Deep Learning Extraction of the Temperature-Dependent Parameters of Bulk Defects.

Area of Science:

• Materials Science
• Semiconductor Physics
• Artificial Intelligence

Background:

• Bulk defects in silicon solar cells significantly reduce efficiency.
• Traditional defect analysis relies on temperature- and injection-dependent lifetime spectroscopy with Shockley-Read-Hall fitting.
• Conventional methods often yield ambiguous results, presenting two possible defect solutions.

Purpose of the Study:

• To introduce a deep learning-based technique for extracting defect parameters in silicon solar cells.
• To address limitations of traditional fitting methods, particularly ambiguity in defect energy spectrum localization.
• To extend defect analysis to temperature-dependent parameters where traditional methods are inapplicable.

Main Methods:

• Utilizing a deep learning approach for defect parameter extraction.
• Employing an alternative representation of lifetime curves: lifetime mapping in temperature and minority carrier concentration space.
• Applying the deep learning model to analyze temperature-dependent defect parameters.

Main Results:

• The deep learning model successfully predicts all defect parameters.
• The method resolves the ambiguity of two possible solutions inherent in traditional approaches.
• Satisfactory prediction of defect parameters is achieved for temperature-dependent analyses.

Conclusions:

• Deep learning applied to lifetime mapping offers a robust alternative for silicon solar cell defect characterization.
• This approach overcomes key limitations of conventional Shockley-Read-Hall fitting methods.
• Image representation and deep learning can significantly enhance solar cell characterization by extracting deeper insights from data.