Self-Driving Fluidic Laboratory for Autonomous Development of Metal Halide Perovskite Nanocrystals

Fully-inorganic lead halide perovskite (LHP) nanocrystals (NCs) have recently been demonstrated to outperform conventional II-VI semiconductor NCs in printed photonic devices. The solution-processability and ionic nature of LHP NCs have resulted in intriguing optical and optoelectronic properties, including near-unity photoluminescence quantum yield (PLQY) and narrow emission linewidth. Despite their potential, the widespread adoption of LHP NCs by clean energy technologies have been limited due to the high content of lead (Pb²⁺) ions in the NCs. One potential solution to reduce the high Pb content of LHP NCs is to incorporate metal cation dopants that have similar optoelectronic properties (<i>e.g.</i>, manganese, Mn²⁺) to not only reduce the toxicity level but also introduce new properties into the host LHP NCs. However, the high-dimensional design space of metal cation-doped LHP NCs impedes comprehensive mapping of the colloidal reaction space and search for the best-performing material with desired phase stability and optical properties one-at-a-time trial-and-error experimentation approach.<br/>The recent emergence of artificial intelligence (AI)-assisted experiment-selection strategies provide an accelerated route to explore the synthesis universe of novel advanced functional (nano)materials. However, the conventional batch reactors utilized for synthesis, development, and characterization of LHP NCs are time-, material- and labor-intensive and their process efficiency is limited due to their irreproducible heat/mass transport rates, high precursor consumption/waste generation rates, and lack of proper NC in-situ characterization probes for the real-time and rapid process control and optimization. Microfluidic platforms with their reproducible and intensified heat and mass transfer rates, small reactor footprint, facile in-situ characterization, and reduced reagent consumption/waste generation provide a miniaturized and reliable alternative to batch reactors for integration with AI-assisted experiment selection strategies. The result of such integration is a closed-loop ‘self-driving fluidic lab (SDFL)’.<br/>In this work, we present a modular SDFL for the accelerated development and fundamental mechanistic studies of pristine and metal cation-doped LHP NCs. The SDFL utilizes, for the first time, a two-stage strategy for the sequential halide exchange and cation doping of LHP NCs to autonomously explore the synthesis and doping design space of LHP NCs. The SDFL utilizes active learning to build an accurate “digital twin” of the multi-stage LHP NC synthesis chemistry (prediction accuracy&gt;85%) within a limited experimental budget (60 experiments). We then employ the digital twin to study the fundamental formation and doping mechanisms of LHP NCs and identify the key process parameters controlling the optical/optoelectronic properties of the in-flow synthesized NCs. The SDFL uses the active learning-guided surrogate model to autonomously manufacture metal cation-doped LHP NCs with the targeted optical properties in less than 90 min per target material. The modularity of the developed SFDL makes it uniquely suited for accelerated formulation-synthesis-property relationship mapping required for discovery and development of next-generation of printable clean energy materials.