Metal halide perovskites are highly promising for photovoltaics, yet their complex physics makes comprehensive device characterization challenging. Here, we introduce a machine learning approach that enables holistic, high-precision characterization of perovskite solar cells from an impedance (EIS) measurement. By training a Random Forests Median Leaf Value model on over 50,000 drift-diffusion simulations, our model simultaneously predicts 12 critical device parameters, including ion dynamics and densities, carrier mobilities, and recombination properties, with high accuracy (R2 > 0.99). This approach advances prior work by extracting a more complete parameter set while alleviating the nonuniqueness problem through information-rich EIS input features. We rigorously validate the model against experimental values and closed-loop verification. Applying this technique to triple-cation perovskites with varying compositions reveals intriguing underlying trends, demonstrating its power to uncover complex structure–property relationships. This work establishes ML-driven frameworks for rapid, reliable device analysis to accelerate photovoltaic optimization.
Diekmann et al. (Mon,) studied this question.