Source-linked AI summary
Accelerating Photovoltaic Materials Development via High-Throughput Experiments and Machine-Learning-Assisted Diagnosis
Shijing Sun, Noor T. P. Hartono, Zekun D. Ren, Felipe Oviedo, Antonio M. Buscemi, Mariya Layurova, De Xin Chen, Tofunmi Ogunfunmi, Janak Thapa, Savitha Ramasamy, Charles Settens, Brian L. DeCost, Aaron Gilad Kusne, Zhe Liu, Siyu I. P. Tian, I. Marius Peters, Juan-Pablo Correa-Baena, Tonio Buonassisi
TL;DR
Rapid materials development requires broader synthesis and faster structural characterization. This study combines a flexible solution-synthesis platform with neural-network-assisted X-ray diffraction analysis to accelerate perovskite-inspired materials development, producing tunable thin films and faster dimensionality classification.
Problem
Structural characterization can limit the learning rate per sample, while precursor solubility constrains broad solution-based composition screening.
Method
The study streamlines solution-based thin-film synthesis and trains a deep feedforward neural network on X-ray diffraction data to classify materials by crystallographic dimensionality.
Results
90% accuracy was achieved for dimensionality classification, while 75 materials enabled bandgap tuning between 1.2 and 2.4 eV and four A3B2Br9 thin films were realized.
Takeaways & Limitations
The combined workflow supports rapid exploration of Bi/Sb halide compositions and machine-learning-assisted structural diagnosis, including multi-site alloying and inorganic layered thin films.
Abstract
from arXiv · showhide
Accelerating the experimental cycle for new materials development is vital for addressing the grand energy challenges of the 21st century. We fabricate and characterize 75 unique halide perovskite-inspired solution-based thin-film materials within a two-month period, with 87% exhibiting band gaps between 1.2 eV and 2.4 eV that are of interest for energy-harvesting applications. This increased throughput is enabled by streamlining experimental workflows, developing a set of precursors amenable to high-throughput synthesis, and developing machine-learning assisted diagnosis. We utilize a 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 novel lead-free compositions. The wider synthesis window and faster cycle of learning enables three noteworthy scientific findings: (1) we realize four inorganic layered perovskites, A3B2Br9 (A = Cs, Rb; B = Bi, Sb) in thin-film form via one-step liquid deposition; (2) we report a multi-site lead-free alloy series that was not previously described in literature, Cs3(Bi1-xSbx)2(I1-xBrx)9; and (3) we reveal the effect on bandgap (reduction to <2 eV) and structure upon simultaneous alloying on the B-site and X-site of Cs3Bi2I9 with Sb and Br. This study demonstrates that combining an accelerated experimental cycle of learning and machine-learning based diagnosis represents an important step toward realizing fully-automated laboratories for materials discovery and development.
I. Workflow Optimization and Precursor Development for a Robust, Flexible Synthesis Platform
The study combines workflow optimization, flexible precursor chemistry, and machine-learning diagnosis to accelerate thin-film perovskite materials development. This platform produced 75 crystalline films, characterized their optical properties, and identified structures across multiple dimensionalities.
- Workflow optimization: 35 high-quality compounds per month was the workflow target, alongside 150 samples per month in the synthesis loop.The throughput target was designed to maximize usable information per unit time and was reported as 35× faster than previous laboratory baselines.
- Flexible synthesis platform: 96 precursor compositions were attempted, yielding 75 crystalline thin-films across Pb-, Sn-, Bi/Sb-, and Cu/Ag/Na-rich material classes.Twenty-one precursor solutions were discarded during deposition, mainly because of low reactant solubility or poor film formation.
- Optical characterization: 65 of 75 measured materials showed direct-bandgap estimates between 1.2 and 2.4 eV, a range relevant to energy-harvesting applications.Bandgaps were deduced from Tauc plots using measurements of freshly made films, with both direct and indirect assumptions evaluated.
- Scientific findings: The platform enabled A3B2Br9 thin films, multi-site Bi/Sb-I/Br alloys, and bandgap tuning from 1.2 to 2.4 eV through compositional engineering.The reported bromide thin films included A = Cs, Rb and B = Bi, Sb, while the alloy series included Cs3(Bi1-xSbx)2(I1-xBrx)9.
III. New, tunable materials in thin-film form
High-throughput synthesis produced previously unreported thin-film perovskites and alloys, while machine-learning diagnostics linked composition-dependent structural transitions to tunable bandgaps. The dual-site alloy series combines a 0D-to-2D transition with anomalous bandgap behavior, including values below 2 eV.
- Twenty Br alloys were synthesized, more than doubling known thin-film all-inorganic Br-based perovskite-inspired materials.
- Four A3B2Br9 compounds were realized as compact layered thin films by one-step solution synthesis without re-dissolving product nanocrystals.The compounds were A = Cs or Rb and B = Bi or Sb.
- The new Cs3(Bi1-xSbx)2(I1-xBrx)9 series exhibits a 0D-to-2D structural transition and non-linear bandgap tunability.The series spans x = 0.1–0.9.
- Machine-learning diagnostics place the crystallographic dimensionality change at 10–20% SbBr3 doping.The classification was applied alongside X-ray diffraction and optical measurements.
- At 20% SbBr3 doping, the bandgap decreases to 1.9 eV assuming an indirect bandgap, below reported values for Cs3Bi2I9 and Cs3Sb2I9.The trend at this composition does not depend on assuming a direct or indirect bandgap during absorptance fitting.
- The accelerated platform investigated 75 unique compounds in two months, with 87% of films spanning the 1.2–2.4 eV bandgap range.The range is described as promising for opto-electronic applications.
- The combined throughput and statistical diagnostics enabled examination of nonintuitive structure-property relationships in a multi-parameter composition space.
I. Workflow Quantification and Optimization
The study quantifies and streamlines a solution-based thin-film workflow across multiple perovskite classes, while using automated analysis to address structural characterization bottlenecks. The optimized process supports broad materials exploration and a future laboratory workflow incorporating robotics, machine learning, and artificial intelligence.
- Workflow optimization: Workflow optimization reduced per-sample fabrication time through equipment investment, parallel sample fabrication, and process-level optimization.The reduction was measured over the laboratory’s previous four years.
- Future laboratory: The optimized workflow was designed as a pathway toward laboratories incorporating robotics, machine learning, and artificial intelligence for experiments and diagnostics.The paper presents this as a future laboratory vision rather than a fully automated implementation in the study.
- Workflow optimization: Structural characterization was identified as a rate-determining step in materials discovery before device fabrication decisions.The time breakdown was quantified during a 260-working-hour testing period over two months.
- Synthesis platform: A singular solution-synthesis environment was used to access Pb-, Sn-, Bi-, Sb-, Ag-, Cu-, and Na-rich material classes.The study describes these classes as spanning established thin-film systems, bulk-explored materials, and lesser-explored compositions.
- Data organization: The study’s processing and characterization results were organized for 75 thin-film samples with structural and optical properties and processing conditions reported separately.The associated supporting tables provide summaries of materials and processing conditions.