Multi-objective Bayesian optimization of carbon nanotube yield and diameter control at synthesis

Carbon nanotube (CNT) synthesis is a ripe area for autonomous experimentation. It is a high-dimensional problem, with both a large number of experimental inputs and critically important outputs, since different single-walled CNT (SWCNT) structures exhibit vastly different properties. In particular, the electronic properties of SWCNTs can vary from metallic to semiconducting based on the diameter and helical twist angle of the structure. Diameter selection has been achieved by post-growth purification at the expense of yield. We show results where both yield and selectivity are increased. This work presents the first multi-objective autonomous experimentation campaign for CNT synthesis, maximizing both overall CNT yield and diameter control. Many attempts at SWCNT diameter control do so via intensive engineering of the catalyst particles: composition, diameter, packing density, etc. Here, we begin with a disordered film of oxidized cobalt on a defective alumina support. By varying the thermodynamic conditions of the synthesis environment, we can control the size distribution and population dynamics of the emergent nanoparticle catalyst bed and, therefore, control the diameters of the resulting CNTs. Using this thermodynamic understanding, we identify regions of experimental space conducive to yield- and diameter control-optimized growth, demonstrating that, for the conditions tested here, these objectives are nearly independent, and with some exception at the extremes, arbitrary degrees of diameter control can be realized for all experimentally achieved yields. In addition, we show that this thermodynamic understanding of CNT catalyst behavior enables effective planning in a vast, information-poor environment, enabling diameter control beyond what might be expected, given the limitations of the in situ diameter characterization methods.