Metasurface for broadband coherent Raman signal enhancement beyond the single-molecule detection threshold

Gap surface plasmon metasurfaces (GSPMs) are a promising technology in the field of photonics and plasmonics due to their simplicity of fabrication and ability to fully control the amplitude, phase, and polarization of reflected light.¹ These metasurfaces are made of a metal film and metal subwavelength elements arranged in a periodic fashion with a subwavelength dielectric spacer in between. GSPMs have numerous potential applications, including beam-steerers,² flat lenses,³ holograms,^4,5 absorbers,^6,7 color printing devices,⁸ polarization controllers,^9,10 surface wave couplers,¹¹ and dynamically reconfigurable metasurfaces.¹² GSPMs can be relatively easily fabricated with a single lithography step and have the ability to function as an absorbing element due to strong near-field coupling between the metallic layers.^13–17 Layered structures are commonly used in the construction of GSPMs due to their simplicity.¹⁸ In addition to their fabrication benefits, GSPMs exhibit broadband behavior.¹⁹ For example, a broadband and highly efficient anomalous meta-reflect array was created using trapezoid-shaped nanorods.² The resulting multifunctional GSP-based metalens was able to split and focus orthogonal linear polarizations into different focal spots. GSPMs may also alter their spectral properties when covered with different dielectric materials, making them useful for sensing applications.²⁰ Overall, the feasibility and simplicity of GSPMs, combined with their broadband resonance character, make them a promising emerging technology with numerous potential applications.¹ 

The ability to amplify weak optical signals makes plasmonics an attractive avenue for advancing nonlinear optical technologies in fields such as laser frequency conversion, generation of ultrashort laser pulses, all-optical signal processing, ultrafast switching, and spectroscopy.^21–23 Plasmonic enhancement of Raman spectroscopy signals has been predicted and applied in a variety of realizations.^24–27 Among numerous Raman techniques, coherent Raman anti-Stokes scattering (CARS) is especially susceptible to plasmonic enhancement due to the underlying high-order nonlinearity. Surface-enhanced CARS has been experimentally realized with thin films,²⁸ nanowires,²⁹ gratings,^30,31 and nanovoid arrays.³² Exploiting GSPMs in this context led to a demonstration of single-molecule signal detection at the gap of plasmonic nanoquadrumers.³³ Employing core–shell structures enabled the time-resolved recording of single-molecule signals.³⁴ Theoretical work on coupled nanohole–slit arrays predicted CARS signal enhancement by an unprecedented 18 orders of magnitude.³⁵ In the supplementary material, Sec. I, we review selected previous literature on various approaches to CARS enhancement with metallic or dielectric nanostructures. This contribution is focused on a broadband GSPM design feasible for fabrication with state-of-the-art technology. The metasurface supports a broadband optical response achieved due to constructive interference of plasmonic modes. These modes sustain significant absorption losses, which are responsible for their large spectral width. In consequence, absorption, usually considered a drawback, is exploited to achieve a crucial advantageous feature: The broadband operation enables the metasurface to simultaneously support Raman signal enhancement of various molecular species, thus waiving the requirement for tunability. The narrow gap employed in the design enables the combination of this broadband response with a substantial electric field enhancement, potentially giving rise to an unprecedented increase in the Raman signal intensity for a zoo of molecules.

The GSPM is designed in a metal–insulator–metal configuration. Its unit cell consists of a metal film at the bottom for reflection and constructive interference of fields in the enhancement volume, a SiO₂ spacer layer, and two nanodisks on top sustaining broadband plasmonic enhancement and separated by a gap [Figs. 1(a) and 1(b)]. The metasurface is periodic in the x and y directions and subject to x-polarized plane wave illumination from the top along the z direction. The unit-cell geometry is optimized numerically to maximize the enhancement factor of CARS for a broad range of molecular Raman shifts (see Methods for simulation details). The resulting geometry has the unit cell period P = 300 nm, the radius r = 39 nm of each nanodisk, the gap g = 3 nm between the nanodisks, the thicknesses h₁ = 28 nm of the gold nanodisks, h₂ = 300 nm of the insulator, and h₃ = 270 nm of the gold film. Fabrication of this and similar structures may be possible, e.g., by combining electron beam lithography with the liftoff technique³⁶ or through pattern transfer nanomanufacturing.^37,38

Figure 1(c) shows the reflectance and absorption spectra, consisting of two overlapping resonances in the visible domain that together form broad plateaux, which is our operational spectral range. This range Δf_EF exceeding 140 THz is marked in Fig. 1(c) with the red dashed-dotted vertical lines. We define Δf_EF based on the requirement of a near-field enhancement factor exceeding 100 at the position in the middle of the gap, as also discussed below. Such strong enhancement is achieved due to constructive interference with the bottom gold reflecting layer: See the supplementary material for simulation results with varying spacer thickness (Sec. II) and in the absence of the reflecting layer (Sec. III). Additionally, the width of the plasmonic resonance of the disks is increased when adding the mirror layer. The transmittance is entirely suppressed in the investigated range due to the presence of the metallic film at the bottom of the structure. Note that we have not assumed vanishing transmittance a priori. The numerically evaluated absorption and reflectance spectra based on the finite element method (see Methods) are in very good agreement with the fit to the semi-analytical transmission line theory (TLT) results. The latter method is described in detail in the supplementary material, Sec. IV. The absorption by the metal bulk makes the resonances spectrally broad, turning the loss into an advantage. The broadband character of the metasurface’s response may be useful in applications for high-resolution displays,³⁹ plasmonic lenses,⁴ spatial phase modulators,^40,41 and various Raman spectroscopy realizations.⁴² Despite the significant absorption loss, the local electric field enhancement can actually exceed two orders of magnitude, as we demonstrate in Sec. III, which results from the reflection at the mirror formed by the bottom metallic layer and the constructive interference of the reflected and the incoming field. In Sec. IV, we discuss how these features can be exploited to achieve an unprecedented enhancement of Raman signals in the coherent anti-Stokes technique.

In the surface-enhanced realization of CARS (SECARS), the signal intensity is improved through several mechanisms, all potentially involving the tailored optical response of plasmonic metasurfaces. We perform our calculations under the weak-excitation assumption, where excitation enhancement and emission enhancement can be treated separately.

1. Excitation enhancement: Subwavelength confinement of electromagnetic fields contributes to the signal at the right-hand side of Eq. (1). In consequence, transitions in molecules driven by locally enhanced pump and Stokes fields occur at improved rates. According to Eq. (1), this results in an enhancement of the CARS signal by the factor

   $gp2gS≡Ex,MS(rm,ωp)Ex,0(ωp)4Ex,MS(rm,ωS)Ex,0(ωS)2.$

   (2)

   Here, the fields at the molecular position in the presence of the metasurface are denoted with the subscript MS and include the incoming and scattered contributions. We have assumed plane-wave illumination, for which the incoming field amplitudes are position-independent. Note that in the presence of the metasurface, the response is not isotropic, and we focus on the x field polarization component that optimizes the performance of the investigated metasurface.

2. The electric near-field enhancement at the highest-symmetry point of the GSPM illuminated with an x-polarized plane wave at normal incidence from the top is shown with the blue curve in Fig. 2(b). This spectrum reflects the double-peak character of the absorption and reflectance from Fig. 1(c), providing strong, over 100-fold field enhancement in a broad spectral range between 378 and 518 THz. This range is wide enough to support different kinds of molecules in a wide range of Raman shifts.

3. Emission enhancement: The resulting molecular signal at the anti-Stokes frequency is additionally improved by the radiative Purcell enhancement factor. For a dipolar transition, it can be evaluated as the ratio of electric-dipole emission powers P(ω_aS) radiated in the presence and absence of the metasurface,⁴⁴ 

   $gaS=Prad,MS(rm,ωaS)P0(ωaS).$

   (3)

   This ratio is plotted as a function of dipole frequency with the orange line in Fig. 2(b). The dipole position has been fixed at the highest-symmetry point of the metasurface marked with the red dot in Figs. 1(a) and 1(b). For simulation details, see Methods. Naturally, the resonances observed in the power ratio are narrower than those characterizing the field enhancement. The power enhancement exceeding 4 orders of magnitude is achieved again in a broad range of frequencies from 399 to 458 THz.
4. In parallel to radiative emission, the electromagnetic energy of the molecular dipole can also be absorbed in the metallic bulk of the nanoparticles forming the metasurface. The ratio of radiated and total (radiated + absorbed) powers is quantified with the efficiency

   $η(rm,ω)=Prad,MS(rm,ω)Prad,MS(rm,ω)+Pabs,MS(rm,ω),$

   (4)

   which is a function of both the position and frequency of the source dipole. Note that this factor is already taken into account in the definition of g_aS as expressed via the radiated power in Eq. (3). The metasurface efficiency is demonstrated in Fig. 2(c) for the dipole position, which varied along the z axis with respect to the highest-symmetry point of the structure (red dot in Fig. 1) and is relatively stable around the level of 50%. We provide more results on near-field dependence on geometry parameters in the supplementary material, Sec. II.

In the usual experimental realizations of CARS, the pump frequency ω_p is fixed while the Stokes frequency ω_S is fine-tuned. On two-photon Raman resonance, ω_p − ω_S = Δ_R, the anti-Stokes signal is efficiently generated. Below, we fix the pump frequency at ω_p/2π = 450 THz at the center of the plateaux spectral region in Fig. 1(c), which allows flexibility of the Stokes and anti-Stokes beam frequencies that ideally should also fit in the plateaux. We evaluate G_SECARS(ω_p, ω_S, r_m) for a given Stokes frequency ω_S = ω_p − Δ_R. The anti-Stokes signal frequency becomes ω_aS = ω_p + Δ_R.

First, we consider a fixed Δ_R = 35 THz = 1167 cm⁻¹, corresponding to a resonant Stokes frequency of ω_S/2π = 415 THz. The field enhancement distribution around the nanostructure is shown in Figs. 3(a)–3(c) for illumination at the pump frequency. Enhancement factors exceed 200 in the gap region. Similar field profiles are found for other illumination frequencies in the range between 360 and 540 THz, in accordance with the spectrum in Fig. 2(b). We evaluate the SECARS enhancement factor G_SECARS(ω_p, ω_S, r) at different positions r in the unit cell with a fixed z = 0 nm [Fig. 3(d)] or y = 0 [Fig. 3(f)]. (See also a depiction in greater detail in the supplementary material, Sec. V.) The simulations involve independent calculations of the Stokes and pump enhancement factors with the plane wave and the anti-Stokes enhancement factor in the dipole illumination scheme, as described in Eqs. (2) and (3). The resulting SECARS enhancement maps demonstrate that the strongest signal enhancement is expected for molecular positions at the highest-symmetry point in the gap between the nanodisks. The enhancement factor predicted at the optimal point reaches an impressive 19 orders of magnitude, slightly exceeding previously obtained literature results.^31,35 This value is also well above the single-molecule detection threshold for the CARS technique, estimated at 10 orders of magnitude.³³ The volume of space where a single molecule could be detected assuming this threshold corresponds to the yellow regions in Figs. 3(d)–3(f). Figure 3(e) shows a zoom of panel (d) on the highest-enhancement region. Note that the saturation threshold for the off-resonant CARS scenario, which is here considered, and for pulsed illumination has been predicted at the pump intensity $Ip∼1022Wm2$ at the Stokes intensity $IS∼1015Wm2$⁠.⁴⁵ These values hugely exceed realistic numbers for plasmonic-based scenarios.

As a result of the broadband character of the near field enhancement spectrum [Fig. 4(a)], the SECARS enhancement factor at a fixed position G_SECARS(ω_p, ω_S, r_opt) is robust for the broad range of Raman shifts up to 1700 cm⁻¹ [Fig. 4(b)]. The supplementary material, Sec. II, illustrates more curves corresponding to other values of z. The same Figs. 4(a) and 4(b) demonstrate that the near field enhancement factor and the resulting SECARS signal enhancement factor may vary strongly as the molecular position is shifted along the z axis. The maximum electric field enhancement is achieved in the gap directly above the glass surface and, depending on field frequency, drops by a factor of 4–5 at a height of 30 nm (supplementary material, Fig. S2). For the highly nonlinear process of CARS, this corresponds to a reduction of the SECARS signal enhancement factor from 19 to 15 orders of magnitude. The most stable response is obtained around $z=h12$ (supplementary material, Fig. S2), i.e., around the middle of the disk height due to a capacitor effect. Above the disks, the performance quickly drops for z > 30 nm. The radiated power ratio at the anti-Stokes frequency proves to be much more robust against position shifts in the xy plane but also sensitive to the position in the z direction [Figs. 4(c)–4(e)]. Panel (f) demonstrates that the power enhancement spectrum is suppressed by a factor of 2 as the position changes from z = 1 nm to z = 20 nm. The averaged SECARS enhancement in the 20 × 20 nm² area of the xy plane centered at z = 0 [Fig. 3(e)] and excluding the area within the metallic disks is $GSECARSav=5×1016$⁠. Similarly, the average enhancement factor in the fragment of the y = 0 plane between the disks is $GSECARSav=1.6×1017$⁠.

Among the geometry parameters of the metasurface, the small gap in which the electromagnetic fields are strongly confined is crucial to achieving such a significant SECARS enhancement. In Figs. 5(a) and 5(b), we study the strong modulation of the near field enhancement factor g(r_opt, ω) (a) and the SECARS signal enhancement factor G_SECARS (b) on the gap size in the range of 2–15 nm. Note the robustness of G_SECARS against the Raman shift again [Fig. 5(b)]. The possibility to further enhance the signal with the same geometry is limited, as gaps smaller than 2 nm are challenging to fabricate. Moreover, one should not expect further signal enhancement with even smaller gaps: larger molecules might be unlikely to fit in the gap, tunneling effects might suppress the field enhancement, and finally, surface adsorption might change the molecular chemical and, hence, Raman properties. On the other hand, the predicted CARS enhancement factors already reach the single-molecule detection threshold for gaps as large as 15 nm [see Fig. 5(b)], which should not pose particular technological challenges and can be fabricated by electron beam lithography.^46–48 Another parameter that strongly influences the field enhancement factor and, therefore, the SECARS signal is the disk radius [Figs. 5(c) and 5(d)]. As the radius increases from 30 to 60 nm, we observe the appearance and splitting of a pair of resonances with an optimized broadband near-field spectrum for radii of around 40 nm [Fig. 5(c)]. The broad response for this radius results in the optimized SECARS enhancement factor [Fig. 5(d)]. The dependence of the enhancement factors on other geometrical parameters of the metasurface is discussed in the supplementary material, Sec. II. In the same section, we characterize the impact of smoothened disk edges.

Finally, we compare the performance of the passive metasurface studied in this work with the previously investigated active design.³¹ The latter was based on the metal-insulator-grating configuration, with the bottom gold layer replaced with a grating. That geometry allowed for optical response in the form of three tunable resonance peaks to quasi-independently control the pump, Stokes, and anti-Stokes modes. Their spectral positions could be adjusted by geometry engineering or the incident angle of the illuminating beams to match selected molecular Raman profiles. The possibility of post-fabrication tuning through the incidence angle yields the “active” characteristic of the metasurface in opposition to the “passive” one, tunable by design. We have demonstrated the tunability of rhodamine 6G with its characteristic Raman peaks at 1314, 1363, and 1512 nm, all enhanced above the single-molecule detection threshold (Fig. 6). In the figure, the red line represents the Raman spectrum S_R6G(Δ) of rhodamine 6G,⁴⁹ while the green dashed line is the predicted spectrum of a rhodamine molecule optimally positioned and oriented on the optimized active metasurface. This result is calculated as a product of the rhodamine spectrum and the predicted signal enhancement $GSECARSactive(Δ)SR6G(Δ)$⁠. Contrary, the passive metasurface described here does not enforce three resonances for the three field modes. Dropping this demanding condition allowed us to achieve a much stronger amplification over a wide spectral range; now, the entire investigated range of the Raman spectrum of the molecule is enhanced by a factor of the order of 10¹⁹ with little modulations (blue line in Fig. 6). This fact makes the metasurface universal, i.e., supporting CARS signal enhancement for a variety of molecular species. Since the metasurface enhances the signal strength but weakly distorts the molecular Raman response spectral shape, it may support both molecular detection and identification. Moreover, the structure’s potential to enhance nonlinear light–matter interaction processes extends beyond the CARS technique. In the supplementary material, Sec. VI, we analyze the potential of the structure to enhance SERS signals.

In summary, this article’s broadband metasurface, combining gap geometry with a reflecting layer, offers a promising solution for Raman spectroscopy applications. The optimized metasurface unit cell has a period of 300 nm, a nanodisk radius of 39 nm, a gap between the nanodisks of 3 nm, and nanodisk, spacer, and gold film thicknesses of 28, 300, and 270 nm, respectively. Due to the broadband optical response, the metasurface is able to simultaneously enhance Raman signals for various molecular species without the need for tunability and holds the potential for an unprecedented increase in signal intensity.

Our study demonstrated excellent agreement between optical responses obtained through numerical simulations and within transmission line theory. Our estimates of the SECARS enhancement factor, based on realistic experimental scenarios, predict significant signal enhancement beyond the single-molecule detection threshold, even in relatively large spatial volumes compared to other plasmonic nanostructures.

Furthermore, the device’s broadband operation makes it compatible with various types of molecules, providing opportunities for molecular detection and identification applications at low concentrations, even at the single-molecule level. Our work provides a promising solution for advancing the study of molecular interactions at the nanoscale, with significant implications for the field of nonlinear optics.

In the supplementary material file, we provide an overview of previous literature on various realizations of SECARS. More results for metasurfaces with varied geometry parameters are provided. Additionally, we have included the effects of mirror metal film in the optical responses of the metasurfaces. Transmission line theory is introduced and the fit parameters that reproduce the numerically obtained data are collected. We discuss in greater detail the electric field distribution in the metasurface unit cell. Finally, we delve into the expanding applications of broadband metasurfaces.

The authors are grateful for the financial support of the National Science Center, Poland, Grant No. 2018/31/D/ST3/01487.

The authors have no conflicts to disclose.

Saeid Izadshenas Jahromi: Conceptualization (equal); Formal analysis (equal); Investigation (equal); Methodology(equal); Software (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Karolina Słowik: Conceptualization (equal); Formal analysis (equal); Project administration (equal); Resources (equal); Supervision (equal); Validation (equal); Writing – original draft (equal); Writing – review & editing (equal).

The data that support the findings of this study are available from the corresponding author upon reasonable request.

The metasurface was modeled using the finite-element method using the COMSOL Multiphysics simulation software. The dispersive properties of gold were modeled using the Johnson and Christie data,⁵⁰ while the spacer layer was modeled as a dielectric with a refractive index of n = 1.45.

For the plane-wave illumination scenario, we exploit the periodic boundary condition along the x and y directions. Two ports were placed in the z direction to calculate the absorption, reflectance, and electric field distribution.

The transmission line theory approach to model the optical response of the proposed metasurface is described in detail in the supplementary material, Sec. IV. The numerical fit to the obtained analytical expressions has been performed using the MATLAB built-in fit function. Custom codes are available upon request.