SCANNING ELECTRON MICROSCOPY/1976 (Paxt III)
Proceedings of the Workshop on
Techniques for Particulate Matter Studies in SEM
IIT Research Institute
Chicago, Tllinois 60616, U.S.A. i
april 1976 ee
MULTICOMPOSITIONAL PARTICLE CHARACTERIZATION USING THE SEM-MICROPROBE
6. Grantl, 5.8. Hail”, A.F, Reid! and M, Zutderwyk?
Division of Mineral Chemistryl
CSIRO, Box 124
Port Melbourne, Vic. 3207, Australia
Julius Kruttschnitt Mineral Research Centre’
University of Queensland
\ustra:
ABSTRACT
An essential requirement for an understanding of industrial proc-
esses such as mineral beneficiation is the classification of multicom-
ponent particles by size and composition. To measure these parameters
for samples consisting of whole particles or for polished cross-sections
of multicomponent materials, a digital scan system has been constructed
which, under mini-computer control, steps the electron beam over the
specimen to generate the necessary data. Algorithms in the computer
program treat these data to provide the geometrical properties of each
particle component. Particle size and compositional distributions,
aodal analyses and the statistics of “coupositeness” of the populations
are then ed.
Self-generated raster patterns from 20 x 20 to 4000 x 4000 points
or computer controlled specific patterns are available from the digi
scanning unit. Dwell time per point can be ted from 10-7 to T sec-
ond. Digital values of the X and Y beam co continuously
imen until
eight-bit composition code and the beam coordinates
memory. Each inage ie recorded as a compressed digital map of Line
segment and points
To reduce the total area to be scanned in detail rectangles bounding
each particle are determined by an initial fast scan. This strategy also
minimizes the size of data arrays. Once all particles within a frame
have been examined, the specimen stage is moved under program control to
centre the next search area under the beam. The software strategy ensures
that all particles on the search boundaries are located, and thereby en-
ables the examination of specinen areas large conpared to the SEM frame
size.
The algorithms determine the particle parameters by linking over-
lapping run lengths on adjacent raster lin
meters, moments and bounds for each partic:
are compiled in FORTRAN IV and are linked to the be:
tion driver, programs incorporated in the assenbly language operatin
system of the mini-computer. The programs occupy in all approximately
Lek of 16-bit word:
Digital Scanning, Image Analysis, Scanning Electron Microscopy,
Particle Dimensions, Particle Composition, Mineral Particle
Classification, Stereology
KEYWORDS
401MILTICOMPOSITIONAL PARTICLE CHARACTERIZATION USING THE SEN-ICROPROBE
G. Grant , 5.5. Hall , A.P. Reid and M. Zuiderwyk
Introduction
‘he determination of the sizes and composite
iona of complex particles, aa well as their peri-
meters and shapes, is a primary requirement for an
understanding of the physical and chenioal behav-
sour of particles in industrial and mineral treat~
ment processes. This topic is the subjectlof ex-
tensive Literature, but automatic methods of part
icle classification yielding compositions and data
in even two dinensions are not yet common, nor of
General application and reliability. The use of
igital images obtained, preferably on-line, from
‘the SEM has been considered for sone years (1,2).
Jones and Shaw (3) have used signals from a micro-
probe to determine intercept lengths over sect-
joned particles, and this data has been inter~
preted (4) to obtain estinates of particle prop-
erties
Wnite et al. (2,5,6/7) have developed the
use of the automated SEN for the measurement of
sizes and chemical composition of vhole particles.
However, their systen requires considerable off-
Line computer capacity, and the image must first
be stored on magnetic tape. For our purposes,
‘the prograns described do not cope with the rath
er complex problen of composite particles, nane-
ly determination of the exposed areas of differ~
fent composition in multicomponent grains.
‘The statistics of such distributions are of
particular inportance in mineral flotation and
in the determination of the degree of Liberation
of minerals during grinding or similar operations.
‘The determination of areas in grain mounts and
“whole-rock" samples is also needed for ore and
mineral assessment.
To treat these various problens a systen vas
required which would alloy digital imaging of
sample areas many tines larger than SEM frane
size, and with the option of treating the data
on-line in a mini-conputer. Measured rather
than inferred areas and perimeters were sought,
‘$0 that shape functions, as for example des-
cribed by Davies (8), could be derived, Tt was
also required that the beam coordinates be add-
ressable by software, and that manual setting
and checking of control paraneters be available.
‘The system developed, satistying these
requirements, 4s the subject of this paper.
Bean Scanning system
‘The electron beam position is controlled by
two addressable 12-bit D/A converters, which
provide a step raster of 4000 x 4000 points.
402
‘The converters are driven by a 10 MHZ digital
pulse generator, vhich linits the dvell tine
Per point to a minimin of 10-7 seconds.
‘The dwell tine can be extended to a maxinun of
1 second, and dwell points can be spaced fron 1
to 200 steps apart. Reduced scan limits
tnt Faax’ Yain’ Yang) 29 A180 be set, These
various paranetars can be software or front panel
controlled, Figure 1. The X and ¥ positions of
the beam are continuously displayed. Aa well.
as being driven in free-running raster node, the
bean may be explicitly positioned by software
for beam patterning, which enables generation
of sampling patterns about given.or detected
features of the specizen.
‘The scan control operating paraneters,in
hexadecimal form,are contained in four 16-bit
words loaded on to a 64-bit buffer at the bes-
Anning of each frame scan. Once this control
buffer ie loaded, the scanner is free-running
until halted by “Line-end", "Frame-end" or a
change in status of the signal channels. The
bean stops and the scanner Interrupts the com,
puter. The X-nd ¥ hexadecimal bean coordin-
Bees and an a-bie digical pateern, cepreseatiog
‘the status of the signal channels, are read into
the computer in a total of 32 bits. This S-bit
status patter ds the result of 2 analogue and
5 X-ray discriminators having found a particular
grey level or particle composition. In the us~
ual mode of operation, the ¥ coordinate is not
stored.
A digital acceptance filter can be set so
that a change to « new signal channel status
must be verified for 1, 2, 4 or 8 successive
‘steps before causing the scanner to halt. This
facility rejects noise signals, and ensures that
composition boundaries have been correctly re-
cognized.
Hexadecimal data transfers are used throug
out to increase operating speed, reduce soft~
ware overheads and to provide the correct format
for assenbly Language logic and arithmetic.
Image storag
‘The image is stored as a Linear array of x
and if required, ¥ coordinates, with associated
‘composition codes. Usually the Y coordinates
fare inferred from the end of line stops (7), and
the number of steps per line is Limited to 256,
0 that the inage can be stored as a simple se~
quence of 16-bit words, with one S-bit byte re-
Presenting the X coordinate and the other the
composition code. The location of the words
in the array give their index value.Figure 1. Digital scan control for SEM bem. A.
Frame. Scan paraneters: step size,
novread interval," C. signal channalat
fnateator
swieche
‘of K and ¥ beam coordinates and frame limita,
"
S.
.
ee
‘Thusbwheel mvitches for manual setting of reduced
jell tine, frame, line, sing
Rows 1-5, x-rays
‘colums 2 and 3, selection indicators and switches. D.
Signal is accepted only after 1, 2, 4 of 8 positive indications. E.
repetitive, read length,
rows’ 6-7, analogue. Column 1, interrupt
Digital acceptance filter
‘Visual atsplay
‘This data format is used directly, without
unpacking or reconstruction of the image, in the
perineter and area algorithas. This imediate
Utilization contributes considerably to mint-
mizing computational tine and storage require
Signal Processing
‘he three available analogue channels aco-
‘opt the outputs from any of the secondary elect-
ron, backscattered electron, or absorbed current
Getectors. Non-linear amplifiers and upper and
lower level discriminators are used to select.
the voltage "window" corresponding to signale
from a given chemical phase. If required, the
sane detector signal can be fed to all three
channels simultaneously to give a total of four
levels of discrimination (0 1, 2, 3). Diacrim
Snator tine constants are sat at 10-4 second.
the X-ray detector channels consist of eight
presettable overflow counters, which can be used
Separately, oF in pairs to provide upper and low
oF level discrimination.
Dwell ties for sufficient X-ray counta
vary from 10? to 107} seconds.
are urually derived from an enersy dispersive
Getector, but are also available from the crys:
tal spectrometers fitted to cur microprobe = SEM.
Any shadoved areas vithin particle boundar
sen, reglatering as background, are assigned
to the major compositional fraction by the sub-
sequent softwar
For a given element, count rates and dvell
‘tines are chosen by the operator, and minimn
count values required during the dwell tine are
set manually on thumbwheel svitches, prior to
403
Particle imaging. The beam scanner has provis-
fon for repetitive scanning of a single line,
alloving the operator to adjust threshold volt-
‘age levels of count rates to suitable values by
Inspection of known standarde or features in
the samples. Once thresholde are set, ond
scanning is begun, a change of signal status 1s
registered if one of the values obtained differs
‘from the value observed at the previous beam
position. Such an occurrence halts the Deas
Scanner, which then sends an interrupt to the
couputer for transfer of X, ¥ and composition
values. Signal processing is thus kept inde-
Pandent of the control functiona of the computer.
pectnen stage X-¥ control
‘The X and ¥ stage positions are controlled
by stepper motors with gearing ratios which give
an X of Y displacenant of 1 aicronetre per motor
Pulse. ‘The stage displacements are digitized
by the number of motor pulses with respect to a
zero position. Pulses are counted by up-down
presettable counters, and the "present™ position
fa continually compared with the value Loaded
conto buffers on the motor control interfaces
Vien the requested position Is reached the motor
stops. Ag with the beam gcanner, the motor
novenents are independent of the computer. This
control system is described in more detail in
reference 9.
Softvare supervisory Systen
In our control programs we have used the
IWTERDATA 70 operating system with all redundant
routines renoved, but with additional software
Grivers which are specific to the various dev-
terthe softvare contra} progras, shich gen-
exaten the necessary parameter words for the
Sean tcanner and notor controllers, and which
ends" fon the scanner the coordinate and’ com
Position inforastion, adresses the sofeare”
Eriver routines by neane of the inetraction,
oe avaliable an the anverdata Aasenbly tang *
Sige.” than the inage analysis precediree for
"Sivan Glemtion ara independent of the soar
tefvand stage control driven sovtineai'and any
enaily be Felonded with a variety of prgrane.
Frovielons for Linking FOREMAN roveines, Sapited
CEE"line ace included in cur eperstiog syetan.
Qperationsl Procedurs
once signal discrininator levels have been
set, and the firet frame of the specinen aren
has been located, the beam is stepped at the
highest useful speed over the vnole frane area,
using a relatively coarse raster. An electron
Signal channel locates the edges of particles
without specific reference to composition. The
Fesulting X-coordinates are then treated by the
perineter algorithn to determine the location of
‘the rectangles bounding each separate particle
in the frane, Figure 2. The position of these
rectangles 1s related to the zero point of stage
ovenent with a precision of #1 micrometres,
‘the reproducibitity of our automated stage.
Bach rectangle, enlarged by one or two
stops widths, is then scamed as slowly and in
as much detail as is necessary, using discrim-
inated secondary electron or X-ray signals.
For many purposes, a raster of 50 x 50 steps
per rectangle gives nore than adequate inform
ation. The particle is closely bounded by the
Fectangle, and sone 1200 “on-particle™ points
would be expected.
In the systen described by Troutnan et al
(7), the beam is driven along a frane Line at.
high speed until a particle edge is located,
‘and then svitched £0 slow speeds during the
traverse of a particle. Our system enables
8 comparable saving of tine but additionally
reduces very considerably the anount of data
Storage required. Pull information for each
particle is retained and the data arrays are
Small compared with whole frame arrays. Adopt-
fon within each rectangle of short and long dvell
times "off" and "on" the particle, would hovever
enable a further saving of time, and is being
investigated.
once the set of Local rectangles within the
whole frane has been fully treated, the stage is
moved to bring the next sample area under the
dean. This movenent ie less than the scan frane
by the maximum expected dimension of the largest
particle. This procedure effectively re-exan-
ines a band around the trailing X and ¥ edges of
‘2 search area delineated by stage movenent,
Figure 2, and ensures the inclusion of every
particle hen all of the sample area is exanined.
404
If a particle near a frane edge is relocated dur-
ing the high-speed scan of an adjacent frane it
ds soft-ware rejected and not re-exanined in
detail.
When search areas are spaced out so as to
obtain only a statistical sampling of the vhole
area, particles vholly within the overlap band
are rejected, while particles on the boundary
between "search" and overlap area are accepted.
‘This procedure ensures that the statistical size
distribution is not skewed .
‘This approach differs fron those in which
particles cutting frames are simply given areas
Gouble those cbserved, or a snali frane enlarge
‘nent is added in one direction only.
Jmage Analysis Prograns and Algori thas
Particle area and perineter determination
are treated as discrete operations based on al~
gorithms which manipulate the digital inage data
for the scanned frane or the rectangles which
contain individual particles. These computat~
onal operations can either be part of the data-
gathering program, or can be performed off-line,
‘They are writeen in ANSIT-standard FORTRAN but
versions of the routines are now being written
in assembly language to increase the speed of
Processing in the mint-computer. Details of
the algorithms and programs will be given in
‘8 separate publication, and only the main points
fare discussed her
‘The approach finally found most useful is
to determine and label firet the connected bound-
ary points of the particle, and then to determin
the area.
‘The perineter algorithm takes advantage of
the Linear nature of the information array, and
from the run lengths defined by the X-coordin~
ates for ON and OFF the particle for a given
raster Line, overlaps between isage Lines are
Getermined, Consideration of the relationships
for a particle boundary cutting or not cutting
across. two adjacent raster lines gives three
potsible topological cases vhich are: simple
concave change of direction; oF
convex change of direction. Taking these sit
‘ations into account, consecutive boundary
coordinates are connected vectorially until the
Doundary is closed. Area is then obtained by
‘summing up the run Lengths between adjacent.
labelled locations on the sane raster Line, with
ue account of individual compositions, and the
contributions of the total run lengths to the
Hirst moment about the X and ¥ bounds of the
particle are also sumed. These latter values
Lead to a simple determination of the X and Y
coordinate values for the centroid of the part-
ele in ite two-dimensional projection. If
required, the perimeter contributions of each of
the different conposition types forming the boun-
dary can be separately sumed.Visual Display of Particles:
As the boundary coordinates of particles or
sub-areas are located, the successive (x,¥) pairs
around the boundary of their indexed manory loce
ations, are stored in an array separate from
that containing the raster information. For
@isplay, pairs of coordinates are loaded se- 7
quentially onto D/A converters, and the result-..
ing X and ¥ analogue voltage values can be
directly applied to an ordinary oscilloscope
‘Alsplay. A perimeter defined by 200 points
loaded into the D/A nodules at one transfer per
millisecond produces a flicker-free display vith
‘2 phosphor having a .25 second persistence.
Such a rate of data transfer poses no problems
for a miniconputer interfa ferineters of
sub-areas can also be displayed in a similar
anner, and we intend to install a 3-color
Gieplay monitor so that particle and sub-area
outlines can be shovn in different colors.
‘mis aystem of data display obviates the
need for the very fast data streams necessary
for TV raster refresh rates, and simplifies the
@isplay equipment required. We believe that
our perimeter determination and storage 1s of
particular value for this reason.
he aigoriehme which we have developed can
be applied to the digital inages of any discrete
objects, elther mlcro- or nacro-sized, provided
that their boundaries are defined and’ continuous,
With the aid of an SEM, particles, particle
cross-sections, cross-sections of ‘continuous
agglomerates such as fine-grained ores, or the
pore holes in sections of ceranics, can equally
Well be characterized.
‘The algorithms are also applicable to mth
spectral optical images, such as those of blood
cells (10) oF those obtained using 3-color 7V
‘scanning of inage sources in general.
Sample Mounting Techni
In determinations of particle characteris
tics, sampling and sample mounting is of import~
ance if unbiased information is to be obtained.
By using the present measuring procedure, which
either examines all particles on a mount or
‘hich examines spaced areas without overall
bias, the problem of obtaining a uniform dist-
ribution on the mount is largely removed.
For flotation concentrates ve have found
that a sample of particles spread onto a filter
paper can be mounted on a flat surface such as
a carbon or aluminium block, aluminized glass
of beryllium sheet by smearing a very thin layer
Of epoxy over the surface of the mount vith the
edge of a razor blade, and then pressing the
40s
inverted mount onto the samples
For sectioned grain mounts, our major pre-
‘cation in sample preparation is the use of the
nifieragraphic technique of placing the sample
“mould in an evacuated chanber imediately after
mixing or pouring in epoxy, so that air entrain-
ment and subsequent formation of holes and the
Plucking of particles during polishing, are
sinimized.
Mounts containing whole grains on a flat
yurface are carbon coated vhilat rotating to
‘endure coating of all exposures of the particlé
‘The nounting of particles for SEM eranin-
ation is a subject treated in detail by sever-
‘al authors (11) and in our vork we have essent-
Aally folloved such published techniques.
Stereology of Particles
‘The backscattered electron detectors in our
SEX give a view of the particles normal to the
pounting surface. This introduces obvious ste:
Logical bias if the particles are mounted on
thelr Largest faces, and a second set of detect~
ora to give biaxial vievs will be fitted around
the collimator of the energy-dispersive X-ray
detector. As vell as giving the sane View for
X-rays an for backscattered electrons, this will
allow better estimates of particle sizes in three
dimensions (12).
Im order to obtain estinates of the probab-
Alities of cbserving the different compositional
‘areas on random particles, we have mounted a
number of particles on a small goniometer, and
have obtained digitized images of ope normal axis
view and five oblique views at a 60° angle fron
‘the normal, spaced 72° apart around the normal
axis. The results of this analysis will be pub-
Lished in the future,
Many stereological problens renain to be
solved, but it is expected that with the iner-
ed ease of obtaining sufficient 2-dinens:
ional statistical data, and with the particular
parameters vhich we are able to obtain, namely
intercept lengths, areas, true perimeters, or-
dered perimeter arrays, centroid positions and
the dimensions of the rectangles bounding the
particles or
positeness distributions can be developed.
Retults
‘The automated scan control and data acqui
ition “hardvare” and the necessary software
control prograns are fully operational. Timing
tests shoved that the operations involved be
tween halting the beam, interrupting the com
puter, transferring the beam position and con
position code values to menory, and restartingthe bean scan take a total of 2.9 milliseconds.
‘mis Includes a deliberate hardvare delay to
allow discriminators to settle, and a software
olay to alloy reliable loading and reading of
the buffers, Tt also inclodes the tine taken by
the operating systen to perform all necessary
vapidtey ‘checks. "in fallablliey tests, up to
sequential interrupt and data transfer
cycles vere performed without ezror.
‘The algorithms and prograns to determine
perimeters, areas and centroids perform reliably
and rapidly Off-Line in FORTRAN, and in the COC
7600 system to which we have access, computation
tines are fractions of a second per particle.
hen the programs are used on-line in our INTER-
DATA 70 system for a particle as shown in Pig~
lure 3a, occupying a 50x 50 raster image, the
perimeter and area computation taki
Ly 8 seconds. It is expacted that this tine
will be considerably reduced once the programs
are vritten in assenbly language.
Signal processing provides most of the re
maining problems in the data acquisition system,
especially vhen vhole particles with rough sur~
faces are being examined. The backscattered
electron signal particularly ie affected by auch
roughness, and discrininator levels are sonetines
aitticult’to set. This aspect of our system 1
being further developed.
In general, digital inayes of moderate -
stzed whole grains, and especially of sectioned
exposures, are readily obeained, and a typical
‘example of an inage from a whole grain is shown
in Figure 3. The raster spacing ie that used
for data acquisition. Much more finely detatl~
fe4 inages can be formed by the 4000 x 4000 raster
wmode, which in fact provides higher spatial dis:
crimination than it is possible to reproduce on
the 1000 line Ckr fitted to our SEM.
‘the paraneters derived for the particle
‘shown in Figure 3 are given in Table 1. Apart
from this dizect numerical information a nunber
of further paraneters can be derived, such as
fitted equivalent ellipses; perimeter to area
ratios and ratio functions and Feret's diameters
land average chord lengths, Information such
Second moments of line intercepts, touch points
of the particle on the bounding rectangle, max~
imum chord in the X-direction and the number of
upward or downward convexities or concavities
‘can also be obtained within the program. Th
particular, the intercept lengths and moments
Fequired for the stereological Interpretations
used by Jones and Barbery (3, 4) would be a sub-
‘et of our information. Such easentially one~
Ginensional data requires severely limiting ass-
unptions for its interpretation, and directly
Getermined areas and perineters carry far high
fer inforaation content.
In the present paper we have described our
autonated SEM control system, and the treatment
405
of the digitized SEM inage to obtain particle pare
wnatars. ‘The application of these measurement
techniques to suites of samples, and the further
development of the perineter - centroid relation-
ships to give shape information such as is dis
cussed by Davies (8) will be the subject of
fature publications.
Acknowleagenen te
‘The authors express their gratitude to
1, G. Dorfler and J.C. Ruse, "A system for
Stereonetric Analysis with the Scanning Electron
Microscope", Scanning Electron Microscopy/1970,
TET Research Institute, Chicago, Illinois, 67=72
(4970).
2. BLM. White, 6.6. Johnson, Jr. and KA.
Mexinstxy, "Computer Processing of SEM Inages",
Scanning Electron Microscopy/1968, IIT Research
Institute, chicago, flinois, 95-103 (1968).
3. MP. Jones and J.L. Shaw, "Automatic Measure-
inant and stereological Assessmant of Mineral Data
for Use in Mineral Technology", 10th Int. Mineral
Processing Congress, MJ. Jones (E4.), The
Institution of Mining and Metallurgy, London, 1974,
pp. 737-75
4. G. Barbery "Determination of Particle size
Distribution from Measurements on Sections” Powder
‘Technology, 9, 231-240 (1974).
5. BM. White, HM. GSEz, G.G. Johnson Je.» and
Rig, MoMillan, “Particle Size Distributions of
jarticulate Aluninas from Conputer-processed SEM
Images”, Scanning Electron Micros
Research Institute, Chicago, 49°64 (1970).
6, E.W, white, K. Mayberry and G.G. Johnson
1 “Computer Analysis of Multichannel SEM
and x-ray Inages from Fine Particles.”
Pattern Recognition, 4, 173-193 (1972).
7.8: Troutman, G.G. Johnson Jr., £.W. white
and J, Lebledzik, “Automated Quantitative
SEM Characterization of Complex Particulate
Samples", International Laboratory, March/April,
1974, 11-26.
8. _R, Davies, "A Simple Feature-space Rapre-
sentation of Particle shapes” Powder Technolozy,
42, 1ni-iz4 (1975).
9. Avr. Rold and M. Zulderwyk, "An Interface
System for Minicomputer Control of Instruments
and Devices” CSIRO Mineral Research Laboratories,
Division of Mineral Chemistry, Investigation
Report 115, August, 1975. (available through
CSIRO, Box'124, Port Melbourne, Vic. 3207,
Australta)407
Figure 2. Schematic of search and scan proced-
re for particle location and examination.
Search areas are defined by stage movenents
Ox, OY, and the scanned area by OX,, O¥;. The
overlap dimensions X,-X, and ¥,-¥, are set at.
the daximm chord of the largest particle. All
particles within the frane are located by an
initial fast scan and the rectangles defining
their minina and maxina are determined. Those
cutting the frane, e.g. 1 and 6, are not exan-
{ned further. Those on the bousdary of the
overlap area are always included, but when
spaced out rather than adjacent areas are
examined, particles wholly within the overlap
band, such as 2, are omitted
‘TABLE 1
(COMPUTED PARTICLE PARWMETERS”
SAMPLE BOUGATNVILLE 7-73-106
PRAME 1
PARTICLE e
RECTANGLE
Location x 210
Locarrow 320
DIMENSION x 50
DIMENSION ¥ “
wo. oF cHoRDS 39
MAX x CHORD as
xKIW 3
XMAK 2
YIN 3
YHAK a
cENTROID x 26.6
CENTROID ¥ 2215
PERIMETER 142.5
AREA 1267
SUBAREA TYPE 2 67
Figure 3. (a) Rastar image of a quartz-chalco-
pyrite composite particle obtained with back-
Scattered electrons. PW = 50 um.
() Digital sap of image (a). 1= quarts, 2=
chalcopyrite. 1 character space = 1 ym.” In the
computer array, only the coordinates of each
change of position are stored.10, 1. Young, “Localization of cellular
Structures", IEEE Transactions on Bicnedical
Engineering, EME~22, 35-40 (1978).
AL. 0. Jahari and P.B. DeNee, "Handling,
Mounting and Examination of Particles for
Scanning Electron Microscopy" Sc
‘Scanning
Electron Microscopy/1972 TIT Research Tastitute,
Chicago Hilinois, 249-256 (1972).
12, ELE. Underwood, “quantitative stereology”
Addison-Wesley, New York, 1970, P. 264-246,"
13. S.B, Gray, “Local Properties of Binary
Images in Two Dinensions™, IEEE Transactions on
Computers, 6-20, 851-561, 1971.
"computer Processing of
60),
14. H. Freeman,
Line-Drawing Taages", Computing Surveys
57-97 (1974).
DISCUSSION WITH REVIEWERS
Reviewer
ysis
How ong does a full picture ana-
Authors: This tine is a simple mitiple of the
tine per particle.
defining fairly complex particl
dvell tine per point for back-scattered electrons
(BSE), the recording tine is 0,25 seconds. For
x-ray, with a dell tine 2x10" seconds, record~
ng tine is 50 seconds. Computing tines have so
far boen tested only for single, although com-
plex, particles. With 50x30 raster envelope,
Gnilne computing time in compiled FORTRAN was
B seconds (off-Line on a CDC-7600, the tine was
Jess than 0.2 seconds). We are still testing
the overall procedure for multiple fields of
particles. Computing tine per particle is
‘essentially proportional to the number of raster
Lines. For simpler composite particles on a
20x20 ‘raster, 20 msec dwell time, and about 2
seconds computing time, we expect that 100 par-
ticles will take approximately 1000 seconds for
x-rays and 280 seconds for BSE.
Reviewer IIT: Can you specify, beam current,
‘spatial resolution, number of fields, number of
particles and the type of output that will be
Obiained tor a specific problem.
Bean currents are typjeal bac
ed electron imaging values, 10°® to 10~
Spatial resolution is dependent on edge
ness, and on the nature of the electron
response. The digital scan system can resolve
secondary electron images dovm to the ultinate
resolution of the SEM, at least 100%. For
fields of 10 to 1000 microns diameter, the
spatial reproducibility as judged by beam tracks
impressed in electron resist materials, is
essentially within the 3 bit resolution of our
high-stability 12 bit D/A convertors, i.e, 21 in
8000. The nunber of fields can be made as great
as desired. Fields are obtained by automat
stage movenents which bring successive sample
408
areas under the beam centre. The contents of
each field are examined, and their paraneters
‘computed and printed (or stored) before the next
field is examined. The nunber of particles which
can be outlined for subsequent exanination is as
Nigh as the computer storage will allow at four
coordinates per particles. The output is as given
in Table 1
Reviever II: What statistical results can be
‘outinely obtained?
Authors: For each particle, parameters are ob-
tained as given in Table 1.” Statistical distri-
butions for suites of particles are obtained from
the populations of single particle paraneters.
Reviewer II: Can all the additional parameters
you mention be derived from the data?
Authors: Yes.
Reviewer III: What criteria are used for proper
Setting OF discriminator levels?
Authors: Of the three methods described by Morton
‘{GSTH Publication STPS04, 1972, pp.£1-94) wo use
manual threshold setting at the $04 maximun-nini-
mun value as displayed on a vaveform nonitor dur-
ing a line scan. The entire inage is then check-
ed'for spatial fidelity in teras of an undiscri-
‘inated inage. The inage, rather than the dis-
crininators themselves, is used as the final
Selection criterion for threshold settings.
hadowed areas" determined?
Reviewer I: How are "
Authors: A shadowed area is recorded as back-
round by the detectors, but is recognized by
Software as lying within a particle boundary. Its
area is either added to that of the major compo-
nent, or is determined if holes are actual f
‘tures of interest.
Reviewer III: What particle configurations are
now al
‘Authors: Any configuration is allowed, provided
that ft has a recognizable boundary. Touching
particles of the sane phase can be discriminated
only if there is a region of contrast at the con-
tract, as usually occurs in secondary electron
images. Touching particles of different phases
are discriminated as a matter of course, lioles
t affect
ina particle are recognizable and do
‘the outcone. They can be treated as subareas,
or equated to the major particle composition.
Discussion continued on p. 450.Additional discussion of paper on '‘Multicompositional Particle Characterization Using the SEN-Microsprobe""
by G, Grant et. al., continued from p. 408.
Reviewer I: what are statistical limits and
fests deteraining whether an element is present
or absent at a given raster location?
Authors: The 90% confidence level is taken as
veak)% - 2(Background)” > (Background), where
Peak and Background values are initially’ deter-
ined from counting tines mich longer than dwell,
times. The acceptance level is then set in
terms of a dwell tine and overflow counter value,
and the inage is tested with these values. If
Correct spatial definition is not obtained, dvell
times and overflow values are increased.
Reviewer IT: Is it not the case that two BSE
jetectors cannot give stereo pairs?
Authors: This is so, and the two sets of BSE
detectors will be used to obtain two distince
views of the sae particl
Wat acguracy of beam positioning
10-7 sec. dvell time and 20x20
point density? To ay knowledge no comercial
SEM can handle that speed, or even close. If
the time constant of apalog discriminators is
set to fixed value 10°45, why vould one change
the dwell tine toa value different than 1045
or so,since this gives the optional filtering of
the analog signal? For exazple, 10-7 dwell tine
will be totally useless since at best at 4009
points per line, the line tine will be 4x10"
(or 10-7x4000) thus any change in contrast will
be filtered out.
Authors:” 10-7 seconds is the inverse of the
digital tine base, 10 12. This perigd is not
used in practice as a dweil tine, 10° seconds
being the usual minimun. However, the slow
speed to the next dwell pgint, say 200 raster
points avay, is 200 x 10°? seconds, and the
Rinimim tine for a complete 4000x4000 raster for
visual or photographic viewing is 1.6 seconds.
What is the cost of the systent
Reviewer
Authors: Approximately, US $75,000 for the
Scanning microscope with energy dispersive X-ray
detector and moter driven stage, US $15,000 for
computer system, and US $10,000’ for the digital
‘beam scanner and computer interfaces.
Reviewer I: How many bits are the D/A conver~
ters!
Authors: The 0/A converters are 12-bit high
‘Stability Burr-Brown bipolar DAC-40 series.
Reviewer I: How is your device different from
one by Yew & Pease (IITRI/SEM/1974 p, 191-197).
Authors: In our system, signals from X-ray or
STectron detectors are digitized as received and
‘these digital values are stored in the computer
450
in a condensed digital array. In the Yow & Pease
System, the inage is physically stored on an array
of 206’ solid state capacitors, and 1s thus able
to be retained for signal processing or enhance.
tent before re-disply as visual oF photographic
image.
Reviewer III: To what real problems has the sys-
‘tem actually been applied?
Authors: Primarily to measuring the perimeters,
‘reas and subareas of a large number of individual
particles such as shown in Figure 3, These data
fre being used to establish statistical and geo-
metrical paraneters for the occurrence probabi-
Lities of particle inclusions. The digital scan
control and data collection programs have been
thoroughly tested in this way. Only a linited
number of test runs have been made with specinens
containing a nunber of particles mounted on sli
oF metal foil, and systenatic and routine seasur-
iments are just beginning. Stage movenent and
control to bring successive fields into view has
been extensively tested and is fully established,
‘The perimeter and area programs have been run off
Line, for narticles of extreme complexity without
Jogical failure, and have been run on-line (in
compiled FORTRAN forn) for particles such as
shown in Figure 3, with output parameters as in
Table 1, The prograns are “on-line” in the sense
of being resident. in the control computer and
operating on the digital inage array once it has
been obtained.