Accelerating the experimental cycle for new materials development is vital for addressing the grand energy challenges of the 21 st century. We fabricate and characterize 75 unique perovskite-inspired compositions within a 2-month period, with 87% exhibiting band gaps between 1.2 and 2.4 eV, which are of interest for energy-harvesting applications. We utilize a fully connected deep neural network to classify compounds based on experimental X-ray diffraction data into 0D, 2D, and 3D structures, more than 10 times faster than human analysis and with 90% accuracy. We validate our methods using lead-halide perovskites and extend the application to lead-free compositions. The wider synthesis window and faster cycle of learning enables the realization of a multi-site lead-free alloy series, Cs 3 (Bi 1-x Sb x ) 2 (I 1-x Br x ) 9 . We reveal the non-linear band-gap behavior and transition in dimensionality upon simultaneous alloying on the B-site and X-site of Cs 3 Bi 2 I 9 with Sb and Br.

X-ray diffraction (XRD) data acquisition and analysis is among the most time-consuming steps in the development cycle of novel thin-film materials. We propose a machine-learning-enabled approach to predict crystallographic dimensionality and space group from a limited number of thin-film XRD patterns. We overcome the scarce-data problem intrinsic to novel materials development by coupling a supervised machine learning approach with a model-agnostic, physics-informed data augmentation strategy using simulated data from the Inorganic Crystal Structure Database (ICSD) and experimental data. As a test case, 115 thin-film metal-halides spanning 3 dimensionalities and 7 space-groups are synthesized and classified. After testing various algorithms, we develop and implement an all convolutional neural network, with cross-validated accuracies for dimensionality and space-group classification of 93% and 89%, respectively. We propose average class activation maps, computed from a global average pooling layer, to allow high model interpretability by human experimentalists, elucidating the root-causes of misclassification. Finally, we systematically evaluate the maximum XRD pattern step size (data acquisition rate) before loss of predictive accuracy occurs, and determine it to be 0.16° 2, which enables an XRD pattern to be obtained and classified in 5.5 minutes or less.

Improving the heat-moisture-light stability of organic-inorganic perovskites, a widely studied semiconductor material class, is a critical challenge. Compositional search within multinary perovskites employing brute force synthesis followed by environmental tests are prohibitively expensive in large chemical spaces. To identify the most stable multi-cation lead iodide perovskites containing Cs, formamidinium (FA) and methylammonium (MA), we fuse results from density functional theory (DFT) calculations and in situ thin-film degradation test within an end-toend machine learning (ML) algorithm to inform the compositional optimization of CsxMAyFA1-x-yPbI3. We integrate phase thermodynamics modelling as a probabilistic constraint in a Bayesian optimization (BO) loop, which effectively guides the experimental search while considering both structural and environmental stability. After three optimization rounds and only sampling 1.8% of the compositional space, we identify thin-film compositions centred at Cs0.17MA0.03FA0.80PbI3 that achieve a 3x delay in macroscopic degradation onset under elevated temperature, humidity, and light compared with the more complex state-of-the-art Cs0.05(MA0.17FA0.83)0.95Pb(I0.83Br0.17)3. We find up to 8% of MA can be incorporated into the perovskite structure before stability is significantly compromised. Cs is beneficial at low concentrations, however, beyond 17% is found to contribute to reduced stability. Synchrotron-based grazing-incidence wide-angle X-ray scattering (GIWAXS) further validates that the interplay of chemical decomposition and phase separation governs the non-linear instability landscape of this compositional space. We reveal the detrimental role of the ẟ-CsPbI3 minority phase in accelerating degradation and it can be kinetically suppressed by cooptimising Cs and MA content, providing insights into simplifying perovskite compositions for further environmental stability enhancement. Our approach realizes the effectiveness of MLenabled data fusion in achieving a holistic, efficient, and physics-informed experimentation for multinary systems, potentially generalisable to materials search in the vast structural and alloyed spaces beyond halide perovskites.

No abstract

Environmental stability of perovskite solar cells (PSCs) has been improved by trial-and-error exploration of thin low-dimensional (LD) perovskite deposited on top of the perovskite absorber, called the capping layer. In this study, a machine-learning framework is presented to optimize this layer. We featurize 21 organic halide salts, apply them as capping layers onto methylammonium lead iodide (MAPbI3) films, age them under accelerated conditions, and determine features governing stability using supervised machine learning and Shapley values. We find that organic molecules’ low number of hydrogen-bonding donors and small topological polar surface area correlate with increased MAPbI3 film stability. The top performing organic halide, phenyltriethylammonium iodide (PTEAI), successfully extends the MAPbI3 stability lifetime by 4 ± 2 times over bare MAPbI3 and 1.3 ± 0.3 times over state-of-the-art octylammonium bromide (OABr). Through characterization, we find that this capping layer stabilizes the photoactive layer by changing the surface chemistry and suppressing methylammonium loss.