Optimization-Based Thickness Estimation and Depth-Resolved FIB-Induced Damage Characterization via Multislice Electron Ptychography
This paper demonstrates that joint Bayesian optimization of specimen thickness and other parameters within multislice electron ptychography enables accurate, self-consistent estimation of local lamella thickness and depth-resolved characterization of FIB-induced amorphous damage from a single 4D-STEM dataset.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
To see the smallest details of the world, scientists often use a beam of electrons instead of light. Because electrons are so small, they can reveal the arrangement of individual atoms inside a material. However, when these electrons pass through a solid object, they do not simply travel in a straight line; they bounce off the atoms and interfere with one another in complex ways. This phenomenon, known as dynamical scattering, means that the final image depends heavily on how thick the object is. If a scientist does not know the exact thickness of the sample, the resulting picture can be misleading, making it difficult to distinguish between a real structural feature and an artifact caused by the material's depth. This uncertainty becomes a major hurdle when studying materials prepared by a focused ion beam, a technique that uses a stream of ions to carve out tiny, thin slices of matter for examination. While this method is essential for creating samples, the cutting process itself damages the surface, creating a layer of disordered material that sits on top of the pristine crystal inside. Until now, measuring the thickness of these slices and the depth of the damage required preparing a separate, special sample just to cut it open and look at the side, a process that was time-consuming and often impossible to do on the exact piece of material being studied.
A team of researchers has developed a way to solve this problem using a single dataset, allowing them to determine both the local thickness of a sample and the depth of the damage without needing a separate specimen. They worked with a thin slice of silicon that was shaped like a wedge, meaning it was thicker at one end and thinner at the other. This slice had been prepared using four different final cutting voltages, creating four distinct regions where the surface damage was expected to vary. The researchers used a powerful imaging technique called multislice electron ptychography, which treats the electron beam as a wave that passes through the material in layers. To make this technique work, the computer model needs to know the total thickness of the sample to begin with, but that is precisely the information the scientists were trying to find. To break this circular problem, they employed a mathematical strategy called Bayesian optimization. Instead of guessing the thickness, the computer tested thousands of different combinations of thickness, focus, and the number of layers in its model. It searched for the specific combination that made the simulated electron patterns match the real patterns captured by the microscope most closely.
The results showed that this method could reliably estimate the physical thickness of the silicon slice. When the researchers compared the computer's estimates to independent measurements taken with a different technique, they found the values were very close, differing by only a small, consistent margin. More importantly, the method successfully tracked how the thickness changed across the wedge-shaped sample, confirming that the computer was measuring a real physical property rather than just finding a mathematical trick. Once the correct thickness was established, the researchers used the same data to look deep inside the slice. Because the technique reconstructs the sample layer by layer, they could see the transition from the disordered, damaged surface to the perfect, ordered crystal in the middle. They measured the thickness of this damaged layer for each of the four regions. The data revealed that the damage was thinnest where the cutting voltage was lowest, measuring 3.2 nanometers at 2 kilovolts, and thickest where the voltage was highest, measuring 26.4 nanometers at 30 kilovolts. These measurements matched previous findings from separate cross-sectional studies, but with the advantage that they were derived directly from the sample intended for atomic analysis.
This approach offers a significant step forward for quantitative microscopy, where precise measurements are critical. By allowing scientists to determine the local thickness and the extent of surface damage from the same dataset used to image the atoms, the method removes the need for extra preparation steps that might alter the sample. It provides a self-consistent way to understand how the material is structured and how it was affected by the preparation process. The researchers found that the damage layer grew steadily as the cutting voltage increased, confirming that lower voltages are indeed better for preserving the crystal structure near the surface. While there remains a small systematic difference between their estimates and other measurement methods, likely due to the complex physics of how electrons scatter, the ability to map these variations within a single sample opens the door to more accurate interpretations of atomic-scale images. This work demonstrates that with the right computational tools, the microscope can tell us not just what the atoms look like, but also how thick the material is and how deep the damage goes, all at the same time.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.