Simultaneously achieving high activity and stability is the primary challenge when engineering (electro)­catalysts. Transition metal perovskite oxides are employed as air electrodes for solid-oxide fuel cells and electrolyzers. However, degradation of oxygen exchange kinetics at the solid–gas interface, often linked to alkaline-earth cation segregation and precipitation, limits widespread commercialization. In this work, we systematically investigated the surface degradation mechanism induced by gas-phase impurities in (La0.5Sr0.5)­FeO3−δ (LSF55) thin-film electrodes by varying the concentration of H2O, SO2, and CO2. Degradation of the area-specific resistance in ambient and humidified synthetic air is significantly greater than in dry ambient and dry synthetic air, pointing to the importance of water vapor. Time-resolved, in situ ambient pressure X-ray photoelectron spectroscopy performed in O2 showed that nonbulk Sr is present on the surface before the exposure to water vapor. Upon introduction of water vapor, neither additional Sr segregation nor precipitation driven by water vapor is a necessary condition for degradation. Rather, hydroxylation of the surface induces irreversible and significant degradation. At the same time, we show that Sr migration driven by water vapor is partially reversible. These fundamental insights can be used for the rational design of electrodes with improved catalytic stability.

A semiconductor bridge (SCB) is an ignition device that provides a safe and efficient method widely used in civilian and military fields. The heating process of an SCB under electrical stimulation has a wide range of applications owing to its unique energy release process. However, the temperature variation of an SCB is challenging to obtain, both experimentally because of the rapid reaction on a microscale and with simulation due to its high demand in nonlinear calculations. In this study, we propose deep learning (DL) approach to study the electrothermal-coupled multi-physical heating process of the SCB initiator. We generated training data with multi-physics simulation (MPS), producing surface temperature distributions of SCBs under different voltages. The model was then trained with partial data in this database and evaluated on a separate test set. A generative adversarial network (GAN) with a customized loss function was used for modeling point-wise temperature dynamics. In the test set, our proposed method can predict the temperature distribution of an SCB under different voltages with high accuracy of over 0.9 during the heating process. We reduced the computation time by several orders of magnitude by replacing MPS with a deep neural network.