MRS Meetings and Events


EL14.02.06 2023 MRS Spring Meeting

Accelerated Discovery of Ligand Molecules for Perovskite Nanocrystals via Machine Learning Guided High-Throughput Experimentation

When and Where

Apr 10, 2023
4:00pm - 4:15pm

Moscone West, Level 3, Room 3014



Min A Kim1,Qianxiang Ai2,Sandra Bueno1,Joshua Schrier2,Emory Chan1

Lawrence Berkeley National Laboratory1,Fordham University2


Min A Kim1,Qianxiang Ai2,Sandra Bueno1,Joshua Schrier2,Emory Chan1

Lawrence Berkeley National Laboratory1,Fordham University2
Surface binding ligands are critical component in nanoparticle studies, from growth to post-synthesis treatment process. Especially for particles with low chemical stabilities, ligands could act as a barrier preventing defect formations to prolong their intrinsic properties. Given vast chemical space available for ligand molecules, successful search will likely discover unseen ligands for the desired effects on nanoparticles. However, searching such large and complex pool of molecules requires high experimental cost and time, while still practically impossible to explore every candidate available. We combined high-throughput (HT) experimental workflow with machine learning model to accelerate exploration of ligand molecules. We implemented automated liquid handing robots to screen surface treatment reactions by exposing CsPbBr<sub>3</sub> perovskite nanocrystal solution to a library of ligands across a range of concentrations. We used perovskite nanocrystals as a model system; Our machine learning guided framework can be easily adopted to screen molecules and their surface interactions with other nanoparticles. Our developed workflow successfully screened 500 reactions in one 8-hour workday demonstrating high efficiency of our experimental method. Each reaction was characterized by HT absorption/fluorescence measurements to observe the change in photoluminescence (PL) intensity as they are exposed to various ligand molecules. Utilized optical measurements provided HT screening tool to correlate the ligand environment and luminescence of nanocrystal solution over time. The surface binding strength of ligand molecules were compared based on the relative change in PL intensity. Ligand selection in HT reactions was prioritized by an active learning (AL) approach with uncertainty quantified by twin regressors. Our model successfully found high performance ligand molecules with initial teaching dataset. Identified surface binding ligand molecules enhanced emission efficiency and chemical stability of perovskite nanocrystals. Feature importance analysis on our models helps the future investigations on uncovering fundamental correlations between molecular properties and their effects on the nanocrystal surface.


perovskites | surface chemistry

Symposium Organizers

Udo Bach, Monash University
T. Jesper Jacobsson, Nankai University
Jonathan Scragg, Uppsala Univ
Eva Unger, Lund University

Publishing Alliance

MRS publishes with Springer Nature