Experimental demonstration of SnO2 nanofiber-based memristors and their data-driven modeling for nanoelectronic applications

This paper demonstrated the fabrication, characterization, datadriven modeling, and practical application of a 1D SnO 2 nano fi ber-based memristor, in which a 1D SnO 2 active layer was sandwiched between silver (Ag) and aluminum (Al) electrodes. This device yielded a very high R OFF : R ON of-10 4 ( I ON : I OFF of-10 5 ) with an excellent activation slope of 10 mV/dec, low set voltage of V SET- 1.14 V and good repeatability. This paper physically explai ned the conduction mechanism in the layered SnO 2 nano fi ber-based memristor. The conductive network was composed of nano fi bers that play a vital role in the memristive action, since more conductive paths could facilitate the hopping of electron carriers. Energy band structures experimentally extracted with the adoption of ultraviolet photoelectron spectroscopy strongly support the claims reported in this paper. An machine learning (ML) - assisted, datadriven model of the fabricated memristor was also developed employing different popular algorithms such as polynomial regression, support vector regression, k nearest neighbors, and arti fi cial neural network (ANN) to model the data of the fabricated device. We have proposed two types of ANN models (type I and type II) algorithms, illustrated with a detailed fl owchart, to model the fabricated memristor. Benchmarking with standard ML techniques shows that the type II ANN algorithm provides the best mean absolute percentage error of 0.0175 with a 98% R 2 score. The proposed data-driven model was further validated with the characterization results of similar new memristors fabricated adopting the same fabrication recipe, which gave satisfactory predictions. Lastly, the ANN type II model was applied to design and implement simple AND & OR logic functionalities adopting the fabricated memristors with expected, near-ideal characteristics.