Automated processing of environmental transmission electron microscopy images for quantification of thin film dewetting and carbon nanotube nucleation dynamics
We report scalable production of carbon nanotubes (CNTs) requires catalysts and reaction conditions that provide high nucleation efficiency. In situ characterization methods such as environmental transmission electron microscopy (ETEM) can reveal fundamental mechanisms of synthesis, but to date have primarily provided qualitative observations on small sample sizes. Here, quantitative analysis is performed using high-resolution, high-rate video capture of ETEM experimentation coupled with automated image processing, involving computer vision algorithms and convolutional neural networks. By this approach, we detect distinct nanoparticle formation from an alumina-supported iron thin film and subsequent CNT nucleation from the nanoparticles. The statistical summary of particles in each video shows that, compared to a H 2 -only atmosphere, pretreatment of the catalyst with carbon added to the H 2 atmosphere results in a smaller average particle diameter, a 2-fold increase in particle density (to 5300 particles/μm 2 ), a 3-fold increase in CNT nucleation efficiency (to 92%), and more than a 5-fold increase in CNT density (to 4800/μm 2 ). Addition of carbon during exposure to H 2 is also more effective than NH 3 at dewetting the catalyst film and increasing the CNT nucleation efficiency, in spite of NH 3 being a stronger reducing agent for iron. Insights from this study are applicable to improving CNT yield and productivity in both batch-style and continuous processes.