view article
Figure 3
(a) Thickness, (b) SLD and (c) roughness predictions derived from the experimental XRR data R(q,t) in Fig. 2[link] agree reasonably well between standard individual curve fits, growth model based CNN and co-refinement results. (d)–(f) For a reduced data set of only 200 reflectivity points in R(q,t), the CNN and the growth model co-refinement fit still fit the data set correctly. A set of CNN prediction curves are shown for different, random selections of 200 points out of the full data set, giving an indication of the CNN prediction variability. (g)–(i) For noisy data, the growth model co-refinement needs a count rate of only ∼1000 counts in the total reflection region to allow for adequate reconstruction, while the CNN can fit data even with ∼300 counts in the total reflection region. Again, CNN predictions of a set of curves with different simulated photon shot noise are shown to indicate the variability of CNN predictions.

Journal logoJOURNAL OF
APPLIED
CRYSTALLOGRAPHY
ISSN: 1600-5767
Volume 55| Part 5| October 2022| Pages 1305-1313
https://doi.org/10.1107/S1600576722008056
Follow J. Appl. Cryst.
Sign up for e-alerts
E-alerts
Follow J. Appl. Cryst. on Twitter
Twitter
Follow us on facebook
Facebook
Sign up for RSS feeds
RSS