- Email: [email protected]
Computer control for X-ray and energy loss line profiles and images
Ultramicroscopy North-Holland
8 (1982) 341-350 Publishing Company
COMPUTER
CONTROL
341
FOR X-RAY AND ENERGY LOSS LINE PROFILES
AND IMAGES
P. REZ and C. AHN * Materials and Molecular Research Division, Lawrence Berkeley Laboratory Engineering. University of California, Berkeley, California 94720, USA Received
8 March
and Department
of Materials
Science and Mineral
1982
A system for computer control and energy loss signals discussed.
of a TEM/STEM Results showing
is described and advantages of computer control for imaging line profiles and images using analytical signals are presented.
1. Introduction It is often desirable to display variations of in a specimen as a line elemental concentration profile or an image. The analytical signals from which this information can be derived are the characteristic X-ray peak intensity or the intensity under an energy loss core edge. In these cases the ability to perform mathematical manipulations on the data might be more important than speed of data acquisition. These signals are also easily available in scanning transmission (STEM) mode. One of the claimed advantages of STEM is the ability to do manipulation on sequential streams of data. There have been a number of cases where computers have been interfaced with STEM instruments. However, the microscope scanning has always been performed by a dedicated digital scan generator [l-3]. It is generally believed that directly driving the scan from a computer would be too slow for most purposes. The minimum time per pixel would be limited by the machine cycle time and execution times for various instructions. For a DEC LSI-11 assuming only reading a fast converter or counter and one movement of the beam for each pixel a time of 30-50 ps would be typical. This means that a 256 X 256 image would * Current
address: University of Bristol, H.H. Wills Physics Laboratory, Royal Fort, Tyndall Avenue, Bristol BS8 ITL, UK.
take3s,a512X512imagewouldtake12s,anda 1024 X 1024 would take 50 s. However, the X-ray and energy loss signals are not coming in fast enough to justify such fast scans. The EDX signal has a maximum rate of about 10 kHz (even with a short amplifier time constant). This includes all elements and a rate of about I-5 kHz might be more appropriate for the element to be mapped. The inner shell energy loss edge signal varies according to spectrometer design and how the spectrometer is matched to the microscope. In a very good system, rates of about 100 kHz are attainable at the carbon K edge. This includes background and the rate for the core loss-r signal for the element being mapped is probably closer to 10 kHz. To ensure adequate statistics at least 100 to 1000 counts are needed per pixel which implies minimum pixel times of at least 10 ms per point. A similar count rate can be expected in scanning Auger systems. This is much longer than the time taken by the computer to move the beam, so for imaging with analytical signals computer control is perfectly adequate. Furthermore it is absolutely necessary in energy loss (and probably desirable for X-ray) to subtract the background. Computer manipulation of the data is therefore essential. A useful equation summarizing image collection is:
&!x
~2
0304-3991/82/0000-0000/$02.75
0 1982 North-Holland
using X-ray
R ’
342
P. Rez, C. Ahn / Computer control for X-ray and energy loss line profiles and images
where T is the image time, N the number of pixels or “resolution”, C the contrast of the feature to be detected, R the rate (in events/s) of the signal being used, and k a constant usually taken to be 5. The probe size should be matched to the pixel size at the operating magnification. Choosing a probe size larger than the pixel size is obviously of little value. It is well known that weak signals require long scan times. Computer control enables the user to easily vary resolution and image time to match contrast needed. Very often high resolution is not required and a 64 X 64 image will give the same information as a 512 X 512 image which might look more acceptable. Computer control of the microscope brings other benefits. For the instrument designer it can make the microscope easier to use. For researchers the ability to control different scans and manipulate different signals by relatively simple software changes makes the microscope a much more flexible instrument.
2. Instrumentation The various experiments described below were performed on a modified Philips 400 TEM/STEM at Berkeley. The STEM unit was modified so that the scan could be switched between the internal scan generator and external scans derived from 2 digital-to-analog Converters (DACs) on the computer bus. Another DAC was used to modulate the intensity of the STEM unit display. Grey levels were generated using a restricted range of the DAC output so that a zero output could be used for beam blanking of the display. As the 12 bit range of the DAC covered the width of the STEM screen, it is necessary to fill in a large number of points to obtain an image of reasonable size. It was decided to fill in an area roughly equivalent to the small scan region. This corresponds to 2048 X 2048 points and takes about 1 min. It would probably have been better to either change the range of the DACs with an operational amplifier or build a special board to fill in pixels by hardware. These refinements were not considered to be necessary for evaluation of the system for taking images. The output of a fourth DAC was used as
an alternative ramp source for the energy loss spectrometer which usually gets its ramp from the Kevex 7000 MCA [4]. The X-ray signal comes in through the Kevex 7000 and is read directly from memory. The energy loss signal is taken from the single electron counting electronics used to produce TTL pulses for the Kevex 5180 [4]. These pulses are also fed into a counter card on the computer bus. It is also possible to scan a diffraction pattern across the spectrometer entrance aperture by operating the microscope in diffraction mode, and connecting the x, y DAC outputs to the external beam deflection system inputs. These features are all shown schematically in fig. 1. The major disadvantage of using DACs in the computer backplane for scanning is high frequency noise pickup. This has a deterimental effect on energy loss scans and resolution is degraded from 3 eV to about 8 to 10 eV. The restricted angular acceptance of the present spectrometer in STEM mode and the relatively poor probe forming capabilities of the present objective lens also seriously limit performance. At present the maximum count
SCAN OEFLECTION INTERNAL
I \ i-
STEM
i
UNIT
COILS
COILS\
I, ,_*-~1
EoX
SCAN
i
(
/
I
-I
PUT
SYSTEM
EELS
RAMP
-
COMPUTER
Fig. 1. System diagram.
+
COUNTING
BUS
-r-
P. Rez, C. Ahn / Computer control for X-ray and energy loss line profiles and images
rate in STEM mode at the smallest camera length is 10 kHz. This could be improved by a factor of about 10 if the entrance aperture were enlarged from 0.6 to 2 mm. To maintain resolution a spectrometer with second order aberration correction such as the GATAN 607 should be used. To increase current for a given probe size the twin lens should be fitted but ultimately energy loss imaging should be performed with a field emission source. It would then probably be necessary to correct for drift and fluctuations in the beam current using analog to digital converters monitoring bright or dark field signals. The procedure used to take an image using either the EDX or the energy loss signal is the same. A spectrum is first taken from a point on the specimen where the element of interest is likely to be present. This gives some indication of the strength of signal to be expected. It is also necessary to decide on the background fitting windows for the energy loss edge or the peak region in EDX. A line profile is then taken across this region and the results are stored in the Kevex MCA. From this profile the range of the signal can be estimated so that when the image is taken black corresponds to the absence of the element and white to the maximum local concentration. If the image is stored on the disk during acquisition it is possible to “play back” the image with an expanded or contracted range of grey levels or to subtract a black level from it.
400
343
..
J0
0-
EV 400
200
600
Fig. 2. Energy loss spectrum of Fe showing dwell times for different regions.
800
the use of different
sufficient to use the number of counts in a channel after the edge as a measure of concentration [7,1 l] as has been attempted in the past [5,6]. The backCOUNTS I
t
400
600
800
700
Cl _ 3UNTS
2000
3. Energy loss imaging Computer control is also very useful for energy loss spectroscopy without imaging. In an energy loss spectrum the zero loss peak can be lOOO10,000 times the height of the core edges. To accommodate all the features on a display it is necessary to use shorter dwell times in the low loss regions or do more sweeps in the high loss part of the spectrum (fig. 2). Another problem is the effect of drift on spectra taken with multiple sweeps. It is possible to get the computer to detect the position of an edge and then shift each successive sweep so that it lines up with previous sweeps (fig. 3). In general, for energy loss imaging, it is not
I/
2N0
1000
0
600
SWEEP
TOTAL
\
700
800
Fig. 3. Energy loss spectrum of Fe L, s edge showing computer correction for drift between successke sweeps. The bottom figure shows that the second sweep has been shifted before being added to the first sweep. The drift was introduced by changing the magnet excitation current.
P. Rez. C. Ahn / Computer control for X-ray and energy loss bne profiles and images
344
ground must first be removed and it has been suggested that it is sufficient to take one point before the edge. This has been implemented in TEM schemes using photographic subtraction [8] or digitization of energy selected micrographs [9]. Although it was found that such a procedure was
adequate for undifferentiated Auger spectra, it was necessary to take two background points for energy loss spectra. A straight line was fitted, extrapolated to the point after the edge and then subtracted. Jeanguillaume et al. [lo] describe a scheme where this is done by dedicated hardware.
:OUNTS
400
EV 200
400
600
Fig. 4. (a) Bright field STEM image of yttrium aluminium garnet (YAG) specimen showing direction of scan for line profiles. (b) Line profile m Y M,,, edge with 400 A probe. (c) Line profile in Y M 2,3 edge with 100 A probe. (d) Line profile in Y M,,, edge with 400 A probe in which thickness effects are cancelled by dividing the edge intensity by the background intensity. (e) Spectrum showing Y M,,, edge and points used for background fitting.
P. Rez, C. Ahn / Computer control for X-ray and energy loss line profiles and images
Over small energy ranges there should not be too great an error in using a straight line which is quick to compute. With an array processor in the system computing time is no longer an important factor and background fitting schemes such as those used for quantitative analysis can be employed [ 141. Examples of energy loss line scans are shown as figs. 4b-4d for an yttrium aluminium garnet specimen. The bright field image is shown as fig. 4a and the direction of the line scan marked. The spectrum showing the yttrium M,,, edge is given as fig. 4e. The first two line scans really show the effect of thickness. The height of the edge is related to the number of atoms and this is proportional both to the local concentration in the area defined by the probe and the local thickness. The thicker region on the left hand edge shows up in both the EELS line profile and the STEM image. The first scan shows the results of the large probe (400A). The resolution is worse but the signal quality is better due to the higher current. Johnson [ 1 l] has suggested that it might be possible to produce a line trace which only shows the effects of changing concentration by dividing the signal by the number of counts in the background. This is shown as fig. 4d and it appears to have worked though the profile is very noisy. If the specimen gets too thick, the energy loss edge changes shape and eventually disappears in a thick specimen as shown in fig. 5a for Si L,,,. This means that energy loss edge intensity is only proportional to thickness in the thinner parts of the specimen. This is shown in fig. 5c which was the Si L,,, edge to image a cross-section through an MOS device. The specimen was prepared from two devices glued together and subsequently thinned. The metal in this case is titanium (it does not show on the image) and the region between the two metal strips is a glue for holding the two devices together. There is not much that can be done about the changes in edge shape for the general case. Although it is possible to recover the single scattering profile the time needed to acquire a complete spectrum for each point would be prohibitive in imaging applications even though the computing time with an array processor might only be of
345
1000..
J EV
0 0
200
400
Fig. 5. (a) Spectrum of silicon showing the effects of thickness on intensity in the L,,, edge; the spectrum in dots is from a thicker specimen. (b) Bright field STEM image of MOS device. (c) Image formed in Si L,., edge of same device; the dark lines are the metal regions. Marker represents 1pm.
order IO-20 ms per point. This problem might not be so important for biological specimens. Fig. 6a shows a spectrum from a ribosome stained with
P. Rez, C. Ahn / Computer control for X-ray and energy loss line profiles and images
346
COUNTS”1000
a
8
200
100
0
COUNTS 4000
2
4
MICRONS
Fig. 6. (a) Spectrum showing U O,,, edge from some ribosome specimen stained with uranyl acetate. (b) Line profile in U O,,, edge. (c) Bright field STEM image of ribosome specimen stained with uranyl acetate; line indicates trace of line scan. (d) Image formed with U O,,, edge. Marker represents 1 pm.
many1 acetate. A line profile fig. 6b and an image, fig. 6a, were taken using the uranium O,,, edge and these can be correlated with the bright field STEM image fig. 6d. The timing of energy loss images can be calculated as follows: Two background points and one point after the edge are taken for each pixel. Usually 20 ms is spent on each point and the images are recorded with 128 X 128 pixels giving a total time of about 15 min. The same scheme that is used to produce energy loss line scans can also be used in diffraction mode to produce energy selected diffraction patterns or display angular distributions of selected edges. An
COUNTSz.1000
1st
12
a
4
s
on
---y-+,
J
1 I
0
-10
Fig. 7. Angular
I
distribution
10
of Fe Me,,, edge.
MRAO
P. Rez, C. Ahn / Computer control for X -ray and energy loss line profiles and images
example, a quick scan (128 points in about 30 s) of the angular distribution for the iron M,,, edge is given in fig. 7. These capabilities were employed by Batson [12] to study the low loss spectra of aluminium and Zaluzec [ 131 has scanned out diffraction patterns (without energy selection) using the ramp from an MCA/MCS.
4. X-ray imaging The biggest limitation in X-ray imaging is the low count rate. Although the software was written to allow for the possibility of background subtraction, this has not been used as there were not enough counts. The specimen shown in fig. 5 is shown again at higher magnification in fig. 81 Line
347
scans across the titanium layer are shown in fig. 9. The top scan is for silicon and the bottom is for titanium. There are complications as the specimen had to be tilted 30” towards the detector and this explains why there is some overlap. The silicon and the titanium images are shown as figs. 8c and 8d. The area at the lower left in the silicon image is thinner (see also the line profile). As with energy loss imaging, the strength of the signal represents the combined effects of thickness and concentration. Unfortunately it seems as if it will not be possible to use a signal derived solely from X-ray spectra to normalize for thickness effects. The background in TEM X-ray spectra is not entirely due to bremsstrablung (which is proportional to thickness) and even if it were, there are still problems with lack of statistics due to low background
Fig. 8. (a) Bright field STEM image of MOS device; line is the trace of the line scan. (b) X-ray dot map showing titanium distribution. (c) X-ray digital image (64X64) showing silicon distribution. (d) X-ray digital image (64X64) showing titanium distribution. Marker represents 1 pm.
P. Rer, C. Ahn / Computer control for X-ray and energy loss line profiles and images
348
quite
weak.
It is therefore
impossible
to do
scans,and imaging using computer controlled
01
fast
scans is quite feasible. A problem with both energy loss and X-ray imaging is the need to correct for thickness effects. There are complications in energy loss imaging due to edge shape changes with thickness. The fundamental limitations derive from the low strength of these signals and the design of the microscope column rather than the computer control.
1
2
MICRONS
Acknowledgements
COUNTS
5001b 400,.
300..
200
”
100..
0-
1
Fig. 9. (a) X-ray titanium.
line scan for silicon.
2
(b) X-ray
MICRONS
line scan for
count rates. It might be worth using energy loss signals to derive thickness information to correct X-ray data [15]. The count rate for the titanium signal could be as low as about 200 counts/s, and times per pixel of 100’ms or more had to be used. Images of 64 X 64 resolution could take as long as 10 min (including the overhead or reading from the 7000 MCA). This could still result in an improvement over the conventional dot map image (fig. 8b) as it is possible to discriminate contrast easier if there are many grey levels available. In a dot map there are only two contrast levels and although they are often recorded at high resolution (see fig. 8b) the noise means that the image quality is inferior to the lower resolution digital image.
5. Conclusion Signals used for displaying variations of chemical composition such as energy loss or X-ray are
We are deeply indebted to Bob Mueller of Philips who modified the STEM unit. We should also like to thank Ian Ward and C.A. Evans for use of facilities at Evans and Associates, M. Sarikaya and P.N.T. Unwin for specimens, O.L. Krivanek, R.D. Leapman, C.E. Fiori, P.E. Batson and J.F. Konopka for helpful discussions. We are most grateful to Professor G. Thomas for his encouragement and providing funds through NSF contract DMR No. 80-23461. Research facilities were provided by the Director, Office of Energy Research, Office of Basic. Energy Sciences, Materials Science Division of the US Department of Energy under Contract No. W-7405-ENG-48.
References [II M. Strahm
and J. Butler, in: Proc. 37th Ann. EMSA Meeting, 1979, p. 598. Ed. 0. 121A.V. Jones and K.C.A. Smith, in: SEM/1978/1, Johari (1978). [31 J.A. Zubin and J.W. Wiggins, Rev. Sci. Instr. 51 (1980) 123. 141 O.L. Krivanek, in: Proc. 37th Ann. EMSA Meeting, 1979, p. 530. K.H. Yang and H.F. Kappert, in: Electron [51 R. Anderson, Microscopy 1978, Vol. I (1978) p. 536. to Analytical Microscopy, WI D.M. Maher, in: Introduction Eds. J.J. Hren, J.I. Goldstein and D.C. Joy (Plenum, New York, 1979) p. 259. 171 G. Zanchi, J. Sevely and B. Jouffrey, in: Electron Microscopy 1978, Vol. I (1978) p. 538. C.W. Porter and J. Bernacki, PI SW. Hui, F.P. Ottensmeyer, in: Proc. 37th Ann. EMSA Meeting, 1979, p. 312. S.W. Hui, J.L. Costa and J.J. Bailey, in: [91 M.A. Douglas, Proc. 37th. Ann. EMSA Meeting, 1979, p. 512.
P. Rez, C. Ahn / Computer control
for X-ray
[IO] C. Jeanguillaume, P. Trebbia and C. Colliex, Ultramicroscopy 3 (1978) 237. [ 1I] D.E. Johnson, in: Introduction to Analytical Microscopy, Eds. J.J. Hren, J.I. Goldstein and D.C. Joy (Plenum, New York, 1979) p. 245.
and energy loss line profiles and images
[ 121 [ 131 [ 141 [ 151
P.E. Batson, PhD Thesis, Cornell University (1976). N.J. Zaluzec, PhD Thesis, University of Illinois (1978). J. Butler, private communication. C.E. Flori, private communication.
349
8 (1982) 341-350 Publishing Company
COMPUTER
CONTROL
341
FOR X-RAY AND ENERGY LOSS LINE PROFILES
AND IMAGES
P. REZ and C. AHN * Materials and Molecular Research Division, Lawrence Berkeley Laboratory Engineering. University of California, Berkeley, California 94720, USA Received
8 March
and Department
of Materials
Science and Mineral
1982
A system for computer control and energy loss signals discussed.
of a TEM/STEM Results showing
is described and advantages of computer control for imaging line profiles and images using analytical signals are presented.
1. Introduction It is often desirable to display variations of in a specimen as a line elemental concentration profile or an image. The analytical signals from which this information can be derived are the characteristic X-ray peak intensity or the intensity under an energy loss core edge. In these cases the ability to perform mathematical manipulations on the data might be more important than speed of data acquisition. These signals are also easily available in scanning transmission (STEM) mode. One of the claimed advantages of STEM is the ability to do manipulation on sequential streams of data. There have been a number of cases where computers have been interfaced with STEM instruments. However, the microscope scanning has always been performed by a dedicated digital scan generator [l-3]. It is generally believed that directly driving the scan from a computer would be too slow for most purposes. The minimum time per pixel would be limited by the machine cycle time and execution times for various instructions. For a DEC LSI-11 assuming only reading a fast converter or counter and one movement of the beam for each pixel a time of 30-50 ps would be typical. This means that a 256 X 256 image would * Current
address: University of Bristol, H.H. Wills Physics Laboratory, Royal Fort, Tyndall Avenue, Bristol BS8 ITL, UK.
take3s,a512X512imagewouldtake12s,anda 1024 X 1024 would take 50 s. However, the X-ray and energy loss signals are not coming in fast enough to justify such fast scans. The EDX signal has a maximum rate of about 10 kHz (even with a short amplifier time constant). This includes all elements and a rate of about I-5 kHz might be more appropriate for the element to be mapped. The inner shell energy loss edge signal varies according to spectrometer design and how the spectrometer is matched to the microscope. In a very good system, rates of about 100 kHz are attainable at the carbon K edge. This includes background and the rate for the core loss-r signal for the element being mapped is probably closer to 10 kHz. To ensure adequate statistics at least 100 to 1000 counts are needed per pixel which implies minimum pixel times of at least 10 ms per point. A similar count rate can be expected in scanning Auger systems. This is much longer than the time taken by the computer to move the beam, so for imaging with analytical signals computer control is perfectly adequate. Furthermore it is absolutely necessary in energy loss (and probably desirable for X-ray) to subtract the background. Computer manipulation of the data is therefore essential. A useful equation summarizing image collection is:
&!x
~2
0304-3991/82/0000-0000/$02.75
0 1982 North-Holland
using X-ray
R ’
342
P. Rez, C. Ahn / Computer control for X-ray and energy loss line profiles and images
where T is the image time, N the number of pixels or “resolution”, C the contrast of the feature to be detected, R the rate (in events/s) of the signal being used, and k a constant usually taken to be 5. The probe size should be matched to the pixel size at the operating magnification. Choosing a probe size larger than the pixel size is obviously of little value. It is well known that weak signals require long scan times. Computer control enables the user to easily vary resolution and image time to match contrast needed. Very often high resolution is not required and a 64 X 64 image will give the same information as a 512 X 512 image which might look more acceptable. Computer control of the microscope brings other benefits. For the instrument designer it can make the microscope easier to use. For researchers the ability to control different scans and manipulate different signals by relatively simple software changes makes the microscope a much more flexible instrument.
2. Instrumentation The various experiments described below were performed on a modified Philips 400 TEM/STEM at Berkeley. The STEM unit was modified so that the scan could be switched between the internal scan generator and external scans derived from 2 digital-to-analog Converters (DACs) on the computer bus. Another DAC was used to modulate the intensity of the STEM unit display. Grey levels were generated using a restricted range of the DAC output so that a zero output could be used for beam blanking of the display. As the 12 bit range of the DAC covered the width of the STEM screen, it is necessary to fill in a large number of points to obtain an image of reasonable size. It was decided to fill in an area roughly equivalent to the small scan region. This corresponds to 2048 X 2048 points and takes about 1 min. It would probably have been better to either change the range of the DACs with an operational amplifier or build a special board to fill in pixels by hardware. These refinements were not considered to be necessary for evaluation of the system for taking images. The output of a fourth DAC was used as
an alternative ramp source for the energy loss spectrometer which usually gets its ramp from the Kevex 7000 MCA [4]. The X-ray signal comes in through the Kevex 7000 and is read directly from memory. The energy loss signal is taken from the single electron counting electronics used to produce TTL pulses for the Kevex 5180 [4]. These pulses are also fed into a counter card on the computer bus. It is also possible to scan a diffraction pattern across the spectrometer entrance aperture by operating the microscope in diffraction mode, and connecting the x, y DAC outputs to the external beam deflection system inputs. These features are all shown schematically in fig. 1. The major disadvantage of using DACs in the computer backplane for scanning is high frequency noise pickup. This has a deterimental effect on energy loss scans and resolution is degraded from 3 eV to about 8 to 10 eV. The restricted angular acceptance of the present spectrometer in STEM mode and the relatively poor probe forming capabilities of the present objective lens also seriously limit performance. At present the maximum count
SCAN OEFLECTION INTERNAL
I \ i-
STEM
i
UNIT
COILS
COILS\
I, ,_*-~1
EoX
SCAN
i
(
/
I
-I
PUT
SYSTEM
EELS
RAMP
-
COMPUTER
Fig. 1. System diagram.
+
COUNTING
BUS
-r-
P. Rez, C. Ahn / Computer control for X-ray and energy loss line profiles and images
rate in STEM mode at the smallest camera length is 10 kHz. This could be improved by a factor of about 10 if the entrance aperture were enlarged from 0.6 to 2 mm. To maintain resolution a spectrometer with second order aberration correction such as the GATAN 607 should be used. To increase current for a given probe size the twin lens should be fitted but ultimately energy loss imaging should be performed with a field emission source. It would then probably be necessary to correct for drift and fluctuations in the beam current using analog to digital converters monitoring bright or dark field signals. The procedure used to take an image using either the EDX or the energy loss signal is the same. A spectrum is first taken from a point on the specimen where the element of interest is likely to be present. This gives some indication of the strength of signal to be expected. It is also necessary to decide on the background fitting windows for the energy loss edge or the peak region in EDX. A line profile is then taken across this region and the results are stored in the Kevex MCA. From this profile the range of the signal can be estimated so that when the image is taken black corresponds to the absence of the element and white to the maximum local concentration. If the image is stored on the disk during acquisition it is possible to “play back” the image with an expanded or contracted range of grey levels or to subtract a black level from it.
400
343
..
J0
0-
EV 400
200
600
Fig. 2. Energy loss spectrum of Fe showing dwell times for different regions.
800
the use of different
sufficient to use the number of counts in a channel after the edge as a measure of concentration [7,1 l] as has been attempted in the past [5,6]. The backCOUNTS I
t
400
600
800
700
Cl _ 3UNTS
2000
3. Energy loss imaging Computer control is also very useful for energy loss spectroscopy without imaging. In an energy loss spectrum the zero loss peak can be lOOO10,000 times the height of the core edges. To accommodate all the features on a display it is necessary to use shorter dwell times in the low loss regions or do more sweeps in the high loss part of the spectrum (fig. 2). Another problem is the effect of drift on spectra taken with multiple sweeps. It is possible to get the computer to detect the position of an edge and then shift each successive sweep so that it lines up with previous sweeps (fig. 3). In general, for energy loss imaging, it is not
I/
2N0
1000
0
600
SWEEP
TOTAL
\
700
800
Fig. 3. Energy loss spectrum of Fe L, s edge showing computer correction for drift between successke sweeps. The bottom figure shows that the second sweep has been shifted before being added to the first sweep. The drift was introduced by changing the magnet excitation current.
P. Rez. C. Ahn / Computer control for X-ray and energy loss bne profiles and images
344
ground must first be removed and it has been suggested that it is sufficient to take one point before the edge. This has been implemented in TEM schemes using photographic subtraction [8] or digitization of energy selected micrographs [9]. Although it was found that such a procedure was
adequate for undifferentiated Auger spectra, it was necessary to take two background points for energy loss spectra. A straight line was fitted, extrapolated to the point after the edge and then subtracted. Jeanguillaume et al. [lo] describe a scheme where this is done by dedicated hardware.
:OUNTS
400
EV 200
400
600
Fig. 4. (a) Bright field STEM image of yttrium aluminium garnet (YAG) specimen showing direction of scan for line profiles. (b) Line profile m Y M,,, edge with 400 A probe. (c) Line profile in Y M 2,3 edge with 100 A probe. (d) Line profile in Y M,,, edge with 400 A probe in which thickness effects are cancelled by dividing the edge intensity by the background intensity. (e) Spectrum showing Y M,,, edge and points used for background fitting.
P. Rez, C. Ahn / Computer control for X-ray and energy loss line profiles and images
Over small energy ranges there should not be too great an error in using a straight line which is quick to compute. With an array processor in the system computing time is no longer an important factor and background fitting schemes such as those used for quantitative analysis can be employed [ 141. Examples of energy loss line scans are shown as figs. 4b-4d for an yttrium aluminium garnet specimen. The bright field image is shown as fig. 4a and the direction of the line scan marked. The spectrum showing the yttrium M,,, edge is given as fig. 4e. The first two line scans really show the effect of thickness. The height of the edge is related to the number of atoms and this is proportional both to the local concentration in the area defined by the probe and the local thickness. The thicker region on the left hand edge shows up in both the EELS line profile and the STEM image. The first scan shows the results of the large probe (400A). The resolution is worse but the signal quality is better due to the higher current. Johnson [ 1 l] has suggested that it might be possible to produce a line trace which only shows the effects of changing concentration by dividing the signal by the number of counts in the background. This is shown as fig. 4d and it appears to have worked though the profile is very noisy. If the specimen gets too thick, the energy loss edge changes shape and eventually disappears in a thick specimen as shown in fig. 5a for Si L,,,. This means that energy loss edge intensity is only proportional to thickness in the thinner parts of the specimen. This is shown in fig. 5c which was the Si L,,, edge to image a cross-section through an MOS device. The specimen was prepared from two devices glued together and subsequently thinned. The metal in this case is titanium (it does not show on the image) and the region between the two metal strips is a glue for holding the two devices together. There is not much that can be done about the changes in edge shape for the general case. Although it is possible to recover the single scattering profile the time needed to acquire a complete spectrum for each point would be prohibitive in imaging applications even though the computing time with an array processor might only be of
345
1000..
J EV
0 0
200
400
Fig. 5. (a) Spectrum of silicon showing the effects of thickness on intensity in the L,,, edge; the spectrum in dots is from a thicker specimen. (b) Bright field STEM image of MOS device. (c) Image formed in Si L,., edge of same device; the dark lines are the metal regions. Marker represents 1pm.
order IO-20 ms per point. This problem might not be so important for biological specimens. Fig. 6a shows a spectrum from a ribosome stained with
P. Rez, C. Ahn / Computer control for X-ray and energy loss line profiles and images
346
COUNTS”1000
a
8
200
100
0
COUNTS 4000
2
4
MICRONS
Fig. 6. (a) Spectrum showing U O,,, edge from some ribosome specimen stained with uranyl acetate. (b) Line profile in U O,,, edge. (c) Bright field STEM image of ribosome specimen stained with uranyl acetate; line indicates trace of line scan. (d) Image formed with U O,,, edge. Marker represents 1 pm.
many1 acetate. A line profile fig. 6b and an image, fig. 6a, were taken using the uranium O,,, edge and these can be correlated with the bright field STEM image fig. 6d. The timing of energy loss images can be calculated as follows: Two background points and one point after the edge are taken for each pixel. Usually 20 ms is spent on each point and the images are recorded with 128 X 128 pixels giving a total time of about 15 min. The same scheme that is used to produce energy loss line scans can also be used in diffraction mode to produce energy selected diffraction patterns or display angular distributions of selected edges. An
COUNTSz.1000
1st
12
a
4
s
on
---y-+,
J
1 I
0
-10
Fig. 7. Angular
I
distribution
10
of Fe Me,,, edge.
MRAO
P. Rez, C. Ahn / Computer control for X -ray and energy loss line profiles and images
example, a quick scan (128 points in about 30 s) of the angular distribution for the iron M,,, edge is given in fig. 7. These capabilities were employed by Batson [12] to study the low loss spectra of aluminium and Zaluzec [ 131 has scanned out diffraction patterns (without energy selection) using the ramp from an MCA/MCS.
4. X-ray imaging The biggest limitation in X-ray imaging is the low count rate. Although the software was written to allow for the possibility of background subtraction, this has not been used as there were not enough counts. The specimen shown in fig. 5 is shown again at higher magnification in fig. 81 Line
347
scans across the titanium layer are shown in fig. 9. The top scan is for silicon and the bottom is for titanium. There are complications as the specimen had to be tilted 30” towards the detector and this explains why there is some overlap. The silicon and the titanium images are shown as figs. 8c and 8d. The area at the lower left in the silicon image is thinner (see also the line profile). As with energy loss imaging, the strength of the signal represents the combined effects of thickness and concentration. Unfortunately it seems as if it will not be possible to use a signal derived solely from X-ray spectra to normalize for thickness effects. The background in TEM X-ray spectra is not entirely due to bremsstrablung (which is proportional to thickness) and even if it were, there are still problems with lack of statistics due to low background
Fig. 8. (a) Bright field STEM image of MOS device; line is the trace of the line scan. (b) X-ray dot map showing titanium distribution. (c) X-ray digital image (64X64) showing silicon distribution. (d) X-ray digital image (64X64) showing titanium distribution. Marker represents 1 pm.
P. Rer, C. Ahn / Computer control for X-ray and energy loss line profiles and images
348
quite
weak.
It is therefore
impossible
to do
scans,and imaging using computer controlled
01
fast
scans is quite feasible. A problem with both energy loss and X-ray imaging is the need to correct for thickness effects. There are complications in energy loss imaging due to edge shape changes with thickness. The fundamental limitations derive from the low strength of these signals and the design of the microscope column rather than the computer control.
1
2
MICRONS
Acknowledgements
COUNTS
5001b 400,.
300..
200
”
100..
0-
1
Fig. 9. (a) X-ray titanium.
line scan for silicon.
2
(b) X-ray
MICRONS
line scan for
count rates. It might be worth using energy loss signals to derive thickness information to correct X-ray data [15]. The count rate for the titanium signal could be as low as about 200 counts/s, and times per pixel of 100’ms or more had to be used. Images of 64 X 64 resolution could take as long as 10 min (including the overhead or reading from the 7000 MCA). This could still result in an improvement over the conventional dot map image (fig. 8b) as it is possible to discriminate contrast easier if there are many grey levels available. In a dot map there are only two contrast levels and although they are often recorded at high resolution (see fig. 8b) the noise means that the image quality is inferior to the lower resolution digital image.
5. Conclusion Signals used for displaying variations of chemical composition such as energy loss or X-ray are
We are deeply indebted to Bob Mueller of Philips who modified the STEM unit. We should also like to thank Ian Ward and C.A. Evans for use of facilities at Evans and Associates, M. Sarikaya and P.N.T. Unwin for specimens, O.L. Krivanek, R.D. Leapman, C.E. Fiori, P.E. Batson and J.F. Konopka for helpful discussions. We are most grateful to Professor G. Thomas for his encouragement and providing funds through NSF contract DMR No. 80-23461. Research facilities were provided by the Director, Office of Energy Research, Office of Basic. Energy Sciences, Materials Science Division of the US Department of Energy under Contract No. W-7405-ENG-48.
References [II M. Strahm
and J. Butler, in: Proc. 37th Ann. EMSA Meeting, 1979, p. 598. Ed. 0. 121A.V. Jones and K.C.A. Smith, in: SEM/1978/1, Johari (1978). [31 J.A. Zubin and J.W. Wiggins, Rev. Sci. Instr. 51 (1980) 123. 141 O.L. Krivanek, in: Proc. 37th Ann. EMSA Meeting, 1979, p. 530. K.H. Yang and H.F. Kappert, in: Electron [51 R. Anderson, Microscopy 1978, Vol. I (1978) p. 536. to Analytical Microscopy, WI D.M. Maher, in: Introduction Eds. J.J. Hren, J.I. Goldstein and D.C. Joy (Plenum, New York, 1979) p. 259. 171 G. Zanchi, J. Sevely and B. Jouffrey, in: Electron Microscopy 1978, Vol. I (1978) p. 538. C.W. Porter and J. Bernacki, PI SW. Hui, F.P. Ottensmeyer, in: Proc. 37th Ann. EMSA Meeting, 1979, p. 312. S.W. Hui, J.L. Costa and J.J. Bailey, in: [91 M.A. Douglas, Proc. 37th. Ann. EMSA Meeting, 1979, p. 512.
P. Rez, C. Ahn / Computer control
for X-ray
[IO] C. Jeanguillaume, P. Trebbia and C. Colliex, Ultramicroscopy 3 (1978) 237. [ 1I] D.E. Johnson, in: Introduction to Analytical Microscopy, Eds. J.J. Hren, J.I. Goldstein and D.C. Joy (Plenum, New York, 1979) p. 245.
and energy loss line profiles and images
[ 121 [ 131 [ 141 [ 151
P.E. Batson, PhD Thesis, Cornell University (1976). N.J. Zaluzec, PhD Thesis, University of Illinois (1978). J. Butler, private communication. C.E. Flori, private communication.
349