Optimization and Neural Network-Based Modelling of Surface Passivation Effectiveness by Hydrogenated Amorphous Silicon for Solar Cell Applications

TL;DR

This paper investigates the material properties of hydrogenated amorphous silicon for solar cell passivation, optimizing deposition parameters, and developing an ANN model to predict carrier lifetime based on these parameters.

Contribution

It introduces a neural network model to accurately predict carrier lifetime from deposition conditions, enhancing the understanding of passivation effectiveness.

Findings

Surface recombination velocity < 50 cm/s with 30 nm a-Si:H(i)

ANN model's prediction varies with hidden layer neurons

Optimal ANN model selected using AIC

Abstract

Intrinsic hydrogenated amorphous silicon films can provide outstanding surface passivation of crystalline silicon wafer surfaces. This quality of Intrinsic hydrogenated amorphous silicon makes it valuable in heterojunction with intrinsic thin layer (HIT) solar cell fabrication. This paper describes the material characteristics and electronic properties of Intrinsic hydrogenated amorphous silicon that affects its passivation quality. A study of passivation quality of intrinsic hydrogenated amorphous silicon layer has been done with respect to deposition parameters in Plasma Enhanced Chemical Vapor Deposition (PECVD), the most commonly used method of its deposition. It was found that very good surface passivation with surface recombination velocity < 50 cm/s can be obtained from thickness of 30 nm of Intrinsic hydrogenated amorphous silicon (a-Si:H(i)), which is better than most other passivation techniques. A mathematical model based on Artificial Neural Network (ANN) is designed to predict the carrier lifetime for a given deposition condition and it is shown that the prediction capability of developed ANN model varies with the number of neurons in the hidden layer using Akaike Information Criterion (AIC), which is a widely accepted model selection method for measuring the validity of nonlinear models.