D. Forchheimer
Intermodulation Products AB,
Sweden
Keywords: atomic force microscopy, AFM, machine learning
Summary:
Conventional single-frequency Atomic Force Microscopy imaging modes such as Tapping Mode or frequency modulated AFM (FM-AFM) operate using the resonance frequency of the AFM cantilever and therefor offer a very sensitive interaction with the surface. The resonance however also works as a frequency filter, making the AFM blind to response at higher frequencies. This is fine for measuring surface topography, but not suitable when it comes to detecting other material properties that can vary across the surface such as elasticity, visco-elasticity, adhesion etc. Those material properties will be “encoded" in the higher harmonics of the AFM signal which the conventional modes can not detect. Intermodulation AFM (ImAFM) combines the high sensitivity of the cantilever-resonance with the material characterization properties of the higher harmonics by using a multi-frequency drive and detection scheme. The cantilever is operated in such a way that the information that exist in the higher harmonics is "folded down" to appear in a comb of frequencies near resonance, where it can be measured with good signal-to-noise ratio. With ImAFM one can routinely measure amplitude and phase of over 20 frequencies above the noise floor, each of these amplitude and phase signals being delicate probes of various aspects of the the surface mechanical response. These new big datasets allow and even necessitate new methods for data analysis. We present methods inspired by big data and machine learning, as well as methods rooted in understanding of the physics and material science at hand to attempt at “decoding” the additional data obtained with ImAFM for improved contrast and insight into the studied surfaces.