Experimental dielectric spectroscopy with validation-guided interpretable machine-learning mapping of CuPc:ZnO-embedded Al/p-Si Schottky structures

Conventional dielectric analysis reveals frequency- and voltage-dependent trends in hybrid Schottky structures but does not directly separate operating-condition and composition effects or establish which response channels support reliable interpretation. Admittance/impedance spectroscopy was therefore combined with validation-guided interpretable machine learning to investigate copper phthalocyanine (CuPc)-incorporated ZnO interlayers in Al/p-Si Schottky structures. Devices containing 0, 0.05, and 0.1 wt% CuPc were characterized from 10 kHz to 1 MHz under ± 4 V bias. The low-frequency response was consistent with Maxwell–Wagner–Sillars interfacial polarization, whereas high-frequency dispersive conduction supported possible hopping-assisted charge transport. The maximum real permittivity increased from approximately 1.5 to 3.7 and 4.7 with CuPc incorporation, accompanied by enhanced AC conductivity. Cole–Cole analysis indicated a dominant relaxation process represented by a series resistance and parallel resistance–capacitance branch. Five-fold voltage-block validation retained real permittivity and real electric modulus as the most reliable attribution responses. Shapley additive explanations (SHAP) identified voltage as their dominant control, with permittivity showing greater frequency sensitivity and modulus greater composition sensitivity. CuPc increased the predicted permittivity while reducing the modulus response. The resulting composition–voltage–frequency maps connect polarization, transport, and relaxation behavior and provide an interpretable basis for operating-region design in tunable dielectric structures.