Searching for what s new- the systematic development of dynamic x-ray microscopy
Abstract
X-ray microscopes offer unique capabilities; they can image millimeter-thick materials with very little multiple scattering background, while delivering 3D spatial resolution in the 10 nanometer range as well as providing sensitivity to specific elements and their chemical states. However, X rays are ionizing radiation, and simple and fundamental considerations of contrast and detectability dictate that high spatial resolution imaging of a completely unknown object involves radiation doses that sometimes affects the sample being imaged. As a result, studies of soft and biological materials at a spatial resolution better than 10 nanometers have been thought to be challenging, and studies of dynamics in such materials have been thought to be precluded. We put forward here a paradigm shift for higher resolution and for the study of dynamics- one can start with knowledge, rather than a blank slate. That is, we usually do not start with a completely unknown object. Instead, we may have imaged similar objects before, and we may also have knowledge of the type of dynamic phenomenon we are looking for (such as a position change of a known feature type, or a change in its size or spectroscopic characteristics). Therefore, searching for a change in a partially known object is fundamentally different, and requires less radiation exposure, than looking for differences between subsequent mappings of an object as if we have never seen it before. We can include this prior information by building it into a cost to be minimized by nonlinear optimization methods. Whereas earlier methods for image reconstruction in x-ray microscopy tended to be based on physicist intuition, the shift to using optimization methods allows one to explicitly include in a cost function to be minimized not only terms for matching the measured data in one measurement, but also terms relating to the similarity of subfeatures against prior knowledge, as well as terms relating to physics constraints on the object as well as on the imaging process. While addition of these terms might seem to make derivatives of the net cost function difficult to minimize, we have shown that automatic differentiation methods included in machine learning toolkits allow one to handle such complexity, and to easily parallelize the algorithms on supercomputers when required. To systematically develop and demonstrate these ideas, we will use x-ray microscopy to study three model systems- microelectromechanical system (MEMS) devices as radiation-robust devices where we can control and then image dynamics, lithium ion battery systems where we will look at the dynamics of charging and discharging, and biological cells where we will concentrate on MEMS-modulated cellular changes as well as chromosome rearrangement during cell division. We will use x-ray ptychography at low dose and low resolution to find the region of a dynamic feature, and then at high resolution in single exposures to capture these dynamics; conventional ptychography images will also be taken of radiation-robust samples such as those in a frozen hydrated state so as to provide the database of similar objects used in our reconstruction methods. This will help the DoD by providing new insights into the function of batteries used in troop-carried electronics, aircraft, and elsewhere; in the response of MEMS systems which are used in the field as sensors; and in the biological response of cells to external stresses and environmental changes.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 29, 2024
- Source ID
- FA95502310284
Entities
People
- George Barbastathis
Organizations
- Air Force Office of Scientific Research
- Massachusetts Institute of Technology
- United States Air Force