Abstract
Semiconducting two-dimensional (2D) materials have emerged as promising candidates for gas sensors due to their exceptional sensitivity and rapid response/recovery times. However, these sensors often face significant challenges, including baseline drift, nonlinearity, cross-sensitivity to multiple gases, and early response saturation, all of which compromise their accuracy and reliability. Conventional resistive sensing approaches, which rely on a single output signal for gas concentration estimation, fail to capture the complex interactions inherent to 2D materials, such as charge carrier generation, transport, and polarization. This work addresses these limitations by utilizing impedance measurements across multiple frequencies for MoS(2)- and WS(2)-based sensors, coupled with machine learning-assisted data processing for accurate relative humidity (RH) quantification. By leveraging the impedance domain, we effectively mitigated baseline drift over extended periods and identified mutually exclusive phase behavior for the WS(2)-based sensor. The MoS(2)-based sensor exhibited long-term stability, motivating the application of a neural network-based multilayer perceptron (MLP), one-dimensional convolutional network (1D-CNN), and long short-term memory (LSTM) models to interpret multifrequency impedance data for precise RH measurements. Our approach enabled robust humidity sensing over a wide range (0-90%) with significantly faster response and recovery times than commercial sensors. Additionally, the neural network-assisted WS(2) sensor effectively minimized cross-sensitivity between humidity and CO(2). This work showcases the potential of multifrequency impedance-based sensing, combined with machine learning, to overcome the traditional limitations of 2D material-based sensors, offering a pathway toward more reliable, stable, and precise gas-sensing technologies.