Optimization-Based Thickness Estimation and Depth-Resolved FIB-Induced Damage Characterization via Multislice Electron Ptychography
Quantitative electron microscopy requires accurate knowledge of specimen thickness because dynamical scattering strongly affects image contrast and diffraction intensities. In multislice electron ptychography (MEP), specimen thickness is a required input parameter to its forward model, yet the local thickness of a transmission electron microscope (TEM) lamella is often precisely what is unknown. Moreover, thickness is coupled to probe defocus and the number of slices, making its determination from the reconstruction itself nontrivial. Here, we assess whether joint Bayesian optimization of these parameters can provide a physically meaningful estimate of local thickness from four-dimensional scanning transmission electron microscopy (4D-STEM) data. Applied to a wedge-shaped silicon lamella, the optimization reproduced similar local thickness variation measured by electron energy loss spectroscopy (EELS), with a systematic offset of 10-17% relative to the EELS estimates. Using the optimized reconstruction parameters, depth-resolved MEP further separated the crystalline silicon interior from focused ion beam (FIB)-induced amorphous surface layers within the same reconstructed volume. Regions milled at final voltages of 2, 5, 8, and 30 kV yielded amorphous-layer thicknesses of 3.2, 5.0, 7.0, and 26.4 nm, respectively, increasing monotonically with milling voltage and agreeing reasonably with previous cross-sectional measurements. These results show that applying MEP coupled with Bayesian optimization on a single 4D-STEM dataset can provide both local thickness estimates and depth-resolved characterization of FIB-induced damage, offering a route toward more self-consistent, thickness-aware quantitative 4D-STEM analysis.