**4. Waveguide-Enhanced Raman Spectroscopy**

## *4.1. Configuration and Integration*

On-chip integration of Raman spectroscopic systems generally requires the enhancement of Raman scattering for greater efficiency, and, at the same time, the scattered light needs to be collected over a small area and a small solid angle (also known as the étendue) to maintain the small size of the device [77]. Single-mode waveguides can provide strong enhancement over a small volume, and therefore constitute an optimal solution for chipintegrated Raman spectroscopic systems. Compared with diffraction-limited systems, waveguide-integrated Raman systems with strong optical field confinement can provide a stronger enhancement of the signal by a few orders of magnitude, allowing for much higher detection sensitivity. Further improvement in signal enhancement can be brought about through the use of nanoplasmonic antennas integrated with waveguides [201].

For Raman spectroscopy, similarly to IR systems, free-space butt coupling through an objective lens, prism-based coupling, and fiber-mediated coupling have been the prominent mechanisms for introducing the light into the waveguides. On the other hand, collection from waveguides can be performed either from the waveguide top surface or from the waveguide facet, which can be in either a back-scattered or forward-scattered configuration (Figure 4a,b). The collection efficiency from the waveguide facet is generally much higher and has been shown to be about 40 times more efficient than that from the surface [202]. As a consequence, this requires less integration time than that from the waveguide surface [28]. However, signals from the surface can provide additional spatially resolved information [52].

While free-space coupling with a high numerical aperture objective provides good coupling efficiency, it also introduces vibrations and critical alignment steps and is unsuitable for use in compact setups. Fiber-mediated coupling solves some of these issues, but it introduces a spurious background signal, including fluorescence and Raman scattering generated from both the input and the output fibers. Kita et al. aimed to eliminate this effect by collecting only the backscattered light. This tactic removes much of the forwardpropagating pump beam, which results in a higher signal-to-noise ratio [47]. The collection of the backscattered beam also has the advantage of having virtually no waveguide length limit due to the propagation loss of the waveguide, even though the contribution to the scattered signal for waveguides of lengths longer than 2/*α<sup>p</sup>* is negligible, where *α<sup>p</sup>* is the propagation loss of the waveguide. In return, the forward-scattered light collection efficiency generally has a maximum for a particular waveguide length, beyond which the propagation loss dominates, thus reducing the collected signal power (Figure 4c). A different approach to eliminating the influence of the background has been to integrate edge couplers and waveguide filters onto the chips, as demonstrated by Tyndall et al. [204,206]

(Figure 4d). An array of polarization maintaining single-mode fibers is aligned directly to the waveguide facets through edge couplers; the subsequent use of lattice filters helped in separating the background and collecting both the forward- and backward-propagated Raman scattered light at separate outputs. Other on-chip elements, such as a grating-assisted contra-directional coupler, have also been proposed to reject the pump beam by directing it to a separate bus waveguide, resulting in a very high extinction ratio [207]. Nonetheless, the use of dielectric waveguides still introduces some photon background, likely arising from localized thermal fluctuations, which has been difficult to get rid of. The use of a nano-plasmonic slot waveguide, combined with a multi-mode interferometer (MMI) and backward Raman collection, has been shown to mitigate this problem [205] (Figure 4e).

**Figure 4.** Different configurations as reported for waveguide-enhanced Raman spectroscopy. (**a**) Free space butt coupling of a laser into a waveguide chip for Raman spectroscopy. The figure shows the collection of the scattered signal from both the top surface and from the end facet in a forward configuration (reproduced from [52] licensed under a Creative Commons Attribution 4.0 International License http://creativecommons.org/licenses/by/4.0/ (accessed on 25 October 2021)). (**b**) Backward collection of the Raman scattered signal from a waveguide (reproduced from [203] © 2021, the author(s)). (**c**) Dependence of the scattered signal on the waveguide length for forward- and backward-scattering (reproduced from [48] © 2021, Optical Society of America). (**d**) On-chip scheme to separate the pump beam, the forward- and the backward-scattered signal through the use of lattice filters (reproduced from [204] © the authors). (**e**) WERS setup with a nano-plasmonic waveguide. The design minimizes the background signal, while the use of MMI helps in separating the input pump and the output Raman signal (reproduced from [205] © 2021, Optical Society of America).

Spontaneous Raman scattering signal intensity grows linearly with the average power of a continuous-wave pump laser; this has been a major bottleneck in improving the Raman signal, particularly in a miniaturized system where it is not possible to increase the pump power indefinitely without causing substantial damage. On the other hand, coherent Raman scattering (CRS) is a third-order non-linear phenomenon involving two laser beams, the pump, and the Stokes. When the difference in frequency between both respective lasers equals that of a specific vibrational (or rotational) transition, the probability of this transition is resonantly enhanced. CRS is capable of improving signal by many orders of magnitude and is typically implemented in two configurations, coherent anti-stokes

Raman scattering (CARS), and stimulated Raman scattering (SRS). Of these two techniques, SRS has been shown to be more promising, particularly in the context of waveguidebased Raman sensors. The reasons behind this are a simpler phase-matching relationship between the two lasers, a linear dependence of the Raman spectra on concentration, and an enhancement of the signal due to self-heterodyned detection. However, in spite of these advantages, increased shot noise severely limits the performance, only providing a modest increase of the Raman signal, as demonstrated by Zhao et al. [208]. On the other hand, other techniques such as cavity-enhanced Raman spectroscopy (CERS) and Purcell-enhanced Raman spectroscopy (PERS) have been recently used to resonantly enhance the laser beam power, as well as to improve the rate of Raman scattering, resulting in having higher laser beam–analyte interaction lengths [209–211]. These allow for sufficiently low pump powers and, in combination with their small size, may play a very important role in constructing miniaturized devices in the future.

Despite the above-reviewed efforts, the complete photonic integration of the Raman spectroscopic system is still in its infancy, primarily due to the fact that WERS requires high-power monochromatic light sources at visible or near-infrared wavelengths, highextinction ratio filters, and sensitive detectors. Beyond this, the suppression of unwanted fluorescence and background from the waveguide material needs to be improved to match the performance of bulk Raman systems [212].
