|
Abstract:
|
Atom Probe Tomography (APT) has emerged as an important materials characterization technique for a wide variety of applications requiring the analysis of small volumes of material at sub-micrometer length scales - e.g., geological, semiconductor, nuclear, and aerospace materials. The technique provides near-atomic resolution, chemically-resolved, 3-dimensional images of the analysis volume. In the context of isotopic analysis, very little methodology exists for robust estimation of isotopic abundances by estimating spectrum peak shapes. The peak shapes vary considerably, with peaks often overlapping one another, making estimating the relative magnitude of each constituent peak difficult. We discuss why estimation using the standard least squares loss performs poorly. We then propose a coordinate-descent based estimation approach, the crux of which involves the estimation of overlapping non-parametric curves over a complicated feasible region facilitated by posing the problem as a linear program. Finally, we demonstrate strong performance on a wide variety of application materials.
|
