Amanda Volk1,Robert Epps1,Kristofer Reyes2,Milad Abolhasani1
North Carolina State University1,University of Buffalo2
Amanda Volk1,Robert Epps1,Kristofer Reyes2,Milad Abolhasani1
North Carolina State University1,University of Buffalo2
Colloidal atomic layer deposition (cALD) is a versatile technique for the solution-phase, room temperature, formation of complex heterostructure nanoparticles.<sup>[1]</sup> In cALD, sequences of independent, self-limiting, half reactions are used to grow multi-compositional shells with monolayer precision. Despite the limited number of existing studies, the versatility of this method has allowed for the formation of a diverse set of complex nanostructures, including Au/CdS, Ag/AlO<sub>X</sub>, CeO<sub>2</sub>/AlO<sub>X</sub>, SiO<sub>2</sub>/AlO<sub>X</sub>, SiO<sub>2</sub>/AlPO<sub>4</sub>, CdSe/CdS, CdSe/ZnS, PbS/CdS, HgSe/CdSe, HgSe/CdS, HgSe/CdSe, and CsPbX<sub>3</sub>/AlO<sub>X</sub>.<sup>[2]</sup> However, a detailed study of this powerful synthetic route remains challenging due to a variety of factors. Primarily, the multistep and dynamic nature of cALD creates an exponentially growing parameter space. In cALD processes which routinely require over 50 sequential reaction steps, the range of tunable conditions becomes impractically large (> 10<sup>100</sup>). Furthermore, optimized cALD reaction conditions at each cycle depend on material properties which cannot easily be measured, such as surface ligand coverage. Experimental designs to elucidate the role of such hidden states, because of an extremely large parameter space, are nearly infeasible as well as highly resource intensive.<br/>In response to the presented challenges, in this work, we designed and developed an autonomous robotic experimentation strategy for accelerated studies of cALD by integrating a modular microfluidic platform with machine learning (ML)-assisted experimental planning.<sup>[3]</sup> Using shelling of cadmium sulfide (CdS) onto cadmium selenide (CdSe) quantum dots (QDs) as a case study, we developed a material-efficient microfluidic platform to automatically investigate cALD cycles in sequence, using only a 10 mL biphasic droplet. The developed autonomous robotic experimentation platform is fully automated and modular in design and is capable of on-demand reagent addition, controlled mixing and reaction, online multimodal spectral monitoring, and in-line phase separation. Using the autonomous cALD platform, reagent addition sequences and purification protocols could be varied without hardware reconfiguration. Reaction progress was monitored through a UV-Vis absorption and photoluminescence flow cell positioned at the end of the reactor. The developed autonomous cALD platform could routinely carry out over 1,000 successful reaction step instructions without user intervention.<br/>To demonstrate machine-driven knowledge discovery of the cALD protocol, we directed the ML algorithms to explore different precursor injection and washing sequences up to 20 total injections per starting reactive precursors. After running an exploratory design space search policy over more than 1,000 total injections, we exploited the ML models to identify the optimized injection sequence. Without any prior knowledge of conventional cALD reaction protocols, the integrated system successfully identified an alternative cycled injection sequence substructure that achieved a larger shift in the first absorption peak wavelength and higher photoluminescence intensity than conventional methods (new knowledge generation). The developed intelligent fluidic robot is the first autonomous system to solve a high-dimensional multi-step material synthesis problem in colloidal nanoscience without user guidance and using only in-house generated data. Further implementation of the technologies developed here can lead to more efficient, machine-driven studies of dynamic, highly complex, multi-step, reactive systems.<br/>[1] <i>J. Am. Chem. Soc.</i> <b>2012</b>, 134, 45, 18585–18590<br/>[2] <i>ACS Materials Lett.</i> <b>2020</b>, 2, 9, 1182–1202<br/>[3] <i>Adv. Mater.</i> <b>2021</b>, 33, 2004495