Beam Cross Sections Create Mixtures: Improving Feature Localization in Secondary Electron Imaging

Vaibhav Choudhary, Akshay Agarwal, Vivek K Goyal · 2025

Read Beam Cross Sections Create Mixtures: Improving Feature Localization in Secondary Electron Imaging by Vaibhav Choudhary, Akshay Agarwal, Vivek K Goyal (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-09942.txt.

Introduction

Vaibhav Choudhary, Akshay Agarwal, and Vivek Goyal improve feature location in electron imaging in this 2025 paper. Microscopes that scan a charged beam across a sample read the electrons the sample throws off. Beam width limits sharpness, and the beam's effect is usually modeled as a blur. The team shows that picture is incomplete: the counts follow a mixture, not a simple blurred average.

That richer model opens room for sharper estimates. In a test task, finding the edge in a two-tone sample, they derive the information limits and the best estimator from time-resolved readings. Simulations cut edge error about fivefold against standard interpolation, reaching below one pixel. Three real ion microscope data sets showed a similar average gain.

Catalog record

CREATORVaibhav Choudhary, Akshay Agarwal, Vivek K Goyal
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE67 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-09942

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Beam Cross Sections Create Mixtures: Improving Feature Localization in Secondary Electron Imaging Vaibhav Choudhary, Akshay Agarwal, Vivek K Goyal 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 at a grid of raster scan positions. Spot size—the diameter of the incident beam’s spatial profile—is one of the limiting factors for resolution, along with various sources of noise in the SE signal. The effect of the beam spatial profile is commonly understood as convolutional. 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 of localizing the edge in a two-valued sample. We derive Fisher information about edge location in conventional and time-resolved measurements (TRM) and also derive the maximum likelihood estimate (MLE) from the latter. Empirically, the MLE computed from TRM is approximately efficient except at very low beam diameter, so Fisher information comparisons are predictive of performance and can be used to optimize the beam diameter relative to the raster scan spacing. Monte Carlo simulations show that the MLE gives a 5-fold reduction in root mean-squared error (RMSE) of edge localization as compared to conventional interpolation-based estimation. Applied to three real HIM datasets, the average RMSE reduction factor is 5.4. I Introduction Nanometer-scale imaging is central to discovery and everyday diagnostics in a variety of scientific and engineering fields. It is essential in semiconductor manufacturing, for example, where the shrinking in device sizes over the last 50 years would not have been possible without precise measurements of each generation of devices [ 1 ] . In biology, nanoscale imaging enables scientists to observe structures and processes at the subcellular level [ 2 , 3 ] . Optical techniques reach nanoscale for specific samples with fluorescent dyes and fluorescent genetic reporters [ 4 ] ; for general samples, however, diffraction precludes such fine resolution. Instead, the most prevalent instruments for nanoscale imaging detect secondary electrons (SEs) generated by a highly focused incident beam of charged particles, raster scanned on a 2D grid [ 5 ] . The micrographs produced by SE imaging instruments show spatial variation of SE yield \eta —defined as the mean number of SEs emitted by the sample per incident particle—subject to various sources of noise. Though the dependence of SE yield on local sample properties is complicated, SE imaging is nevertheless ubiquitous in semiconductor metrology, where it is used for critical dimension measurements [ 6 ] , line width roughness analysis [ 7 , 8 , 9 , 10 , 11 , 12 ] , characterizing cross-sectional samples [ 13 , 14 ] , and analyzing and classifying sample defects [ 15 , 16 , 17 ] . Biological applications of SE imaging have been abundant for a half century [ 18 ] and are especially popular when combined with focused ion beam (FIB) sectioning for volumetric imaging [ 19 , 20 , 21 ] . The noise in SE images comes from three mechanisms [ 22 , 23 ] : source shot noise , variation in the number of incident particles in a fixed time period; target shot noise , variation in the number of emitted SEs for a given incident particle; and detector noise , variation in measured response to a number of SEs (including the efficiency in emitted SEs reaching the detector). Additionally, the spot size or cross-sectional profile of the incident particle beam is not generally considered a source of noise but it too influences the image quality. Despite the widespread use of SE imaging, works that examine the information content in SE detector signals and seek statistically optimal estimation of SE yield have appeared only recently. Peng et al. [ 24 , 25 ] introduced the concept of time-resolved measurement (TRM) to SE imaging under an assumption of ideal SE detection. TRM can be understood to almost completely mitigate source shot noise; see Section III . Agarwal et al. [ 26 ] introduced modeling of actual detector voltage signals and experimentally demonstrated noise reduction in a helium ion microscope (HIM) [ 27 ] . These previous works do not include any modeling of beam cross section. In this paper, we complement the framework of previous works with detailed physics-based probabilistic modeling of the beam cross section. Our main assertion is that nonzero beam cross-sectional area causes the SE count due to a single incident particle to follow a mixture distribution. This contrasts with a convolutional model that is intuitive because, naïvely, the effect of the beam seems analogous to blurring in a linear optical system. Indeed, convolutional modeling is conventional in scanning electron microscopy [ 28 , 29 , 6 ] ; we seek to expose the limitations and suboptimality of this modeling while providing an operational alternative. As a prototypical problem in semiconductor metrology, we consider here the localization of an axis-aligned straight edge between SE yield levels in a two-dimensional sample. This simplified imaging-based inspection problem allows us to use root mean-squared error (RMSE) to explicitly compare the accuracies of estimates produced from mixture modeling and from convolutional modeling. Monte Carlo simulations show RMSE reduced by a factor of 2.5 by using TRM with a mixture model as compared to TRM with a convolution model; improvement relative a conventional interpolation method is by a factor of 5. On experimental HIM data, we observe an an average RMSE reduction factor of 1.5 with respect to a convolution model and 5.4 with respect to interpolation. The paper is organized as follows. Section II introduces models of the beam, the sample, and their interaction that underlie this work. We establish there a physical justification for a two-component Poisson mixture model; the estimation advantage presented by this model motivates our entire study. Analyses in Section II are for observations of SE counts due to a single incident particle and for deterministic repetitions of such hypothetical observation trials. In practice, particle incidences are random and not perfectly inferrable. Mitigating this effect was the motivation for time-resolved measurement. Section III reviews—and in some cases slightly generalizes—results from [ 25 , 26 ] on the advantages from TRM when beam cross-sectional effects are neglected. Section IV brings together the spatial ( Section II ) and temporal ( Section III ) aspects of our incident beam model. Section V analyzes estimation of edge location based on the mixture model when combining data acquired from a line of scan locations across an edge. We show that accuracy of localization can be finer than scan grid spacing, thereby achieving super-resolution. We furthermore establish a method for optimal selection of beam width. In Monte Carlo simulations of estimator performance, we observe that the mixture estimator has lower mean-squared error (MSE) than other estimators of the edge location. The same performance trends are also observed on experimental datasets acquired on a helium ion microscope. II Underlying Models and Problem Setup SE imaging modalities involve data collection

Read the complete text