Machine Learning-aided High-throughput Synthesis and Development of Optoelectronic Materials

In recent years, researchers have increasingly turned to machine learning (ML) techniques to help accelerate research spanning disparate fields. This has largely been motivated by the advent of readily accessible high-performance computing techniques; and equally by the need to decrease the time in which lab-scale products reach the market. The development of metal-halide perovskites (MHPs) in particular can benefit from accelerated techniques: MHPs have emerged as excellent optoelectronic materials, but despite seeing unprecedented research intensity in the last decade, only a small fraction of the available chemical space has been explored. In this thesis I present workflows to accelerate the discovery and understanding of new optoelectronic materials.I begin by developing a ML-accelerated workflow for the synthesis and characterization of perovskite single crystals. I demonstrate that a Protein Crystallization Robot can be modified to prepare autonomously 288 independent perovskite single crystal growths. The robot acquires iii images as the crystal growth proceeds, allowing for direct measurement and classification of each experiment. I then use this method, aided by ML, to discover the growth conditions for a new Cl-based MHP. I characterize this perovskite and find that it emits light in the deep-blue region. Next, I use Quasi-Elastic Neutron Scattering (QENS) to identify the dynamics of the multiple cations in state-of-art perovskite solar cell compositions as a function of bromine incorporation. I find that the suppression of one of the cations, FA, correlates with an increased carrier lifetime. I find that when the fraction of bromine that is incorporated reaches 0.15 – a composition used extensively in literature for single-junction solar cells – the FA rotation is suppressed by more than 25% compared to the pure iodine composition. Lastly, I demonstrate how the high-throughput method I developed can be applied to discover new EO modulating materials. I use the robotic workflow to synthesize high quality perovskite single crystals. I train an ML model to classify the space group of a material from its pXRD spectra, which allows for the rapid evaluation of non-centrosymmetry, a key consideration in EO modulating materials. With this workflow I screen > 30 different ligands for new perovskite single crystals and discover three new noncentrosymmetric perovskites.

Department dc:contributor.department

Electrical and Computer Engineering

Year dc:date.issued

2021

Author dc:creator

• Johnston, Andrew

Advisor dc:contributor.advisor

• Sargent, Edward

• High Throughput Experimentation
• Machine Leaning
• Optoelectronics

Handle dc:identifier.uri

http://hdl.handle.net/1807/108733

OAI identifier oai:identifier

oai:utoronto.scholaris.ca:1807/108733