Self-referencing surface plasmon sensor for resolution enhancement

Abstract

In this paper, a self-referencing evanescent field sensor based on surface plasmon resonances is designed and fabricated. The sensor is based on sub-wavelength two-dimensional gold gratings and is optimized to detect changes in the surrounding refractive index for a water-like material. The sensor has a dedicated mode for self-referencing, which is isolated from the surrounding environment and can be used to correct errors due to temperature variations. To understand the important design parameters and optimize the sensor for best performance, many variations were fabricated and measured experimentally. Using a localized surface plasmon resonance dominant mode, a high sensitivity of 435 nm/RIU was achieved experimentally, while the self-referencing mode was successfully isolated from the surrounding environment within a refractive index range of 1.34 to 1.39. Further, we show that by incorporating the self-referencing mode into the sensitivity measurements, the resolution of the sensor can be improved by a factor of 3.6. This approach can be employed effectively for resolution enhancement of the plasmonic sensors in the presence of environmental variations.

Introduction

Metallic nanostructures exploiting surface plasmon resonances have attracted a great deal of attention in the past decade^1–3. Plasmonics is now a part of nanophotonics, which enables numerous applications, including sensing⁴, disease diagnosis⁵, waveguiding⁶, and active photonic devices⁷. In particular, lots of research has been done on plasmonic sensing, and nowadays, plasmonic sensors are offering a robust and label-free sensing platform and are being employed in many fields such as food safety⁸, medicine⁹, homeland security¹⁰ and more¹¹.

The working principle of plasmonic sensors is based on manipulating the surface plasmon’s propagation constant^12–14. Any change that happens to the surrounding refractive index of the sensor or any surface attachment manipulates the propagation constant of the surface plasmons and thus changes the resonance conditions of the plasmonic resonances, e.g. resonance wavelength, intensity, coupling angle, etc. One can monitor these changes and measure the difference in the surrounding refractive indices^12–14. There are two types of plasmonic sensors: surface plasmon resonance (SPR) sensors and localized surface plasmon resonance (LSPR) sensors. In SPR sensors, the surface plasmon-polaritons are propagating at the interface of a metallic thin-film and a dielectric layer. In contrast, in the LSPR sensors, surface plasmon-polaritons are localized in sub-wavelength nanostructures^12–14. In an SPR sensor, a prism or grating must be used to provide the required momentum for exciting the surface plasmons. However, the nanostructure of the LSPR sensors support the required momentum for the excitation of the surface plasmons, and thus, surface plasmons can get excited with direct excitation from the source without the use of prisms or a grating. Further, if a nanostructure is periodic in two directions, the LSPR sensor would be polarization independent i.e. the direction of the electric field in the excitation doesn’t need to be aligned with the nanostructure. Thus, two-dimensional LSPR sensors require a simpler optical set up, compared to the SPR sensors^12–14.

Plasmonic resonances are very sensitive to temperature variations and mechanical vibrations¹⁵. Thus, high-sensitivity plasmonic sensors can face challenges while performing in an unstable environment. Further, there could be power fluctuations of the incident light in practical applications, which causes further errors in the measurements of the sensor^16,17. The accuracy of the sensor is significantly vital when it is being employed in some fields such as disease diagnosis and food safety. Accordingly, self-referenced plasmonic sensors have been the topic of many recent research works^18–21. Self-referencing in the sensing context is defined as the ability of the sensor to correct certain errors using an internal reference. In plasmonic sensors, the internal reference could be a plasmonic resonance feature other than the one being used for the sensitivity measurement. Thus, a plasmonic sensor with two resonance features could be potentially used as a self-referencing platform^18–24,. One can isolate one of the resonance features from the surrounding environment and uniquely use it for the self-referencing measurements^19,20. For instance, in reference¹⁸ a self-referencing plasmonic sensor was experimentally presented and used for error correction. The sensor was based on one dimensional gold gratings, fabricated on top of a glass substrate. Another experimental work²⁰ reported a self-referenced plasmonic sensor based on multilayered gratings, in which plasmonic resonances of the top grating layer was used for sensing and resonance modes excited in a buried grating layer was used for self-referencing. Further, in a recent experimental work²¹, an aluminum capped nanoslit array was optimized as a self-referencing plasmonic sensor. The self-referencing feature of the sensor was used to remove the variations in the bulk refractive index from the surface sensing measurements, thus only using the sensor for precise biolayer thickness measurement. In addition, two recent studies^22,23 proposed and theoretically evaluated self-referencing plasmonic sensors based on the metallic grating on silicon and metallic grating on multilayered dielectric, respectively.

In addition to being vulnerable to the temperature variations and environmental factors, the current technology of the self-referencing plasmonic sensors usually require complex geometries with extensive fabrication steps, expensive readout technology and bulky optical setups required for the plasmonic mode excitation²². Realization of easy-to-fabricate structures which are polarization independent is necessary for improving the portability of the sensors and lowering the cost of the sensing systems. Further, designing platforms that operate in visible spectrum may lower the cost of readout technology through realization of plasmonic colorimetric sensors¹⁹.

In this paper, we design and fabricate a self-referencing plasmonic sensor based on metallic nano-gratings. The structure was fabricated using standard nanofabrication techniques and needed only one lithography step. Thus, it is easy to fabricate, and due to the periodic nature of the grating layer, the sensor is polarization-independent. The structure is able to excite two plasmonic modes in the visible to near IR spectrum and can potentially be used for the correction of temperature errors. One of the modes is sensitive to the refractive index of the surrounding environment and the other mode is isolated from the surrounding environment and can be used as a self-referencing mode. The maximum sensitivity achieved for the optimized array was 435 nm/RIU. Further, we show that if the wavelength difference of the two modes is used for the sensitivity measurements, the resolution of the sensor is improved by a factor of 3.6. The paper is organized as follow; first we briefly review the design, improving the simulations that previously reported¹⁹ by including all the layers required in fabrication. Next, we explain the fabrication steps and characterize the dielectric spacer layers. Finally, we present the sensitivity results and show that the self-referencing feature can improve the resolution of the sensor.

Design and simulation

Figure 1a shows the structure of the self-referencing plasmonic sensor. The sensor is based on cubic nano-gratings, which are separated from a thin gold film using a dielectric spacer layer. The whole structure is placed on a low-cost Corning EAGLE GX Glass substrate. Depending on the dimensions and refractive index of the dielectric spacer layer, the structure can support several LSPR-dominant, SPR-dominant, Fabry–Perot, and grating modes. The design is optimized in a way that at least two Fano resonances get excited within the structure so the sensor could be used in a self-referencing manner. Further, the structure was optimized toward isolating one of the modes from the surrounding environments. For design optimization, we assumed the surrounding refractive index was close to that of water i.e., 1.33, as is often the case for biomedical applications. A detailed design analysis of the structure was previously discussed in¹⁹ using ideal conditions. The simulations here include the effect of layers needed in fabrication, like the titanium layers for adhesion, and better represent a fabricated structure. The structure was simulated using the Rigorous Coupled Wave Analysis (RCWA) method. A commercial software was used²⁵, and the results were also verified by the Finite Difference Time Domain (FDTD) method. For the RCWA simulations, 14 harmonics with periodic boundary conditions were used. Figure 1b shows the simulated reflection spectrum of the proposed sensor. We confirmed that the structure was polarization independent by changing the polarization angle with the grating lattice and observed that the reflection remains the same no matter what the polarization angle is. The refractive indices of the dielectric layer and glass substrate were set to 1.8 and 1.5, respectively. The thickness of the gold film, the dielectric layer and the nano-cubes was 40 nm each. The side length of the nano-cubes was 200 nm and pitch values between the nano-cubes was 400 nm. A 3 nm thick titanium (Ti) layer was also placed between the gold layers and substrate or dielectric spacer layer as an adhesion layer. The refractive indices of gold and Ti used for the simulations were extracted from the experimental data^26,27. As seen in Fig. 1b, two modes are excited in the reflection spectrum of the sensor, which are labeled as modes 1 and 2 in the figure. The magnetic field distributions of the mode 1 and 2 at their resonance wavelengths are also shown in Fig. 1c,d. As shown in Fig. 1c, mode 1 is a Fano resonance of the LSPR mode of the top grating layer and SPR modes of the underneath gold film. Due to significant field being localized on the top surface of the grating, mode 1 is sensitive to the changes of the surrounding refractive index and can be used for sensing the detectant. However, the field distribution of the mode 2 in Fig. 1d shows very little field in the surrounding medium. Mode 2 is the result of a Fano resonance of strong SPR modes of the gold film and substrate, along with some weak coupling of SPR and LSPR at other interfaces. The grating does play a role in exciting the mode but very weakly. In the design optimization, not only the refractive index but also the thickness of the spacer layer plays an important role in these weak grating interactions. The dependency of mode 2 to the SPR modes in the substrate is very strong and thus mode 2 is more or less isolated from the surrounding environment and can be used as a self-referencing resonance.

Fig. 1.

(a) The structure of proposed self-referencing plasmonic sensor, (b) reflection spectrum of the optimized structure, (c) magnetic field distribution of mode 1, and (d) magnetic field distribution of mode 2.

The refractive index of dielectric spacer layer was a critical factor in our design, as the isolation of mode 2 from the surrounding medium only happens for a limited range of dielectric refractive index (1.8 ± 0.05) when the surrounding material is water. The self-referencing mode is being excited through the same top grating in the weakly coupled grating way and the spacer layer is required to isolate this mode from the surrounding medium. Thus, a limitation of our design is that it becomes application specific. The refractive index of the spacer layer changes monotonically with the refractive index of the surrounding medium. For example, simulations revealed that when the surrounding material was changed to air (n = 1), a dielectric spacer layer with a refractive index of 1.5 was required to generate two distinct resonance features. Therefore, if the sensor is to be used as a gas sensor, the dielectric spacer layer must be SiO₂ (n = 1.5). Further, the refractive index of the dielectric layer was set to a constant value for the simulations and did not include the chromatic dispersion within the layers. Silicon-oxy-nitride (SiON) and silicon nitride (SiN) have similar refractive index values in visible spectrum. However, in reality, the value of the refractive index depends on the deposition recipe of the dielectric material and changes with the wavelengths, due to chromatic dispersion and the experimental results could be different. Further, in the simulations, a plane wave excitation was used due to the computational ease of using periodic boundary conditions. However, during the characterization, the spot size will be limited to a few tens of microns. The limitation of the spot size affects the strength and the quality factor (Q) of the resonance modes. To summarize, several factors including the chromatic dispersion of the dielectric layer, size variations during the fabrications, smaller beam spot size and surface roughness are going to change the plasmonic resonance conditions such as location and strength of the modes and more importantly the contribution of the LSPR and SPR modes in formation of the Fano resonances. The contribution of the LSPR and SPR modes in the construction of Fano resonances significantly affects the sensitivity of the modes to the changes in the surrounding refractive index. For instance, in our previous work¹⁹, we theoretically showed that the highest sensitivity is achieved for a cube side length of 200 nm and pitch of 400 nm. However, since the simulations have many limitations, the optimal design could be different experimentally. Thus, an experiment was designed to fabricate many variations of the sensor and determine the best design through the design of the experiment.

Fabrication

The structure was fabricated on a Corning EAGLE GX glass substrate. A 3 nm thick Ti layer with a deposition rate of 0.5 Å per second, and subsequently a 40 nm thick gold layer with a deposition rate of 1 Å were deposited on the sample in an Intlvac electron beam deposition chamber. Next, Plasma Enhanced Chemical Vapor Deposition (PECVD) was used to deposit a dielectric layer on top of the gold film using Oxford PlasmaLab System 100. As mentioned, the refractive index of the dielectric spacer layer was a critical parameter for separation of the modes and isolating the self-referencing mode. To experimentally understand the effect of the refractive index to isolate the SPR-dominant mode, 40 nm thick layers of SiON and SiN were deposited on two different samples. Figure 2a,b show the refractive indices of the deposited SiON and SiN versus wavelength, extracted using a Woollam M-2000 ellipsometer, respectively. As seen in Fig. 2, in the range of 400–1000 nm wavelength, the refractive index of SiON changes between 1.83 and 1.74. Similarly, the refractive index of SiN ranges from 1.89 to 1.83, which is close to the value we used in simulations and also higher than that of SiON. Thus, using two different spacer layers on two different samples allows us to study experimentally the importance of the refractive index of the spacer layer and its effect on the resonance excitations.

Fig. 2.

The refractive indices of the deposited (a) SiON and (b) using PECVD extracted using an ellipsometer.

The samples were spin-coated with 950 K poly (methyl methacrylate) (PMMA) A4 resist at a speed of 4000 rpm. Subsequently, the samples were baked on a hot plate for 3 min at 180 °C. To avoid the charge build-up problem, PMMA Electra92 was spin-coated at a speed of 6000 rpm on top of the photoresist. Electra92 is a conductive protective coating that is used for dissipation of e-beam charges on insulating substrates and can easily be removed after lithography by deionized water (DI) water. Electron Beam Lithography (EBL) was done at 20 kV using RAITH150^two lithography system. Periodically arranged patterns were designed with a 100 µm × 100 µm dimension in an individual array. Also, arrays were separated by 100 µm distance to make sure each array can be measured independently during the characterization. Five different values of 400, 450, 500, 550, and 600 nm were chosen for the pitch. Further, five different values of 170, 180, 200, 220, and 250 nm were chosen for the side length of the cubes. An e-beam dose of 280 µC/cm² was chosen to pattern the nano-cube arrays. As the pitch was increased to 550 nm and 600 nm, the e-beam dose needed to be enhanced to 300 µC/cm², in order to form highly ordered nano-cubes. Due to the lack of the sharp corners, the sensitivity of the sensor is expected to be lower as compared to simulations; however, the Fano resonance are still expected to form at the wavelengths close to the values given by simulation. For smaller pitch sizes, there is a proximity effect where exposure in one nano-cube affects the exposure of the neighboring ones. Thus, a lower dose is required. This effect becomes smaller as the pitch is increased. The same e-beam dose values were used for both samples with SiON and SiN spacer layers. The patterned samples were soaked in DI water for 2 min to remove the Electra92 PMMA and dried with nitrogen. The samples were developed in MIBK and isopropyl alcohol (IPA) solution in 1:3 proportions respectively for 30 s, and then IPA as a stopper for 30 s and dried with nitrogen. Next, the samples were installed in the Intlvac e-beam deposition chamber. Again, a 3 nm thick Ti layer was deposited as the adhesion layer, along with a 40 nm thick gold layer. Finally, lift-off was done by soaking the samples in a PG remover bath overnight. Figure 3a,b shows the scanning electron microscopic (SEM) of two of the representative samples with the SiON spacer layer, 400 nm pitch and side lengths of 220 nm and 250 nm, respectively.

Fig. 3.

SEM images of two representative cubic arrays with 400 nm pitch and (a) 220 nm side length and (b) 250 nm side length. The scale bars are 1 micron.

Figure 4 shows a microscopic image of the cubic arrays taken by an optical microscope. The top left array is for the nano-cubes with 170 nm side length and 400 nm pitch. As we go to the right side, the pitch values increase to 450, 500, 550, and 600 nm. As we go to the bottom, the side lengths of the nano-cubes change to 180, 200, 220, and 250 nm. As seen in Fig. 4, each array reflects a different structural color, showing that the reflection spectra are being manipulated due to the excitation of different plasmonic modes within each array.

Fig. 4.

A microscopic image of the cubic arrays taken by an optical microscope.

Measurement

The measurements of the reflection spectra were done using a benchtop F40-UV thin film confocal measurement system. F40-UV excites a sample with a broadband light source spanning ultra-violet, visible and near-infrared wavelengths (190–1100 nm) and measures the reflection to determine the thickness of thin films. We used this capability to measure the reflection spectra and see how it changes as different refractive index fluids are used for the surrounding medium. A 15 $\times$ microscope objective was used to create a beam spot of approximately 16 microns. This value was chosen through experimentation. At this low magnification, the presence of longitudinal fields in the focused beams is still low, and the beam can be approximated as linearly polarized. This should allow experimental results to match simulations where plane wave approximation is used.

To evaluate the sensing performance of the arrays, a series of Cargille index liquids²⁸ were used, each having a known refractive index. The refractive indices ranged from 1.3 to 1.39, with increments of 0.01. This range was selected because the refractive index of water falls within it, as do the buffer liquids used for DNA hybridization. Additionally, the refractive indices of many common solvents, including ethanol, methanol, isopropanol, and acetone, also lie within this range. In addition, Cargille fluids are oil-based liquids which are very stable, with no polarity and thus, the constituents do not electromigrate near gold surfaces. As such, the experiments exhibit high measurement repeatability when compared to conventional water-based solutions used for sensor characterization like sodium chloride or glucose.

Characterization

Effect of dielectric spacer layer

The samples were tested using F40-UV while the surrounding refractive index was 1.33. Figure 5 shows the reflection spectra measured from nano-cube arrays with 250 nm side length and 400 nm pitch, with the two different dielectric spacer layers. Two plasmonic resonance features are excited for both cubic arrays with different spacer layers. The resonance features of the cubic array with the SiN spacer layer are slightly red-shifted compared to the cubic array with the SiON spacer layer. This is due to the higher refractive index of the SiN spacer layer and follows the trend we had observed in simulations. For both arrays, mode 1 and mode 2 are blue shifted compared to the simulations for the same structure. We believe, this is mainly due to the chromatic dispersion of the dielectric spacer layer which was not considered in the simulations. Further, the Q of mode 1 is close to that of simulations. However, the Q of mode 2 is lower than what was predicted. The main reason is the limited spot size of the beam in the F40-UV, which reduces the contribution of grating/lattice mode in mode 2 more than in mode 1. This is due to the weak grating coupling in mode 2, thus reducing the Q. Other reasons include surface roughness and variations in thickness and refractive indices of different layers. We observe that both modes are excited for both the samples with a red-shift in wavelength with the higher refractive index spacer dielectric layer.

Fig. 5.

Reflection spectra measured from nano-cube arrays with two different dielectric spacer layers of SiON and SiN. Cube side length is 250 nm pitch is 400 nm.

Effect of cube side length

Figure 6 shows the reflection spectra for the nano-cube arrays with SiON spacer layer and different side lengths. The pitch is 400 nm, and surrounding refractive index was 1.33. As seen in Fig. 6, by increasing the cube side length, mode 1 gets stronger, and its resonance wavelengths shifts to longer wavelengths. The mode is getting stronger because of increased near-field coupling and increased surface area of the cubes. However, mode 2 is qualitatively stronger for 200 nm side length as compared to the 250 nm side length. The SPR dominant mode (mode 2) is a Fano resonance, and the constituent resonances seems to be getting separated in wavelength as the side length increases to 250 nm and thus the Q decreases. The two modes re well defined for the side lengths in the range from 200 to 250 nm.

Fig. 6.

Reflection spectra for the sample with SiON spacer layer for cubic arrays with 400 nm pitch and different side lengths.

Effect of pitch

Figure 7 shows the reflection spectra for different pitch values for the nano-cubic array with 220 nm side length and SiON spacer layer. While different side lengths were fabricated and tested, we are showing representative results for 220 nm side length as these arrays were showing higher Q for the modes excited across different pitch values. The reflection spectrum for the underneath layers, off the arrays is also shown. As shown in Fig. 7a, for a 400 nm pitch, two distinct modes (Mode 1 and Mode 2) are excited. When the pitch increases to 500 nm, four modes appear in the reflection spectrum at 642, 726, 800, and 868 nm, respectively. These modes are weaker than Modes 1 and 2 and result from the splitting of modes within the Fano resonance. To understand the origin of each mode for the 500 nm pitch, the magnetic field distributions along the y-direction are shown in Fig. 7b–e. The mode at 642 nm, shown in Fig. 7b, arises from the coupling of a Fabry–Perot mode between the gold layers and weak LSPR and SPR modes of the grating and gold thin film. The mode at 726 nm, illustrated in Fig. 7c, results from the coupling of the LSPR modes of the top grating layer with the SPR modes of the thin gold film, where the LSPR is stronger. Additionally, SPR modes at the gold–dielectric interface are stronger than those at the gold–substrate interface. In Fig. 7d, the mode at 800 nm is due to stronger coupling between the LSPR of the top layer and the SPR modes of the gold film compared to the mode at 726 nm. Finally, the mode at 868 nm, shown in Fig. 7e, is also a result of coupling between the LSPR and SPR modes. For this mode, the SPR modes at the gold–dielectric and gold–substrate interfaces dominate over the LSPR of the top grating layer. For a pitch value of 600 nm, a single mode gets excited between the 600 nm and 700 nm wavelengths which is very similar to the mode in the off-the-pad reflection spectra. The rest of the excited mode are very weak. This shows that the near field coupling is negligible for these pitches, and the arrays are allowing the light to transmit without absorption for shorter wavelength. For the purpose of sensitivity, we are mainly interested in those arrays that can produce Fano resonances with higher Q. Thus, a pitch below 500 nm is needed.

Fig. 7.

(a) Reflection spectra measured from the nano-cubic array with the SiON spacer layer and 220 nm side length with different pitch, (b–e) magnetic field distribution in y-direction for nano-cubes with 500 nm pitch at resonance wavelengths of 642, 726, 800 and 868 nm, respectively.

Bulk sensitivity measurements

Nano-cubic arrays were tested for bulk sensitivity by changing the surrounding refractive index. Cargille refractive index fluids were used with different values of the refractive indices ranging from 1.33 to 1.39. The reflection spectrum of each array was measured using the F40-UV. After each test, samples were cleaned with IPA and Acetone multiple times, and dried with nitrogen, and then the next oil was applied. For quantitatively measuring the sensitivity of the sensors, we only consider those arrays with high Q resonance features and strong Fano resonances, i.e. cubic arrays with 200, 220, and 250 nm side lengths and pitch values equal to or smaller than 500 nm. As the experimental reflection spectra acquired from the F40-UV were noisy, there is an uncertainty in measuring the resonance peaks. To overcome this problem, we first used Savitzky-Golay algorithm²⁹ to smooth the data. Savitzky-Golay algorithm fits a polynomial to a small window of neighboring data points and then estimates the smoothed value at the center point of the window. In this way, high-frequency noise is removed while the essential features of the spectra are preserved. While it has not been used in plasmonics characterization before to our knowledge, it has been successfully applied to spectroscopy data for machine learning, especially in food analysis³⁰.

Figure 8a–c shows an example of how we calculated the bulk sensitivity. As seen in Fig. 8a, first the reflection spectra for different surrounding refractive indices are collected from F40-UV. Only two surrounding refractive indices are shown in this example for clarity. In the actual measurements, the samples were tested for several surrounding refractive indices and reflection spectra were collected. The smoothed curve is also shown in the figure. Mode 1 and 2 are labeled in Fig. 8a, along with a Fabry–Perot (FP) mode which is coupled with LSPR modes. Another point is labeled on the reflection spectra in Fig. 8a as a local maximum. This local maximum is created between an FP mode and mode 1 and it has a very similar quality factor to mode 1. Mode 1 is a Fano resonance between the LSPR of the grating and the reflections from the bottom gold layer. The blue side of the spectra is dominated by the LSPR feature and the maxima also moves with increasing refractive index. Thus, this feature can also be used for sensing applications. During the experiment, it was observed that for some designs, the sensitivity of the local maximum to the changes in the surrounding refractive index is higher than that of mode 1. Thus, the resonance wavelength of the local maximum was also used to measure sensor’s sensitivity.

Fig. 8.

An example of bulk sensitivity calculations, (a) reflection spectra collected using F40-UV and corresponding smoother curves using Savitzky-Golay algorithm, (b) first derivate of the reflection spectra, and (c) resonance wavelengths of the modes versus refractive index.

After smoothing the data, we calculated the first derivative of the reflection spectra for each array with different surrounding indices. Any local minimum and maximum in the reflection spectra will appear in the first derivative as zero. In this way, we can find the wavelengths of the features and calculate the sensitivity accurately. Figure 8b shows the first derivative of the reflection spectra, corresponding to the Fig. 8a. Finally, the sensitivity is measured by fitting a line to the resonance wavelengths of each mode versus the surrounding refractive index, as shown in Fig. 8c.

In total, 50 different designs were fabricated, and only those arrays with high Q resonance features were tested for the bulk sensitivity. Here, we show the sensitivity calculations for some representative arrays, and then we summarize of the results. Figure 9a–c shows the changes in the resonance wavelengths of the modes versus surrounding refractive index for three cubic arrays with 200, 220, and 250 nm side lengths, respectively. The pitch is 400 nm, and the dielectric spacer layer is SiN. For the cubic array with 200 nm side length in Fig. 9a, sensitivity of mode 1 is 169 nm/RIU. However, the local maximum has a higher sensitivity of 293 nm/RIU. Further, the resonance wavelength of mode 2 seems to be completely isolated from the surroundings when the refractive index is 1.35 and higher. In simulations, we observed that mode 2 was isolated from the surrounding medium for refractive indices higher than 1.33. However, due to the chromatic dispersion of the dielectric spacer layer, mode 2 starts to get isolated from the surrounding after refractive index is higher than 1.35 for this array. For the 220 nm side length in Fig. 9b, the sensitivities of mode 1 and the local maximum are 108 and 317 nm/RIU, respectively. Increasing the cubes’ side lengths to 220 nm, reduced the sensitivity of the mode 1 and enhanced the sensitivity of the local maximum. This suggests that the contribution of SPR resonances of the gold film in formation of Fano resonance of mode 1 is increased, compared to that one of 200 nm side length. Also, mode 2 is again almost isolated from the top for surrounding refractive index of 1.35 and higher. For the cubic array with 250 nm side length in Fig. 9c, the sensitivity of the mode 1 is almost zero which suggests that mode 1 is now an SPR dominant mode and is almost isolated from the surrounding environment. However, the sensitivity of the local maximum (which is dominated by the LSPR feature) is further increased compared to the smaller size arrays and is 422 nm/RIU. This is slightly lower than the sensitivity value we achieved in simulations for the cubic array with 200 nm side length and 400 nm pitch. Mode 2 is only stable after a refractive index of 1.36, though.

Fig. 9.

Changes in the resonance wavelengths of the modes versus surrounding refractive index for three cubic arrays with (a) 200, (b) 220 and (c) 250 nm side lengths, respectively. The pitch is 400 nm and dielectric spacer layer is SiN.

For the sensor arrays with 500 nm pitch and 250 nm side length, the quality factors of the resonance modes were still high and thus the arrays were tested for the bulk sensitivity. However, three resonance modes were excited within these arrays and thus the bulk sensitivity was measured for the resonance wavelengths of three modes along with two local maximums. As an example, Fig. 10a–c shows the reflection spectra with different surrounding indices, first derivatives of the reflection spectra and changes in the resonance wavelengths of the modes, respectively. The dielectric spacer layer is SiN and pitch is 500 nm with cube side length of 250 nm. As seen in Fig. 10a, three major modes are excited within the array. Mode 1 is attributed to an FP-LSPR mode. Mode 2 is an LSPR dominant mode (similar to mode 1 of other arrays), and mode 3 is an SPR dominant mode (similar to mode 2 of other arrays). Further, there are three maximums happening in the reflection curve, on the left side of mode 1 (local maximum 1), one between mode 1 and 2 (local maximum 2), and one between mode 2 and 3. We will only use the first two maximums for the sensitivity measurement as the third maximum between mode 2 and 3 is very similar to mode 3 and it is not very sensitive. These modes and local maximums are labelled in Fig. 10b. It is clear from the first derivative curves that local maximums are moving faster than the modes when the surrounding refractive index changes. The changes in the resonance wavelengths of the modes and local maximums versus the surrounding refractive index are plotted in Fig. 10c. Mode 1 and 2 show 135 and 219 nm/RIU sensitivity, respectively. Further, Local maximum 1 and 2 show 435 and 327 nm/RIU sensitivity, respectively. Mode 3 as a self-referencing mode is fully isolated i.e. its wavelength doesn’t change with changing refractive index, from the surrounding environment for a range of 1.34–1.39.

Fig. 10.

(a–c) Reflection spectra with different surrounding indices, first derivatives of the reflection spectra and changes in the resonance wavelengths of the modes, for the cubic arrays with 250 nm side length and 500 nm pitch, respectively. The spacer layer is SiN.

To compare the sensitivity results of the samples, a summary of the results is presented in Fig. 11a,b for the samples with SiON dielectric spacer layer and SiN spacer layer, respectively. The maximum sensitivity achieved from arrays are plotted versus the cube side length for different values of pitch. For the sample with SiON spacer layer in Fig. 11a, the maximum sensitivity is achieved for the array with 500 nm pitch and 250 nm side length. For both 400 nm and 450 nm pitch values, as the side length increases, the sensitivity also increases. However, along with the sensitivity, the quality factor of the resonance features is important and must be considered when choosing an array. Also, when the side length is constant, arrays with 400 nm pitch show higher sensitivity compared to the arrays with 450 nm pitch. This also confirms that for 400 nm and 450 nm pitch values, as the distance between two cubes is decreased, the sensitivity is improved. As the pitch increases to 500 nm, the array with 250 nm side length shows higher sensitivity compared to the arrays with similar side length and lower pitch values. This is due to the addition of a strong FP-LSPR mode to the system which increased the sensitivity. For the sample with SiN spacer layer in Fig. 11b, similar to the SiON sample, for a constant pitch value, the sensitivity increases by enhancing the cube side length. Also, arrays with 400 nm pitch show higher sensitivity compared to arrays with similar size and 450 nm pitch. The highest sensitivity is achieved in the cubic arrays with 500 nm pitch and 250 nm side length.

Fig. 11.

The sensitivity versus cube side length for, (a) the sample with the SiON spacer layer and (b) the sample with SiN spacer layer.

Comparing the performance of the cubic arrays in SiON and SiN samples, the highest sensitivity in both samples is achieved in the array with SiN spacer layer, 250 nm side length and 500 nm pitch. The only cubic array in the sample with SiON spacer layer which is showing higher sensitivity compared to a similar array with SiN spacer layer, is the array with 450 nm pitch and 250 nm side length. For this combination, the array with the SiON has a sensitivity of 343 nm/RIU bulk sensitivity, while the SiN sample has 308 nm/RIU sensitivity to the changes in the surrounding refractive index. It was also observed that the variation in the sensitivity with changing side length was lower with the SiN spacer layer as compared to the SiON layer. Based on these results, due to the higher refractive index provided by the SiN spacer layer, we can claim that SiN spacer layer is performing better in case of mode separation and also sensitivity.

Comparison with other experimental works

Table 1 summarizes the highest sensitivity values achieved for the different designs. The highest sensitivity achieved was 435 nm/RIU sensitivity for 250 nm cube length and 500 nm pitch with SiN as the dielectric spacer layer. Values of Q in the sensing peak, figure of merit (FOM) defined as the sensitivity/full-width half maximum of the sensing peak³¹, and Q of the self-referencing peak are also summarized in the table. Mode 2 (the self-referencing peak) was stable for the refractive index values ranging from 1.34 to 1.39 for this sensor. There are few theoretical works^22,23 related to the self-referenced plasmonic sensors which have been proposed contemporary to this work, however, we will only consider the experimental works for comparison. Table 2 provides a comparison between the current study and previously reported experimental works. A self-referenced plasmonic sensor based on one dimensional rectangular gold gratings was reported in¹⁸. The sensitivity of this sensor was 470 nm/RIU, which is slightly higher than the value we achieved for our design. It also has a higher Q and FOM of the sensing mode compared to our device. However, the self-referencing mode in this reported sensor was not very well-defined mode. More importantly the sensor is polarization dependent and required careful alignment of the electric field of the incident field. Another self-referenced plasmonic sensor was fabricated in reference²⁰ using multilayered cubic gratings. The structure consists of two cubic gratings layers which are separated by a nitride layer, in which the top grating layer is used for sensitivity and the underneath grating layer is used for self-referencing measurement. The structure has several extra nanofabrication steps which require careful alignments compared to our structure. The demonstrated sensitivity of this sensor was only 122 nm/RIU and is significantly lower than what we achieved with a much simpler structure to fabricate. Further, the FOM is also lower by a factor of 4. In reference²¹, an aluminum capped nanoslit array was fabricated and used as a self-referencing plasmonic sensor for biolayer thickness measurement. The bulk sensitivity of the sensor was not reported; however, the structure is polarization dependent. Our proposed structure achieves comparable sensitivity while having a well-defined self-referencing mode and being polarization sensitive. Of the two structures which are polarization independent, our proposed structure has sensitivity larger by a factor of 3.5.

Table 1.

Summary of the highest sensitivity values achieved for different designs.

Spacer layer	Pitch (nm)	Cube length (nm)	Sensitivity (nm/RIU)	Q	FOM (RIU−1)	Mode 2 stability range
SiON	400	250	367	4	2.4	1.36–1.39
SiON	500	250	395	13.5	7.3	1.34–1.39
SiN	400	250	422	4.5	2.8	1.35–1.39
SiN	500	250	435	8	5.5	1.34–1.39

Table 2.

Comparison of this work with other experimental self-referencing plasmonic sensors.

References	Structure	Ease of fabrication	Polarization independence	Bulk sensitivity (nm/RIU)	Q of sensing mode	FOM (RIU−1)	Q of Self-referencing Mode	Complete isolation of self-referencing mode
18	1D rectangular grating	Single lithography but needs high aspect ratio over long lengths	No	470	47	31	Not measured in reference, the mode is not well defined	Yes
20	Multilayered cubic gratings	Two lithography steps with critical alignment	Yes	122	Not measured in reference	1.3	Not measured, but a well defined mode	Yes
21	Aluminum capped nanoslit array	1-step Nanoimprint lithography	No	Not reported	Not measured but is low	Not measured	Not measured in reference, but has a well defined peak	Yes
This work	2D Cubic grading	Single lithography and liftoff	Yes	435	8	5.5	14	Yes

Advantage of self-referencing

An experiment was designed to show one of the advantages of the self-referencing structure proposed in this work. A cubic array with 250 nm cube side length and 400 nm pitch, from the sample with the SiN spacer layer was tested over multiple days over a month using F40-UV and the reflection spectra were collected. The surrounding refractive index was set at 1.35. Each day the temperature of the environment was slightly different. Further, initial settings of F40-UV such as focusing and baseline measurements were different for each day, as they were done by a human. Due to these changes in the measurement conditions, the reflection spectrum for each day differs and the resonance wavelengths of LSPR and SPR modes were shifted by a few nanometers. The reflection spectra which show the largest variation across the month are shown in Fig. 12. Table 3 summarizes the extracted resonance wavelengths of SPR, LSPR and local maximum for day 1 and 2. Further, the difference between the resonance wavelengths of the local maximums and the SPR modes, Δλ, is shown for both days. As can be seen in this table, the resonance wavelengths are slightly different for each day.

Fig. 12.

Reflection spectra of the cubic array with 250 nm side length and 400 nm pitch from the sample with SiN spacer layer for two different days.

Table 3.

Resonance wavelengths of SPR, LSPR and local maximum for day 1 and 2.

	λLocal_Max (nm)	λLSPR (nm)	λSPR (nm)	Δλ = (λSPR − λLocal_Max)
Day 1	675.85	757.44	859.16	183.31
Day 2	676.91	758.14	859.84	182.93

The sensitivity of the local maximum to the changes of the surrounding refractive index is 422 nm/RIU for this array. If we use the local maximum for the sensitivity measurements, the minimum amount of change in the surrounding refractive index which can be detected by the sensor can be calculated as follows:

$$Minimum\mathrm{\Delta }n=\frac{\left|{\lambda }_{Local\_max}\left(Day1\right)-{\lambda }_{Local\_max}\left(Day2\right)\right|}{\mathit{Sensitivity}}=\frac{\left|675.85-676.91\right|}{422}=2.5\times {10}^{-3}$$ 1

Thus, if we only use the local maximum for the sensitivity measurements, the minimum amount of change in the surrounding index needs to be $2.5\times {10}^{-3}$ so the sensor can detect it. If the change in the surrounding refractive index is below this number, it would be considered as noise and the sensor cannot detect it reliably. Instead of using the resonance wavelength of the local maximum, we can use Δλ for the sensitivity measurements. If we use Δλ, the minimum change in the surrounding refractive index detectable by the sensor, can be calculated as follows:

$$Minimum\mathrm{\Delta }n=\frac{\left|\mathrm{\Delta }\lambda \left(Day1\right)-\mathrm{\Delta }\lambda \left(Day2\right)\right|}{\mathit{Sensitivity}}=\frac{\left|183.31-182.93\right|}{422}=9\times {10}^{-4}$$ 2

Thus, the minimum detectable Δn is improved by a factor of 3.6 when we use Δλ for measuring the sensitivity. This means that self-referencing has improved the resolution of the sensor. Note that due to temperature change, there are refractive index changes not accounted for in this calculation. By measuring the sensitivity of the sensor to temperature, a matrix can be developed which can further improve the sensitivity measurements.

Conclusion

A self-referenced evanescent field sensor based on surface plasmon resonances was experimentally fabricated. The plasmonic nanostructure was based on two-dimensional gold gratings and many variations of the sensor were fabricated to optimize the sensor for the highest sensitivity. Two different dielectric spacer layers were used on two different samples to provide enough refractive index for the mode separation. The sensor has a dedicated mode for self-referencing, which was isolated from the surrounding environment in the refractive index range of 1.34–1.39 and can potentially be used for correcting temperature errors. Further, the best design showed a sensitivity of the 435 nm/RIU to the changes in the surrounding refractive index. It was also shown that by incorporating the self-referencing mode in the sensitivity measurements, the resolution of the sensor can be improved by a factor of 3.6. To our knowledge, this is the first demonstration of high sensitivity self-referencing plasmonic sensor with a well-defined self-referencing mode.

Author contributions

R.K. and S.S. contributed equally to this work.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no Competing interests.