A novel optically gated thin-film transistor sensor for real-time chemical differentiation using machine-learning analysis.

We present an optically gated thin-film transistor sensor that distinguishes chemicals through illumination-induced transient electrical responses. The device consists of a p-type silicon substrate with native oxide and an amorphous Ge2Se3 photogating layer, and operates without reference electrodes or surface functionalization. Structured light-pulse sequences produce time-dependent electrical signatures whose features differ across chemical environments. These transient responses are analyzed using supervised machine-learning models to demonstrate real-time chemical differentiation in controlled solvent-prepared samples. As a demonstration, we classify three alcohols (methanol, ethanol, and isopropanol) and three perfluoroalkyl substances (perfluorooctane, perfluoropentanoic acid, and perfluoropropionic acid), each measured at 1 ppm in methanol. Waveform-derived features enable machine-learning classification with accuracies of 95-99% and F1 scores of 0.85-0.96, while a total organic fluorine classifier achieves 94% accuracy. These results show that a simple optically gated Si-based transistor can be used as a compact and generalizable sensor architecture for real-time chemical identification.