1. Introduction

Waveguide enhanced Raman spectroscopy (WERS) offers long interaction lengths and strong interaction with an analyte by using the evanescent field of a high-index contrast waveguide. Coated nanophotonic silicon nitride waveguides have been used for detecting traces of chemical warfare agent stimulants down to 5 ppb [1]. Micromolar levels of cyclohexane have been probed in aqueous solutions with slot waveguides, again coated and made of silicon nitride [2]. This demonstrates the applicability to both gasses and liquids. The waveguides can be made by standard fabrication methods, are robust and can easily be integrated with microfluidics to make a lab-on-a-chip. Optical components like directional couplers, wavelength filters and grating input couplers can be fully integrated, and further integration or hybrid assembly with lasers and detectors is possible. As a first step towards integration, a packaged, fiber-coupled sensor has been demonstrated, with an integrated directional coupler for splitting the forward-propagating pump and the backward-propagating signal [3]. A recent and excellent review of WERS gives more details about the technique [4]. For biological applications, Raman spectroscopy offers label-free detection and chemical analysis. Currently, confocal Raman microscopy is gaining importance in the fields of biochemistry and microbiology [5–7] due to its ability to selectively sample cells and smaller volumes. For biological particles, the combination of Raman microscopy with optical trapping can analyse particles down to the nanoscale [8]. The extension of WERS to applications in microbiology and for the analysis of biological nanoparticles may significantly increase the capabilities of lab-on-a-chip systems for these field. As a first step in this direction, the use of TriPleX waveguides has been proposed [9]. However, there are some significant hurdles to overcome. WERS is an alternative to surface enhanced Raman spectroscopy (SERS), which gives significantly higher Raman enhancement. A comparison of the two methods is beyond the scope of this article and we refer to the many good reviews of SERS, e.g. [10–12].

Among the limitations and challenges for waveguide enhanced Raman spectroscopy are propagation losses that limit the interaction length and the Raman background of the material. The Raman background induced from the waveguide material acts as a noise-signal that overlays the signal from the analyte/sample, thus the detection performance of a WERS device is fundamentally limited by the characteristics of the waveguide material. The application of waveguides for Raman spectroscopy can be divided into three general concepts:

While both evanescent and direct field interactions give high coupling efficiencies and sensitivity, they are both most efficient for homogeneous analytes. Our work is focused on nanoparticles, for which field projection into a micron-sized volume is suitable. Thus, an embedded waveguide with negligible evanescent field interaction and high transmission is optimal. Optical fibers have extremely low propagation losses and it has been demonstrated that they also have a very low Raman background [13]. In this work, we investigate silica-based waveguides that are designed and fabricated to mimic the performance of silica fibers. The waveguides are made by UV-writing in doped silica, as will be described later. The waveguides have low refractive index contrast, a relatively large core, low propagation losses and low losses when end-coupling to an optical fiber [14–16]. However, as the waveguides are buried, with silica on all sides, there is no evanescent field available to do Raman spectroscopy of an analyte. Our approach will be to etch structures into the waveguide core (e.g. holes or trenches) for access to the field, for analysis of biological nanoparticles. Using a waveguide to project the trapping/exciting rather than using the evanescent field to excite Raman scattering enables the creation of a compact chip device with one or more micron-scale trapping sites suitable for nanoparticles.

In this article, we investigate the intrinsic properties of the waveguides for Raman scattering, without an etched interaction volume and without an analyte. The waveguides are measured using an in-line measurement scheme with a high-power laser acting as pump and a secondary, low power laser acting as a reference to obtain measurements on an absolute scale. The measurements are repeated for two pump wavelengths ($660$ nm and $785$ nm) to evaluate the wavelength sensitivity of the background. This information will decide the choice of pump laser for future applications. Raman scattering increases with $1/\lambda ^4$, possibly making a shorter wavelength favourable, but the background from the waveguide also depends on the wavelength. Thus, an experimental study is necessary to find which wavelength gives the best signal-to-noise ratio for a given waveguide and sample. The results are compared with Raman scattering in an optical fiber and in a polystyrene bead (diameter $7\;\mathrm {\mu }$m), with the first serving as a ’gold’ standard and the second to estimate the achievable signal-to-noise ratio for an easily reproducible case. The results are also compared with a model and with published values. A limited study of the influence of waveguide dimensions, composition, and cladding material is included for the UV-written silica waveguides.

The Raman background of four common waveguide materials, $Al_2O_3$, $Si_3N_4$, $Ta_2O_5$, and $TiO_2$, has been compared previously [17], and its wavelength dependence for $Si_3N_4$ and $Ta_2O_5$ was studied by D. Coucheron et al. [18]. Only N. Le Thomas et al. [13] has previously reported absolute values for the Raman background (of $Si_3N_4$). Here, we report the absolute values for the Raman background of UV-written silica waveguides in comparison to previous measurements of $Si_3N_4$. Given the similarity between the silica waveguides and silica fibers in terms of mode size and composition, the absolute values for the Raman background of select fibers are also reported here.

The UV-written silica waveguides studied here have a low index contrast relative to those considered in previous works. The relative similarity between the investigated silica waveguides and silica fibers, which have been successfully implemented as Raman-probes, promotes such waveguides as an integrated optics platform for on-chip Raman spectroscopy.

Membrane waveguides has recently emerged as a new waveguide geometry [19], with a thin core surrounded by an analyte. The results presented here are also relevant for silica membrane waveguides, where the index contrast is between (undoped) silica and water. Thus, although the present results are for burried silica waveguides, several approaches can be envisioned to exploit the low Raman background measured by modifying the waveguide geometry locally or along the entire length, by making a membrane waveguide.

2. Model and expectations for Raman background in waveguides

Waveguide enhanced Raman spectroscopy requires the pump laser to propagate through a waveguide core made of a dense material with small cross-section, and it is thus expected that this propagation will generate substantial Raman scattering in the device itself. As a consequence, the Raman scattering collected from an analyte will also contain the Raman scattering from the waveguide as a background signal. This background represents the fundamental noise limit for Raman spectroscopy of the analyte, as stated in the introduction. The Raman spectrum of SiO$_2$ is readily available [20,21], but its intensity relative to the pump laser is necessary, for the waveguide considered, to compare it with the spectrum of an analyte. Before preceding to this measurement, a model for the background is useful for interpreting the results, although several of the parameters must be obtained by fitting to the measurements. N. Le Thomas et al. have proposed a model for the Raman scattering in optical waveguides [13]. The main equation of the model will be described and used here. The model aims to express the fundamental level of the Raman scattering in a dielectric waveguide by considering the stochastic fluctuations of the induced thermal field and the subsequent noise induced in the guided wave. In contrast to previous models based on standard diffusion [22,23], this model is derived from the concept of "frozen" thermal diffusion, where the decay time of a diffusion-driven heat flux is considered to be significantly longer than the decay of spontaneous heat fluxes induced in the medium. The shift in perspective from the "slow" diffusion to the much faster stochastic heat fluxes allows the effect of the fluctuations at higher frequencies, such as those relevant to Raman spectroscopy, to be considered. The fundamental level of frequency noise induced in a wave propagating through a medium can then be predicted using knowledge of the behaviour of the stochastic heat fluxes through their temporal and spatial correlation in the medium. The model is given by Eq. (10) in [13]:

As shown in Eq. (1), the inverse square dependence of the Raman intensity on the radial frequency shift $\Omega$ as seen in Eq. (35) in [23] is replaced by an exponential dependence due to the explicit consideration of temporal correlations in the thermal field. In the case of low frequency shifts $\Omega$, the thermal fluctuations in the mode are assumed to be governed primarily by diffusion in the guiding medium. This results in an approximation of the Raman spectrum that is proportional to the inverse square of the frequency shift $\Omega$. However, when $\Omega$ becomes large, the period of the propagating wave becomes much shorter than the correlation time of the diffusion, i.e. the inducing field oscillates faster than diffusion can propagate the generated heat. Using this assumption, the diffusion can be considered as a steady state phenomenon rather than reactive to the propagating wave. In this setting, both the temporal and spatial correlations of the thermal field can be considered as strong influences. Pursuing this assumption leads to an approximation of the Raman spectrum that is exponentially dependent on $\Omega$ when $\Omega$ is large. The influence of the temporal correlation is accounted for through the introduction of the characteristic time $\gamma$ of the correlations. Similarly, the spatial correlations are accounted for through a characteristic length $\ell$ which has a linear dependence on $\gamma$ (see Eq. (9) in [13]). Using these variables, along with the length L and width W of the guided mode, the thermal field $\delta T$ can be modeled and connected to the optical field through the expected variance of the refractive index change $\left \langle {\delta n}^2 \right \rangle$. As we intend to filter out the pump wavelength ($\Omega = 0$), we see that the dirac-delta term $\delta \left ( \Omega \right )$ becomes zero for all relevant $\Omega$, allowing us to remove it from the expression. By also allowing the exciting field intensity $A_0^2$ to be an input variable, the model in Eq. (1) can be rewritten as:

As the waveguide length $L$ and pump wavelength $\lambda _0$ are known a priori, these can be collapsed into a known quantity $\alpha = 4\pi ^2L/\lambda _0^2$ for convenience. Of the four remaining variables, all but the mode width $W$ are strongly dependent on the characteristics of the temperature field $T$ and its fluctuations $\delta T$. Given the known dependencies on the material parameters listed in Table 1, it is possible to estimate these variables given the findings of N. Le Thomas [13]. However, as the exact characteristics of the thermal field fluctuations $\delta T$ are not known, the variables dependent on it are collapsed into a fit parameter $\beta$. For convenience, the unknown characteristic time $\gamma$ is replaced by a fit parameter $\epsilon$ such that the model can be expressed as a function of the wavenumber shift $\tilde \nu$ in cm$^{-1}$ rather than the radial frequency shift $\Omega$ in rad/s. The model in Eq. (2) is thus rewritten as:

Table 1. Material parameters for $SiO_2$ [25–29] and $Si_3N_4$ [13]: thermal conductivity $\kappa _0$, density $\rho$, heat capacity $C_V$, thermal expansion coefficient $\alpha _L$, thermo-optic coefficient $\partial n /\partial T$, and refractive index $n_phi$. (*Mean values)

View Table | View all tables in this article

The thermal conductivity of SiO$_2$ is substantially lower than for Si$_3$N$_4$, which is expected to shorten the characteristic length $\ell$ of SiO$_2$ due to its square root dependence on thermal conductivity [13]. Factoring in the slightly lower density $\rho$ and significantly higher heat capacity $C_V$ we estimate the thermal diffusivity of SiO$_2$ to be 55 times weaker than for Si$_3$N$_4$, implying a correlation length $\ell$ that is $86.5{\%}$ shorter for the same relaxation time $\tau$. Considering that both the thermal expansion coefficient $\alpha$ and thermo-optic coefficient $\frac {\partial n}{\partial T}$ of SiO$_2$ is almost an order of magnitude smaller than for Si$_3$N$_4$ as well as the lower refractive index $n_\phi$, it is implied that the variance of the refractive index $<\delta n^2>$ is approximately $96{\%}$ smaller for SiO$_2$ than for Si$_3$N$_4$.

These material properties of SiO$_2$ compared to Si$_3$N$_4$ implies that the induced background of SiO$_2$ is significantly lower than for Si$_3$N$_4$. Factoring in the significantly larger mode field diameter of the UV-SiO$_2$ waveguides [15], it is implied that the Raman intensity of the UV-SiO$_2$ waveguide which is $18.3$ dB weaker than for Si$_3$N$_4$, further implying a peak Raman intensity of $-113.5$ dB at $\tilde \nu =530$cm$^{-1}$. This is in agreement with the measurements made by N. Le Thomas et al. on the PM-fiber (see Fig. S1 in [24]), as its peak corresponds to $\approx -114$ dB when scaled to the same length.

As the materials of a PM-fiber and the silica waveguides are very similar, it is expected that the Raman spectrum of the waveguide will follow the same pattern as the PM-fiber as shown in Fig. S1 in the supplementary work of N. Le Thomas et al. [24]. However, given that the PM-fiber has a stated mode field diameter (MFD) of $5.3\;\mathrm {\mu }$m while the UV-written waveguide has a MFD of approximately $5\;\mathrm {\mu }$m, and since it is assumed that $\ell \ll W$, it is expected that the $\alpha \beta$-scalar for the waveguide is approximately $12{\%}$ smaller than for the PM-fiber and its characteristic time $\approx 6{\%}$ longer, assuming all other factors are equal. From this it is expected that the fiber and the waveguides will perform approximately the same, with the waveguide having a peak that is implied to be $1.1$ dB lower than the fiber.

Regarding wavelength dependence, the equation derived by N. Le Thomas shows an inverse square dependence on the pump wavelength which implies a $1.5$ dB stronger Raman background induced by a $660$ nm pump compared to a $785$ nm pump.

3. Experimental design

3.1 Waveguide fabrication

As depicted in Fig. 1, the fabrication begins by growing an oxide layer on a 150 mm diameter, 1 mm thick silicon wafer using a wet thermal process, common in microfabrication. On all the devices presented in this work, this layer is at least $15\;\mathrm {\mu }$m thick and is high-purity silica. This layer will eventually serve as the bottom cladding of the waveguide, isolating the guided mode from the silicon underneath. Flame hydrolysis deposition (FHD) is then used to deposit a thick layer of doped silicate glass to serve as the core layer of the waveguide. The core is doped with germanium to provide UV photosensitivity, which is further enhanced by adding boron. Germanium and boron co-doping is conventionally used in UV-photosensitive optical fibers and planar waveguides. However, in this work, we have also investigated germanium and phosphorous co-doping. Lastly, an optional layer is added on top of the core to serve as a top cladding to the waveguide. This layer is doped with phosphorus and boron to match the refractive index of the bottom silica-clad layer. The wafer is then diced into dies of approximately 10 x 20 mm and hydrogen loaded at 120 bar for several days to further enhance the UV photosensitivity [15,30]. The chip is then irradiated with two focused beams from a frequency-doubled argon-ion laser operating at a wavelength of 244 nm, see Fig. 1. UV exposure induces a localised increase in refractive index, thereby forming a channel waveguide.

 

Fig. 1. Fabrication process for UV-SiO$2$ waveguides

Download Full Size | PDF

Due to the weakness of the photorefractive effect, the $\Delta n$ of the waveguide is low (typically $5\cdot 10^{-3}$ [31] ), resulting in a large mode and low NA, similar to standard optical fibers. The low $\Delta n$, along with the intrinsic smoothness of an etch-free waveguide, contribute to low propagation losses in the guided mode. The low loss, combined with the low NA ($\approx 0.1$ [15]) allows these waveguides to project light with high power and low divergence into free space, but with lower intensity than a waveguide with high $\Delta n$. The low NA gives low loss across a gap in the waveguide, e.g. a hole or a trench, enabling several gaps in series for analysis of several samples along the waveguide. The large mode of these singlemode waveguides (MFD $\approx 5\;\mathrm {\mu }$m) contributes to a low background in itself, as described in sec. 2.

3.2 Samples

Five waveguide chips are considered in this work to give a limited study of the sensitivity to waveguide design, all waveguides are of length $\approx 20$ mm and are single mode for 660/785 nm wavelengths:

3.3 Experimental setup

The experimental setup, depicted in Fig. 2, consists of three main sections (a-c) and one auxiliary section (d).

 

Fig. 2. Sketch of experimental setup for acquiring power Raman spectra of a sample waveguide. The path of the pump beam (red) moves through the entire setup untill removal by a long pass filter (LP) after out-coupling from the waveguide. A longer wavelength reference laser (brown) is joined to the pump path by a beam-splitte (BS 1) such that it bypasses the long pass filter and can be recorded by the spectrometer for calibration.

Download Full Size | PDF

Section a

Section a is the origin of the pump beam, containing a high-power laser along with beam conditioning optics. To achieve best coupling to the waveguide, the output beam of the pump laser is expanded and collimated by a Galilean beam expander (L1 and L2) such that it yields a plane wave field with gaussian profile and width that is compatible with the back aperture of the in-coupling objective. The beam is also passed through a narrow band-pass filter (BP) such that the side-bands of the laser are suppressed, with special attention to the longer wavelengths.

Section b

Section b is the origin of the reference beam, accepting a fiber coupled laser source of wavelength longer than the pump beam and merging it to a common path with the pump beam. The fiber output is collimated by an appropriate lens (L3) such that it yields a gaussian plane wave that is coupled to a common path with the pump beam by a 90:10 beam splitter (BS 1).

Section c

Section c is the central part of the setup, containing the waveguide as well as the in-coupling (obj. 1) and out-coupling (obj.2 ) objectives. The output of the waveguide is, after collimation by the out-coupling objective, passed through a long-pass filter (LP) that removes the transmitted pump beam, leaving the Raman scattering and the reference beam. The filtered output is then coupled to a fiber by a final objective (obj. 3) and passed to the spectrometer.

Section d

Section d is a microscope for imaging the surface of the chip and for assisting in coupling to and from the waveguide, thus is an auxiliary section of the setup and does not contribute to the Raman measurements. This tower consists of a white light source (WLS) focused on the back focal plane of the imaging objective (Im. Obj.) by a lens (L5) relayed via a pellicle beam splitter (BS 2). Images are obtained by focusing the backscatter by a tube lens (L4) onto a camera (CMOS) for acquisition.

Components

3.4 Power calibration

Determining the spectrum of the Raman scattering induced in the waveguide is a relatively trivial task, requiring only subtraction of the background and correcting for the spectral sensitivity of the setup. However, in order to compare different waveguides and infer the intensity of the background in a Raman-on-chip device built from those waveguides requires a more thorough calibration such that the spectra can be expressed in absolute intensity rather than arbitrary units. This is why the reference laser (see Fig. 2) is necessary, providing a power reference that bypasses the long-pass filter such that it can be used as an intermediate bridge to compare the Raman spectra with the pump intensity. This requires a set of common measurement points (P1 and P2 in Fig. 2) where the pump and reference beams can be compared and a common point where the reference beam and the Raman scattering can be compared (P3 in Fig. 2). The goal of this is to obtain a calibration spectrum $C\left ( \tilde \nu \right )$ such that the intensity spectrum $I \left ( \tilde \nu \right )$ can be expressed from the measured spectrum $S \left ( \tilde \nu \right )$ as:

With the spectral elements of $C\left ( \tilde \nu \right )$ being separated into the sensitivity curve $A^{-1}\left ( \tilde \nu \right )$, the remaining calibration coefficient $c_1$ can be obtained by measuring the intensity of the pump and reference laser at in-coupling to the waveguide (P1 in Fig. 2) and at the fiber output (P3 in Fig. 2). This is done by first determining the scaling between mW and CCD counts through the coefficient $c_0$ using the measured intensity of the reference laser at the spectrometer (P3 in Fig. 2) $P3_{Ref}$ and a measured spectrum of the reference laser $S_{Ref}\left ( \tilde \nu \right )$ along with the known attenuation factor $A_{ND}$ of the neutral density filter (ND in Fig. 2). The coefficient $c_0$ can thus be expressed as:

Finally, since the setup will have a difference in coupling and propagation losses for the pump and reference beams, this must also be taken into account. Using measurement points P1 and P3, the pump transmission can be expressed as:

3.5 Composite spectra

To fully take advantage of the spectrometer’s capabilities and the fact that the fiber-coupled input (ThorLabs, SM-830) gives an effective slit width of approx. $5\;\mathrm {\mu }$m, we choose to use the finest grating available (1200g/mm) to achieve a dispersion of $2.30$ nm/mm at the focal plane. With a CCD pixel size of $20\;\mathrm {\mu }$m, this yields a per pixel resolution of $0.05$ nm but limits the spectral range of each acquisition to $52.3$ nm. Therefore, the entire range of the Raman scattering ($240-365$ nm) cannot be captured in a single acquisition without compromising resolution. A one-shot acquisition also demands the dynamic range of the spectrum is within the dynamic range of the CCD ($48.2$ dB) and above the noise floor. Given the expected exponential decay of the signal with increasing wavenumber shift, as discussed in Sec. 2 a uniform spectral sensitivity risks either over-saturating the CCD at low wavenumber shift, where the signal is strong, or losing the signal to noise at high wavenumber shift, where the signal is weak.

To solve this, we propose dividing the spectral range into segments that can be individually measured and later be merged in post-processing to obtain a composite spectrum covering an arbitrary spectral range. This allows us to acquire the entire spectral range without sacrificing the resolution afforded by the fine grating. This also allows us to dynamically select the exposure time and number of repeat exposures for each segment separately such that the SNR and dynamic range usage can be normalized for each segment separately.

This method is implemented through an automated script (Python 3.8.10) that partitions the desired range into a set of overlapping segments, each with their own exposure time $t_e$ and number of averaged acquisitions $n_{avg}$. The acquisition parameters $t_e$ and $n_{avg}$ are then estimated using an initial guess for the exponential decay of the spectra such that the expected signal fills $\approx 10{\%}$ of the dynamic range of the CCD. A test acquisition of three spectra per segment is then made such that the parameters SNR and dynamic range usage can be estimated. This is achieved by a rough separation of the signal and CCD noise by low-pass filtering the measurement, using the low-frequency elements as a signal estimator and the high-frequency elements as a noise estimator. The signal estimator is then used to tune the dynamic range usage through the exposure time $t_e$ while the noise estimator is used to tune the SNR to an acceptable level ($\geq 10$dB) through increasing the number of averaged spectra $n_{avg}$ in that segment. Using the determined parameters, the spectra of each segment is acquired and cleaned for background and cosmic rays. Using a least squares fit of their overlap, the segments are adjusted for their varying sensitivity and level of dark signal such that they are brought to a uniform scale with the first segment and can then be merged to form the composite spectrum.

4. Results

4.1 Pump wavelength: 600 nm vs. 785 nm

As previously mentioned in Sec. 1, one of the objectives of this work is to evaluate the sensitivity of the Raman scattering of the UV-SiO$_2$ waveguides to the wavelength of the pump lasers. To this end, the experiment is repeated with two sets of pump and reference lasers, first using a $660$ nm pump complemented by a $686$ nm reference and then using a $785$ nm pump complemented by a $826$ nm reference. These two pump wavelengths were chosen because of the availability of high-power lasers with high spectral purity and that both wavelengths are commonly used in Raman spectroscopy.

As described by N. Le Thomas et al. [13] and shown in the $\alpha$ component in Eq. (3), the intensity of the Raman scattering in the guided mode is expected to have an inverse square dependence on the pump wavelength $\lambda _0$. From this, it is expected that using $\lambda _0=660$ nm will induce Raman scattering approximately $1.5$ dB stronger than using $\lambda _0=785$ nm, indicating that $785$ nm may be favourable for a Raman-on-chip device. The use of $785$ nm may also help reduce undesirable fluorescence in the waveguide and/or analyte compared to $660$ nm, thus separating the Raman spectrum from the flourescence spectrum. However, the signal from a particle intersecting the beam path is expected to be proportional to $\lambda ^{-4}$, implying a $3$ dB increase in signal when using $660$ nm instead of $785$ nm as the pump wavelength, potentially compensating for the increased background in the waveguide. Another benefit of using a $660$ nm pump is that it allows a longer range of wavenumber shift to be measured using a high-quality silicon-based CCD.

This enables us to acquire measurements up to 5 400cm$^{-1}$ without exceeding the useful range of the spectrometer camera ($\lambda \leq 1025$ nm) while using $785$ nm only allows for measurements up to 3 000 cm$^{-1}$. A pump of $660$ nm thus allows for measurement of features in a wavenumber range where the background from the waveguide is expected to be negligible. One significant challenge with using shorter pump wavelength is the increased potential to induce fluorescence, both from the analyte and the waveguide material, which would mask the Raman signal. The measurements are obtained from two waveguide chips, chip A for $785$ nm and chip B for $660$ nm. For reference, the same fiber as used by N. Le Thomas et al. [24] is measured with both pump wavelengths, as is a $7\;\mathrm {\mu }$m polystyrene bead to demonstrate the achievable SNR compared to a particle. As shown in Fig. 3, both the waveguides and the PM-fiber exhibit stronger Raman scattering when excited with $660$ nm, notably so with the appearance of a flattening of the spectra as the shift exceeds $\approx 1500$ cm$^{-1}$. Given that this flattening forms a wide bulge and is only present when excited with $660$ nm, this is more consistent with fluorescence than Raman. Aside from this, we can see that the levels and features of the two pairs remain almost identical prior to the flattening and that the features of the PS-bead are almost identical.

 

Fig. 3. Raman background spectra for waveguide chips A and B (with top cladding) compared to a PM-fiber with a single $7\;\mathrm {\mu }$m polystyrene bead for reference, measured for two separate pump wavelengths ($660$ nm and $785$ nm).

Download Full Size | PDF

In terms of features, we see that both waveguides exhibit a peak at $\approx 920$ cm$^{-1}$ that is absent in the spectrum of the fiber, this can be readily assigned to the stretching mode of Ge-O-Si [32] due to the known high concentration of germanium in the waveguide cores. We can also observe that both waveguides produce peaks at $\approx 1310$ cm$^{-1}$ that corresponds well with B-O$^{-}$ [33] as is also expected due to the boron-doping of the core. The remaining features at $\approx 1440$ cm$^{-1}$ and $\approx 1710$ cm$^{-1}$ may also be due to B-O-B and [BO$_3$]$^+$ modes, but due to the ambiguity of features near those shifts, we are hesitant to make the assignment. One other noteworthy observation in the Raman spectra is the significantly weaker feature at $\approx 580$ cm$^{-1}$ in the waveguides compared to the fiber. This is commonly assigned as a defect mode of Si-O-Si [34], which diminishes with increasing dopant concentration as observed here.

Because of the higher background produced by the waveguide when excited using $660$ nm compared to $785$ nm and the fact that the PS-bead, serving as an analogue for future particles, shows only a weak increase in feature intensity when excited using $660$ nm, it is concluded that $785$ nm is the preferred pump wavelength for this type of waveguide device.

4.2 Dependence on doping and cladding

Another objective is to determine the sensitivity of the Raman scattering on the doping of the core layer and the use of a top cladding. In this section, three additional chips (C,D$_1$ and D$_2$) are measured using a $785$ nm pump. For reference, both the fully clad NIR-waveguide (chip A) and the PM-fiber shown in Fig. 3 are included. As mentioned in Sec.3.2, both chips C and D are manufactured without top cladding while chips A and B are manufactured with a top cladding ($>10\;\mathrm {\mu }$m B+P doped SiO$_2$). Chip C is doped with phosphorous instead of boron while chips D are doped similarly to chips A and B. As shown in Fig. 4, the results from chip A and chips D$_1$ and D$_2$ match well, as expected, and it can be seen that the measurements obtained from two waveguides from D$_1$ and two from D$_2$ group well, supporting consistency of the power calibration. We also see that the phosphorous doped chip C performs significantly poorer than the other chips, producing a noticeably flatter Raman spectrum with a higher baseline than the other samples. The measured transmission through chip C was up to 5.6dB lower than either chip D and the mode at the output was poorly defined with significant slab guiding in the core layer relative to the guiding in the UV-written waveguide. The high degree of slab guiding indicates poor lateral confinement, likely due a low $\Delta n$ being induced by the photorefractive effect without the presence of boron doping. Several of the chips, notably chip D$_1$, also displayed a significant variance in transmission between waveguides, varying as much as 5.4dB. The variance in transmission among the topless waveguides suggests chipping or defects at the facet because of a lack of the protective top cladding. The low and varying transmission for some of the waveguides may thus be due to poor in-coupling and surface defects, rather than absorption or scattering in the core itself.

 

Fig. 4. Measurement of all chips with PM-fiber for comparison. Previously reported measurements of Si$_3$N$_4$ and similar PM-fiber are also included for reference.

Download Full Size | PDF

4.3 Fitting the spectra to the model

In this section, the measured spectra are fitted to the model described in eq. (3) and the parameters of the fit are compared between the measured waveguides and previously reported measurements. The fits are shown in Fig. 5. The general level of the measurements follow the model in Eq. (3) well, with the majority of deviance being due to specific features in the Raman spectra. We also see from the parameters and peak intensity in Table 2 that our measurements of the PM-fiber agree with those made by N. Le Thomas et al. [24] in both profile and intensity. From the listed intensities we see that all of the Ge+B-doped SiO$_2$ (Chips A, B, D$_1$ and D$_2$) have a negligible difference in peak amplitude, while the P+Ge-doped SiO$_2$ (Chip C) has a peak $6$ dB higher than the rest, emphasizing the negative effect of phosphorous in the core. Lastly, we see that the Raman scattering in our UV-SiO$_2$ waveguides is more than $12$ dB weaker than what is reported for Si$_3$N$_4$, reinforcing their advantage for Raman-applications.

 

Fig. 5. Fit of the model in eq. (3) to measurements of chips A, B, C, D$_1$ and D$_2$ with a similar PM-fiber measured in this work and in the work of N. Le Thomas et al. as a reference. The fit of reported measurements of a $Si_3N_4$ waveguide is also added for comparison with the UV-written SiO$_2$ waveguides.

Download Full Size | PDF

Table 2. Table of fit parameters for spectra in Fig. 5 to Eq. (3) as described in Sec. 2. *All measurements are scaled to emulate a length of 1cm

View Table | View all tables in this article

5. Conclusion

The Raman background of UV-written silica waveguides has been measured and compared to other platforms, notably Si$_3$N$_4$ waveguides and (silica) optical fibers. To obtain results in absolute terms, the acquired spectra were calibrated to the input intensity. This was achieved using a separate laser source, acting as a reference and coupled into a common path with the pump laser, such that both the Raman spectra and the reference laser could be measured with the same configuration. Furthermore, by combining the spectrum of a reference source measured by a spectrometer with the power of the pump and reference source measured by a photodiode power meter, the acquired Raman spectra were calibrated to absolute terms. To take full advantage of the setup, a high-resolution grating (1200 lines/mm) is used and the spectra are acquired in adjoining segments, allowing the full spectral and dynamic range of the spectrometer to be used. The acquisition time and number of repetitions was set independently for each section, to exploit the full range of the CCD spectrometer and to tailor the sensitivity to the expected intensity for the section. This allowed a high signal-to-noise ration for a very large dynamic range (from −100 to −160 dB relative to the pump laser).

The background of the UV-written SiO$_2$ waveguides was measured for two excitation wavelengths, $660$ nm and $785$ nm. Waveguides written into five chips were characterized, and for reference two optical fibers and a $7\;\mathrm {\mu }$m polystyrene bead were also measured. This revealed a Raman intensity of less than $-107.4$ dB in the biochemical fingerprint region for a waveguide excited by $785$ nm, and less than $-106.5$ dB when excited by $660$ nm (normalised to 1 cm length). The difference increased for increasing wavenumber shifts, leading to the conclusion that $785$ nm is better suited than $660$ nm when using these waveguides. This conclusion depends on signal-to-noise ratio, which depends on the analyte, and it was shown that it holds for a polystyrene bead as it gave the same Raman signal for both wavelengths. The largest peak in the Raman spectrum of a $7\;\mathrm {\mu }$m PS-bead was $10.4$ dB higher than the waveguide background. This shows that a good signal-to-noise ratio can be obtained for microparticles with all the background from a 1 cm long waveguide collected as noise. The signal scales with the intensity and the interaction volume, with the first depending on the waveguide structure for illuminating the particle (e.g. hole, trench or taper) and the second on the diameter of the particle. The noise depends on how much of the waveguide background is collected (e.g. by a microscope objective or a collection waveguide). The achievable signal-to-noise ratio for nanoparticles thus depends on the interaction structure and the collection method. We will investigate this in future work.

The background induced in the best waveguide was $8.7-10.3$ dB higher than for optical fibers and approximately $15$ dB smaller than for Si$_3$N$_4$ waveguides [13]. UV-written waveguides thus present a very promising alternative for on-chip Raman spectroscopy, but there is still room for improvement when comparing with optical fibers.

A limited study of the impact of doping was made, with phosphorous doping giving significantly higher background than boron, with $-100.0$ dB and $-106.5$ dB, respectively, for a wavenumber shift of $410$ cm$^{-1}$ and $436$ cm$^{-1}$. In addition, the phosphorous doping resulted in a significant flattening of the spectrum and poor waveguide confinement. This may have influenced the result. The presence or not of a top-cladding did not influence the background significantly (for boron doped samples). The background of four top-clad waveguides from two chips deviated by less than $5.7$ dB over the entire fingerprint region, showing good repeatability considering the mean level of $-121$ dB.

This work has only considered the noise related to Raman background of the waveguides, and the logical next step will be to modify the waveguides to obtain signals from an analyte. Several procedures will be tested, notably etching trenches across the waveguides. Approaches for incorporating more complex structures, such as tapers and/or nanoantennas will also be explored.