Near-infrared gas spectroscopy based on plasmonic photodetector applied for multiple gas species

The proposed photodetector absorbs the corresponding incident wavelength at a specific incident angle and outputs a current. Figure 1(a) shows a schematic diagram of the photodetector used in this study. The photodetector consists of a gold diffraction grating formed on the surface of n-type silicon and an aluminum thin film on the backside. When a collimated TM wave on the incident plane of wavelength λ is incident on the gold diffraction grating, diffraction of the incident light occurs in the direction dependent on the pitch of the diffraction grating. In this case, the grating grooves are perpendicular to the incident plane of the light. When the matching condition Eq. (1) is satisfied in terms of the wave number component in the direction parallel to the device surface, the incident light couples with the surface plasmon resonance. ²⁸⁾

$$\begin{eqnarray}&&\frac{\omega }{c}\sqrt{{\varepsilon }_{{\rm{air}}}}\sin \theta +\frac{2l\pi }{a}=\frac{\omega }{c}\sqrt{\frac{{\varepsilon }_{{\rm{Au}}}{\varepsilon }_{{\rm{air}}}}{{\varepsilon }_{{\rm{Au}}}+{\varepsilon }_{{\rm{air}}}}}.\end{eqnarray} \tag{ 1 }$$

Here, ω represents the angular frequency of the incident light, θ is the incidence angle, a is the pitch width of the gold diffraction grating, l is the order of diffraction, c is the speed of light in vacuum, ε_air is the dielectric constant of air, and ε_Au is the dielectric constant of gold. When this condition is satisfied, the energy of the incident light is absorbed by the gold as surface plasmon resonance. The energy of this incident light excites the electrons in gold through the surface plasmon resonance. As a result, hot electrons with high energy are generated within the gold.

On the other hand, a Schottky barrier is formed at the interface between the gold film of the diffraction grating and the n-type silicon, as shown in Fig. 1(b). The energy supplied to the electrons allows them to overcome this Schottky barrier, converting them into a photocurrent I_ph. This photocurrent is generated due to the surface plasmon resonance, so the photocurrent shows a peak at incident angles and wavelengths that satisfy Eq. (1). By changing the incident angle θ, the resonance condition in Eq. (1) can be changed, thus changing the resonant wavelength λ. Generally, since the band gap E_g of silicon is 1.1 eV, only IR light shorter than a wavelength corresponding to that energy, 1.1 μm, can be measured. The Schottky barrier Φ_B between gold and n-type silicon shows 0.7–0.8 eV, and NIR light up to a wavelength of 1.5–1.7 μm can be detected.

The spectral reconstruction procedure using the plasmonic photodetector is detailed in other literature, ^29,30) so we briefly describe it here. The spectroscopy step consists of a two-stage process: (1) measuring the wavelength sensitivity characteristics of the photodetector and (2) reconstructing the spectrum of the incident light based on that composition data. First, the procedure starts with the calibration of the photodetector. Light with a known wavelength ${\lambda }_{i}$ and intensity $P\left({\lambda }_{i}\right)$ is input into the photodetector, and while sweeping the incident angle θ, the photocurrent $I\left({\theta }_{k},{\lambda }_{i}\right)$ at each incident angle ${\theta }_{k}\,\left(k=1,2,\cdots ,m\right)$ is measured [Fig. 1(c)]. At this time, the responsivity ${R}_{{\lambda }_{i}{\theta }_{k}}$ of the photodetector at a specific angle ${\theta }_{k}$ for the wavelength ${\lambda }_{i}$ $\left(i=1,2,\cdots ,n\right),$ that is, the amount of generated photocurrent per unit incident light, can be expressed by the following formula.

$$\begin{eqnarray}&&{R}_{{\lambda }_{i}{\theta }_{k}}=I\left({\theta }_{k},{\lambda }_{i}\right)/P\left({\lambda }_{i}\right)\end{eqnarray} \tag{ 2 }$$

As shown in Fig. 1(c), the responsivity is measured for each wavelength ${\lambda }_{i}$ within the wavelength range used for spectroscopy. This responsivity for each wavelength and incident angle forms an m × n matrix, which is used in the next step.

Spectrum reconstruction can be considered as the process of deriving the unknown light intensity vector ${\boldsymbol{P}}\,={\left({P}_{{\lambda }_{1}},\cdots ,{P}_{{\lambda }_{n}}\right)}^{t}$ of the incident light from the photocurrent vector ${\boldsymbol{I}}={\left({I}_{{\theta }_{1}},\cdots ,{I}_{{\theta }_{m}}\right)}^{t}$ measured by changing the angle of the plasmonic photodetector relative to the incident light. When multiple wavelength-monochromatic lights are irradiated at the same time, the photocurrent ${I}_{{\theta }_{k}}$ at a certain incident angle ${\theta }_{k}$ can be expressed as the linear sum of the photocurrents of each monochromatic light (depicted in Fig. 1(d) as "Without gas"), and is expressed as:

$$\begin{eqnarray}&&{I}_{{\theta }_{k}}={R}_{{\lambda }_{1}{\theta }_{k}}{P}_{{\lambda }_{1}}+{R}_{{\lambda }_{2}{\theta }_{k}}{P}_{{\lambda }_{2}}+\cdots +{R}_{{\lambda }_{n}{\theta }_{k}}{P}_{{\lambda }_{n}}\end{eqnarray} \tag{ 3 }$$

Therefore, when photocurrents are measured at multiple incident angles ${\theta }_{k},$ the following matrix equation holds:

$$\begin{eqnarray}&&\left[\begin{array}{c}{I}_{{\theta }_{1}}\\ \vdots \\ {I}_{{\theta }_{1}}\end{array}\right]=\left[\begin{array}{ccc}{R}_{{\lambda }_{1}{\theta }_{1}} & \cdots & {R}_{{\lambda }_{n}{\theta }_{1}}\\ \vdots & \ddots & \vdots \\ {R}_{{\lambda }_{1}{\theta }_{m}} & \cdots & {R}_{{\lambda }_{n}{\theta }_{m}}\end{array}\right]\left[\begin{array}{c}{P}_{{\lambda }_{1}}\\ \vdots \\ {P}_{{\lambda }_{n}}\end{array}\right]\end{eqnarray} \tag{ 4 }$$

Therefore, if the responsivity matrix is represented by R , it can be written in a simple form as I = RP . Here, the spectrum P of the incident light can be obtained from the generalized inverse matrix of the responsivity matrix R and the photocurrent vector I during the incident light irradiation using the following equation:

$$\begin{eqnarray}&&P={\left({R}^{{\rm{T}}}R\right)}^{-1}{R}^{{\rm{T}}}I.\end{eqnarray} \tag{ 5 }$$

Thus, the unknown light intensity vector P is reconstructed from the pre-constructed responsivity matrix R and the measured data, the photocurrent vector I , enabling NIR spectroscopic measurement. When light passes through a gas, the transmittance decreases due to absorption by the gas, which slightly changes the vector I as shown in Fig. 1(d) "With gas." This allows the wavelength spectra for cases with and without gas to be obtained [Fig. 1(e)]. By dividing the intensity measured in the "With gas" condition by that of the "Without gas" condition at each wavelength, the gas transmittance spectrum can be measured [Fig. 1(f)], where the absorbance of the gas will appear as the transmission drop.

Using the setup in Fig. 3(a), we measured the transmission spectra of water vapor, two different concentrations of ethanol gas, and 10% NH₃ gas. The absorption of light by gas is explained by the Beer–Lambert law. Suppose the intensity of the incident light is A₀, the intensity of the transmitted light is A, and the optical path length of the gas cell is l. In that case, the molar absorptivity is ε, and the molar concentration of the gas is c, the following equation holds, A = A₀ exp(−εcl). This allows for the measurement of changes in gas concentration through the intensity of transmitted light. During this measurement, gas was enclosed in the gas cell that had previously been opened to the atmosphere. However, to calculate the transmission rate of the gas, it is necessary to take a reference value with a gas that does not absorb NIR light. Therefore, we used nitrogen gas to measure the reference value. The light spectrum, when nitrogen gas at 1 atm is enclosed in the gas cell, is P_Reference, and the light spectrum, when the gas sample is introduced, is P_Sample. The transmission spectrum T of the gas sample at this time was calculated using the following formula.

$$\begin{eqnarray}&&T={P}_{\text{Sample}}/{P}_{\text{Reference}}.\end{eqnarray} \tag{ 6 }$$

First, nitrogen gas was enclosed in the gas cell, and six wavelengths (1352, 1392, 1422, 1442, 1472, and 1502 nm) were simultaneously incident on the plasmonic photodetector. In this study, we selected six wavelengths close to 1392 nm, where ethanol gas has an absorption peak. We obtained the light intensity P of the transmission spectrum by spectroscopy and used it as a reference.

Then, water vapor, ethanol gas of different concentrations, and 10% NH₃ gas were enclosed in the gas cell, and the light spectrum was obtained similarly. Water vapor was generated by putting pure water in the gas-collecting bottle and bubbling it with nitrogen. The experiment was conducted at a room temperature of 20 °C. Water vapor is assumed to be the saturated vapor at room temperature, which corresponds to 2% at room temperature of 20 °C. Ethanol gas was generated by putting anhydrous ethanol (99.5% ethanol, Wako, Japan) in a gas collecting bottle and bubbling it with nitrogen similarly. To generate ethanol gas of different concentrations, bubbling was performed with the same amount of pure water and anhydrous ethanol in the gas collecting bottle, or with only anhydrous ethanol in the gas collecting bottle. As a result of the measurement with a gas detection tube, we were able to obtain ethanol gas with concentrations of 2.7% and 4.5% measured with a gas detector tube [a gas detection tube: 112 (for ethanol) and gas sampling pump kit GV-100S, GASTEC, Japan)]. The NH₃ gas stored in a gas cylinder of which concentration was adjusted to be 10% was employed. Each gas was filled into the gas cell. The gas generated was introduced into the gas cell with the inlet and outlet open for 5 min, and then the inlet and outlet were sealed to confine the gas inside the cell. The pressure was 1 atm at each measurement. To obtain the precise gas transmittance and absorbance, we also measured the same gas cell introduced into the Fourier transform IR spectroscopy (FTIR) (IRTracer-100, Shimadzu, Japan).

Figure 4(a) shows the transmission spectrum of N₂ gas and the water vapor. Peaks can be seen at the six wavelengths irradiated. This transmission spectrum of N₂ gas was used as a transmission rate of 100% at the corresponding wavelength in the following, so the transmission spectrum of N₂ itself is not shown in the graph. The calculated transmission rate based on these results is shown in Fig. 4(b). Figure 4(b) also includes the FTIR measurement results for water vapor. The error bars of the photodetector's result indicated by "Device" correspond to maximum and minimum values, and the data point corresponds to the average value of the four experiments. In the case of water vapor, it can be seen that the transmission decreased by about 30% at all wavelengths compared to that of N₂ gas. Also, no specific peak was observed within the measurement range. This characteristic observed with the plasmonic photodetector is consistent with the FTIR measurement result, in which a broad decrease in the transmission by about 30% was observed. Since there is consistency between the two, it can be concluded that the proposed spectroscopy method could measure the water vapor response.

Zoom In Zoom Out Reset image size

Figures 4(c) and 4(d) show the ethanol gas measurement results with the plasmonic photodetector and the FTIR as a reference. The measurements were repeated four times, the data points represent the average of the four experiments, and the error bars indicate the maximum and minimum values. As shown in Fig. 4(c), in the case of 4.5% ethanol gas, a transmission dip was observed at λ = 1392 nm, which could be attributed to the ethanol gas light absorption. The transmission decreased by 6%–7% at λ = 1392 nm from the baseline. Comparing this measurement result with FTIR, absorption was observed at the same wavelength, and the amount of the transmission decreased and also presented consistency. When the concentration was changed to dilute 2.7% ethanol gas, a decrease in transmission was observed at the same wavelength λ = 1392 nm as the 4.5% ethanol gas. Because of the low gas concentration, the decrease in the transmission was 4%, which was also consistent with the transmittance drop in an FTIR plot from the baseline. This transmittance difference depending on the gas concentration suggests the plasmonic photodetector can be applied to quantitative measure of the gas concentrations.

Lastly, we conducted measurements on NH₃ gas. Figure 4(e) shows the transmission characteristics of NH₃ gas at a concentration of 10%. It should be noted that the data is preliminary and limited with measurements at only 3 wavelengths. The data from the FTIR is also presented for comparison. Measurements were taken 10 times by irradiating at wavelengths of 1494, 1514, and 1534 nm, and the room temperature at the time of measurement was 18 °C. The error bars indicate the maximum and minimum values. The spectrum shape is more complex than that of the ethanol gas due to many small absorption peaks, the FTIR showed a transmittance drop band around 1514 nm and was ∼3% from the baseline. In the case of the measurement with the plasmonic photodetector, the transmittance drop at the central wavelength compared with the other two wavelength transmittances was ∼2%–3%, which was also coherent with the FTIR analysis. Due to the limited measured wavelengths, a full spectrum evaluation remains a subject for future study.

Based on the above experiments, we have demonstrated the possibility of gas sensing with the plasmonic photodetector for gases that show a gentle absorption spectrum like water vapor, those that present an absorption dip near specific wavelengths like ethanol gas, and those like NH₃ that exhibit absorption at longer NIR wavelengths. Although the range of gases measured is limited, we have shown that gas sensing using the proposed photodetector can be expanded to a variety of gases. However, at this point, there are challenges to improving the accuracy of the measurement method.

Firstly, there is a problem with the interpretation of the water vapor spectrum. Based on the data from the HITRAN database, for water vapor, an absorption peak should be observed in the 1350–1400 nm range, but a broad absorption trend was evident in this case. Regarding this, it seems reasonable to interpret that the spectrum obtained this time is due to the fusion of absorption by water vapor and the gentle absorption band around the longer wavelength of 1500 nm by condensed liquid water resulting from the saturated water vapor. Therefore, in this measurement, there is a possibility that the absorption of pure water vapor in the gas phase has not been accurately measured.

Moreover, compared to the absorption wavelength band of ethanol, that of NH₃ is relatively situated on the longer wavelength side. As a result, the energy of the incoming photons is reduced, causing a decrease in the signal-to-noise ratio, which affects the accuracy of measurements. The current photodetectors have not been structurally optimized yet. By enhancing the structure of plasmonic photodetectors, for example, an appropriate choice of a Schottky and plasmonic metal, both the quantum efficiency and signal-to-noise ratio can be improved. This will enable more precise measurements of gases like NH₃, which have absorption bands in the NIR region of longer wavelengths.

In addition, spectroscopy was conducted using a special light source capable of irradiating monochromatic light this time. It is also desirable to replace the light source with a more affordable and easily accessible broadband light source. However, to make use of such a light source, it is necessary to improve the sensitivity of the photodetector. This device exhibits limited sensitivity, which limits its application in spectrometry using white light as a light source. Nonetheless, progress has been made in spectrometry by employing commercially available halogen lamps as light sources, as reported in Ref. 33. The light source used in this case had a power density of 8 μW nm⁻¹ at wavelengths around 1300 nm (SL S201L, Thorlabs, USA). In terms of sensitivity, we can improve the responsivity by reducing the Schottky barrier height and optimizing the structure. Although the microstructures are different, previous measurements by our group presented a sensitivity of 10 mA W⁻¹ at a wavelength of 1500 nm using a Cu/n-Si Schottky barrier. ³⁴⁾ Therefore, while optimization of the device's structure is necessary, we anticipate a potential sensitivity improvement of two to three orders of magnitude in principle. This improvement can facilitate its application to the broadband light source.