Optimization of Reliable and Reproducible Growth of Carbon Nanotube Forests and Microstructures based on The Machine Learning-Assisted Control of Catalyst Nanoparticle Morphology

Recently, the realm of carbon nanotube (CNT) applications has been expanding in accordance with the rapid growth of cutting-edge industries including secondary batteries, semiconductors, sensors, and so on. For instance, CNTs are actively used in a powder or solution form as a conductive agent for both the anodes and cathodes of secondary batteries, which, however, typically entails various physical and chemical post-treatment processes. Utilizing the CNT forests and lithographically patterned CNT microstructures in their as-grown forms may obviate the labor for harvesting (off the substrate) and subsequent post-treatment (e.g., solution preparation) and may provide more functional frameworks with controlled 3D morphology, density, and aspect ratio. This can be achieved by optimizing the morphology of catalyst nanoparticles that are converted from the thin-film catalyst layer during the annealing step of the chemical vapor deposition (CVD)-based CNT growth process. Each catalyst nanoparticle offers a growth site of each CNT, suggesting that the catalyst nanoparticle’s size and density control the diameter, density, and degree of alignment of the resultant CNT forests – and thus their physical, chemical, and electrical properties. Here we propose a machine learning-based thermal and fluid interaction analysis for the control of catalyst nanoparticle morphology during the CVD process through thermal energy-driven Ostwald ripening and subsurface diffusion. We use the thermal fluid-structure interaction (TFSI) method to analyze the heat transfer characteristics of the catalyst layer system (Fe/Al₂O₃/SiO₂). We pay more attention to the micropatterned catalyst system which exhibits considerably different heat transfer characteristics depending on the micropattern geometry as compared to the non-patterned thin-film catalyst system. Additionally, we conduct machine learning to optimize the CVD parameters for reliable growth of CNT forests on micropatterned catalyst systems with various geometries. This optimization is based solely on preprocessed SEM images of the catalyst nanoparticles. The morphological data (density, diameter, and roundness) of catalyst nanoparticles are learned, and the global optimum values for CVD parameters - temperature and time during the annealing step, as two factors dominating catalyst nanoparticulation - are proposed. The experimental growth results are presented as a validation. We demonstrate that such a reliable and reproducible CNT growth – either to a forest or microstructure - realized by taking the global optimum values can be utilized in several cutting-edge applications including neural signal recording (using 3D CNT microelectrode arrays) and stretchable strain sensors (using one-directionally aligned CNT thin films engineered from microscale CNT blade patterns).<br/><br/>Acknowledgment<br/>This work was supported by the National Research Foundation of Korea (NRF) grants (No. 2021M3H4A3A02099204, and 2022M3C1A3081178 (Ministry of Science and ICT) and No. 2022R1I1A2073224 (Ministry of Education)) funded by the Korean Government.