(339d) An Autonomous, Closed Loop Research System for Scalable Carbon Nanotube Synthesis

Autonomous experimentation systems are becoming increasingly prevalent in the field of materials science as a means to accelerate discovery and development of new materials. Such systems combine three key capabilities: robotic execution of experiments, in situ or in line analysis techniques, and the use of machine learning to plan follow-up experiments based on the collected data. By iterating through this experiment-analyze-plan cycle, autonomous research systems can pursue a specified goal an order of magnitude more efficiently than traditional design of experiment techniques, as benchmarked by the number of required experiments, while also consuming less material and researcher time.[1] These productivity gains have been demonstrated across diverse materials science problems, ranging from carbon nanotube (CNT) synthesis to topological optimization of 3D printed components, to the exploration of material phase transitions.[2,3,4]

Previous closed-loop research on CNT synthesis from our group has used the autonomous research system (ARES) to investigate on the surface growth of CNT in a cold-wall chemical vapor deposition (CVD) reactor using laser heating.[2,5,6] While this system has provided valuable insight into factors affecting the growth rate of CNTs, the results are not directly applicable to the scale of up controlled CNT production to commercially viable levels. To address this shortcoming we demonstrate the operation of a new ARES systems based around the floating catalyst chemical vapor deposition (FCCVD) process for CNT synthesis. FCCVD is a substrateless process that can be operated continuously and is thus a promising route for the scaled production of high-quality single or few-walled CNTs.[7]

This floating catalyst ARES (FC-ARES) consists of a benchtop FCCVD reactor with computer control of reactor temperature and the flow rates of up to three gas phase and four liquid phase feedstock species. CNT samples are collected on a reel-to-reel tape at the reactor outlet.[8] Samples deposited on this tape are advanced through one or more analysis stations to collect experimental metrics such as the presence and distribution of CNT radial breathing modes or the yield of the experiment. Experimental control, data collection, and experimental planning are centralized using our open-source ARES OSâ„¢ software. Building upon our previous work in the in cold-wall CVD reactor ARES, we utilize a machine learning planner based on the expected improvement decision policy to demonstrate the usefulness of the FC-ARES in optimizing both CNT yield and controlling SWCNT diameter.[5] Additionally, we compare the results of the autonomous experiments to traditionally planned experimental campaigns as a benchmark.

References

• Stach, E. et al. Matter 4, 2702–2726 (2021).
• Nikolaev, P. et al. npj Comput Mater 2, 16031 (2016).
• Gongora, A. E. et al. Science Advances 6, eaaz1708 (2020).
• Kusne, A. G. et al. Nat Commun 11, 5966 (2020).
• Rao, R. et al. npj Comput Mater 7, 157 (2021).
• Waelder, R. et al. Carbon (In Review)
• Weller, L. et al. Carbon 146, 789–812 (2019).
• Rao, R., and Maruyama, B. S. Patent No. 10,994,990. (2021).