Abstract Gas sensors play a critical role in safety assurance, environmental monitoring, and health diagnostics, requiring high sensitivity, fast response, and low power consumption—especially in portable applications. This study presents graphene‐based chemiresistive gas sensors fabricated on MEMS microheaters and functionalized with atomically thin layers of vanadium pentoxide or copper‐manganese oxide. In these heterostructures, the metal oxide serves as the gas receptor while graphene functions as the transducer. Operated in a pulsed heating mode (115–205 °C for 0.05–1 s every 10 s), the sensors demonstrated ultra‐low power consumption ranging from 13 to 520 µW. Ammonia (NH 3 ), a hazardous industrial gas and a biomarker in exhaled breath, is used as the target analyte. Transient conductance profiles at 4–32 ppm NH 3 are analyzed using machine learning. Feature extraction via discrete Fourier transform and prediction using a compact neural network enables NH 3 concentration estimation within 10–20 s, achieving a mean absolute error below 1% (or below 0.1 ppm at low concentrations). Despite the raw signal's sensitivity to relative humidity (RH), the model accurately predicts NH 3 concentrations without RH data. The highest accuracy and humidity robustness are achieved using signals from two sensors with different oxide coatings.
Vafaei et al. (2025) studied this question.