Hydrogen plasma-functionalized graphene–silicon photodetectors: enhanced photoresponse and machine-learning-assisted performance prediction

Graphene’s broadband absorption and ultrafast carrier dynamics make it a strong candidate for next-generation photodetectors; however, its lack of an intrinsic bandgap limits photosensitivity. Here, a graphene–silicon heterostructure combined with electronic and interfacial modification via short-time hydrogen-plasma functionalization is employed to address this limitation. The results show a significant increase in photocurrent, particularly at shorter wavelengths. A short-time (15 s) functionalization of monolayer chemical vapour deposition (CVD) graphene produced measurable structural and electronic modifications, as evidenced by Raman spectroscopy and direct ATR-FTIR signatures of C–H bond formation (sp 3 C–H stretching bands at ∼2830 and 2900 cm −1 ), together with the accompanying optoelectronic response, while preserving the monolayer structure, reduced surface residues, reduced trap states, and significantly lowered the Schottky barrier. These effects collectively enhanced carrier separation, charge transfer, and carrier lifetime, resulting in a substantial photocurrent increase. Optoelectronic measurements under visible laser irradiation revealed an average photocurrent enhancement of approximately 4.46× across seven devices, with individual enhancements ranging from 3.0- to 4.8-fold, along with improved responsivity, and detectivity; For example, the responsivity in the diode configuration increased from 0.232 to 0.804 A/W, corresponding to an approximately 3.5-fold enhancement. The intrinsic response time, estimated from the low-frequency limit of chopper measurements, was 90 μs, consistent with reduced interfacial trapping after functionalization. A Random Forest model was further developed to predict photocurrent beyond the measured conditions by learning nonlinear relationships among bias voltage, laser power, wavelength, functionalization state, and contact type. This data-driven approach enables rapid optimization of device performance without extensive experimental iterations.