Machine Learning Guided Device-Level Design for High-Efficiency Tunnel Oxide Passivating Contact Solar Cells

Area of Science:

• Materials Science
• Renewable Energy
• Semiconductor Physics

Background:

• Tunnel Oxide Passivated Contact (TOPCon) solar cells are dominant in crystalline silicon photovoltaics due to excellent passivation and contact properties.
• Further efficiency improvements in TOPCon solar cells require comprehensive optimization beyond traditional methods.

Purpose of the Study:

• To develop a data-driven methodology for optimizing n-type TOPCon solar cells by overcoming experience-driven limitations.
• To identify critical physical and photovoltaic parameters influencing solar cell efficiency and establish coordination principles for design.

Main Methods:

• A dual-directional framework combining forward prediction and reverse optimization with numerical simulations was employed.
• A multidimensional database of 13,000 datasets, including 13 physical and 4 PV parameters, was constructed.
• Ensemble learning for a high-accuracy regression model and genetic algorithm-based optimization were utilized, alongside SHapley Additive exPlanations (SHAP) for mechanism analysis.

Main Results:

• Data-driven analysis confirmed that excessive tunnel oxide thickness and interface recombination degrade efficiency.
• An optimal parameter combination was identified, predicting a solar cell efficiency of 28.10%.
• SHAP analysis revealed front emitter parameters as most impactful, with poor coordination causing low efficiency and complex front-rear interactions.

Conclusions:

• A scalable, data-driven methodology for TOPCon solar cell optimization was established.
• Key insights into parameter coordination, such as front-heavy and rear-heavy doping strategies, were uncovered.
• The findings provide essential principles for future photovoltaic device design and efficiency enhancement.