July 1996 REPORT FOR
W. CHARLES SYMONS, III
Recently a single point Near-Field Scanning Optical Microscope (NSOM)
model was further developed into a fully scanning model utilizing the
Finite Difference Time Domain (FDTD) method. In order to fully develop a
fast useable code, the FDTD NSOM code was parallelized for the Convex
Exemplar. However, this parallelized code did not perform as expected.
Upon closer examination, it was discovered that the parallel version of
this code simply requires too much memory. That is, because the problem
involves solving for complex field points distributed throughout a three
dimensional region, a large amount of memory is required to iteratively
solve for these fields. Furthermore, the current parallel implementation
of the code calls for each separate processor to run the code for varying
aperture placements. Therefore, the parallel code requires the field
points to be stored separately for each processor so as not to overwrite
data from separate parallel excitations. In order to alleviate this memory
problem, solutions involving both smaller problem regions and new
parallelization methods are being explored.
In addition, to the parallel NSOM code described above, the simple study of
tapered probes discussed in the January 1996 report was enhanced by
incorporating exponentially tapered probes in addition to the linearly
tapered probes studied earlier. For further information, a report
summarizing this study is available online.