Prediction of nano etching parameters of silicon wafer for a better energy absorption with the aid of an artificial neural network

Kayabasi H., Ozturk S., ÇELİK E., KURT H., Arcaldioglu E.

Solar Energy Materials and Solar Cells, vol.188, pp.234-240, 2018 (SCI-Expanded) identifier identifier


To enhance energy absorption of photovoltaics, several etching experiments with various parameters were performed. In addition, an Artificial Neural Network (ANN) simulation was utilized to predict chemical nano etching parameters such as masking and etching durations for Silicon (Si) solar cell applications to reach minimum surface reflectance in an optimum etching duration. Experiments were performed with different masking and etching durations to determine the characteristics of surface reflectance of micro textured n-type single crystalline Si wafers in 25mmx25mm width and 300 gm thickness to provide training data for ANN. For this purpose, solutions with identic properties including Ag nanoparticles were applied with different application durations on the surfaces of n-type single crystalline Si wafers to cover partially the Si surfaces with Ag nano-particles at masking step. After, partially masked Si surfaces were exposed to chemical nano etching to develop nano-sized porous structures under different etching durations in an identic acidic etching solution. For the etching of Si wafers, 32 masking and etching processes were performed. Reflectance measurements and SEM images were evaluated to determine the optimum etching duration resulting the best surface quality with minimum reflectance. In addition, reflectance values were utilized as input data for training, testing and validation steps of developed ANN. In the ANN simulation, 70% of reflectance values were used as training, 15% of reflectance values were used as validation and 15% of reflectance values were used to test data in the ANN. Consequently, surface reflectance values under different masking and etching durations were predicted with the new parameter set by using the trained ANN with a success level above 99%.