Feasible atomic-resolution electron tomography for general crystal surfaces by quantitative reconstruction from a high-resolution image

Whether or not the 3-dimensional surface morphologies of a crystal sample can be reconstructed at atomic-scale from a single 2-dimensional image becomes an interesting issue in high-resolution transmission electron microscopy, after the work by Jia et al. [1]. Here we propose an improved and self-validated algorithm to enhance such an electron tomography method and to make it applicable to more general crystal surfaces even with thin amorphous layers. Our study shows that a resolution in the beam (z) direction and a confidence level have to be defined and estimated after performing tomographic reconstruction in order to evaluate the quality and the reliability of its result. Applying the proposed procedure to the Si[110] image to recover the surface morphologies of a silicon crystal with amorphous contamination, the obtained results show that an atomic-resolution of 0.384 nm in the z-direction and a high confidence level of 95% are achieved for imaging the Si-surface structures, quantitatively described by tomographic parameters, i.e., the height (defocus) and the thickness (atom number) of Si-atomic columns.

An improved iterative wave function reconstruction algorithm in high-resolution transmission electron microscopy

Exit wavefunction reconstruction is a powerful image processing technique to enhance the resolution and the signal-to-noise ratio for atomic-resolution imaging in both aberration uncorrected and corrected transmission electron microscopes. The present study aims to improve the performance of the iterative wavefunction reconstruction algorithm in comparison not only with its conventional form but also with the popular commercial Trueimage software for exit wavefunction reconstruction. It is shown that by implementing a wave propagation procedure for refining its image alignment, the iterative wavefunction reconstruction algorithm can be greatly improved in accurately retrieving the wavefunctions while keeping its original advantages, which allow the reconstruction be performed with less images and a larger defocus step in the data set of through-focus image series. In addition, calculations of this algorithm can be accelerated drastically by the graphic processing unit (GPU) hardware programming using the popular computer unified device architecture language, whose computing speed can be 25-38 times as fast as a central processing unit (CPU) program.