ECE PhD Prospectus Defense: Vaibhav Choudhary

ECE PhD Prospectus Defense: Vaibhav Choudhary

Title: Physics-Informed Computational Methods for Secondary Electron Imaging

Presenter: Vaibhav Choudhary

Advisor: Professor Vivek Goyal

Chair: Professor Prakash Ishwar

Committee: Professor Vivek Goyal, Professor Prakash Ishwar, Professor David Bishop

Google Scholar Profile: https://scholar.google.com/citations?user=GHJBQKwAAAAJ&hl=en&authuser=1

Abstract: Secondary electron (SE) imaging techniques, such as scanning electron microscopy and helium ion microscopy (HIM), use electrons emitted by a sample in response to a focused beam of charged particles incident on a grid of raster scan positions. Several factors fundamentally limit the quality of images produced by these microscopes. Previous works have primarily focused on mitigating source and target shot noise, as well as detector noise. This thesis explores two less-studied factors: the incident beam’s spot size and beam-induced sample damage. First, we seek improvements rooted in a better understanding of the effect of the beam spatial profile, which is conventionally treated as a convolution. We show that under a simple and plausible physical abstraction for the beam, though convolution describes the mean of the SE counts, the full distribution of SE counts is a mixture. We demonstrate that this more detailed modeling can enable resolution improvements over conventional estimators through a stylized application in semiconductor inspection, specifically in localizing an edge in a two-valued sample. Further, we aim to extend this model to an arbitrary number of edges with different shapes, thereby creating new physics-informed estimators for segmenting materials in a micrograph. Second, we address the trade-off between sample damage and image quality. The standard techniques for sample damage reduction generally involve dose reduction, in the form of reduced dwell time or reduced beam current, which also reduces the quality of images. In contrast, we investigate the post-scanning recovery of an image of the undamaged sample from a sequence of images of the sample taken under high doses. We aim to develop a differentiable forward process to model the damage, enabling a neural network to learn the inverse mapping from these high-signal images. We hope that this approach facilitates high-fidelity imaging of delicate samples.