Artificial neural networks in the analysis of the fine structure of the SWCNT Raman G-band

Although the diameter dispersion of the phonons composing the Raman G‐band of single wall carbon nanotubes (SWCNTs) is well understood theoretically, systematic experimental studies on the subject are scarce. We investigated 6 different diameter samples between d=1.05−1.57 nm with several excitation lasers and used artificial neural networks (ANN) to explore if there is a connection between the fine structure of the G‐band and the sample diameter. An initial screening by a Kohonen self‐organizing map revealed that ANN technology is able to identify spectra measured on the same sample. Based on this result several supervised learning algorithms were tested and finally we succeeded in building a resilient propagation ANN with one hidden layer which is able to predict the diameter distribution of a macroscopic SWCNT sample from the structure of its Raman G‐band with acceptable accuracy. We suggest that with more extensive calibration this method could be developed into a useful auxiliary technique of SWCNT characterization.