Classifiable Limiting Mass Change Detection in a Graphene Resonator Using Applied Machine Learning

Abstract

Nanomechanical resonator devices are widely used as ultrasensitive mass detectors for fundamental studies and practical applications. The resonance frequency of the resonators shifts when a mass is loaded, which is used to estimate the mass. However, the shift signal is often blurred by the thermal noise, which interferes with accurate mass detection. Here, we demonstrate the reduction of the noise interference in mass detection in suspended graphene-based nanomechanical resonators, by using applied machine learning. Featurization is divided into image and sequential datasets, and those datasets are trained and classified using 2D and 1D convolutional neural networks (CNNs). The 2D CNN learning-based classification shows a performance with f1-score over 99% when the resonance frequency shift is more than 2.5% of the amplitude of the thermal noise range.