A Robust Neural Network for Extracting Dynamics from Electrostatic Force Microscopy Data.

Area of Science:

• Materials Science
• Physics
• Chemistry

Background:

• Scanning probe microscopy (SPM) methods like time-resolved electrostatic force microscopy (trEFM) offer high-resolution spatiotemporal mapping of dynamic processes.
• Traditional analysis and calibration of trEFM data are often time-consuming and complex.

Purpose of the Study:

• To design and train a regression neural network (NN) for accelerated and simplified extraction of local dynamics from SPM data.
• To develop a cantilever-independent method for analyzing trEFM data, enabling processing of data from different cantilevers.

Main Methods:

• A regression neural network was designed and trained to directly extract local dynamics from SPM data.
• The NN's performance was validated using simulated and real microscopy data, comparing its fidelity to conventional calibration methods.
• SHapley Additive exPlanations (SHAP) were used to evaluate model robustness and interpret feature importance.

Main Results:

• The NN successfully accelerated and simplified the extraction of local dynamics from SPM data.
• The method achieved comparable or superior fidelity to conventional, time-consuming calibration techniques.
• Accurate photoinduced carrier dynamics were extracted for butylammonium lead iodide, a halide perovskite semiconductor.

Conclusions:

• The developed NN provides an efficient and robust tool for analyzing trEFM data, overcoming limitations of conventional methods.
• The cantilever-independent approach broadens the applicability of the NN across different experimental setups.
• This technique facilitates the study of carrier dynamics in perovskite materials for solar cells and quantum light sources.