Exploring optimal pyramid textures using machine learning for high-performance solar cell production

Abstract

The pursuit of increasingly efficient and cost-effective solar energy solutions has driven significant advancements in photovoltaic (PV) technologies over the past decade. Among these innovations, bifacial solar cells, which capture sunlight from both the front and back surfaces, with front surface texturing and rear-side optimization playing crucial roles, present a promising avenue for enhancing efficiency compared to conventional designs. The effectiveness of these cells, however, is largely dependent on the optimization of rear surface properties and the material characteristics employed. This study investigates into the pivotal role of surface texture, particularly on silicon wafers, in shaping key performance metrics such as open-circuit voltage, short-circuit current, fill factor, and overall efficiency. Given the complex interdependencies among these parameters, machine learning (ML) tools, specifically random forest regression models, have been utilized to decode these intricate relationships. The findings underscore the significance of surface texture in modulating reflectance from both the rear and front surfaces, which in turn influences the overall performance of the solar cells. By applying ML models, this research provides an improved understanding of the impact of surface characteristics, thereby offering valuable insights into the optimization of design and material selection for next-generation high-performance solar cells. This ML optimization study indicates that the pyramid structures with a height of 3 μm and a base angle of 62° can significantly reduce reflectance to 9% while maximizing solar cell efficiency to 23.61%, marking a substantial advancement over existing designs. This model achieves 75% accuracy on synthetic test data and 78% on experimental data reinforcing model’s applicability despite typical ML limitations in PV systems.