METHOD FOR FAST IDENTIFICATION OF ORIENTATION PARAMETERS IN MULTICRYSTALLINE SILICON
S. M. Peshcherova1, E. A. Osipova2, A. G. Chueshova1, S. S. Kolesnikov2, M. Yu. Rybyakov1, A. A. Kuznetsov2, V. L. Arshinsky2
1Vinogradov Institute of Geochemistry, Siberian Branch, Russian Academy of Sciences, Irkutsk, Russia
2Irkutsk National Research Technical University, Irkutsk, Russia
Keywords: multicrystalline silicon, grain orientation parameters, neural networks, machine learning, SiView algorithm, backscattered electron diffraction
Abstract
This work demonstrates that digital technologies can be successfully applied to image analysis and prediction of the properties of functional materials. As an example, a new method is used to rapidly identify the crystallographic orientation parameters in multicrystalline silicon. The proposed method is based on machine learning technologies. The analysis of textured multicrystalline silicon wafers is carried out using the original single-crystal grain clustering algorithm, and the crystallographic orientation parameters are identified using a neural network model. The principle of identification is based on the correlation of the contrast of the macrostructure display associated with the reflective features of the grains and their orientation parameters. The architecture of the neural network - a multilayer perceptron - is chosen taking into account the restrictions on the number of input data. However, in conjunction with the algorithm, the optimal amount of training data satisfies the requirements of the neural network training process and ensures high efficiency in identifying orientation parameters on scanned images of textured multicrystalline silicon wafers.
|
