Machine Learning, Artificial Intelligence and High-throughput Experimentation for Material Discovery of Hybrid Organic Inorganic Perovskites
Date Issued
May 2026
Author(s)
Sanchez, Sheryl Lea
The University of Tennessee
Advisor(s)
Mahshid Ahmadi
Sergei Kalinin
Additional Advisor(s)
Claudia Rawn
Yongtao Liu
Abstract
This dissertation establishes a closed-loop, artificial intelligence–driven framework for accelerated discovery and optimization of hybrid organic–inorganic perovskites through the integration of automated experimentation, probabilistic modeling, and autonomous decision making. The central objective is to transform conventional trial-and-error materials development into an adaptive self-driving laboratory capable of navigating complex chemical design spaces spanning ligands, organic spacers, and compositional engineering across multiple perovskite dimensionalities.
The experimental foundation combines robotic high-throughput synthesis with multimodal characterization to generate structured datasets describing crystallization dynamics, morphology evolution, optical response, and phase stability. In CsPbBr₃ nanocrystals synthesized through ligand-assisted reprecipitation (LARP), systematic variation of ligand chain length, acid–amine ratios, and solvent environments reveals how interfacial chemistry governs nucleation pathways, morphology control, and defect formation. Machine learning interpretability approaches quantify parameter importance, providing mechanistic insight into ligand-mediated phase transitions and photoluminescence stability while identifying synthesis regimes that suppress non-perovskite phases.
A central contribution is the development of autonomous artificial intelligence agents that guide experimentation through Monte Carlo decision tree strategies. These agents evaluate prediction uncertainty, expected information gain, and experimental constraints to select subsequent synthesis conditions in real time. Rather than performing static optimization, the system continuously balances exploration of unexplored chemical regions with exploitation of promising candidates, forming a dynamic feedback loop linking synthesis, characterization, and model refinement. Large language models further augment this process through automated literature interpretation and hypothesis generation, enabling contextual reasoning that complements statistical learning.
The closed-loop methodology is extended to layered perovskites through AI-assisted organic spacer discovery, including FuDMA-based Dion–Jacobson systems designed to stabilize the photoactive α-FAPbI₃ phase. Structural and optical characterization demonstrate that spacer chemistry modulates lattice strain, dimensional coupling, and crystallization pathways, improving phase persistence under ambient conditions. Complementary studies employ structured Gaussian process models and latent spectral representations to capture nonlinear bandgap behavior and time-dependent degradation across mixed perovskite compositions.
Together, this work demonstrates a scalable paradigm for materials discovery in which AI agents, probabilistic modeling, and robotic experimentation operate cooperatively to generate experimentally validated insight, enabling rational ligand and spacer design for stable optoelectronic perovskite materials.
Degree
Doctor of Philosophy
Major
Material Science and Engineering
Embargo Date
May 1, 2027