Surface sensitivity of Rayleigh anomalies in metallic nanogratings

Sensing schemes based on Rayleigh anomalies (RAs) in metal nanogratings exhibit an impressive bulk refractive-index sensitivity determined solely by the grating period. However, the surface sensitivity (which is a key figure of merit for label-free chemical and biological sensing) needs to be carefully investigated to assess the actual applicability of this technological platform. In this paper, we explore the sensitivity of RAs in metal nanogratings when local refractive-index changes are considered. Our studies reveal that the surface sensitivity deteriorates up to two orders of magnitude by comparison with the corresponding bulk value; interestingly, this residual sensitivity is not attributable to the wavelength shift of the RAs, which are completely insensitive to local refractive-index changes, but rather to a strictly connected plasmonic effect. Our analysis for increasing overlay thickness reveals an ultimate surface sensitivity that approaches the RA bulk value, which turns out to be the upper-limit of grating-assisted surface-plasmon-polariton sensitivities.


Introduction
Spectral anomalies in the response of diffraction gratings constitute a subject of longstanding interest in optics, which originated from the observations by Wood in 1902 [1] and their physical interpretations provided by Rayleigh [2,3] and Fano [4]. The reader is referred to a recent book chapter by Maystre [5] (and references therein) for a nice historical perspective and a comprehensive review of the quantitative phenomenological theory. Basically, two types of anomalies can be identified: sharp anomalies, due to the passing-off of a spectral diffraction order, and diffuse anomalies arising from the excitation of surface waves. The former, typically referred to as Rayleigh anomalies (RAs), occur at wavelengths ( ) m R λ given by the well-known grating formula [5] where a one-dimensional (1-D) grating of period Λ immersed in a medium with refractive index (RI) a n is assumed, and i θ denotes the angle of incidence (measured anticlockwise from the normal to the grating). Examples of diffuse anomalies are surface plasmon polaritons (SPPs) occurring when the collective charge oscillations on the surface of a metallic grating are excited, or guided-mode resonances, which may occur, for instance, in dielectric-coated metallic gratings [6]. Both types of anomalies (sharp and diffuse) strongly affect the transmission/reflection of light through/by sub-wavelength metal gratings. They typically manifest themselves as rapid intensity variations in the spectra, and, depending on the type of periodic structure, may occur either separately or jointly (almost superimposed) [7].
It is worth pointing out that RAs and SPPs have very different near-field distributions. More specifically, for RAs, the field enhancement extends far from the grating surface, whereas, for SPPs, the field distribution is exponentially bound to the grating surface. Although both RA and SPPs are inherently sensitive to changes in the surrounding RI, this distinction bears important implications in connection with possible applications of these gratings to sensing, as it will be discussed hereafter.
An approach for using RA-based nanogratings (specifically, periodic hole arrays in thin gold films) for RI sensing was proposed in [8], where it was essentially demonstrated that the coupling of an SPP with an RA may lead to a large increase in the transmission of a narrow linewidth peak over a small RI range.
Based on the observation that the Rayleigh wavelengths in Eq. (1) are sensitive to changes of the surrounding ambient RI, a novel sensing scheme was proposed in a recent work [9]. Essentially, this approach is based on the monitoring of the wavelength shift of a reflection peak associated with an RA of a concentric gold ring nanograting laid on the end facet of an optical fiber. Among the potential advantages of this type of sensors, the authors highlight the polarization insensitivity (due to the rotational symmetry of the grating), the linear relationship between the RA-peak wavelength and the ambient RI [cf. Eq. (1)], with a bulk sensitivity determined solely by the grating period, namely 900 nm per RI unit (RIU). Given these impressive properties, one may be led to consider RA-based grating nanosensors as a valid alternative to standard SPP-based sensors [10,11] in label-free chemical and biological sensing applications. However, in these typical scenarios of practical interest, the RI changes are mainly restricted to surface modifications occurring at the sensor interface (e.g., due to the binding of a biological layer, physical or chemical adsorption of analyte molecules). Consistently with the inherently diffractive character of the RA phenomena (with extended field distributions) [12], the figures of merit predicted in [9] are expected to deteriorate when local (rather than bulk) RI changes are considered, as typical in all sensing schemes when the sensing volume is not entirely filled up. However, a quantitative assessment of such deterioration cannot be attained via simple considerations, since local surface modifications significantly enrich and complicate the underlying phenomenology. Therefore, an in-depth investigation of the optical response of the structure is required, which motivates our present study. are deposited on the top surface of the nanograting in order to estimate its surface sensitivity. Also shown is the unit cell considered in the numerical simulations, as well as the illumination from air.
Accordingly, in what follows, we numerically and experimentally study the surface sensitivity of RA-based metallic nanograting sensors, starting with nanosized overlays mimicking local RI changes. We show that, in this case, these devices exhibit a rather low (up to two orders of magnitude below the bulk figures) sensitivity, which is however in line with what observed in other grating-assisted sensing platforms [13]. Interestingly, we show that such residual sensitivity is not attributable to the RA wavelength shift, but rather to a strictly connected SPP effect. As a consequence, we found that the grating duty-cycle (DC) value that optimizes the surface sensitivity is generally different from that considered in the bulk case in order to maximize the RA visibility.
Moreover, we show how, for increasing overlay thickness (up to wavelength-sized values comparable with the typical sensing-volumes of plasmonic modes), this sensitivity approaches as an ultimate limit the RA bulk sensitivity, which represents an upper-bound for gratingassisted SPP sensitivities.

Experimental study on the RA surface sensitivity
Compatibly with our fabrication facilities [13], but without loss of generality, we begin our analysis by considering the 1-D metallo-dielectric structure schematically depicted in Fig. 1. It essentially consists of a dielectric (ZEP) 370 nm thick layer backed by a silicon substrate and patterned with a 1-D lattice with period 1500nm Λ = and duty-cycle A 30nm thick film of gold is deposited on both the ridges and grooves of the dielectric grating.

Fabrication
The nanograting was fabricated via electron beam lithography (EBL) followed by gold sputtering. The first step was the deposition of 370nm of e-beam positive resist (ZEP 520A, Zeon Chemicals) via spin coating onto a cleaned square silicon substrate (10×10 mm 2 ). After spinning, the sample was baked on a hot plate for 3 minutes at 180°C. The periodic pattern was then defined on an area of 400×400 µm 2 by using a Raith 150 EBL system with a voltage of 20kV and a dose of 55 µC/cm 2 . The grating lines were fabricated by using the Fixed Beam Moving Stage mode, which yields high precision and flexibility in defining thin but extended lines (or paths) with no field stitching boundaries. This operation allows the fabrication of well-defined line widths with no need to rely on a defocused beam. After the exposure, the sample was developed in a ZED N50 solution (Zeon Chemicals) for 55 sec. and rinsed in isopropanol. A nitrogen flow was subsequently utilized to remove the possible resist residue. The resulting dielectric grating was then coated with a 30 nm thick gold layer by DC magnetron sputtering in a vacuum system at a base pressure of 5 8 10 − × Torr. The 30 and 60 nm thick SiO 2 layers were laid on top of the structure by RF sputtering at a pressure of 3mTorr with a rate of 3.3 Å/sec.
A scanning electron microscope (SEM) image (top view) of the fabricated device is shown in Fig. 2. The inset contains a magnified detail of the grating, showing the high degree of uniformity down to few nanometer length-scales.

Characterization
The experimental setup for the out-of-plane reflectivity measurements comprises a broadband white light source directly coupled to a standard fiber-based reflection probe (AVANTES -Standard RP). Such probe illuminates the sample (placed on a XYZ positioning stage with 10 µm absolute on-axis accuracy) via an optical fiber bundle with six outside fibers in a ring-shaped configuration, and collects the reflected light by means of a central fiber (with a collection spot diameter ~350 µm) connected to an optical spectrum analyzer (with 0.4 nm resolution). The fibers' numerical aperture is 0.22. A 9 mm distance between the reflection probe and the sample was chosen as a suitable tradeoff between a reasonable signal-to-noise ratio and the fulfillment of the paraxial conditions (normal incidence). Finally, the measured reflectivity spectra were normalized with respect to the reference response of an unpatterned gold film (on the same substrate, but away from the grating area).  1)], at which the first ( 1 m = ± ) diffracted order reaches the grazing condition [5]. Also shown (black-dashed curve) in the same figure is the numerical response computed via a Rigorous Coupled Wave Analysis (RCWA) algorithm [14], assuming as RI for ZEP and silicon 1.54 ZEP n = and 3.4 s n = , respectively, and relying on a Lorentz-Drude model [15] to account for the dispersion of gold. The grating period, DC and thickness were set to their nominal design values (1500 nm, 0.25 and 370 nm, respectively), and normal-incidence conditions were assumed. For a meaningful comparison, since no polarization control is implemented in our measurement setup (cf. Section 2.2), the numerical response was obtained by averaging the contributions of the transverse-electric (TE) and transverse-magnetic (TM) polarizations (i.e., electric field parallel and orthogonal to the grating lines, respectively). Moreover, consistently with the experiment, the response was also normalized with respect to the reflectivity response of a gold mirror. A good agreement between experimental and numerical results can be observed although, around the reflectivity peak, the experimental response exhibits a smoother profile by comparison with the numerical counterpart. This may be attributable to fabrication tolerances as well as slight deviations from the normal incidence condition in our experimental setup. It should also be noticed that our simulations assume an idealized conformal deposition of the SiO 2 overlay (i.e., with same thickness on both the grating ridges and grooves) which is likely different from that actually attained in the fabricated device. For this configuration, a bulk sensitivity

Results and discussion
is theoretically predicted [5,9], with (1) and 60nm , respectively) can be observed when the overlay thickness is increased. Also in this case, the experimental responses match the numerical ones, demonstrating a good agreement. In essence, the main effect of the overlay deposition is a broadening of the RA peak. This may be quantified by looking at the 80% linewidth of the reflectivity peak, which increases from 42 nm (no overlay), to 57nm ( 30 d t nm = ) and 73nm ( 60 ), resulting in a maximum broadening up to 73.8%. In addition to the resonance linewidth broadening, also a red shift of the reflection peak can be observed. Consequently, we estimated the surface sensitivity by monitoring the wavelength shift of the reflectivity peak from ( )

Remarks on the general validity of our results
It should be stressed that the above results and related observations are not limited to the specific metallo-dielectric nanograting structure in Fig. 1 (which was chosen only for fabrication convenience), but are instead quite general, and hold for different nanograting configurations. To prove this point, in what follows, we proceed with a numerical study the RI surface sensitivity of the device considered in [9]. Besides confirming the general character of our findings, this will also allow us to consider, in the rest of the discussion, a structurally simpler configuration which greatly facilitates a clean-cut interpretation of the underlying phenomenologies. The analyzed device, schematically shown in Fig. 4 ). Differently from our initial configuration in Fig. 1, a fiber-optic-type illumination (i.e., impinging from the substrate) is considered here. Moreover, in view of the fused-silica substrate, a different dielectric overlay

Nature of the surface sensitivity
For a deeper understanding of the observed surface sensitivity, it is insightful to look at the numerical reflectivity spectra pertaining to the TM and TE polarizations, shown in Figs. 5(b) and 5(c), respectively. Such spectra, which were averaged in Fig. 5(a), turn out to be quite different. First, the dynamic range is quite different (note the different scale). Moreover, while both responses exhibit a quite visible spectral feature (peak and dip, respectively) around the 1st-order RA wavelength ( ) , the presence of the overlay induces only a slight offset on the reflectivity scale, without any shift of the spectral dip. From this evidence, it is possible to recognize that the small residual surface sensitivity observed is not attributable to a genuine RA wavelength shift, but rather to the interplay with other phenomena that occur only for the TM-polarization. The most intuitive interpretation of such interplay would involve the excitation of a plasmonic (SPP) mode at a wavelength close to ( ) 1 , R air λ , which is responsible for the falling front of the spectral reflectivity peak [16], and cannot occur for the TE-polarization.
This even better illustrated in Fig. 6, which compares the reflectivity spectra and relevant field-maps in the presence and absence of a thicker ( 150 ) overlay, for the TMpolarization. Basically, in the absence of the overlay [ 0 d t = , cf. Fig. 6(a)], only the steep rising front of the reflectivity peak is actually attributable to an RA, whereas the falling front exhibits a plasmonic character. This is evident from the comparison between the extended field distributions at the RA wavelength [ Fig. 6(b)] and the localized character observed at slightly higher wavelengths [ Fig. 6(c)]. Our conclusions are further supported by the responses in the presence of the overlay. From the reflectivity spectrum shown in Fig. 6(d), we observe that the spectral feature at the 1st-order RA wavelength is clearly de-emphasized but still visible at ( ) To sum up, our observations above indicate that RAs are per se completely insensitive to surface modifications of the ambient RI. The residual small sensitivity observed is actually attributable to their interplay with SPP modes excited at nearby wavelengths.
Within this framework, we recall that the particular choice of the grating DC in [9] (and also utilized here) was the result of an optimization process aimed at maximizing the visibility of the RA spectral feature in the form of a sharp peak, rather than as a moderate jump or cusp which would be difficult to detect experimentally (cf. Fig. 2 in [9]). Our interpretation above suggests that such optimization basically corresponds to exciting an SPP close to an RA wavelength. However, in view of the actual plasmonic nature of the residual surface sensitivity, and recalling that the grating DC also affects the SPP field-decay in the direction orthogonal to the metal-dielectric interface (which determines the sensing volume and hence the surface sensitivity), it turns out that the DC value that optimizes the RA visibility may not necessarily be the most convenient choice here. Figure 7 shows the surface sensitivity [computed via Eq. (3)] as a function of the overlay thickness d t varying within the range 40-200nm, for DC=0.35, 0.45 and 0.55. It is evident that the value DC=0.35, chosen in [9] to maximize the RA visibility (in their bulk configuration), does not guarantee the best surface sensitivity, which is instead attained for larger DC values. Further improvements of the surface sensitivity based on plasmonic effects are also possible via the deposition of functional high refractive index overlays at nanoscales (so as to enhance the field content at the metal-dielectric interface) [17], or by selecting sensing configurations involving localizedsurface-plasmon-resonance (LSPR) effects characterized by much stronger field localization at the metal interface [18].

Ultimate sensitivity
It is now interesting to explore the connection between the RA and SPP-based sensing mechanisms that coexist in this type of nanograting devices. In particular, one might wonder whether both mechanisms essentially boil down to similar sensitivities when the associated sensing volumes are completely filled by the analyte.