[17p-W9_325-1~11] 16.3 Bulk, thin-film and other silicon-based solar cells

3:30 PM - 3:45 PM JST(6:30 AM - 6:45 AM UTC)Presentation by Applicant for JSAP Young Scientists Presentation Award

[17p-W9_325-8] Improving the performance of multilayer emitter films for silicon heterojunction solar cells using a multistage regression model incorporating physical knowledge

〇(M2)Soma Kondo¹, Yasuyoshi Kurokawa^1,3, Kentaro Kutsukake^1,2, Shion Takeno¹, Ryoji Katsube¹, Noritaka Usami^1,2,3 (1.Grad. Eng. Nagoya Univ., 2.InFuS Nagoya Univ., 3.IMasSS Nagoya Univ.)

Keywords:

solar cells,Machine Learning,silicon

In this study, we propose a cascade-type Gaussian Process Regression (GPR) framework for predicting the performance of multilayer nc-SiOx:H emitter structures for silicon heterojunction (SHJ) solar cells. Instead of directly mapping deposition conditions to device performance, physically meaningful intermediate parameters—such as recombination current density, contact resistivity, and short-circuit current density—are introduced to enable stepwise prediction. Both single-layer and multilayer emitter structures are utilized to increase the effective training dataset and reduce the dimensionality of the input parameter space. By comparing multiple cascade model architectures, we demonstrate that independently estimating intermediate physical parameters and integrating them at the final stage significantly improves prediction accuracy. The results highlight that not only the introduction of physical parameters but also their hierarchical integration plays a critical role in achieving accurate and interpretable performance prediction for multilayer thin-film systems.