Image reconstruction of two clusters of gold balls showing the
convergence of both the reconstructed image and the support as
the number of iterations increases from 1 to 1000. For comparison,
a scanning electron microscope (SEM) image is also shown. The
Rayleigh resolution of the reconstructed image is 10 nm.
In Fienup's hybrid input-output algorithm, one starts with a diffraction
pattern with random phases and then iteratively transforms between
real space (the image) and back to reciprocal space (the diffraction
pattern with phases). Each transform to reciprocal space provides
improved phases for the next cycle. An essential constraint is the
requirement that the intensity of the diffraction pattern be zero
outside the boundary of the object (the support). The better the
support is known, the faster the iterations converge to an accurate
image. Most researchers have relied on x-ray microscopy or other
techniques to supply this information.
Effect of noise on reconstruction error. The shrink-wrap algorithm
(blue) is superior to the hybrid input-output algorithm with fixed
support (all other colors), except for the case when the support
is known perfectly (green). The accuracy of the supports decreases
from support 1 to support 4.
The Livermore/ASU/ALS collaboration has been working to eliminate
the need for supplementary experiments. Previously, the team had
succeeded by preparing on a silicon nitride substrate clusters of
gold balls 50 nm in diameter together with an isolated single gold
ball as reference that generated the information needed to construct
the support. Now, the researchers have done away with even that
requirement with a new "shrink-wrap" algorithm. They use a transform
of the diffraction pattern as the initial support. At intervals,
they generate a new support from the transform of the current diffraction
pattern. In this way, the support converges to a tight boundary
around the cluster of balls, and the image also emerges.
Work on three-dimensional images from a series of diffraction patterns
obtained at many illumination angles is underway. In the meantime,
the researchers believe that a resolution of 10 nm will be possible
for life-science samples, where radiation damage is an issue, and
2 nm for solids. A dedicated CXDI beamline at the ALS after a planned
brightness upgrade could improve the figure to 1 nm, owing to an
increase in imaging speed. The ultrabright, femstosecond x-ray pulses
expected when SSRL's Linac Coherent Light Source comes on line around
2008 may enable atomic-resolution imaging of single molecules.
Research conducted by S. Marchesini, H.N. Chapman, S.P. Hau-Riege,
and A. Noy (Lawrence Livermore National Laboratory); H. He and M.
Howells (ALS); and U. Weierstall and J.C.H. Spence (Arizona State
University).
Research funding: U.S. Department of Energy, Office of Basic Energy
Sciences (BES), and National Science Foundation. Operation of the
ALS is supported by BES.
Publication about this research: S. Marchesini, H. He, H.N. Chapman,
S.P. Hau-Riege, A. Noy, M.R. Howells, U. Weierstall, and J.C.H.
Spence, "X-ray image reconstruction from a diffraction pattern
alone," Phys. Rev. B 68, 140101(R)
(2003). |