An official website of the United States government
Precise alignment of the electron beam is critical for successful application of scanning transmission electron microscopes (STEM) to understanding materials at atomic level. Despite the success of aberration correctors, aberration correction is still a complex process. Here we approach aberration correction from the perspective of accelerator physics and show it is equivalent to minimizing the emittance growth of the beam, the span of the phase space distribution of the probe. We train a deep learning model to predict emittance growth from experimentally accessible Ronchigrams. Both simulation and experimental results show the model can capture the emittance variation with aberration coefficients accurately. We further demonstrate the model can act as a fast-executing function for the global optimization of the lens parameters. Our approach enables new ways to quickly quantify and automate aberration correction that takes advantage of the rapid measurements possible with high-speed electron cameras. In part II of the paper, we demonstrate how the emittance metric enables rapid online tuning of the aberration corrector using Bayesian optimization.
more »
« less
Emittance minimization for aberration correction II: Physics-informed Bayesian optimization of an electron microscope
Aberration-corrected Scanning Transmission Electron Microscopy (STEM) has become an essential tool in understanding materials at the atomic scale. However, tuning the aberration corrector to produce a sub-Ångström probe is a complex and time-costly procedure, largely due to the difficulty of precisely measuring the optical state of the system. When measurements are both costly and noisy, Bayesian methods provide rapid and efficient optimization. To this end, we develop a Bayesian approach to fully automate the process by minimizing a new quality metric, beam emittance, which is shown to be equivalent to performing aberration correction. In part I, we derived several important properties of the beam emittance metric and trained a deep neural network to predict beam emittance growth from a single Ronchigram. Here we use this as the black box function for Bayesian optimization and demonstrate automated tuning of simulated and real electron microscopes. We explore different surrogate functions for the Bayesian optimizer and implement a deep neural network kernel to effectively learn the interactions between different control channels without the need to explicitly measure a full set of aberration coefficients. Both simulation and experimental results show the proposed method outperforms conventional approaches by achieving a better optical state with a higher convergence rate.
more »
« less
- Award ID(s):
- 2039380
- PAR ID:
- 10597694
- Publisher / Repository:
- Elsevier
- Date Published:
- Journal Name:
- Ultramicroscopy
- Volume:
- 273
- Issue:
- C
- ISSN:
- 0304-3991
- Page Range / eLocation ID:
- 114138
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Pilat, Fulvia; Fischer, Wolfram; Saethre, Robert; Anisimov, Petr; Andrian, Ivan (Ed.)At the Facility for Advanced Accelerator Experimental Tests (FACET-II) accelerator, a pair of 10 GeV high-current electron beams is used to investigate Plasma Wakefield Acceleration (PWFA) in plasmas of different lengths. While PWFA has achieved astonishingly high accelerating gradients of tens of GeV/m, matching the electron beam into the plasma wake is necessary to achieve a beam quality required for precise tuning of future high energy linear accelerators. The purpose of this study was to explore how start-to-end simulations could be used to optimize two important measures of beam quality, namely maximizing energy gain and minimizing transverse emittance growth in a 2 cm long plasma. These two beam parameters were investigated with an in-depth model of the FACET-II accelerator using numerical optimization. The results presented in the paper demonstrate the importance of utilizing beam-transport simulations in tandem with particle-in-cell simulations and provide insight into optimizing these two important beam parameters without the need to devote significant accelerator physics time tuning the FACET-II accelerator.more » « less
-
Plasma-based acceleration (PBA) is being considered for a next generation linear collider (LC). In some PBA-LC designs for the electron arm, the extreme beam parameters are expected to trigger background ion motion within the witness beam, which can lead to longitudinally varying nonlinear focusing forces and result in an unacceptable emittance growth of the beam. To mitigate this, we propose to use quasi-adiabatic plasma density ramps as matching sections at the entrance and exit of each stage. We match the witness electron beam to the low density plasma entrance, where the beam initially has a large matched spot size so the ion motion effects are relatively small. As the beam propagates in the plasma density upramp, it is quasi-adiabatically focused, and its distribution maintains a non-Gaussian equilibrium distribution in each longitudinal slice throughout the process, even when severe ion collapse has occurred. This only causes small amounts of slice emittance growth. The phase mixing between slices with different betatron frequencies leads to additional projected emittance growth within the acceleration stage. A density downramp at the exit of an acceleration section can eliminate much of the slice and projected emittance growth as the beam and ion motion adiabatically defocuses and decreases, respectively. Simulation results from QuickPIC with Azimuthal Decomposition show that within a single acceleration stage with a 25 GeV energy gain, this concept can limit the projected emittance growth to only ∼2% for a 25 GeV, 100 nm emittance witness beam and ∼20% for a 100 GeV, 100 nm normalized emittance witness beam. The trade-off between the adiabaticity of the plasma density ramp and the initial ion motion at the entrance for a given length of the plasma density ramp is also discussed.more » « less
-
Plasma based acceleration (PBA) is being considered for a next generation linear collider (LC). In typical AsmPBA-LC designs, the extreme beam parameters are expected to trigger background ion motion, which can lead to longitudinally varying nonlinear focusing forces and result in emittance growth of the beam. While various schemes have been proposed to mitigate this at low beam energies, a solution to minimize the emittance growth in the later high energy stages of a multistage electron acceleration arm is yet to be found. In this paper, we propose to use an adiabatic plasma density ramp as a matching section that is able to match the witness electron beam to the low-density plasma entrance, where the beam initially has a large matched spot size so the ion motion effects are relatively small. As the beam propagates in the plasma density upramp (downramp), it is adiabatically focused (defocused) and its distribution maintains an equilibrium distribution throughout the entire process even when severe ion collapse has occurred. Simulation results from QPAD show that within a single acceleration stage, this concept can limit the projected emittance growth to only ∼2% for a 25 GeV, 100 nm normalized emittance witness beam and ∼20% for a 100 GeV, 100 nm normalized emittance witness beam.more » « less
-
We propose a simple architecture for deep reinforcement learning by embedding inputs into a learned Fourier basis and show that it improves the sample efficiency of both state-based and image-based RL. We perform an infinite-width analysis of our architecture using the Neural Tangent Kernel and theoretically show that tuning the initial variance of the Fourier basis is equivalent to functional regularization of the learned deep network. That is, these learned Fourier features allow for adjusting the degree to which networks underfit or overfit different frequencies in the training data, and hence provide a controlled mechanism to improve the stability and performance of RL optimization. Empirically, this allows us to prioritize learning low-frequency functions and speed up learning by reducing networks' susceptibility to noise in the optimization process, such as during Bellman updates. Experiments on standard state-based and image-based RL benchmarks show clear benefits of our architecture over the baselines. Code available at https://github.com/alexlioralexli/learned-fourier-featuresmore » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript
- Journal Article:
- https://doi.org/10.1016/j.ultramic.2025.114138
