[14a-A205-1~10] 23.1 Joint Session N "Informatics"

[14a-A205-2] Features of Generation Points of Dislocation Clusters in Multicrystalline Silicon Ingot based on Transfer Learning of Convolutional Neural Network

〇Hiroaki Kudo¹, Tetsuya Matsumoto¹, Kentaro Kutsukake², Noritaka Usami³ (1.Grad. Sch. of Inf., Nagoya Univ., 2.Center for AIP, RIKEN, 3.Grad. Sch. of Eng., Nagoya Univ.)

Keywords:machine learning, multicrystalline silicon, dislocation clusters

It is important to clarify crystallographic features and underlying physics of generation of dislocation clusters in multicrystalline silicon ingot for further improvement of crystal quality for solar cells. In this paper, we attempted to extract latent features in generation points of dislocation clusters by a machine learning. As results, we obtained results that the predicted images contain grain boundaries with complex structures. It suggest that triple points of grain boundaries with complex structures would be probable generating points of dislocations.