High-precision machine learning analysis of photoluminescence and I-V measurements in SiNWs-based Schottky diodes

Silicon nanowires (SiNWs) structures were obtained by Ag-assisted chemical etching (Ag-ACE) method for different etching durations (d_etch). Ag/SiNWs form Schottky barriers and are highly sensitive to surface states, which make them excellent for photodetectors Schottky diodes. Mathematical (MT) and decision tree (DT) algorithms were used to predict and analyze the photoluminescence (PL) intensity and current-voltage (I-V) characteristics of Ag/SiNWs/Si junctions as a function of d_etch. The developed models were validated using experimental measurements. A comparative analysis based on statistical evaluation criteria was also carried out. The MT model shows outstanding performance. This model based on a 7th-order Gaussian function was trained to analyze the PL curves at times of 30, 90, and 300 s. It achieves a MSE of 1.47 × 10⁻⁹, a MAE of 2.9 × 10⁻⁵, a RMSE of 3.8 10⁻⁵, and an R² of 0.9988 for d_etch of 300 s. For the prediction of I-V characteristics, the mathematical model based on a 9th generation polynomial function was trained at the same etching times. For d_etch equal to 300 s, this model achieves an MSE of 1.00 10⁻¹⁴, an MAE of 8.34 × 10⁻⁸, an RMSE of 1.00 × 10⁻⁷, and an R² value of 0.9998, demonstrating its high predictive accuracy for this time. The obtained results demonstrate superior capabilities of the mathematical models and highlight the performance of the applicability of machine learning to the validation of opto-electrical experimental data of SiNWs-based Schottky diode.