K9 glass substrates (20 mm × 10 mm × 1.5 mm) were used to provide a good light transmissivity and a physical support for the triangular nanoprisms. To ensure hydrophilicity and cleanness, the substrates were prepared as follows: first, immersed in a piranha solution (30% H₂O₂/H₂SO₄ = 1/3) at 80 °C for 30 min; second, after cooled in N₂ and rinsed with copious amounts of double distilled water, the substrates were sonicated in 5:1:1 H₂O/NH₃·H₂O/30%H₂O₂; third, rinsed repeatedly with de-ionized water and stored in it before nanostructures fabrication process.

As a transducer of light signals and vapor concentration, silver triangular nanoprisms present a good extinction peak and a large surface area to facilitate quantitative adsorption of organic vapors. The NSL technique, one of the most popular means to fabricate large area periodic nanostructure arrays, was employed to create the triangular nanoprisms in the experiment due to its time and money cost effectiveness. The fabrication process primarily consists of three steps: first, a layer of ∼399 nm polystyrene (PS) nanosphere colloidal suspension (1.8%, 3400A, Thermo Scientific Inc., Canada) was spin coated (speed: 3,000 rpm) onto the substrate to form a large-area close-packed nanosphere monolayer pattern; second, a layer of ∼50 nm silver was deposited on the monolayer template in a thermal evaporator (400-I, C-Vac Inc., China); and third, the PS nanospheres were removed by dissolving them in ethanol with the aid of sonication for 20 s∼60 s. Following that, a well ordered two-dimensional triangular nanoprisms array was finally obtained on the substrate.

The schematic diagram of the detection system is shown in Figure 1. A white light beam radiated by a halogen lamp was propagated through a multimode optical fiber to illuminate the sensing chip. On the other side of the chip, it was collected by another multimode fiber. The LSPR spectra were measured at room temperature (20 °C) by using a Scientific-grade UV-Vis Spectrometer (QE65000, Ocean Optics Inc., USA), with associated data processing software (SpectraSuite, Ocean Optics Inc.). The spectrometer was operated at a wavelength resolution of 1 nm in our experiment, and the extinction can be evaluated by: (1) E ( λ ) = − log 10 ( S ( λ ) − D ( λ ) R ( λ ) − D ( λ ) )where S(λ), D(λ) and R(λ) denote the sample intensity, the dark intensity, and the reference intensity at the wavelength of λ, respectively. An airproof gas cell was used in this experiment to provide a platform where the test vapor was adsorbed on the transducer and the vapor concentration was transmitted into the optical signal. The gas cell was made of a specific kind of polymer and specially designed and fabricated to enhance the light transmission.

The test vapor of a defined concentration was prepared by injecting a specific volume of saturated vapor into the gas cell, where the saturated vapor was produced by stewing a certain amount of the corresponding matter of liquid phase in a big sealed container for several hours. The injection volume in experiment was defined as: (2) V i = C d C s • V rwhere C_d, C_s and V_r refer to the test vapor concentration, the saturated vapor concentration and the volume of gas cell, respectively.

According to the test results for the sensor chip exposed to ethanol vapor of different concentrations presented in Figure 3(a), the peak wavelength of the spectrum shifts from 537 nm to 547 nm gradually when the ethanol concentration varies from 0 mg/L (in air) to 105 mg/L. Comparing Figure 2(b) with Figure 3, it is apparent that the variation of peak wavelength for ethanol in gas phase is much smaller than that in liquid phase, indicating a much smaller refractive index of ethanol in gas phase than in liquid phase. This phenomenon is an interior interpretation of the difficulty of gas molecule detection for those refractive index based sensors.

In order to better resolve the overall relationship between the resonant wavelength and the ethanol vapor concentration, we performed more experiments on different vapor concentrations. The extinction spectra peak wavelength shift (0 mg/L as the base point) as a function of ethanol vapor concentration isplotted in Figure 3(b). As the concentration increases from 0 mg/L to 105 mg/L, in increment of 15 mg/L, the peak shift amount increases gradually, suggesting a linear relationship. Hence, we define “vapor sensor sensitivity” m to directly evaluate the sensing performance of metal nanostructures based volatile organic vapor sensors: (3) m = Δ λ p ( nm ) Δ C d ( mg / L )where Δλ_p and ΔC_d denote the peak wavelength shift in a certain vapor concentration relative to that in air and the corresponding concentration change. For the ethanol vapor test, the sensitivity is approximately 0.1 nm/mg L⁻¹, as shown in Figure 3(b), indicating a 1 mg/L increment in ethanol vapor concentration will induce a 0.1 nm red shift in spectrum.

The detection limit, determined by the sensor sensitivity as well as the operational resolution of the applied detector, is approximately 10 mg/L for ethanol test in our experiment based on the results in Figure 3(b). It will be improved if a spectrometer with greater wavelength resolution is used.

The reversibility, stability, and response time of the LSPR spectra were examined by repetitive intermittent exposure of the silver triangular nano-transducers to ethanol vapor. The saturated ethanol vapor, approximately 105 mg/L in room temperature, was switched on for 15 s and then switched off for 120 s by closing or opening the riser of the gas cell during the three test cycles. A response of the LSPR spectra was then observed upon the repetitive variation of the environment in the gas cell between clean air and saturated ethanol vapor-containing air.

Figure 4 shows the dynamic peak wavelength shift changes as time increases in the three test cycles. After the 105 mg/L ethanol vapor was injected into the gas cell, a considerable peak wavelength response (10 nm) can be observed. This is quantitatively consistent with the experiment results in Section 3.1. As indicated in Figure 4, the peak wavelength response was rapid, stable, and reproducible. It should be noted that the recovery time (∼66 s for the first cycle) lasts much longer than the response time (∼4 s for the first cycle). This can be attributed to the difference of mixing time for ethanol molecule: when switched on, the concentration in the tiny closed gas cell grows from 0 mg/L to saturated in a short time; when switched off, the saturated ethanol vapor diffused into the exterior environment with a concentration in gas cell growing from saturated to approximately 0 mg/L in a much longer time.

It also should be noted that the peak wavelengths after the second test cycle and the third test cycle are a little bigger than that before and after the first test cycle. This is because the concentration of ethanol vapor in the gas cell cannot return to 0 mg/L (clean air) after two test cycles. The reversibility will be further improved when the test gas cell is created of vacuum after each test cycle.

The LSPR spectrum response of silver triangular nanoprisms to other four analyte volatile organic vapors (i.e., acetone, benzene, propanol and hexane) was tested. The sensor chip applied in this section was the one that applied in ethanol vapor test in Sections 3.1 and 3.2 for comparison. Figure 5 describes the LSPR spectra measured in situ during the sensor’s exposure to the four analyte vapors of various concentrations. The peak wavelengths vary as the concentration increases for the four test vapors, indicating different increments in refractive index in the gas cell. As indicated in Figure 5(a), the saturated acetone vapor induces a red shift of the LSPR spectrum of approximately 17 nm, much larger than that for propanol, 3 nm as shown in Figure 5(d). Also, both the maximum red shift for benzene vapor and hexane vapor is 10 nm, the same with ethanol vapor shown in Figure 3(a).

In order to better resolve the selective characteristics of the nano-sensor, we further made a comparison of the peak wavelength response to the concentration of the five test vapors, as shown in Figure 6. All the five test vapors, ethanol, acetone, benzene, propanol and hexane, showed a good relative linear response, with the slope of ethanol being the greatest. This reveals selectivity for ethanol of the LSPR sensor, and indicates the physical interaction between nanostructures and vapor molecules. On the one hand, as is well known that the LSPR spectrum responds only to the ambient refractive index change, it is more sensitive with compounds of a higher refractive index than that of a lower refractive index. On the other hand, changes in refractive index induced by compounds in gas phase are relatively smaller than that in liquid phase. This indicates that vapors with poor volatility (i.e., low saturated vapor pressure), which are easy to liquefy on the surface of metal nanostructures, will exhibit a good LSPR sensitivity.

As described in Table 1, both the refractive index and saturated pressure contribute to the sensor response. Ethanol, with a relatively higher liquid refractive index and a lower saturated vapor pressure, exhibits a higher sensitivity than four other vapors. The detection limit of the ethanol is approximately 10 mg/L, much lower than that of the other four test vapors. Thus, the fabricated nano-transducer has potential applications in detection of ethanol in environmental monitoring and food safety.