A diverse portfolio of renewable energy technologies is required to limit global warming to less than 2 ◦C. Of the possible emissions-free options, photovoltaic (PV) technologies can be most widely deployed, given the abundance of the solar resource compared. As with all power generation sources, PV adoption is predicated on the availability of technology solutions that are both inexpensive and highly efficient. One solar cell material, the metal halide perovskites (MHP), may provide the ideal combination, with > 25% efficiency devices within the first decade since their invention fabricated through simple spin coating. Despite the unprecedented rise in MHP performance, stability remains a critical challenge with the most stable devices at the 1-year benchmark compared to the >25-year lifetime of Si-based PV. Further progress concerning enduring power output will require a fundamental understanding of the impact of environmental stressors (light, temperature, bias, oxygen, and water) on the basic physical processes governing solar cell operation. Therefore, my dissertation elucidates the interplay between the ambient environment and MHP composition on both the optical and electrical behavior using in situ methods.

The first part of my thesis elucidates the time-dependent optical and elec- tronic response of different MHP compositions using different in situ microscopy techniques. I capture the transient photovoltage of both Br- and I-containing per- ovskites for different photon energies using heterodyne Kelvin probe measurements. My measurements demonstrate that the voltage rise (light ON) is 104× faster than the subsequent decay (light OFF). Uniquely, the decay time for the residual voltage depends on the excitation wavelength, but only for the MAPbBr3 thin film. Next, I spatially and temporally resolve the relationship between radiative recombination and relative humidity (rH) for multi-cation films. The time-dependent photolumi- nescence (PL) indicates that the Cs-Br ratio impacts the magnitude of light emission hysteresis across an rH cycle. Further, I establish the existence of a repeatable and reversible ≈25× PL gain for multiple moisture cycles up to 70% rH. The second part of my thesis establishes the ability of machine learning (ML) models to predict the time-dependent behavior of perovskite material properties. I collect a comprehensive set of humidity-dependent PL data for both MAPbBr3 and MAPbI3 perovskites. I then use that data to train recurrent neural networks to forecast light emission based on only the recorded rH values. Using Echo State Networks, I achieve a normalized root-mean-squared error of <11% for both compositions for a 12+ h prediction win- dow. Further, I use a Long Short Term Memory network to predict the PL from a degrading sample, achieving <5% error. My in situ measurements and predictive ML models provide a powerful framework for identifying structure-property rela- tionships and can help accelerate the development of long-term stable perovskite materials.