Surface defect–mobility correlation in graphene FETs unveiled by Raman spectroscopy and machine learning

Correlating surface defect characteristics with charge transport behavior is essential for evaluating the performance of graphene-based electronic devices. Here, we present a spectroscopy-informed machine learning framework that predicts electrical transport properties in graphene field-effect transistors (GFETs) by correlating Raman spectral features with device performance metrics. By extracting peak parameters—including intensity, position, full width at half maximum (FWHM), and area—from the D, G, D′, and 2D bands, a neural network regression model was trained to predict the field-effect mobility and charge-neutral Dirac point (CNDP) of chemically functionalized GFETs. Feature importance and partial dependence analyses revealed that the D′-band FWHM, a sensitive indicator of surface defect density, exhibits the strongest correlation with the predicted mobility values. Furthermore, spatial mapping of predicted mobility over device areas shows a high degree of correspondence with the D’-band width distribution, highlighting the capability of this approach for non-invasive, spatially resolved evaluation of device performance with high sensitivity to interfacial and surface defect variations. This work demonstrates a robust and scalable strategy for linking surface spectral fingerprints with charge transport behavior in 2D material-based transistors.