Frequency-dependent electrical characteristic reconstruction with physics-based evaluation for interface-state analysis in CeO₂NPs-doped GQDs device

Abstract Frequency-dependent capacitance–voltage ( C–V ) and conductance–voltage ( G–V ) measurements provide important information on charge transport and interface properties in semiconductor devices, but dense frequency sweeps increase experimental measurement requirements. This study systematically evaluates frequency-domain reconstruction strategies for a CeO₂NPs-doped PEI N-GQDs/n-Si heterojunction using physics-based post-reconstruction evaluation. Experimental C–V and G–V data acquired at 17 frequencies and 81 bias-voltage points per frequency (1377 measurement points) were analyzed using classical interpolation, ensemble machine-learning, recurrent deep-learning, and analytical equivalent-circuit approaches. Leave-one-frequency-out validation was complemented by sparse-frequency experiments using reduced subsets of 13, 9, 7, and 5 frequencies from the original 17-frequency dataset, separation of interior interpolation from endpoint extrapolation, and repeated stochastic-model evaluation. Reconstruction accuracy was assessed for C–V and G–V characteristics and physically derived R s , G p /ω , and N ss . Classical interpolation remained highly competitive and frequently outperformed machine-learning methods under reduced frequency sampling. With only five measured frequencies, PCHIP achieved R 2 = 0.9959 for G -V reconstruction, while Linear interpolation and Extra Trees achieved R 2 = 0.9869 and 0.9852, respectively, for C-V reconstruction. Sensitivity analysis further showed that reconstruction errors can be amplified during N ss extraction and that insufficient frequency resolution around the G p /ω peak can degrade physical-parameter recovery. These findings demonstrate that reconstruction performance depends on measurement density, interpolation versus extrapolation conditions, and the subsequently derived physical quantity. Accurate reconstruction of C–V and G–V characteristics therefore does not necessarily guarantee equally accurate recovery of interface-related parameters.