1. Introduction

Waveguide-enhanced Raman spectroscopy (WERS) [1–3] is a powerful analytical technique that utilizes integrated photonics for chemical [4,5] and biological [6–9] sensing. For trace-chemical vapor sensing, a sorbent material acts as the top cladding of the waveguide core to concentrate analyte molecules in the evanescent field. As the laser propagates down the waveguide, it continuously interacts with analyte molecules that emit Stokes-shifted Raman light, which is recollected into the guided modes of the nanophotonic waveguide. The sorbent film can be targeted to a class of molecules of interest and optimized (in regard to thickness and absorption) for a particular laser wavelength [10].

WERS benefits from long (and thus low-loss) waveguides, as well as small (subwavelength) feature sizes to define components such as edge couplers and directional couplers. Thus, fabrication in state-of-the-art photonic foundries is highly beneficial. These foundries also provide a path to fiber-attach, die assembly, packaging, and volume production. To date, efforts to transition waveguide fabrication to integrated photonics foundries have focused on wavelengths in the near-infrared (NIR, 1064 nm and 785 nm). At these wavelengths, single-mode silicon nitride (SiN) waveguides provide very low propagation loss and very low background emission, including autofluorescence [11]. Extending WERS into the visible to take advantage of the $\lambda ^{-4}$ scaling of the Raman scattering cross-section has been demonstrated [12,13], but the waveguide materials and processes utilized in those works are not available in a state-of-the-art 300 mm silicon photonics foundry.

In this work, we demonstrate visible WERS in SiN waveguides fabricated at a 300 mm wafer silicon photonics foundry (AIM Photonics). We weigh the higher background fluorescence and propagation loss, compared to the NIR, against the $\lambda ^{-4}$ Raman cross-section dependence. We compare two visible WERS configurations (532 nm and 633 nm laser pumping) to the current NIR state-of-the-art (785 nm) to assess the trade-offs of foundry-quality SiN WERS in the visible compared to the infrared. We experimentally demonstrate a foundry-compatible photonic design and wavelength range that optimizes SiN WERS based on both Stokes signal strength and on signal-to-background ratio. We also demonstrate that these optimal designs and wavelengths are consistent with the predicted WERS efficiency, so long as optical loss is accounted for.

2. WERS efficiency

WERS signal is proportional to the waveguide interaction length function, which accounts for propagation losses at the laser and Stokes wavelengths. Waveguide propagation loss arises from sidewall scattering, material absorption, or substrate or bend losses. The WERS signal is also proportional to the evanescent field penetration into the part of the cladding used for sensing. Here, we focus on a sorbent material designed for the sorption of trace vapors as the top cladding. The evanescent field penetration is enhanced by single-mode waveguiding and weakly-guided modes. Optical insertion losses comprise facet-fiber coupling and sensing trench-waveguide coupling. In order to compare the three different sensing wavelengths, we first calculate the efficiency of each. Together, the WERS efficiency for a given waveguide geometry and pump wavelength can be described by [2,14]:

For forward-scatter, the length function [1] is

Figure 1 shows the result of the mode-solver calculations for $\beta$, as well as images of the z-component of the Poynting vector (the propagating modal power) for the laser wavelength in its respective waveguide geometry in the TM$_{00}$ mode. We focus exclusively in this work on the TM$_{00}$ mode for WERS since this mode has greater evanescent overlap with the top (and bottom) cladding, less overlap with the (fluorescent) core, and has lower propagation loss due to weaker overlap with the core sidewalls, compared to other waveguide modes [4]. For each WERS laser wavelength and waveguide geometry, we calculated $\beta$ for the SiN core, the SiO$_2$ bottom cladding, and the AMS-191 (described in more detail below) top cladding of the waveguide. As expected, $\beta$ for the core of the waveguide generally decreases as the wavelength increases and is less confined. Similarly, the overlap with the cladding increases with the wavelength. The stronger overlap with the bottom cladding than the AMS-191 is due to the larger refractive index of SiO$_2$ (1.46), than that of AMS-191 (1.41) [15].

 

Fig. 1. Calculated $\beta$-factor of the TM$_{00}$ mode given A) 532 nm pump in a 0.5 $\mu$m waveguide, B) 633 nm pump in a 0.5 $\mu$m waveguide, and C) 785 nm pump in a 0.8 $\mu$m waveguide. Above each plot is an image of the z-component of the Poynting vector, with labels for the top cladding (AMS-191), bottom cladding (SiO$_2$), and waveguide core (SiN). The thick black lines correspond to physical material interfaces in the waveguide layer stack.

Download Full Size | PDF

3. Experimental methods

3.1 Waveguide fabrication

AIM Photonics fabricated the silicon nitride waveguides utilized for this work in a process very similar to the SiN Passive PIC Multi-Project Wafer (MPW). This platform provides very low propagation losses and low background fluorescence at wavelengths shorter than 1000 nm [11], making it an excellent choice for visible WERS. In order to improve performance at shorter wavelengths, we use a thinner silicon nitride waveguide core (150 nm) than what is standard in the MPW. Spirals were patterned and etched into a deposited LPCVD SiN layer using single-mode 0.5 $\mu$m wide (for 632 nm WERS) or 0.8 $\mu$m wide (for 785 nm WERS) waveguides. 532 nm WERS was carried out using 0.5 $\mu$m wide waveguides which, despite supporting a TE$_{10}$ mode, have lower loss than narrower, single-mode waveguides. The SiN layer sits on a 5 $\mu$m thermal bottom oxide, and a 5 $\mu$m PECVD top oxide clads the waveguides. For spirals that will be used for WERS, a sensing trench was defined wherein the top oxide is etched away such that the top surface of the SiN waveguide core is exposed.

We measured spiral waveguides both in a sensing trench, and clad with SiO$_2$. In trenches, we examined air-clad spirals (no top cladding) as well as polymer-clad spirals. The polymer cladding is AMS-191 (https://www.gelest.com/product/AMS-191/), a (9-11${\%}$ aminopropylmethylsiloxane) - dimethylsiloxane copolymer that can be used as a general purpose sorbent for hydrogen-bonding vapors. We employed 0.5 $\mu$m wide waveguides in spiral geometries for wavelengths from 530 nm to 750 nm. For direct comparison to the state-of-the-art NIR WERS configuration, we also fabricated 0.8 $\mu$m wide waveguide spirals, the single-mode width for 785 nm WERS in the TM$_{00}$ mode. The waveguide spirals have a minimum bend radius of 150 $\mu$m for the 0.5 $\mu$m wide waveguides, and 175 $\mu$m for the 0.8 $\mu$m wide waveguides.

3.2 White light spectroscopy

We measured propagation loss ($\alpha (\lambda )$) and coupling loss ($K(\lambda )$) via white light spectroscopy [16]. These measurements required two broadband sources. The first, a tungsten-filament light source (Thorlabs SLS201L), emits efficiently at wavelengths above approximately 630 nm. The second, an LED source (Thorlabs MCWHLP2), is more effective for measurements throughout the visible. In both cases, we couple the light through a linear polarizer into a polarization maintaining (PM) fiber. A lensed PM fiber couples the light into the waveguide. After propagating through the waveguide edge coupler and the length of the spiral, a second lensed PM fiber collects the light at the output edge coupler and sends it via PM fiber to a VIS-NIR spectrometer with a thermo-electrically-cooled (TEC) detector array (Wasatch Photonics). After measuring several lengths of waveguide, increasing from 0.5 to 10 cm, a linear regression analysis separates the propagation losses from the facet coupling and trench insertion losses. We measured waveguide samples with an oxide top cladding, no top cladding, and an AMS-191 top cladding.

3.3 Waveguide-enhanced Raman spectroscopy

The WERS measurements were performed with Witec Alpha 300 Raman chemical imaging microscopy systems, which here operates in the backscattering configuration. The measurements were performed with a thick (droplet) coating of AMS-191 in the sensing trench. A diagram of the experimental setup is shown in Fig. 2. Three different laser wavelengths were used: 532 nm, 633 nm, and 785 nm. The lasers were fiber-coupled into the microscope tower via polarization-maintaining single-mode fibers. Laser-line filters were used to to remove Raman scattering in the fibers and any spontaneous emission or sidemodes from the lasers. Laser powers of 5 mW (532 nm excitation), 4 mW (633 nm excitation), and 10 mW (785 nm excitation) incident on the back aperture of the microscope objective were used. For all measurements a 10X, 0.25 NA microscope objective (Zeiss EC Epiplan-Neofluar DIC) was used to couple light into and out of the waveguide facet. The laser polarization was oriented using half-wave plates to excite the TM$_{00}$ mode of the waveguides. The Raman backscattered light was recollected through the same objective, and passed through two notch filters and a linear polarizer oriented to pass the TM$_{00}$ mode. It was then focused into a photonic crystal (PC) fiber for collection and sent to a spectrometer for analysis.

 

Fig. 2. Schematic diagram of the setup used to measure the WERS spectra in the backscattering configuration.

Download Full Size | PDF

For 532 nm excitation, a Witec UHTS 600 VIS spectrometer with a 600 mm focal length and a 300 g/mm grating was used. To detect the Raman scattered light, an Andor Newton DU970P-BVF electron-multiplying charge-coupled device (EM-CCD) camera operating in conventional mode was used. The quantum efficiency of this camera is approximately 95${\%}$ in the wavelength range of interest for Raman spectroscopic measurements with 532 nm excitation. The sensor was cooled to -60 $^\circ$C, which reduces the dark current of the detector. For 633 nm and 785 nm excitation, a Witec UHTS 400 NIR spectrometer with a 400 mm focal length and a 300 g/mm grating was used. For these excitation wavelengths an Andor DU401-BR-DD back-illuminated deep-depletion CCD camera optimized for longer wavelengths was used. The sensor in this camera was also cooled to -60 $^\circ$C to reduce the dark current. The quantum efficiency of this camera is approximately 80-90${\%}$ in the wavelength range of interest for Raman spectroscopic measurements with 633 nm and 785 nm excitation. The integration time used for all spectral measurements was 0.1 s per spectrum.

The brightfield camera of the Witec Raman imaging microscope is used to coarsely align the focus of the laser on the waveguide facet in the out-of-plane direction, as well as along the waveguide axis. A second compact microscope oriented perpendicular to Raman microscope objective is used to coarsely align the focused spot in the lateral (in-plane) direction and verify rough coupling into the spiral via observation of elastically scattered laser light. Optimization of the alignment is then achieved by using the motorized xyz microscope stages to maximize Raman scattered signal as measured by the spectrometer.

4. Results

4.1 Optical transmission

We measured the TM$_{00}$ waveguide propagation loss spectra $\alpha (\lambda )$ for both waveguide widths via white light spectroscopy in order to determine $L(\lambda _p,\lambda _s)$. Figure 3 plots this measured loss for different wavelength ranges. For each band, the waveguide spirals were measured with an oxide cladding, with a trench etched down to the top of the waveguide, and with the trench back-filled with AMS-191. Understanding the losses in these three different layer structures provides information about where losses originate. For example, Fig. 3(A) indicates excessive bend loss when the trench has been etched down to the top of the waveguide due to the low index of the air top cladding. Bend loss prevents the propagation of the TM$_{00}$ mode at larger wavelengths in the air-clad spirals. This measurement also indicates that the etch process does not significantly increase the roughness-induced scattering from the waveguide core, nor does the AMS-191 contribute significantly to absorption. These spectra, combined with other measured loss spectra at different modes and different waveguide thicknesses [17], together suggest that the loss is dominated for all waveguide widths and wavelengths by absorption in the SiN core, with the exception of the the longest wavelengths (>900 nm) in the 0.8 $\mu$m wide AMS-191 coated waveguides.

 

Fig. 3. Propagation loss ($\alpha (\lambda )$) of the TM$_{00}$ mode in A) the green/yellow band (0.5 $\mu$m waveguide), B) the orange/red band (0.5 $\mu$m waveguide), and C) the i/z band (0.8 $\mu$m waveguide). The black curve shows loss in oxide-clad waveguide spirals, the red curve shows loss in waveguide spirals with no top cladding in a sensing trench, and the blue curve shows loss in waveguide spirals in a sensing trench clad with AMS-191. The standard error of the loss fitting technique is shown by the transparent region surrounding each curve.

Download Full Size | PDF

In the green/yellow band, which is appropriate for a laser wavelength of 532 nm, we measure a loss of 2.25 dB/cm at the laser wavelength, decreasing to 1.75 dB/cm at 600 nm, or 2000 cm$^{-1}$ Stoke-shift, for a waveguide spiral clad in AMS-191. For a 633 nm WERS pump laser, the AMS-191-clad waveguide spirals have a loss of 1.5 dB/cm, which decreases to 1.0 dB/cm at 740 nm. The NIR WERS platform coated with AMS-191 in the I/Z band has a loss of 0.28 dB/cm at 785 nm, which decreases to 0.08 dB/cm at 860 nm. Bend loss increases the loss to 0.6 dB/cm at 950 nm.

White light spectroscopy yields both the propagation loss (the slope of the linear regression) and the coupling loss (the y-intercept). These coupling losses include edge coupling and trench insertion loss. The edge couplers contribute approximately 3-5 dB of loss per facet, dominated by imperfect mode-matching. Trench losses result from scattering at the trench edge, and range from 1-5 dB. These losses, cumulatively, are represented by $K(\lambda _p)$ and $K(\lambda _s)$ for the pump and signal, respectively, in Eq. (1).

4.2 WERS and fluorescence

Raman chemical imaging microscopy was used to measure the WERS response of thick AMS-191 coated spirals in a sensing trench for comparison of the relative efficiencies of the WERS process. Three different spiral waveguide lengths were used for each excitation wavelength: 1.8 cm, 2.9 cm, and 4.0 cm. For 532 nm and 633 nm excitation, 500 nm wide waveguides were used, while for 785 nm excitation, 800 nm wide waveguides were used. For each case the optimal signal was obtained by using the alignment process described above. Figures 4(A) and 4(D) show plots of these optimized spectra obtained with 532 nm excitation, Figs. 4(B) and 4(E) show spectra obtained with 633 nm excitation, and Fig. 4(C) shows spectra obtained with 785 nm excitation. Figures 4(D) and 4(E) show the intensity of the CH-stretching bands; bending loss in the 800 nm wide waveguide combined with the decreased quantum efficiency of the CCD camera at wavelengths >900 nm (Stokes shifts > 1500 cm$^{-1}$) led to very small CH-stretch signal with 785 nm excitation (not shown here). The spectra shown in Fig. 4 correspond to the facet location of the focus that has the strongest Raman signal.

 

Fig. 4. WERS spectra from 200-2000 cm$^{-1}$ for A) 532 nm excitation, B) 633 nm excitation, C) 785 nm excitation. Additionally, WERS spectra from 2820-3050 cm$^{-1}$ for D) 532 nm excitation and E) 633 nm excitation.

Download Full Size | PDF

The three sets of WERS spectra show extremely high signal and signal-to-background levels even though integration times of only 0.1 s were used. Figures 4(A) and 4(D) show the largest fluorescence background, from the 532 nm laser, likely due to the presence of impurities or nanocrystals in the SiN core excited at higher energy. The WERS signal from the two longest lengths at shorter Stokes shifts (< 1000 cm$^{-1}$) are similar, indicating that after the first centimeter or so, waveguide loss significantly reduces the backscattered signal strength. This is consistent with the measured loss in Fig. 3(A): Propagation losses of 1.7 dB/cm reduce the “effective” waveguide length $L$ to approximately 1.2 cm for actual waveguide lengths $L_0$ in excess of 1.2 cm. We can use Eq. (4) to calculate the effective length for a 633 nm and 785 nm WERS backscatter configuration given an infinitely long spiral. For 633 nm, we expect a maximum effective length of 1.75 cm, and for 785 nm we expect a maximum effective length of 9.25 cm. However, the maximum physical spiral length measured in this work was 4.0 cm. For 633 nm pumping, this corresponds to an effective length of 1.6 cm, and for 785 nm pumping this corresponds to 3.5 cm. In addition to having the highest effective waveguide length, using a 785 nm laser generates the lowest fluorescent background, with Raman scattering from the SiN and SiO$_2$ contributing virtually all of the background [11]. WERS using a 633 nm laser generally falls between these two cases.

5. Analysis and discussion

The measured loss spectra shown in Fig. 3 are inserted into Eq. (4) and then into Eq. (2) along with measured $K(\lambda )$ values to obtain predictions of the relative WERS backscattering efficiency for a single Stokes peak. We can compare these estimates to measured WERS signal at specific peaks for various waveguide widths, waveguide lengths, and laser wavelengths. Because the background contains contributions from both fluorescence and SiN/SiO$_2$ Raman scattering, we isolate the measured signal from the background by extracting the peak height (in counts/s/mW). We then compare this peak height between each WERS configuration for the three different AMS-191 Raman peaks over three different spiral lengths. Similarly, we compare the calculated efficiency between each WERS configuration for each peak over each spiral length.

The results are plotted in Fig. 5. Here, the two y-axes (for the measured and calculated peak strengths) are scaled such that the data and calculation from the 633 nm excitation in the 2.9 cm long spiral overlap. The relative agreement between the calculated signal and measured signal is excellent, and suggests that, based purely on signal strength, and with current SiN waveguide losses, the optimal excitation laser wavelength is 633 nm, and that even longer spirals than the 4.0 cm used here would be optimal. These results also suggest that Eq. (1) and Eq. (2) accurately predict relative WERS signal strengths for different photonic designs and wavelengths. This prediction requires careful measurement of the waveguide losses (propagation and coupling) since accurately predicting these losses for different foundry processes and external coupling configurations a priori is not yet possible.

 

Fig. 5. The measured Stokes peak height compared to the calculated Stokes efficiency ($\eta$). Measurements are represented by the left y-axis and the black, red, and blue squares. Calculations are represented by the right y-axis and the orange, green, and purple triangles. Comparison is presented for three different peaks: A) 490 cm$^{-1}$, B) 1440 cm$^{-1}$, and C) 2900 cm$^{-1}$. The theoretical and measured y-axes are aligned such that the points at 633 nm in the 2.9 cm spiral overlap.

Download Full Size | PDF

We can qualify our theory by extracting the percent deviation of our measured peak heights from the theoretical model. First, we normalize each measured peak height to the case of 633 nm excitation in a 2.9 cm spiral, for consistency with the data we have reported up to this point. Then, we normalize the theoretical efficiency in the same way. Now, we have a relative measured value and a relative theoretical value that can be directly compared. Table 1 reports this comparison for each pump, peak, and spiral combination. Most of the measured data deviates from the theory by less than a factor of two, which indicates excellent agreement given the uncertainty in the propagation loss ($\alpha$) values used to define the loss function. The peaks at 490 cm$^{-1}$ when pumping at 785 nm deviate more dramatically from the theoretical model, which could be due to a peak in the CCD efficiency at these wavelengths.

Table 1. Percent deviation of measurement from calculated efficiency.

View Table

Figure 6 shows the measured signal-to-background ratio for the same spiral lengths, laser wavelengths, and Stokes peaks as shown in Fig. 5. The background was determined by averaging data-points on either side of the peak. Though many techniques exist for accurate background removal, such as using multiple lasers to separate fluorescence from Raman scattering [18], technical requirements may prevent their implementation. For example, simple background subtraction cannot isolate background noise from signal when using short waveguides or low laser powers. Also, background signal resulting from waveguide or fiber Raman scattering cannot be removed with the use of multiple lasers. Figure 6 shows that signal-to-background ratios favor a 785 nm laser to both 633 nm and 532 nm lasers, for Stokes peaks up to 1500 cm$^{-1}$, a limit imposed by the quantum efficiency of our detector at wavelengths beyond 900 nm.

 

Fig. 6. The measured Stokes peak height divided by the measured background for each spiral length. Comparison is presented for three different peaks: A) 490 cm$^{-1}$, B) 1440 cm$^{-1}$, and C) 2900 cm$^{-1}$.

Download Full Size | PDF

Here, overall signal strengths are so strong (tens of thousands of counts with only a 0.1 s exposure time) that noise present in the background is unlikely to degrade a measurement’s fidelity. Separate measurements of WERS signal using compact spectrometers [19] have shown similar signal strength, with only a slight increase in noise. Thus, we expect background noise for WERS measurements of top claddings to be insignificant even with the use of miniature, TE-cooled spectrometers. On the other hand, observing low concentrations of trace vapors sorbed into these layers could still be limited by background noise.

6. Conclusions

State-of-the-art 300 mm photonic integrated circuit foundries are now capable of fabricating high-quality silicon nitride waveguides and components for waveguide-enhanced Raman spectroscopy (WERS) in the visible. This includes laser wavelengths of 532 nm and 633 nm, in addition to the more mature 785 nm in the near-infrared. Our measurements of propagation loss and coupling loss combined with calculations of the WERS efficiency agree with our WERS measurements to suggest that both 633 nm and 785 nm lasers provide outstanding Stokes spectral signal strengths. A 633 nm laser is optimal for Stokes signal strength, including at the C-H stretch bands, whereas a 785 nm laser provides slightly better signal-to-background ratios in the fingerprint region, and the very low propagation loss enables the use of long spirals (up to 10 cm) to increase maximum WERS signal. Further research will increase the effectiveness of the SiN platform at visible wavelengths by addressing the higher-energy absorbance and fluorescence inherent in these films. In summation, AIM Photonics is capable of direct fiber integration, system assembly, and volume manufacturing, which will enable the production of compact analytical systems for biological, environmental, and chemical detection.