A Strategy for Accurate SERS Gas Detection: Skillful Integration of Mass-Productive Wafer-Scale SERS Substrate and Machine Learning-Assisted Multifeature Profiling

The highly sensitive and precise detection of gaseous molecules continues to pose a significant challenge in surface-enhanced Raman scattering (SERS) to this day. To address this issue, a novel strategy is proposed here, which skillfully integrates mass-productive wafer-scale particle-in-quasi cavity (PIQC) optical structure and machine learning (ML)-assisted multifeature profiling method. This PIQC optical structure not only demonstrates low limit of detection for gaseous molecules (H2S: 10(-10) v/v) but also exhibits high SERS stability (lowest relative standard deviation <10%) and yield (261 chips or 29 sensor matrix for one wafer), enabling it to provide massive high-quality Raman data for ML model (>72000 spectra for one sensor matrix), which coincidentally aligns with the utmost necessity in ML analysis. Benefiting from this, the ML model can conveniently and comprehensively integrate all Raman characteristic peaks of multiplex reporter molecules. In contrast with the common analytical method reliant solely on a single or limited Raman characteristic peaks of a singular reporter molecule, this approach can encompass more vibration information on the targeted gaseous molecule, thereby significantly augmenting the detection precision. Finally, we take H2S detection as an example, comprehensively verifying both the feasibility and advancement of this SERS strategy under various real-life scenarios.