Flexible Dual-Modal Sensing Transistor Enabled by Deep Learning Decoupling for Independent Light and Temperature Reconstruction

Abstract Herein, a flexible dual-modal sensing transistor (FDST) is reported, based on zinc oxide nanofibers (ZnO NFs) integrated onto an indium–gallium–zinc–oxide thin-film transistor, and combined with a deep learning-based signal decoupling strategy. Defect-mediated subgap excitation and thermally activated interfacial potential modulation enable high sensitivity dual-modal responses, delivering a broadband photoresponsivity ( $$R$$ R ) up to 2.69 A W −1 and a temperature coefficient ( $${TC}$$ TC ) of 0.071 °C −1 . To enable reliable discrimination and simultaneous reconstruction of light and temperature, a multibias readout physically encodes the coupled stimuli into a high-dimensional current fingerprint, which is decoded by a lightweight multilayer perceptron. This synergistic approach enables accurate and independent reconstruction of light intensity and temperature, achieving coefficients of determination ( R 2 ) around 0.99. The FDST exhibits exceptional mechanical robustness under 10,000 bending cycles and severe bending (2 mm radius). Furthermore, a wearable system based on a low-power microcontroller demonstrates real-time monitoring with negligible cross-interference between optical and thermal modalities under uncontrolled outdoor conditions. This work establishes a general strategy for resolving cross-sensitivity, paving the way for robust and intelligent artificial perception systems.