Machine learning assisted multifrequency AFM: Force model prediction

Abstract

Multifrequency atomic force microscopy (AFM) enhances resolving power, provides extra contrast channels, and is equipped with a formalism to quantify material properties pixel by pixel. On the other hand, multifrequency AFM lacks the ability to extract and examine the profile to validate a given force model while scanning. We propose exploiting data-driven algorithms, i.e., machine learning packages, to predict the optimum force model from the observables of multifrequency AFM pixel by pixel. This approach allows distinguishing between different phenomena and selecting a suitable force model directly from observables. We generate predictive models using simulation data. Finally, the formalism of multifrequency AFM can be employed to analytically recover material properties by inputting the right force model.