials as it allows for conformal growth and accurate control of the material thickness [8–10].

Another approach to enhance the Kerr nonlinearity of silicon-based devices is to design a structure with resonant features such as Bragg gratings or photonic crystals that can increase the nonlinearity through slow light effects. Recently, combining the two aforementioned approaches in one single structure has been proposed. The idea is to periodically modulate the refractive index along the light propagation direction of a silicon slot waveguide. This type of a nanostructure then acts as a 1D photonic crystal, in which light is tightly confined providing a possibility for enhanced interaction with a nonlinear material deposited in the slot waveguide whilst simultaneously creating slow light effects in the periodic refractive index lattice [11]. This approach could then allow to reach ultrahigh nonlinearities even at low power levels [12–14]. Other schemes have also been proposed to integrate slot waveguides with photonic crystals [15–17], usually by creating a groove in the center of a photonic crystal.

Here, we demonstrate a nanoscale merged photonic crystal slot waveguide (MPCSW) on a silicon on insulator (SOI) platform that operates as a bandpass filter in the near infrared region of the spectrum. The structure on top of a SOI substrate integrates a silicon slot waveguide together with a photonic crystal cavity. Efficient in- and out-coupling is ensured by the integration of nano-waveguides and adiabatic couplers. Importantly, the device is conformally embedded in a layer of amorphous titanium dioxide (TiO₂) using atomic layer deposition (ALD), which allows to reduce surface roughness and linear losses [18] whilst still allowing for strong light-matter interactions with the filling material having low TPA. One motive to use ALD grown amorphous TiO₂ is its low optical losses, which helps to demonstrate linear functionality of the device. The structure is optimised using a 3D Finite Difference Time Domain (FDTD) method and the experimentally measured characteristics are in excellent agreement with those predicted by the simulations. Specifically, the measured effective refractive index experienced by the propagating light is found to be close to that of TiO₂ showing evidence that light is indeed confined in the ALD layer. The concept is very flexible and our MPCSW thus opens a new route to combine nonlinear guided-wave optics resulting from the strong field confinement in the slot region with slow light effects in the photonic crystal cavity.

2. Device design

A schematic of the merged photonic crystal slot waveguide concept is illustrated in Fig. 1 which shows the geometry and key parameters of the structure. The structure consists of a vertical silicon slot waveguide connected to a photonic crystal and a cavity. The slot waveguide provides strong optical confinement of the electric field in between the silicon ridges whilst the photonic crystal lattice consists of 20×2 silicon rectangular pillars that open a photonic bandgap in the waveguide transmission. A nanocavity is then created in the center of the photonic crystal lattices by connecting two consecutive pillars with a silicon ridge. The cavity resonance induces a sharp transmission peak within the bandgap and the device thus acts as a bandpass filter. Varying the period of the silicon pillars allows to vary the spectral position of the bandgap whilst varying the cavity length affects the location of the resonant peak within the bandgap. The whole structure is dry etched on top of a SOI wafer and embedded into a layer of titanium dioxide so as to confine light within an ALD grown material. To ensure efficient coupling into the MPCSW device a three-step coupler is also implemented in form of a silicon taper followed by a nanowaveguide and an adiabatic nanowaveguide to slot waveguide coupler, which allows to reduce significantly the insertion losses [19]. A similar coupler is added at the other end of the MPCSW for efficient out-coupling.

 

Fig. 1 Schematic illustration of the merged photonic crystal slot waveguide conformally coated with amorphous TiO₂. The geometrical parameters are the slot width W_S = 80 nm, the rail width W_R = 200 nm, the period D, the length of the silicon nano block d, the cavity length L_C, and the thickness t of the TiO₂ layer.

Download Full Size | PDF

Confinement is achieved in the vertical dielectric slot waveguide which implies that the device operates for the quasi-TE polarization (i.e the electric field is oriented along the x direction). The thickness of the waveguide layer is fixed at 220 nm. The thickness of the TiO₂ coating layer is set to 180 nm. The input waveguides are 3 μm wide and are tapered down to 480 nm wide nanowaveguides. The length of the classical tapers and the adiabatic couplers is 18 μm and 4 μm, respectively. Optimization of the design for operation around λ = 1350 nm was performed by the 3D-FDTD method [20]. The period of the photonic crystal lattices D, fill factor f defined as f = d/D where d is the length of a silicon rectangular pillar, and cavity length L_c are used as optimization parameters. The wavelength-dependence of the different materials refractive indices is accounted for in the design with the specific values n_Si = 3.4, n_TiO2 = 2.3, and n_SiO2 = 1.44 at λ = 1350 nm.

Color plots of the transmission spectrum of the structure when varying the fill factor, the cavity length, and the lattice period are illustrated in Fig. 2. We first examine the effect of the fill factor (Fig. 2(a)) on the bandgap characteristics. For this purpose, the cavity length and lattice periods were chosen to be identical and equal to 320 nm. We first remark that a large bandgap is only obtained for a fill factor within 20–80 % range as manifested by the dark areas. We can also see how the fill factor plays a key role in the position of the bandgap with a net red-shift of c.a. 70 nm when the fill factor increases from f = 25 % to f = 50 %. At this point it is important to bear in mind that the resolution in the fabrication process is around ±10 nm thus corresponding to approximately ±8.5 nm variation in the exact location of the gap. We clearly observe the presence of a sharp transmission peak within the bandgap associated with the resonant wavelength of the cavity and we note a non-negligible dependence of the resonance amplitude on the fill factor. More specifically, for low or high values of f the overall amplitude of the resonance decreases. This may be counter-intuitive as in principle the maximum amplitude of the peak should increase for low or large values of the fill factor. But because the amplitude of the bandgap decreases as well, the net result is a reduction of the resonance amplitude. On the other hand, for values of f close to 50%, the opposite occurs and the overall amplitude of the resonance peak appears larger.

 

Fig. 2 Simulation results obtained by 3D-FDTD. Normalized transmission spectra as a function of the fill factor f (a), the cavity length L_C (b) and the photonics crystal period D (c). (d), (e), (f): Group index corresponding to the transmission spectra for the same parameters. For fill factor variations, L_C = D = 320 nm; for cavity length variations, f = 50 %, and D = 320 nm; for period variations f = 50 %, L_C = D.

Download Full Size | PDF

We next studied the effect of changing the cavity length as shown in Fig. 2(b). The fill factor and the lattice period in this case were set to 50 % and 320 nm, respectively. As can be expected, varying the cavity length allows to tune the resonant wavelength and the associated sharp transmission peak across the entire bandgap (nearly 200 nm). Moreover, it is well-known that for L_C = D the resonance falls exactly at the center of the bandgap and it is indeed what is seen in Fig. 2(b).

When studying the spectral transmission dependence on the period the fill factor was kept at 50 % and the cavity length taken to be identical to the period so as to maintain the resonance at the center of the bandgap. As shown in Fig. 2(c), it is clear that the lattice period affects drastically the spectral position of the bandgap with a detuning of the bandgap by as much as 500 nm for a 100 nm change in the photonic crystal period. At this stage we emphasize that from the fabrication point of view, the period is in fact the most stable parameter in terms of reproducibility, which then allows us to set precisely the bandgap spectral position. In this work we have chosen a period D = 320 nm resulting in a resonant peak (and bandgap) centered at λ = 1350 nm and a bandgap width of c.a. 150 nm. Note that this particular choice was based on the spectral range of the available broadband source in our laboratory for the device characterization but the spectral features of the structure could be easily tuned to a different wavelength by simply scaling the geometrical parameters. Finally, we point out that the slot waveguide width (W_S = 80 nm) which is the same as that of the silicon bridge in the photonic crystal cavity as well as the number of lattice periods were optimized so as to increase the localization of the electromagnetic field in the slot region. The former affects slightly the position of the bandgap and the resonance peak amplitude whilst the latter acts on the amplitude of the bandgap.

In order to evaluate the performance of the device regarding the potential use of slow light effects in the photonic crystal cavity we have also computed in Figs. 2(d)–2(f) the group index n_g = c/v_g = c(dk/dω) = (c/L)(dφ/dω) from the group velocity of light in the structure compared to that in vacuum c and examined its sensitivity to the structure parameters (fill factor, cavity length and lattice period) [21]. As expected, large amplitude variations of the group index are observed and follow that of the sharp cavity resonance, allowing theoretically for a factor of 10 in light speed reduction comparable to the design proposed by Riboli et al. [11].

3. Fabrication

The device was fabricated on top of a SOI wafer with a 2 μm thick buried oxide and patterned into a 220 nm thick silicon layer. The designed structure was exposed to an electron beam patterning tool (Vistec EBPG 5000+ ES HR) on a spin coated negative tone hydrogen silsesquioxane electron beam resist (Dow Corning XR-1541) applying an area dose of 5000 μC/cm² with an acceleration voltage of 100 kV. The patterns were then developed with a water diluted Microposit 351 developer. Excellent quality of the vertical profile and sidewall smoothness were achieved by using a hydrogen bromide and oxygen based inductively coupled plasma etching process (Oxford Instruments Plasmalab 100). The residual resist layer was subsequently removed by buffered hydrofluoric acid applied for a carefully calibrated time to avoid etching of the buried oxide. Finally, the sample was cleaved by hand and coated by a 180 nm thick TiO₂ layer that fills both the slot waveguide and the photonic crystal cavity. The TiO₂ layer was grown by ALD (Beneq’s TFS 200) from TiCl₄ and H₂O precursors at a 120 °C temperature [22]. Here, the particular choice of ALD is crucial so as to deposit conformally the TiO₂ in nanoscale structures allowing the shape of the nanodevice to be perfectly preserved and the structure to be completely filled with a material of nonlinear properties that can also be tailored [23, 24]. In addition, low-loss anatase TiO2 waveguides have recently been demonstrated [25] and anatase TiO2 can easily be grown with ALD [24].

The different sub-sections of the fabricated merged photonic crystal slot waveguide are shown in Fig. 3 prior to deposition of the TiO₂ layer. Specifically, Figs. 3(b) and 3(c) show a scanning electron microscope image of the top view of the etched MPCSW before applying the ALD coating and of the adiabatic coupler used for efficient coupling light from the nano-waveguide to the slot waveguide, respectively. For completeness, zoomed images over the slot waveguide section and the cavity are also displayed in Figs. 3(d) and 3(e). The high quality and excellent smoothness of the various sub-sections surfaces can be clearly observed with the structure dimensions matching closely those of the intended design.

 

Fig. 3 (a) Field distribution in the photonic crystal slot waveguide on resonance (a). SEM images of the photonic crystal slot waveguide (b), the adiabatic coupler (c), transition from the slot waveguide to a merged photonic crystal slot waveguide (d), and the cavity of the photonic crystal (e).

Download Full Size | PDF

We then proceeded to evaluate the optical confinement in the actual merged photonic crystal waveguide through numerical simulation as illustrated in Fig. 3(a). One can see how most of the field is indeed guided by the slot waveguide and well confined within the TiO₂ region. This means that we can expect (i) strong interaction between light and the filling material and (ii) significantly reduced two-photon absorption, which are both pre-requisites for the implementation of nonlinear functionalities.

4. Characterization

The transmission measurements of the merged photonic crystal waveguide were performed using a broadband supercontinuum source spanning the 1100–1600 nm wavelength range and which allows to capture the full transmission spectrum of the fabricated structure in a single measurement. The source is composed of a gain-switched fiber laser operating at λ = 1547 nm (Keopsys KPS-KULT2-1550) producing 1 ns pulses with a peak power of 6 kW at a repetition rate of 75 kHz. The laser output is connected to a 6 m long dispersion shifted fiber (Corning LEAF, G.655) with 1510 nm zero-dispersion wavelength. A dichroic mirror was used to filter out part of the long wavelength spectral components that typically corresponds to high energy solitons in the time domain and could damage the sample. The supercontinuum source is then injected into the merged photonic device using a tapered lensed fiber with a 2.5 μm focus spot size. The output light is collected using a similar taper fiber connected to an optical spectrum analyzer (OSA). In order to eliminate possible coupling aberrations from the lensed fiber, the measured transmission spectrum was normalized to that measured for a nano-waveguide without the slot waveguide and photonic crystal cavity. The results are shown in Fig. 4 where we compare the transmission spectrum from the 3D-FDTD simulation from the ideal design (Fig. 4(a)) with the measured transmission spectrum without and with the cavity (Figs. 4(b) and 4(c)). Several observations can be made. First we note that the photonic crystal opens a bandgap in the transmission between λ = 1280 nm and λ = 1450 nm, in excellent agreement with the prediction of the numerical simulation. We also see clearly how the cavity further introduces a 12 nm wide transmission peak centered at λ_peak = 1350 nm within the bandgap (compare (b) and (c)). The spectral amplitude of the peak corresponds to approximately 50 % of the maximum intensity transmitted through the device with an extinction ratio of 8 dB. We remark that the characteristics of the resonance peak (wavelength, amplitude and spectral width) also agree well with the simulations except for a small red-shift, which we attribute to the variation of the fill factor during the fabrication process. Note that the observable decrease in transmission below the short wavelength edge of the bandgap is due to the slot waveguide itself.

 

Fig. 4 Transmission spectrum of the merged photonic crystal slot waveguide. Simulated by 3D-FDTD (a), measured for a 15 periods photonic crystal without the cavity (b), and measured for a 10 periods photonics crystal on each side with the cavity (c). All spectra were filtered using a Fourier transform band pass filter.

Download Full Size | PDF

The wavelength of the resonant peak and the length of the cavity allow us to calculate the effective index of the mode: n_eff = λ_peak/2L_C ≃ 2.11 which is close to the refractive index of the TiO₂ layer (n_TiO2 =2.3 at λ = 1350 nm) and confirms that light is indeed highly confined in the TiO₂ layer. Finally, an estimation of the losses in the photonic crystal was determined by measuring the total losses in the device with and without the structure using the Fabry-Perot method. This method allows us to eliminate the coupling loss at the input and output of the waveguide. In this measurement we consider the full device, as describe in Fig. 1, as the Fabry-Perot cavity. The additional losses induced by the nanostructure are α = 0.8 dB.

5. Discussion and conclusions

We have demonstrated the concept of integrating on the same nanodevice a slot waveguide and a photonics crystal cavity embedded in titanium dioxide. Tapers, nanowaveguides and adiabatic couplers were further integrated in the device so as to minimize coupling losses. The resulting merged photonic crystal slot waveguide fabricated with CMOS compatible processes exhibits a remarkably high quality both in terms of targeted dimensions and surface roughness. Significantly, the atomic layer deposition employed for coating the device with titanium dioxide allows for an accurate control of the thickness of the conformal layer. The fabricated device was designed to operate as a bandpass filter at 1350 nm with an extinction ratio of 8 dB and a spectral width of 12 nm, but operation of the device can be easily tailored for the 1.55 μm telecom wavelength range simply by scaling the design parameters. Also, the ALD coating of the device is not limited just to TiO₂ but the material can be chosen from a versatile set of ALD materials, with an ultimate goal of employing a highly nonlinear material. Numerical simulations further show that the precise characteristics of the bandgap and transmission resonance can be manipulated by fine-tuning the geometrical structure parameters. Because the electric field is essentially guided in the titanium dioxide layer, two-photon absorption is expected to be drastically reduced compared to silicon waveguides. By filling the structure with some high nonlinear coefficient material, not necessarily limited to the ALD technique, the device may be used advantageously for nonlinear optical signal processing thus opening a new route to integrated devices including both linear and nonlinear functions such as tunable band pass filters, switches, modulators and wavelength converters, which can be controlled with low power.