Machine Learning-Assisted Optimization of Iodide Electrolytes for Efficient Indoor Dye-Sensitized Solar Cells with Engineered Photoanodes.

Although dye-sensitized solar cells (DSSCs) have lower power conversion efficiencies than other emerging photovoltaic technologies under standard test conditions, they perform remarkably well under low light or artificial lighting. This makes them particularly well suited to indoor photovoltaics (IPVs) and powering Internet of Things (IoT) devices. The most efficient indoor DSSCs rely on copper-based redox shuttles to achieve high photovoltages; however, their long-term stability is still uncertain. In contrast, iodide-based electrolytes are robust under harsh and outdoor conditions, but under indoor illumination, they often deliver limited photovoltage. Here, we propose a strategy that combines an innovative photoanode architecture, designed to suppress charge recombination, with a design of experiments (DoE) approach that is integrated with machine learning (ML). This strategy is intended to optimize iodine-based electrolytes for high-performance IPVs. Support vector machines and Bayesian optimization were employed to accelerate development, achieving efficiencies surpassing 23% at 1000 Lx. As photovoltage is crucial for IoT applications, it was specifically targeted during ML optimization, achieving up to 683 mV at 500 Lx using an iodide/triiodide redox couple. Bayesian optimization also enabled the exploration of electrolyte formulation, achieving 713 mV at 1000 Lx. These results are among the highest reported for iodide/triiodide-based DSSCs under indoor lighting.