Support Vector Machine Classification of Single Walled Carbon Nanotube Growth Parameters. Issue 1700 (2014)

• Record Type: Journal Article
• Authors: Westing, N.; Clark, J.; Hooper, D.; Nikolaev, P.; Maruyama, B.
• Editors: Futaba, D.; Loiseau, A.; Yap, Y.K.; Zheng, M.
• Is Part Of: MRS proceedings. Issue 1700:(2014)
• Page Start: 3
• Page End: 8
• Publication Date: 2014
• Subjects: chemical vapor deposition (CVD) (chemical reaction), -- nanostructure, -- nucleation & growth; Electrical engineering -- Congresses; Physics -- Congresses; Materials -- Research -- Congresses; Materials science -- Congresses; 620.11
• DOI: 10.1557/opl.2014.604
• Languages: English
• ISSNs: 0272-9172

Abstract

ABSTRACT: Selective single-walled carbon nanotube (SWNT) growth is a challenging problem, limiting their use in a wide variety of applications. Significant degrees of freedom in these experiments may lead to synthesis of multi-walled carbon nanotubes (MWNTs), which are less preferred. Thus, a method for constraining the synthesis results to only SWNTs is desired. A machine learning based approach for selectively growing SWNTs using a laser-induced chemical vapor deposition growth system is introduced. This approach models the complex relationships between the associated synthesis parameters to predict SWNT growth. The parameters under consideration include argon, ethylene, hydrogen and carbon dioxide partial pressures, growth temperature, and water vapor concentration. The catalyst consists of 10 nm of alumina and 1 nm of nickel deposited onto 10 µm diameter silicon pillars with a height of 10 µm. Determination of SWNT growth is performed through in-situ Raman spectroscopy using a 532 nm excitation laser. A total of 121 experiments are used to train a SWNT vs. MWNT classifier with a resulting model accuracy of 94.21%. The classifier model is applied to a range of simulated inputs, and the subset of these inputs that meet a >90% probability of SWNT growth are investigated further. The simulated inputs consist of 531, 201, 645 unique growth parameter combinations spanning the entire parameter space. A reduced dataset of 449, 117 growth parameter combinations define 90% probability of SWNT growth according to the model. Randomly selected input parameters from this reduced dataset were tested experimentally, resulting in SWNT growth for all performed experiments validating the classifier model. This approach maps input growth conditions to SWNT growth selectivity using a limited set of experimental data and allows for further investigation into SWNT growth rates and chiral dependencies.