Predicting the bandgap energy of distorted GaSbxAs1-x and InSbxAs1-x using design of experiment (DoE) and artificial intelligence (AI): A comparative study

Abstract

The band anticrossing (BAC) theory is widely used to model the bandgap energy of GaSbxAs1-x and InSbxAs1-x materials and is based on two effects: impurity-host interaction in the range rich in antimony and arsenic content, and intraband coupling within the conduction and valence bands in the moderate range. In this study, the response surface method (RSM) was applied to the data and an attempt was made to develop an artificial neural network (ANN) based on the Levenberg-Maquardt backpropagation algorithm to predict the bandgap energy in a more precise manner. The predicted results were verified using experimental data. Using R2 and RMSE, we found that the neural network model provided more accurate losses, with correlation coefficients of 99.28% and 97.42% for GaSbAs and InSbAs, respectively. Subsequently, we exploited the predicted values of the bandgap energy to study the effect of the lattice mismatch on the optoelectronic properties of the materials. We have shown that the non-degenerate bandgap energies of the deformed layer associated with heavy holes (hh) and light holes (lh) calculated from the values predicted by the ANN undergo a highly nonlinear variation compared to their BAC and RSM counterparts, and the energy range narrows with increasing Sb(x) content. These materials are of scientific interest for infrared photodetectors, which are used in a wide variety of applications including medical diagnostics, thermal imaging for industrial process control, and military applications.