Raman spectroscopy is a proven analytical technique for the detection and identification of condensed phase materials. However, the use of Raman scattering for vapor- and gas-phase analytes has proven to be more difficult due to the combination of weak Raman scattering cross sections and dilute molecular densities. Benchtop systems that utilize a long optical path length [1] or hollow-core optical fibers [2] have been demonstrated, but the development of a sensitive, handheld Raman system for trace vapors has proven challenging. Previously, we reported the measurement of Raman spectra of trace vapors using a 1 cm long sorbent-functionalized nanophotonic waveguide [3]. This technique, waveguide-enhanced Raman spectroscopy (WERS) [4], is enabled by the use of highly evanescent, low-loss waveguides with a sorbent material as a top cladding. Strong partitioning of trace vapors into the sorbent, combined with the overlap of the waveguide’s evanescent field with the sorbent cladding, enable Raman scattering from sorbed analyte molecules to be efficiently collected into the waveguide’s propagating modes. Since its initial demonstration, WERS has subsequently been used for the demonstration of waveguide surface-enhanced Raman scattering [5], the detection of biological monolayers [6], and waveguide stimulated Raman scattering [7].

In this Letter, we show the unique differential WERS spectra at parts-per-billion detection levels of four different chemical warfare simulants: dimethyl methylphosphonate (DMMP), diethyl methylphosphonate (DEMP), trimethyl phosphate (TMP), and triethyl phosphate (TEP). An enhancement over previously reported detection capabilities is enabled by the use of the quasi-TM waveguide mode and a thinner silicon nitride (SiN) core layer (110 nm), which results in significantly improved signal-to-noise ratio than the quasi-TE mode and a thicker core (175 nm). This is because the TM mode in a thin-core waveguide increases the WERS signal by increasing the modal overlap with the sorbent, and decreases the fluorescence and Raman background [8,9] from the SiN core. Additionally, we describe the molecular origins of the measured spectral features of both the analyte and the sorbent, including peaks in both that are perturbed by hydrogen bonding.

The 110 nm thick SiN core is deposited by low-pressure chemical vapor deposition onto a 5 μm thick thermal ${\mathrm{SiO}}_2$ bottom cladding. The waveguides were patterned using a fixed-beam moving stage electron beam to expose a thin-film resist. After patterning, the SiN was etched $\sim 55\mathrm{nm}$ via inductively coupled plasma reactive-ion etching to create 1.5 μm wide waveguide ribs. After the etch, 1 μm of ${\mathrm{SiO}}_2$ was deposited by plasma-enhanced chemical vapor deposition as a top cladding. This top oxide was then patterned via photolithography and etched with buffered hydrofluoric acid to form a 7.6 mm long trench, as shown schematically in Fig. 1. This trench exposes the top of the waveguide for functionalization. The top ${\mathrm{SiO}}_2$ is retained at locations where facets are cleaved to create the 9.6 mm long waveguides. A scanning electron micrograph (SEM) of the facets (after etching and metalization to enhance image contrast) is shown in the top of Fig. 1.

 

Fig. 1. Top, an SEM image of a waveguide facet, shown after etching and metalization (to enhance image contrast); bottom, the waveguide sample in the measurement setup. BPF, bandpass filter; LPF, long-pass filter; LP, Linear polarizer.

Download Full Size | PDF

The waveguides are functionalized by depositing an approximately 700 nm thick layer of a hyperbranched carbosilane fluoroalcohol polymer [10], a custom-designed, transparent hypersorbent material that serves as the upper cladding of the waveguide in the trench. The polymer has hydrogen bond acidic functional groups that selectively bond with hydrogen bond basic groups of organophosphonates such as chemical warfare agents. The HCFSA2 was deposited by spin-coating with an acetone/methanol solution containing the dissolved polymer, which ensured a uniform film with a controlled thickness. The sorption/desorption times in this naturally reversible sorbent have been previously measured to be on the order of minutes [3].

The sample was mounted in a custom flow cell for chemical and optical characterization. The flow cell has windows on three sides for optical access and input and output tubing to flow trace analytes in a nitrogen carrier gas. The flow cell was placed between two refractive objectives. A 15 mW, linearly polarized CW pump laser passed through a laser line filter and was focused by the input objective onto the waveguide facet. The pump light was polarized either normal to the sample surface (vertical) to excite the ${\mathrm{TM}}_{00}$ mode, or parallel to the sample surface (horizontal) to excite the ${\mathrm{TE}}_{00}$ mode. As shown in Fig. 1, forward-propagating Raman emission and pump light traveled through the output facet and were collimated by the collection objective. The light passed through a long-pass filter, designed to block any remaining pump light, and a linear polarizer, oriented to block orthogonal fluorescence or Raman scattered light. The signal was then refocused into a single-mode optical fiber with an off-axis parabolic mirror and sent to a 0.75 m spectrometer with a 300 groove/mm, 1.3 μm blazed grating and a liquid-nitrogen cooled silicon (for 785 nm pumping) or indium gallium arsenide (for 1064 nm pumping) detector. Exposures were taken over 100 s.

Mass flow controllers set the flow rate of nitrogen gas through a bubbler containing the liquid-phase analyte and a second path for dilution. These flows were then combined and sent to the sample in the flow cell. The chemical warfare simulants used in this Letter were chosen for their structural similarities to chemicals such as sarin and tabun. Specifically, they share the hydrogen bond basic P=O site, which binds to the hydrogen bond acidic HCSFA2.

Figure 2(a) shows a comparison between the background spectra (no analytes present) obtained under three different conditions: a horizontally polarized pump to excite the ${\mathrm{TE}}_{00}$ mode at 785 nm [finite-element model of the propagating power shown in Fig. 2(b)], with the output polarizer also horizontal (in blue); a vertically polarized pump to excite the ${\mathrm{TM}}_{00}$ mode at 785 nm [shown in Fig. 2(c)], with the output polarizer vertical (in red); and a horizontally polarized pump at 1064 nm [shown in Fig. 2(d)], with the output polarizer horizontal (in black). The ${\mathrm{TM}}_{00}$ mode at 1064 nm suffers excess loss due to substrate leakage. We observe a large, broad background centered around 920 nm when pumping at 785 nm [shown by the dashed brown line in Fig. 2(a)]. Because this broad background is absent with the 1064 nm pump, we attribute it to fluorescence. The intensity of this fluorescent background increases by almost three times when exciting the TE mode versus the TM mode at 785 nm. The TE mode is concentrated much more heavily in the SiN core of the waveguide than in the TM mode, as shown by the finite-element modal images. Therefore, we attribute the majority of this fluorescence to the SiN core. Though this broad fluorescence background is absent from some SiN waveguides, [4,8] many types of SiN are known to fluoresce in the near-infrared [11,12].

 

Fig. 2. (a) Measured background spectra for the three different pumping conditions. The arrows show the primary Raman lines from the sorbent. Finite-element surface plots of the longitudinal component of the Poynting vector for the ${\mathrm{TE}}_{00}$ mode pumped at 785 nm (b), the ${\mathrm{TM}}_{00}$ mode pumped at 785 nm (c), and the ${\mathrm{TE}}_{00}$ mode pumped at 1064 nm (d).

Download Full Size | PDF

All of our spectra also show a background Raman emission at Stokes shifts below $1500{\mathrm{cm}}^{-1}$ [shown in Fig. 2(a)], as well as a more pronounced edge below $500{\mathrm{cm}}^{-1}$ that could originate from the oxide cladding [13]. However, the broader Raman background signal below $1500{\mathrm{cm}}^{-1}$ could originate from the SiN, the ${\mathrm{SiO}}_2$, or both.

The spectra show two clear narrow features, near $775{\mathrm{cm}}^{-1}$ (a doublet) and $1610{\mathrm{cm}}^{-1}$, which originate from the HCSFA2 layer [3] [shown by the arrows in Fig. 2(a)]. Compared to TE WERS pumped at 1064 nm in 175 nm thick SiN [3], WERS from 110 nm thick SiN significantly enhances the strength of the collected signal from the HCSFA2 compared to the SiN and/or ${\mathrm{SiO}}_2$ waveguide background. Since the strength of the HCSFA2 Raman peaks directly correlates to the strength of our observed analyte WERS peaks, these HCSFA2 peaks can be used to estimate the WERS signal-to-noise ratio of analytes. Assuming that the background contributes noise dominated by shot noise, the signal-to-noise ratio for TM WERS is enhanced by a factor of 2.5 at 832 nm compared to that of TE WERS. The ${\mathrm{TM}}_{00}$ mode is also characterized by a larger single-mode cutoff width, compared to the ${\mathrm{TE}}_{00}$ mode, permitting a wider waveguide (with less sidewall-scattering loss) to be used. Thus, the ${\mathrm{TM}}_{00}$ mode was used for the subsequent WERS spectra shown in this Letter.

WERS spectra were obtained for each analyte by first measuring the background spectrum, then measuring spectra while flowing a chemical species with a known concentration. The background spectrum was then subtracted from the analyte’s spectrum to produce the differential Raman spectra shown in Fig. 3. The concentrations are estimated from the analyte’s vapor pressure and the dilution ratios in our vapor generator. Also shown in Fig. 3 is a reference liquid-phase Raman spectrum for each analyte. The one-$\sigma$ limit of detection is found by plotting the signal from the strongest feature of each analyte’s spectrum versus the analyte concentration and interpolating a linear fit through the spectral noise floor. As shown in Fig. 4, the detection limits for DMMP and DEMP are approximately 5 and 10 ppb, respectively, and are approximately 50 ppb for TMP and TEP.

 

Fig. 3. Measured differential Raman spectra of (a) DMMP, (b) TMP, (c) DEMP, and (d) TEP compared to reference liquid-phase Raman spectra [14].

Download Full Size | PDF

 

Fig. 4. Measured differential peak signal strength, along with a linear fit, and one-$\sigma$ spectral noise versus analyte concentration for each analyte.

Download Full Size | PDF

The Raman scattering efficiency ($\eta$) is the internal (in the waveguide) Stokes power ($P_s$) generated at optical frequency ${\nu }_s$ for a given pump power ($P_p$) at optical frequency ${\nu }_p$. It depends on a waveguide modal factor, $\beta$, which accounts for overlap between the Stokes and pump fields and the analyte molecules, as well as modifications to the scattering rate due to the presence of guided modes [15]. The efficiency is determined by [3,15]

We used Comsol Multiphysics to calculate via finite-element modeling the $\beta$ factor for our geometry for both the ${\mathrm{TE}}_{00}$ and ${\mathrm{TM}}_{00}$ modes. The TM efficiency is significantly larger than the TE efficiency, corresponding to 0.0086 and 0.0049, respectively. This is consistent with the images of the fields shown in Fig. 1. For the peak DMMP feature at $715{\mathrm{cm}}^{-1}$ shown in Figs. 3(a) and 4, we use ${\lambda }_s=832\mathrm{nm}$, $N_{\mathrm{vapor}}=1.7\times {10}^{18}{\mathrm{m}}^{-3}$ (70 ppb), $K\sim 1\times {10}^8$, $\beta =0.0086$, and $\sigma =2.6\times {10}^{-30}{\mathrm{cm}}^2$ [16] to give $\eta =2.8\times {10}^{-12}$. We can compare this to our measured efficiency by estimating the Stokes waveguide power in the collection fiber that corresponds to 10 counts/s at 832 nm ($2.1\times {10}^{-15}\mathrm{W}$) divided by the input fiber pump power (15 mW) and accounting for the total fiber-to-fiber coupling losses ($\sim 18\mathrm{dB}$ at 785 nm): ${\eta }_{\mathrm{meas}}=1.1\times {10}^{-12}$. Given the uncertainty in the exact partitioning of DMMP into HCSFA2 and the waveguide coupling at 832 nm, this agreement is very good and shows the validity of our model in predicting the efficiency of WERS.

There are several notable features in the spectra present in Fig. 3. First, each analyte shows a strong peak between $715{\mathrm{cm}}^{-1}$ and $750{\mathrm{cm}}^{-1}$. These peaks, which correspond to the ${\mathrm{PCO}}_2$ and ${\mathrm{PO}}_3$ stretches [17–19], are clearly resolved between these analytes and are consistent with the peak energy in the liquid-phase reference spectra. Secondly, many other peaks in the WERS spectra clearly correspond to analyte peaks in the reference spectra: the peak at $1100{\mathrm{cm}}^{-1}$ is present for both DEMP and TEP; and the peaks near $1465{\mathrm{cm}}^{-1}$ and $500{\mathrm{cm}}^{-1}$ are present in all four spectra. The analyte molecules clearly contribute significantly to the measured differential WERS spectra, enabling high-fidelity molecular identification.

There are also important differences between the measured spectra and the reference spectra. Some of the peaks appear consistently shifted (and quenched) compared to their corresponding reference peaks. For example, the reference frequency of the peak near $1250{\mathrm{cm}}^{-1}$ in DMMP and DEMP, and near $1275{\mathrm{cm}}^{-1}$ in TMP and TEP, is decreased by $50{\mathrm{cm}}^{-1}$ to $75{\mathrm{cm}}^{-1}$ and weakened in the WERS spectra. Additionally, the intensity of the peak near $1460{\mathrm{cm}}^{-1}$ in TMP and TEP is diminished. This effect may be due to a decrease of the polarizability and the stiffness of the P=O stretch [17] of the analyte due to hydrogen bonding.

Negative peaks in the differential WERS signal result from a reduction in peak strength of the HCSFA2 upon analyte binding. Two such features are clear in the WERS spectra, at $775{\mathrm{cm}}^{-1}$ (a doublet) and at $1610{\mathrm{cm}}^{-1}$, with a stronger effect from DEMP and TEP than from DMMP and TMP. The feature at $775{\mathrm{cm}}^{-1}$ is assigned to ${\mathrm{CF}}_3$ deformation [20], and the feature at $1610{\mathrm{cm}}^{-1}$ is assigned to the C=C stretch [21]. Figure 5 is a molecular representation of the hydrogen bond-based sorption of an analyte into HCSFA2. The data suggest that hydrogen bonding by the analyte reduces the polarizability of both of these vibrational modes. As the electron-rich phosphoryl group of the analyte binds to HCSFA2, the distribution of electron density associated with the C=C and ${\mathrm{CF}}_3$ bonds is altered. Larger modifications are expected from analytes with a larger hydrogen bond basicity such as TEP and DEMP [22]. These attributes of the differential WERS spectra provide distinguishing features for species identification, while also clarifying the nature of the analyte-sorbent hydrogen bond.

 

Fig. 5. Hydrogen bonding between the sorbent (HCSFA2) and analyte (TEP) molecules.

Download Full Size | PDF

We have shown that WERS can be used to detect and identify parts per billion level vapor-phase organophophonate analytes: the chemical warfare agent simulants DMMP, DEMP, TMP, and TEP. By using the ${\mathrm{TM}}_{00}$ mode of a 110 nm thick SiN waveguide, we have extended the wavelength range of the pump to 785 nm for compatibility with commercial, low-cost Raman systems. The measured WERS spectra show Raman features associated with the both the analyte and sorbent molecules, as well as features that indicate perturbations to both molecules due to hydrogen bonding. Current efforts are focused on adapting this technique to photonic integrated circuit-based fabrication, [23] including integration with waveguide filters and direct fiber-to-waveguide attachment. These steps, combined with advancements in waveguide-based spectrometry and direct on-chip laser integration would lead to a fully integrated chip-scale Raman spectrometer for trace chemical vapor detection.