Mid-Infrared Gas Classification Using a Bound State in the Continuum Metasurface and Machine Learning

Mid-infrared (mid-IR) spectroscopy enables nondestructive and real-time chemical identification. Emerging demand for in situ sensing has motivated the development of novel sensing technologies that minimize size, weight, power, and cost. These technologies include mid-IR microspectrometers that interface a detector array with an array of discrete spectral filters. This specific approach is compatible with many different filter technologies that provide distinct tradeoffs in spectral selectivity, fabricability, and optical efficiency. A family of all-dielectric, single-layer, coupled waveguide-grating structures excels in all these aspects but has not yet been considered. These filters permit spectrally isolated, linewidth-tunable transmission features via quasi-bound state in the continuum resonances. Here, we study a filter-array detector-array spectrometer to exploit these filters, for the first time to the best of our knowledge. Through simulations that incorporate a machine learning classifier (MLC), we predict accurate classification of common acyclic hydrocarbons down to concentrations of 75 ppm using a 10-cm optical path length, as well as a single-gas limit of detection (LoD) of 32 ppm.