Fast ptychographic imaging of semiconductor structures at grazing incidence

Coherent Diffraction Imaging (CDI) is an interesting candidate option for the metrology of EUV masks and wafers because of its contained costs, ease of implementation and scalability. The imaging step in CDI is performed algorithmically by phase retrieval. The missing phase is reconstructed from the diffraction dataset by iterating back and forth among the real space and the Fourier space. This usually implies the use of Fast Fourier algorithms, which is an efficient method used in different light propagation models, that makes it ideal in many CDI applications. However, when the illumination beam is not perpendicular to the sample, the diffraction orders no longer lie on a circle in reciprocal space centered at the origin, but on a cone centered around the incident wavevector. This distorts the spatial frequency map so that the diffraction angles no longer correspond linearly to spatial frequencies under standard FFT assumptions. In order to solve for this issue, one usually processes the data so to correct for this distortion. The correction is performed by an interpolation step which can be numerically expensive to evaluate. In this work, we explore the possibility to image the object directly from the skewed data, without the need for an expensive numerical correction of the dataset.