A South Korean research team has developed a novel method integrating machine learning into the "intelligent manufacturing" of solar cells. Their constructed prediction model enables precise control of sheet resistance during the phosphorus oxychloride (POCl₃) doping process. The study utilized 3,420 sets of experimental data collected from industrial-grade equipment, covering 10 key variables including pre-deposition temperature/time, drive-in conditions, and process gas parameters. The Bayesian optimization algorithm achieved precise control of the target resistance at 150 Ω/□.

This technical pathway comprises two core modules: First, the SHAP (SHapley Additive exPlanations) method was employed to analyze the influence weight of each process parameter on sheet resistance, with its game-theoretic algorithms quantifying feature importance and predictive contributions. Subsequently, Bayesian optimization replaced traditional trial-and-error methods. Based on the trained machine learning model, it rapidly identified the optimal combination of process parameters within 100 random samples and 100 optimization iterations.

Experiments confirmed that this method significantly accelerates the process compared to existing solutions in the photovoltaic industry, with model predictions showing strong alignment with physical theory. Project leader Lee Seungtae emphasized to *PV Magazine*: "This framework not only advances Industry 4.0 but will also pave the way for Industry 5.0. Its applications can extend to a wide range of industrial scenarios, including semiconductor manufacturing." The findings have been published in the journal *Semiconductor Processing and Materials Science* under the title "Bayesian Optimization-Based Sheet Resistance Control Method for Silicon Wafers Enables Automated Solar Cell Manufacturing."
url:https://www.pv-magazine-latam.com/2025/08/06/como-integrar-el-aprendizaje-automatico-en-la-fabricacion-de-celulas-solares/