Prediction of Electronic parameters of Carbon Nanotube random network Field-effect Transistors under liquid gated conditions using a machine learning approach

In this work, we employed a machine learning approach to predict the performance parameters of CNT bundle network field-effect transistors (FETs). The CNT bundle network FETs are complex system not only due to the random nature of the network but also due to the conductivity and work functions of bundles with both metallic and semiconducting tubes. It is difficult to model a whole network of CNT bundles with unknown conductivity and work function. In a recent work we examined 119 devices for investigate the role of CNT junctions in the electrical conduction and gating of CNT network FETs under liquid gated conditions. In this work we used the experimental data to train and predict the performance using the Support vector machine (SVM) technique. The CNT bundle density of the film measured by atomic force microscope images was used as the input parameter and the on-current, off-current and threshold voltage were the target. We were able to estimate the on current with an accuracy better than 90 percentage. However OFF current and threshold voltage prediction has an accuracy about 82 and 77 percent respectively. The 90-percentage accuracy for the on-current prediction can be attributed to the strong dependency of on-current with the CNT bundle density. However, the off-current and threshold voltage are depending on other parameters namely the presence of metallic CNTs within the bundles and the overall variations in the composition of network. Based on the experimental results, despite the unknown variables, we managed to predict the electronic parameters of a CNT network FETs with using the SVM techniques with significantly high accuracy.