**5. Summary of Current Technology–Comparison with Refractive Index Sensing**

We summarize the reviewed literature with a technology map (Figure 6). The technology map is a 3D plot, where each data point represents one particular sensing structure, plotted against its LOD, operation wavelength l, and propagation loss, aprop. Different sensing techniques are color-coded to show the general trends for each sensor family. Absorption spectroscopy sensors based on strip or rib waveguides (red) are located in the upper left corner, with LODs above 100 ppm. Lower LODs have been achieved with suspended waveguides (purple) at longer wavelengths, with the most sensitive among them capable of detecting a gas concentration of 7 ppm. Lower LODs are owing to clad systems. Raman measurements (grey) of VOC (liquid at normal standard conditions) present limits of detection two orders of magnitude lower, due to the enrichment properties of the cladding. These devices operate at a shorter wavelength region, closest to the left corner. Subwavelength gratings and photonic crystal-based devices (orange and blue, respectively) were able to detect gases such as NH3 with a LOD down to 0.150 ppm and are located at the back of the technology map due to high losses (non-reported propagation losses were estimated based on the length of the used waveguide).

**Figure 6.** Technology map. A 3D plot of various representative sensing structures, categorized according to their working wavelength l (x-axis), propagation losses αprop (y-axis), and LOD (z-axis). Legend: strip/rib waveguides (strip/rib), suspended waveguides (SW), photonic crystal waveguides (PCW), subwavelength gratings (SWG), clad waveguides used for WERS (WERS).

#### **6. Outlook and Future Perspectives**

With this work, we strove to provide a comprehensive review of on-chip waveguidebased IR-absorption and Raman spectroscopy sensors for gas sensing. We discussed the main components, including new and sophisticated waveguide designs, as well as the latest advances in the domain of spectroscopic sensor integration.

From the IR absorption-sensing perspective, traditional simple rib and strip waveguides have evolved in design and processing into self-standing designs, with a corresponding increase in confinement factor from one digit to more than 100%. This dramatic enhancement in confinement factors, together with transition to MIR wavelengths and compatible nanophotonic components, brought the limit of detection down by several orders of magnitude since the first waveguide-based sensor reports. In return, integrated sensors capable of detecting small gas molecules below 10 ppm were reported (see Table 1). To reduce the limit of detection even further, slotted photonic crystals, capable of reducing the speed of light while maintaining a high air confinement factor, were proposed and fabricated. Nevertheless, the performance is still limited by the high propagation losses of the waveguides, allowing for only centimeter- or, in the case of photonic crystals, millimeter-long pathlengths.

On another front, the use of enrichment cladding that is compatible with integrated waveguide platforms showed promise for enhancing the sensitivity of WIRAS by the selective absorption and up-concentration of volatile analytes. So far, cladding (applied on WIRAS) has been only used to sample solvents in aqueous environments; however, no restrictions exist to use them with gas matrices as long as the cladding remains transparent within the wavelength range of interest. Further development will imply the use of recognition sites in the cladding to boost the specificity and allow for the enrichment of other gases than VOCs. Enrichment remains a challenging topic for small molecules and, particularly, for gases with critical temperatures below room temperature, such as the majority of greenhouse gases including methane and CO2. Some progress has been made in this direction by developing composite cladding, with cage-like molecules as trapping sites, but the specificity, the transparency, and the processing still need to be matched.

Waveguide-enhanced Raman spectroscopy greatly profits from the high intensity of tightly confined guided modes and from Raman signal collection along the entire waveguide length, increasing both the signal-to-noise ratio and the sensitivity. Nevertheless, the limit of detection of air-clad sensors remains around 50,000 ppm in solution and could not be applied to gases. In a quest to improve this figure, plasmonic structures coupled to waveguides for the surface enhancement of Raman scattering (SERS) were tested, but no great improvement in the performance was reported other than partial background suppression. Only the use of enrichment cladding significantly pushed down the limit of detection. Depending on the Raman cross-section values, the analyte and the cladding, the LOD in a solution can drop down to parts-per-billion.

Comparing WIRAS and WERS, the weak Raman cross-section has limited the WERS sensor performance in comparison with absorption spectroscopy. The absence of direct gas sensing with air-clad waveguides in a Raman configuration and orders of magnitude lower detection limits for absorption spectroscopy, even with the use of cladding, are a direct consequence. The performance of waveguides for both techniques is still limited by losses, while Raman sensors additionally require the careful selection of materials to suppress the spurious Raman background. WERS sensors, however, have a certain advantage over WIRAS for the detection of larger gas molecules, molecules in cladding, or in liquid matrices. Namely, WERS offers the spectral coverage needed to capture the broad spectral features of such analytes and maintain the sensor specificity. Another advantage of WERS is its lower sensitivity to water interference, as Raman spectra typically overlap with a water window spanning from 600 to 2600 cm–1, with a minor band at 1600 cm–1 [225]. Moreover, WERS systems operate in VIS-NIR, where the photonic materials and fabrication processes are mature, and traditional cladding materials such as polymers or mesoporous oxide films are transparent. In contrast, in WIRAS, strong MIR water absorption bands must be avoided, e.g., by measurement at low pressure and between water absorption lines, or water needs to be excluded by sample preconditioning or hydrophobic functionalization. WIRAS waveguides are also subject to spurious absorption due to residual OH, NH groups, or organic cladding, which is currently the major factor limiting their performance. Finally, both WIRAS and WERS can handle small sample volumes in combination with microfluidics, which is a major advantage of both techniques compared to the bulk systems.

At present, the miniaturization of WIRAS gas sensors is a large step forward ahead of WERS, with the first systems integrating both laser, waveguide, and detector being successfully demonstrated in both the NIR and MIR. The testing of such sensors in practical applications, followed by commercialization efforts, is underway, and first reports on the deployment of integrated absorption spectroscopic sensors within new platforms such as networks or UAVs will likely emerge within a few years. WERS sensors are, on the other

hand, more suitable for large-molecule detection and operation with enrichment cladding. We expect an increasing number of works exploring chemical-spectroscopic detection using such sensors, eventually translating into commercial applications in chemical, biological, and biomedical research.

**Author Contributions:** All authors contributed to the literature review, writing and editing of the manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was funded by the European Research Council (grant no. 758973) and Tromsø Research Foundation (project ID 17\_SG\_JJ).

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Acknowledgments:** The authors would like to thank Olav G. Hellesø for a careful review of the manuscript.

**Conflicts of Interest:** The authors declare no conflict of interest.

## **References**

