1. Introduction

Plasmonic waveguides have received much attention in the past few years owing to their ability to spatially confine light well below the diffraction limit [1]. Recent investigations have been carried out to compensate for metal losses using amplifier media [2–4] as well as to integrate plasmonic structures in large-scale-integration technology [5–11]. Combining plasmonic structures with silicon photonics thus represents an elegant way to bridge the gaps between macroscopic optics and nanodevices with either optical or (opto)electronic functionality. So far, essentially two main types of plasmonic waveguide geometry have been investigated on a silicon-on-insulator (SOI) platform: (i) slot waveguides composed of two metal strips separated by a nanoscale dielectric slot [5–9] and (ii) hybrid metal-silicon waveguides consisting of a thin metal film deposited on a narrow SOI waveguide section [10], [11]. Strong confinement of light was indeed demonstrated for both structures by channeling the electromagnetic field of an SOI waveguide mode into a plasmonic guide of reduced dimensions. However, this in turn required strong modifications of the SOI platform as for instance the insertion of either tapered sections or guide interruptions.

Recently, we have demonstrated the integration of a different plasmonic waveguide geometry, which consists of a long metal nanoparticle (MNP) chain deposited on top of a SOI waveguide [12]. Energy transfer via dipolar interactions between closely spaced MNPs supporting localized surface plasmons (LSP) leads to the formation of a waveguide that can confine light at smaller scales than previous plasmonic guides [13–15]. Because the MNP chain is excited through the evanescent field of the TE waveguide modes, standard SOI waveguides can be used without requiring any specific modification of the SOI structure. Giant coupling effects were demonstrated between such long waveguide-like MNP chains and SOI waveguides [12].

Short MNP chains represent a somewhat different situation since they do not follow the dispersion relation of an infinite chain [16], and can behave in a different manner. They combine both waveguide and cavity features. Here we show that even in the case of chain lengths as small as ≈700 nm (<λ/2), almost all the TE mode energy can be transferred to the MNP chain at resonance. It is also shown that the transmission spectrum of the MNP-loaded SOI waveguide can be significantly modified by slightly changing the external environment of nanoparticles. This opens the way towards ultra-compact bio- or chemical- sensors [17] as well as to optical tweezers [18] in guided plasmonics on SOI. Because of the small size of MNPs, one could envisage the detection of a very small number of molecules of interest. The paper is organized as follows. Device fabrication is presented in Section II. Transmission measurements performed on SOI waveguides with short MNP chain deposited on top are described in Section III. The results of transmission measurements are compared to those obtained from 3D finite-difference time-domain (FDTD) calculations, which also provide mapping of the electric field along the SOI waveguide and MNP chain. It is then shown both numerically and experimentally in Section IV how the grafting of molecules to metal nanoparticles can modify the transmission spectrum of the MNP-loaded SOI waveguide. Experiments are carried out at near-infrared wavelengths. This is followed by a conclusion.

2. Device description and fabrication

Figure 1(a) shows a schematic view of the fabricated structure, which consists of five gold nanoparticles deposited on top of a SOI waveguide. The 500 × 250 nm² waveguide cross-section allows operating the waveguide on its fundamental TE mode over the spectral range of interest from 1260 to 1630 nm. The size and shape of nanoparticles are accurately determined in such a way that the LSP resonance can be excited by the evanescent field of this mode. The MNP shape is ellipsoidal (Fig. 1 (b)) with the long axis (D1 = 203 ± 5nm) parallel to the electric field and the short axis (D2 = 83 ± 5nm) parallel to the propagation direction of the guided wave. The center-to-center distance between adjacent particles is chosen equal to d = 150nm so as to provide a sufficiently strong coupling between metallic dipoles. It is worthwhile noticing that the precision achieved in the fabrication allows us to control the LSP resonance with wavelength accuracy better than ≈15nm. The fabrication of the MNP-loaded SOI waveguide is similar to that previously used for long MNP chains [12]. It includes two main steps. Firstly, a standard deep-UV lithography process followed by RIE etching and photoresist removal is used to fabricate the SOI waveguide. Secondly, gold nanoparticles are fabricated on top of the waveguide by e-beam lithography and lift-off process. A 30 nm thick gold layer is deposited by electron beam evaporation. A 1 nm titanium (Ti) adhesion layer is deposited prior to the deposition of gold.

 

Fig. 1 (a) Schematic view of the structure that we consider. (b) Picture showing the ellipsoidal shape of gold nanoparticles with D1 (respectively D2) the long (respectively small) axis size. (c) Scanning Electron Microscopy image of five gold MNP centered on the of SOI waveguide.

Download Full Size | PDF

3. Transmission measurements and numerical simulations

The transmission spectrum of the MNP-loaded SOI waveguide was measured by injecting a wavelength-tunable TE polarized light at the waveguide entrance. A reference waveguide without MNP was used on the same chip for transmission normalization. Both as-cleaved waveguides were ended by tapered sections to optimize the light injection. The input light was delivered by a tunable laser scanned by steps of 1nm over the 1260-1630 nm range. A lensed polarization maintaining fiber was used to couple the laser light to the entrance facet of the SOI waveguide. The light at the sample output was collected by an objective with a × 20 magnification and a 0.35 numerical aperture, and was focused onto a power meter. Figure 2(a) (blue curve) shows the normalized transmission spectrum measured between 1260 and 1450 nm. A transmission minimum close to zero is obtained at 1335 nm (inset of Fig. 2(a)), which corresponds to the maximum excitation of nanoparticles and to the highest ohmic losses. Almost the entire energy of the TE waveguide mode is then transferred to the MNP chain. In contrast, far from LSP resonance, MNPs have a weak in〉uence, and the waveguide transmission approaches 100%. Oscillations in the transmission spectrum are due to Fabry-Perot (FP) resonances caused by re〉ections at the as-cleaved Si waveguide facets. For an overall waveguide length of 4 mm, the period of FP oscillations is ≈0.06 nm. A complete resolution of these oscillations would require time-consuming measurements (additional information in [12]).

 

Fig. 2 (a) Normalized transmission spectrum (linear scale) of the SOI waveguide with gold nanoparticles deposited on top. Blue curve is for measurements. Red curve is for FDTD calculations. Inset shows the signal dynamics (log scale) around the transmission minimum. The minimum detection level is below −40dB. (b) Calculated re〉ection (black) and transmission (red) spectra.

Download Full Size | PDF

A 3D FDTD model from Lumerical with a uniform mesh of 3 nm × 3 nm × 3 nm was used to model the spectral behavior of the fabricated structures. Accurate dispersion data were introduced for deposited gold after fitting a Drude model to experimental ellipsometric measurements. The presence of a thin layer of native oxide between Si and Ti was also accounted for in calculations. In contrast, re〉ections at the Si waveguide facets were disregarded to avoid extremely time-consuming calculations. Figure 2(a) (red curve) shows the calculated transmission spectrum, which is in very good agreement with the experimental one. Both the general shape of the spectral response and the position of the transmission minimum are well reproduced by calculations. Figure 2(b) provides additional information about the re〉ection spectrum (black curve) calculated from the FDTD model. As seen, the level of re〉ection never exceeds 15% over the entire spectral range. It is ≈12.5% at the LSP resonance (λ≈1320 nm) while the transmission is ≈25%.

The FDTD model was further exploited to calculate the field intensity along the MNP chain and the SOI waveguide for three wavelengths: 1250, 1325 and 1450 nm. Figure 3 (left column) shows the field intensity maps along the propagation direction (x-z plane). For each wavelength, a substantial fraction of the SOI waveguide mode intensity is transferred to the MNP chain, which thus forms a short plasmonic waveguide. The propagating wave experiences ohmic losses before re-coupling to the dielectric guide at the end of the chain. In each case, interference patterns are observed in the first part of the guide due to re〉ections at the guide discontinuity. The three field maps essentially differ from each other by the amount of energy deposited in the chain and by the field distribution along the different particles of the chain. At 1250 nm, the energy transfer from the SOI waveguide to the chain is only partial, thus suggesting that dielectric and plasmonic modes have different wavevectors. The field maximum in the chain is located in the first particles. At 1325 nm (transmission minimum, Fig. 2(a)), the energy transfer to the MNP chain is almost total, indicating similar wavevectors for the two modes. The field maximum occurs at the middle of the chain.

 

Fig. 3 Left column: maps of the field intensity (| E|²) calculated in the vertical symmetry plane of the structure for three wavelengths, 1250 nm (a), 1325 nm (b) and 1450 nm (c). Middle column: intensity profiles calculated along the nanoparticle chain axis (blue) and the SOI waveguide axis (red). Right column: top view maps of the field intensity (|E|²) calculated in the mid-plane of the nanoparticle chain.

Download Full Size | PDF

At 1450 nm, the energy transfer to the chain is decreased, and the field is concentrated in the last particles of the chain. The spatial shift of the field maximum with wavelength presents similarities with that recently reported for a particle array excited by an external plane wave [19]. Figure 3 (middle column) illustrates previous evolutions from field intensity profiles calculated along the chain axis (blue curve) and along the SOI waveguide axis (red curve). Two field maxima are observed for each particle corresponding to the two air/metal interfaces. Figure 3 (right column) shows top-view maps of the field intensity calculated in the mid-plane of the MNP chain. Again, previous evolutions are verified for the three wavelengths. More importantly, field maxima are found at the two extremities of each nanoparticle as expected for a dipolar excitation. The localization of the field in the MNP chain suggests a strong dependence of the waveguide behavior with the external environment of nanoparticles.

4. Toward molecular detection in guided optics

4.1. Modeling and experiments

The sensing performances of the structure were numerically investigated by artificially incorporating a thin dielectric coating at the air/gold interface of the metallic chain to simulate a molecular environment of particles. As the cubic mesh used in FDTD calculations imposed coating thicknesses of at least 3 nm, smaller thicknesses were simulated by using low values of refractive index. Figures 4(a-b) represent the evolution of the waveguide transmission for dielectric coatings with a 5 (10) nm thickness and a refractive index varying from 1.125 to 1.5. Only gold nanoparticles are coated with dielectric to mimic the chemisorption of octadecanthiol molecules used for biomolecular detection (inset of Fig. 4(a)) [20–23]. For a thickness of 5 nm and refractive indexes of 1.125 and 1.25, the transmission minimum is redshifted by ≈25 and 40 nm, respectively (Fig. 4(a)). The wavelength shift per refractive index unit is then 160 nm/RIU. A transmission redshift is also observed when increasing the coating thickness at a constant value of refractive index (Fig. 4(b)). The transmission spectrum slightly broadens whether the index or the coating is increased. More importantly, Fig. 4(b) shows that the transmission spectrum calculated for n = 1.25 and t = 10 nm is almost identical to that calculated for n = 1.5 and t = 5 nm. A refractive index variation, ∆n = 0.25, induces a wavelength shift of the transmission minimum equivalent to that produced by a thickness variation ∆t = 5nm. This equivalence was verified for different values of ∆n and ∆t. More generally, wavelength redshifts of the transmission minimum calculated with respect to the uncoated structure (t = 0, n = 1) were found to be well fitted by the following formula:

 

Fig. 4 (a-c): Calculated transmissions of waveguides with a chain of 5 gold nanoparticles coated with dielectric material. (a) 5 nm thick coating with refractive index n = 1.125, and 1.25, (b) 5 and 10 nm thick coatings with either n = 1.25 or n = 1.5 refractive index, (c) 5 nm thick, n = 1.125 coatings with or without absorption. Red curve in (a), (b) and (c) is the reference for uncoated gold. Inset in (a) schematizes a monolayer of thiol molecules deposited on gold. (d): Transmission minimum redshift calculated versus ∆n = n-1 using Eq. (1). Circles are FDTD calculations. The triangle is for experimental measurements with a 2 nm thick thiol layer.

Download Full Size | PDF

First experiments towards molecular detection were carried out by depositing a monolayer of octadecanthiol (ODT) probe molecules on gold nanoparticles. ODT molecules are commonly used in plasmonic bio-sensors. The deposition process was the same as the one described in [21]. The nanoparticle dimensions were D1 = 210 nm and D2 = 65 nm (instead of D1 = 205 nm and D2 = 85 nm in Fig. 2(a)) leading to a plasmon resonance at 1375 nm (instead of 1335 nm). The thiol layer thickness and refractive index were estimated to be t≈2nm and n≈1.4 like in recent infrared experiments [22, 23]. Figure 5 shows the results of waveguide transmission measurements for the same device without (blue curve) and with ODT molecules (red curve). Averaged curves are also depicted to smooth out FP oscillations caused by reflections at the waveguide ends. As in Fig. 2(a), a reference waveguide (without gold nanoparticles) was used on the same chip to normalize the transmission of the waveguide with gold nanoparticles before and after deposition of thiols. This reference SOI waveguide had exactly the same characteristics as the measured waveguide except for the absence of gold nanoparticles and then of thiols. It was also subjected to the same chemical process. Therefore, the spectral changes in Fig. 5 only reflect the influence of thiols deposited on the device with gold nanoparticles. As seen, the transmission minimum is redshifted by ≈22 nm compared to the waveguide without thiol. This result is in very good agreement with the predictions of the formula reported above for t = 2nm and ∆n = 0.4. The lower transmission level as well as the spectral broadening reveals the presence of optical losses. Near-infrared absorption of OTD molecules is in principle small, but could be explained in part by absorption tails associated to electronic excitations in the visible. So far, near-infrared absorption of thiols has not been explored in details. The transmission decrease in Fig. 5 (bottom) can also be attributed to scattering losses induced by the surface roughness of gold nanoparticles coated with thiols. Such a surface roughness is not explicitly treated in FDTD calculations, but scattering losses are accounted for through the imaginary part of the refractive index, n_i, in Fig. 4(c). A value of 0.1 for n_i corresponds to a loss coefficient of 10000 cm⁻¹.

 

Fig. 5 Waveguide transmission measured without octadecanthiols (blue curve) and with octadecanthiols grafted on metallic nanoparticles (red curve). As in Fig. 2(a), oscillations in the transmission spectrum are due to Fabry-Perot (FP) resonances caused by re〉ections at the Si waveguide facets. Averaged thick curve are guides for the eyes.

Download Full Size | PDF

4.2. Discussion

Numerical results of Fig. 4 and experimental ones of Fig. 5 show the very high sensitivity of our waveguide structures. A ≈22 nm redshift of the waveguide transmission minimum is found for a ≈2 nm layer of thiol molecules deposited on gold nanoparticles. For a number of five ellipsoidal particles with ≈0.013 µm² cross-section and 0.030 µm thickness, this represents a detection volume as small as 0.26 attoliter (≈2.6 × 10^—4 µm³) ! Figure 4(d) represents the wavelength shift of the transmission minimum calculated versus the refractive index variation using Eq. (1) for different thicknesses of molecular layers deposited on gold nanoparticles. This allows us to quantify the sensitivity of our device in terms of wavelength shift per refractive index unit. If molecular sensing is performed in air, the sensitivity scales from 80 to 270 nm/RIU for layer thicknesses varying from 2 to 10 nm (slopes of curves near ∆n = 0 in Fig. 4(d)). If molecular sensing is performed in aqueous solutions (n≈1.33), the sensitivity scales from 43 to 167 nm/RIU for the same layer thicknesses (slopes of curves near ∆n = 0.33 in Fig. 4(d)). The deposition of a 2nm thick thiol layer in an aqueous medium is then expected to produce a wavelength shift of ~3nm. Such a shift will be easily detectable using an improved device with anti-reflecting coatings at waveguide ends, thereby avoiding Fabry-Perot resonances. Previous values of sensitivity are comparable to those recently reported for photonic crystal based waveguiding configurations. For instance, a sensitivity of ~130nm/RIU was found at ~1550nm for a system including a 1D photonic crystal coupled to a waveguide [24]. A sensitivity of ~103nm/RIU was obtained at ~1500 nm for a photonic crystal waveguide coupled to nanocavities [25]. Advantages of our device are its higher compactness and a simpler fabrication process.

5. Conclusions

In summary, we have reported the fabrication and optical characterization of very short metal nanoparticle chains on standard SOI waveguides at near-infrared wavelengths. Waveguide transmission measurements have been found to be in very good agreement with FDTD calculations, which also revealed the strong coupling of chains to the TE waveguide mode at localized surface plasmon resonance. Almost the entire waveguide mode energy can be transferred to chains as short as ≈700 nm at this resonance. Thanks to the accuracy achieved in fabrication, the resonance wavelength itself can be controlled with a relative precision better than 5%. These results are believed to be key steps towards the integration of nanometer-size plasmonic functions in silicon photonics. Localized surface plasmons potentially offer a wide variety of guiding configurations since metallic nanoparticles can be arranged on demand on SOI waveguides.

Molecular sensing based on the plasmonic resonance of short metallic chains in guided optics has been investigated both numerically and experimentally. Results have shown the high sensitivity of such waveguide structures for an ultrasmall detection volume of 0.26 attoliter (≈2.6 × 10^—4 µm³) ! This result was reproduced by FDTD calculations. An important advantage of using waveguides instead of free space configurations stems from the fact that optical sources and detectors can be positioned far from the plasmonic section, thus providing an independent access to this section for optical detection, microfluidics and biosensing. Moreover, channeling the excitation light in a waveguide ensures better efficiency and energy saving. Further extension to visible wavelengths is expected with the use of different waveguide systems including doped glass and silicon nitride waveguides.