Electron probe microanalysis EPMA

1
Electron probe microanalysisEPMA

• Thin Film Analysis
• and Particles

2
Whats the point?
EPMA is traditionally done for bulk material.
What are the issues for thin films? How
precise/accurate are such analyses? Can
unpolished particles be analyzed by WDS or EDS?
3
Bulk vs thin film

• Normal EPMA assumes that the electron beam is
  exciting a homogeneous volume, i.e. there is no
  difference either laterally or vertically. Thus,
  the matrix correction is being applied in a
  uniform matter, and there is one applicable f(rz)
  profile for each element .
• As research has improved the accuracy of the
  f(rz) profiles, it is now possible to take thin
  films (including multiple films) and apply f(rz)
  models and calculate best fits for unknown
  parameters. For example, if you know that there
  is a TiO2 skin atop your Ti metal, you can
  acquire Ti and O X-ray counts at several E0
  values, and then try to match them by modeling
  various film thicknesses with 3rd party software
  programs. Or if you have been able to measure a
  film thickness, you could determine what the
  phase stoichometry is.

4
MC Simulation TiO2 on Ti
It can be helpful to run Monte Carlo simulations
of thin films. Here, the new Casino software is
used to model 15 keV electrons hitting a 1 mm
layer of TiO2 on Ti. Red trajectories to top left
are BSEs. At bottom left is a model of the O Ka
f(rz) profile (blue), plus the profile of X-rays
predicted to escape (red) and get to the detector.
Here the range of various energy incident
electrons are modeled.
5
Thin Film Software
Thin films can be studied with the electron
microprobe, although the acquired data cannot be
run through the normal probe software (which only
works for homogeneous volumes). Standard counts
are acquired on normal standards, and then
K-ratios acquired from the thin films at several
E0 values (minimally 3, preferably more e.g. 5,
10, 15 keV). There are two software packages the
costly STRATAGem, above left (6K) and the
freeware GMRFilm (below written by R.Waldo of GM
). STRATA-Gem is very slick and has a Windows
interface, whereas GMRFilm runs under DOS and
requires manual tabulating. In the bottom
example, we see that a 0.1 mm oxide coat on Ti
robs the metal of 6 of the Ti Ka counts it
should yield.
6
Oxygen on Boron metal (2 standards)
This and the next slide demonstrate the utility
of thin film software. We needed to verify that
our boron standard was pure, but there was a
small peak at O ka. I ran it at 2, 3, 7 15 keV
(red and black symbols), and then tested various
interpretations of the data. Oxygen as bulk did
not fit, whereas a 12Å oxide film did.
Experiment
Models
Not bulk, but 12 Å film B2O3 (2 different Boron
standards)
Thin film modeled with GMRfilm
7
Carbon on Boron metal
Not bulk, but 12-18 Å C film (2 different Boron
standards)
Thin film modeled with GMRfilm
8
Particles - 1

• Mass effect/error electrons escape from sides
  of small particles if E0 is large enough (left)
• Absorption effect of non-flat upper surface
  different path length from normal flat geometry
  (middle)
• Variable effect of geometry of trajectory
  between beam impact area on non-uniform surface
  and the location of the detector (right)

Goldstein et al, 1992, p. 479, 481
9
Particles - 2

• Traditional approach normalize numbers but
  this is not very good (above left table)
• Armstrong and Buseck (1975) developed a
  procedure based upon a regular geometric shape
  factor, where the different path length and other
  effects could be used. Method is based on
  bracketing particle and beam overscanning during
  collection of spectrum by EDS and modeling of
  electron path and x-ray propagation out through
  several shapes sphere, hemisphere, squared
  pyramid, and rectangular, tetragonal, cylindrical
  and right triangular prisms. Correction factors
  are given in terms of predicted k-factors for
  pairs of particular elements, vs particle
  thickness along e beam. This is not easy, takes
  much trial and error, but apparently can yield
  fairly good results (see table above).

Goldstein et al, 1992, p. 488, 489