Egerton R.F. Electron Energy-Loss Spectroscopy in the Electron Microscope

Подождите немного. Документ загружается.

308 5 TEM Applications of EELS

Fig. 5.10 Low-loss ﬁne

structure in the energy-loss

function measured from

evaporated thin ﬁlms of three

pyramidines (cytosine, uracil,

and thymine) and two purines

(guanine and adenine). The

effect of electron irradiation

is also shown. Reprinted with

permission from Isaacson

American Institute of Physics

of around 10

−12

C, allowing measurements on areas down to 100-nm diameter with-

out bubbling or devitriﬁcation if the specimen was held at −160

◦

C. More recently,

Yakovlev et al. (2010) applied similar methods to study the water distribution in

frozen hydrated skin tissue, achieving a spatial resolution of 10 nm, largely deter-

mined by radiation damage; the measurement dose was of the order of 1 C/cm

Even at 1-eV resolution, different chromophores are distinguishable on the basis

of their distinctive low-loss spectrum (Reimer, 1961) and some of these dyes can

be used to selectively stain biological tissue. Electron spectroscopy or energy-

selected imaging can then provide spatial resolution superior than obtainable with

light-optical techniques (Jiang and Ottensmeyer, 1994). Since the energy losses

involved are in the visible or near-UV region of the optical spectrum (below 5 eV),

the spatial resolution will be limited by delocalization of the inelastic scattering

(Section 5.5.3); measured values down to 1.6 nm have been reported (Barfels et al.,

1998).

Spectral ﬁne structure is more easily distinguished if the spectrometer system

offers sub-electron volt energy resolution, making a monochromated TEM a pre-

ferred option. Alternatively, a ﬁeld-emission tip can provide 0.3-eV resolution at

low emission current or 0.16 eV with Fourier sharpening (Batson et al., 1992).

5.2 Low-Loss Spectroscopy 309

5.2.2 Measurement of Plasmon Energy and Alloy Composition

When an alloying element is added to a metal, the lattice parameter and/or valency

may alter, leading to a change in valence electron density and plasmon energy E

as discussed in Section 3.3.1.TheshiftinE

can be determined experimentally for a

given alloy system, using calibration samples of known composition. In some cases,

the plasmon energy varies linearly with composition x:

(x) = E

(0) +x(dE

/dx) (5.13)

For Al and Mg alloys, whose sharp plasmon peaks make small energy shifts eas-

ier to measure, dE

/dx is in the range –4 to +6 (Williams and Edington, 1976).

Least-squares ﬁtting enables the mean energy of the peak to be determined with

good accuracy (Wang et al., 1995). For aluminum and magnesium alloys, Hibbert

and Eddington (1972) achieved better than 0.1 eV accuracy using photographic

recording and a tungsten ﬁlament electron source.

To maximize the signal/noise ratio, the specimen thickness should be roughly

equal to the plasmon mean free path (Johnson and Spence, 1974), so ultrathin spec-

imens should be avoided. But in thicker specimens, the tail of the double-scattering

peak displaces the ﬁrst-plasmon peak toward higher energy loss; deconvolution is

necessary to remove plural scattering and ensure repeatability. Surface oxide or con-

tamination layers also cause a shift in peak position but are minimized by careful

specimen preparation and clean vacuum conditions. If the specimen is crystalline,

strongly diffracting orientations should be avoided (Hibbert and Edington, 1972).

The plasmon-shift method has been used to demonstrate solute depletion at

grain boundaries, to estimate diffusion constants, and to examine solute redistri-

bution in splat-cooled alloys (Williams and Edington, 1976). A spatial resolution

better t han 10 nm was achieved, partially limited by the localization of inelastic

scattering. From processing of spectrum image data, Hunt and Williams ( 1991) con-

structed plasmon-shift images to directly reﬂect elemental composition. Tremblay

and L’Esperance (1994) used the same technique to measure the volume fraction

of Al(Mn, Si, Fe) precipitates in aluminum alloys. McComb and Howie (1990)

used low-loss analysis to study the de-alumination of zeolite catalysts, which are

damaged by electron doses beyond about 6 C/cm

The plasmon-shift method has also been applied to metal–hydrogen systems,

where hydrogen usually introduces an upward shift in the plasmon energy. For these

systems, the free-electron formula gives plasmon energies that are generally too

low by 1–3 eV, but predicts differences between the metal and hydride free-electron

values that sometimes agree well with observation (Colliex et al., 1976b; Zaluzec

et al., 1981; Zaluzec, 1992). Woo and Carpenter (1992) investigated the zirconium

hydride system and found the plasmon energy to be higher in δ- and ε-hydrides

than in the γ -hydride, allowing them to identify small precipitates in the Zr/Nb

alloys used in nuclear reactor pressure tubes. The importance of these plasmon-shift

studies lies in the fact that dispersed hydrogen cannot be detected by WDX or EDX

spectroscopy, while quantiﬁcation of lithium is difﬁcult by core-loss EELS or EDX

310 5 TEM Applications of EELS

methods. Hydrogen in Ti, V, and Nb specimens may also introduce weak energy-

loss peaks in the 4–7 eV region, presumably by creating a band of states below the

conduction band (Stephens and Brown, 1980; Thomas, 1981).

Oleshko et al. (2002) have shown that the plasmon energy displays a good cor-

relation with mechanical properties s uch as elastic, bulk, and shear modulus, and

they used this fact to determine the properties of small precipitates in Al–Cu alloys

(Oleshko and Howe, 2007). The plasmon energy E

depends on the effective mass

of the participating electrons, as well as their concentration n, so a measurement of

and n (from Kramers–Kronig analysis) can provide an estimate of effective mass

(Gass et al., 2006a).

5.2.3 Characterization of Small Particles

Because small particles have a high surface/volume ratio, their excitations are

dominated by localized surface plasmon modes, as discussed in Section 3.3.7.

Surface-mode scattering involves energy losses below that of the volume plasmon

peak (Section 3.3.5); the details of this scattering depend on the geometry of the

interface and the mismatch in the energy-dependent dielectric constant. In the case

of a small probe and a single spherical particle, the low-loss spectrum depends on

the probe position and is different for a metallic sphere and a sphere covered with an

oxide layer. Ugarte et al. (1992) observed an additional peak in the 3–4 eV region,

which they attributed to a thin conducting spherical shell outside the oxide layer,

perhaps caused by oxygen depletion by the electron beam.

Plasmonic modes in small particles have important applications in optical signal

processing, biosensing, and cancer therapy, and a modern (S)TEM–EELS system

provides a good tool studying these modes. In the case of triangular silver particles,

Nelayah et al. (2007) showed that the lowest energy (1.75 eV) mode has maxi-

mum amplitude at the corners of the particles, a 2.7 eV mode at the edges, and a

3.2 eV mode around the geometrical center, as shown in Fig. 5.11a. Simulation

using a boundary element calculation gave very similar intensity distributions

(Fig. 5.11b).

Similar STEM data on silver nanorods are shown in Fig. 5.12. Plasmon reso-

nance peaks are observed at energy losses ranging from the infrared (0.55 eV) to

the ultraviolet (3.55 eV) region of the electromagnetic spectrum. In Fig. 5.12d,the

EELS and ADF image information are combined to show the resonance modes with

m = 1 − 5. These modes were in good agreement with those predicted using the

CYLNDRCAP option of the discrete-dipole DDSCAT Fortran code (Draine et al.,

1994).

Analogous images can be obtained in ﬁxed-beam EFTEM mode, as demonstrated

for gold nanoparticles by Schaffer et al. (2008). The STEM technique is generally

preferable in terms of energy resolution, whereas EFTEM allows a larger ﬁeld of

view in a short acquisition time (Schaffer et al. 2010). Surface plasmon modes

Freeware available from http://www.astro.princeton.edu/~draine/DDSCAT.html

5.2 Low-Loss Spectroscopy 311

Fig. 5.11 (a) STEM energy-ﬁltered images of silver particles recorded at energy losses of 1.8, 2.9,

and 3.2 eV. The white bars are 30 nm long. (b) Intensity maps computed by the boundary-element

method for similar energies. Reproduced from Nelayah et al. (2007), copyright Nature Publishing

Group

within arrays of holes (made by FIB machining) have also been visualized (van

Aken et al., 2010).

Silicon nanoparticles could provide a light emission system compatible with sil-

icon processing technology. Yurtsever et al. (2006) fabricated a dispersion of Si

Fig. 5.12 (a) T hickness map of a silver rod (465 nm long, 24 nm diameter), obtained from (b)the

STEM-ADF image. (c) Summed spectrum from the entire nanoantenna, showing resonance modes

(m = 1−5) and a bulk plasmon loss just below 4 eV. (d) Intensity along the rod as a function of

energy loss. Courtesy of D. Rossouw and G. Botton, McMaster University. See also Rossouw et al.

(2011)

312 5 TEM Applications of EELS

particles, by annealing s ilicon oxide ﬁlms formed by chemical vapor deposition of

silane. Cross-sectional TEM samples showed little contrast, the mean free paths

for Si and SiO

being very similar, but the plasmon energies differ substantially:

17 eV in Si compared to 23 eV in SiO

. Energy-selective tomography was used

to provide a three-dimensional visualization, by recording a series of TEM images

at specimen tilts up to ±60

◦

in 4

◦

increments. Computer software allowed align-

ment and reconstruction of the 17 ± 2 eV images in three dimensions, as shown in

Fig. 5.13a. The particles themselves had soft outlines, as expected from the delo-

calization of plasmon scattering, but they were delineated by choosing a threshold

intensity whose contour is represented in Fig. 5.13 as a mesh image, revealing the

irregular shape of the particles. This complex morphology is consistent with the

broad spectral range of photoluminescence and electroluminescence observed in

this material, while the large surface area of each particle may account for the high

efﬁciency of light emission.

Similar tomographic techniques, exploiting differences in plasmon energy, have

been applied to generate three-dimensional images of carbon nanotubes within a

polymer (nylon) matrix (Gass et al., 2006b). It is also possible to combine the

plasmon signal from a light element with a simultaneously acquired HAADF sig-

nal (from a heavy element) to show the distribution of both. This is illustrated in

Fig. 5.13b, which is a composite tomographic image recorded from a needle-shaped

specimen prepared by FIB milling (Li et al., 2009).

Alexander et al. (2008) used TEM-EELS to study the properties of individual car-

bon particles (130–600 nm diameter) collected at various altitudes above the Yellow

Sea. Kramers–Kronig analysis revealed that most particles could be categorized as

brown rather than black (strongly absorbing) carbon, with implications for climate

Fig. 5.13 (a) Tomographic reconstruction of silicon particles in silicon oxide. White “fog” repre-

sents the 17-eV plasmon-loss intensity; the Si particles are rendered as mesh images at a constant

American Institute of Physics. (b) Composite image showing the three-dimensional distribution

of silicon (green, from plasmon signal) and erbium (red, from ADF signal) in an Er-doped silicon-

rich oxide ﬁlm. Courtesy of P. Li, X. Wang, A. Meldrum, and M. Malac, National Institute of

Nanotechnology and University of Alberta

5.2 Low-Loss Spectroscopy 313

Fig. 5.14 Real part ε

of the

dielectric function for various

ﬁlm thicknesses (in

monolayers) within a

molybdenum/vanadium

superlattice, compared with

the results of an optical

thick-layer calculation. From

Zaluzec (1992), copyright

TMS Publications

change models. K-edge intensity measurements gave a range of 1.4–1.6 for the spe-

ciﬁc gravity. Aerosol particles are usually characterized by light-optical techniques,

but often within a limited spectral range and only as an average over many particles.

If transmission measurements are made with the electron beam parallel to an

interface, surface-mode contributions are maximized. Differences in scattering are

further ampliﬁed by performing Kramers–Kronig analysis, as shown for the metal

multilayer system in Fig. 5.14. As the layer spacing decreases, the structure departs

from that calculated from bulk properties, possibly indicating structural transfor-

mation to a strained layer superlattice (Zaluzec, 1992). Turowski and Kelly (1992)

recorded low-loss spectra as a function of position across Al/SiO

/Si ﬁeld-effect

transistor structures and computed the dielectric function at each position of the

STEM probe, as well as the electronic polarizability α

(E), which may be a measure

of dielectric strength. The maximum polarizability and the energy E

max

of this max-

imum were lower near the Al and Si interfaces (Fig. 5.15), suggesting that contact

materials reduce the dielectric strength in very thin oxides.

Fig. 5.15 Molar

polarizability α

within a

Al/SiO

/Si heterostructure

and the energy position E

max

of the maximum in α

derived from energy-loss

spectroscopy. From Turowski

and Kelly (1992), copyright

Elsevier

314 5 TEM Applications of EELS

Fig. 5.16 Low-loss spectra

recorded using the specular

(400) reﬂection from a (100)

MgO surface heated to

various temperatures in the

TEM. From Wang (1993),

http://iopscience.iop.org/

0034-4885/56/8/002

As discussed i n Section 3.3.6, energy-loss spectra can be obtained in reﬂection

mode within the TEM. Figure 5.16 shows reﬂection low-loss spectra from a MgO

specimen raised to successively higher temperatures. Above 500 K, the 10 eV peak

started to disappear; reﬂection K-loss spectra indicated a depletion of surface oxy-

gen atoms. Above 1400 K, the energy of the main plasmon peak and core-loss

spectroscopy suggested formation of a surface layer of MgO

, a few nanometers

thick, which remained after subsequent cooling and exposure to air.

5.3 Energy-Filtered Images and Diffraction Patterns

As demonstrated in Chapter 2, the electron spectrometer in a TEM can act as a band-

pass energy ﬁlter: inserting an energy-selecting slit at the spectrum plane results in

an electron image or diffraction pattern from a chosen range of energy loss. This pro-

cedure provides information in a convenient form; for example, a core-loss image

can indicate the spatial distribution of a particular element.

As discussed in Section 2.6, energy ﬁltering is possible not only in a ﬁxed-beam

(conventional) TEM instrument, by using an imaging ﬁlter within or below the

TEM column, but also in a scanning transmission (STEM) microscope. The various

modes of operation are discussed in Chapter 2 and by Reimer et al. (1988, 1992). A

detailed discussion of the contrast mechanisms in energy-ﬁltered images is given by

Spence (1988), Reimer and Ross-Messemer (1989, 1990), Bakenfelder et al. (1989),

and Reimer (1995). Colliex et al. (1989) discuss energy-ﬁltered STEM imaging of

thick biological specimens. Energy-ﬁltered diffraction is treated by Spence and Zuo

(1992).

5.3 Energy-Filtered Images and Diffraction Patterns 315

5.3.1 Zero-Loss Images

By operating the spectrometer in a diffraction-coupled mode (the TEM set for imag-

ing) and adjusting the spectrometer excitation or accelerating voltage so that the

zero-loss peak passes through the energy-selecting slit, a zero-loss image is pro-

duced with greater contrast and/or resolution than the normal (unﬁltered) image;

see Figs. 5.17 and 5.20. The main factors responsible for such improvement are

listed below; their relative importance varies according to the type of specimen and

whether a conventional TEM or STEM is involved. Some factors can be identiﬁed

as affecting the resolution and others the contrast of an image, although in the case

of small-scale repetitive features these two concepts are closely related (Nagata and

Hama, 1971).

5.3.1.1 Chromatic Aberration and Contrast-Reducing Effects

In the conventional (ﬁxed-beam) TEM, an energy-selecting slit centered on the zero-

loss peak eliminates most of the inelastically scattered electrons, greatly reducing

chromatic broadening of the image. For a very thin specimen and a low energy

loss E, the inelastic image corresponding to an energy loss E is blurred relative to

the elastic image by a Lorentzian point-spread function whose half-width is 2r

≈

2θ

(E/E

), and since θ

≈ 0.5E/E

, the image resolution is approximately

≈ C

(E/E

)

(5.14)

Fig. 5.17 (a) Unﬁltered and (b) zero-loss micrographs of a 40-nm epitaxial gold ﬁlm, recorded

with 80-keV electrons and 10-mrad objective aperture. Energy ﬁltering increased the crystallo-

graphic contrast by a factor of two

316 5 TEM Applications of EELS

where C

is the chromatic aberration coefﬁcient of the objective lens. In the case of

100-keV incident electrons, C

= 2 m m and E ≈ 37 eV (average energy loss per

inelastic event for carbon), d

≈ 0.3 nm. For a large energy loss (e.g., core loss),

the chromatic point-spread function is given by Eq. (2.25); the chromatic broaden-

ing depends on the semi-angle β of the objective aperture in CTEM but is several

times less than the overall diameter βC

(E/E

) of the chromatic disk, as indicated

in Fig. 2.17.

As the specimen thickness increases, an increasing fraction of the electrons

are inelastically scattered and plural scattering causes the average energy loss

to increase, while plural elastic/inelastic scattering further broadens the inelastic

angular distribution. The objective aperture is then “ﬁlled” with scattering and the

chromatic broadening is closer to d

≈ βC

E/E

(≈ 7nmforβ = 10 mrad).

As a result of these various factors, chromatic aberration becomes more serious in

thicker specimens. Zero-loss ﬁltering then substantially improves the image contrast

and resolution, particularly in organic specimens where inelastic scattering is strong

relative to elastic scattering (Section 3.2.1).

In the case of high-resolution phase-contrast imaging, inelastic scattering is

often assumed to produce a structureless background that reduces image contrast,

although if the low-loss spectrum contains sharp plasmon peaks, this plasmon scat-

tering could produce image artifacts (Krivanek et al., 1990). In general, energy

ﬁltering permits a more quantitative comparison of image contrast with theory

(Stobbs and Saxton, 1988), especially if allowance is made for the point-spread

function of the image-recording CCD camera (Thust, 2009). The s ituation should

be further improved with the deployment of multipole devices that correct for both

spherical and chromatic aberration of TEM imaging lenses.

For the examination of thick specimens, energy-ﬁltered microscopy (EFTEM)

with 80 or 100-keV electrons is therefore an alternative to the use of higher accel-

erating voltages, where chromatic aberration is reduced in proportion to 1/E

according to Eq. (5.14). However, zero-loss ﬁltering reduces the image intensity

by a factor of exp (t/λ), where λ is the total inelastic mean free path, limiting the

maximum usable specimen thickness to about 0.5 μm at 80-keV incident energy.

Staining of biological tissue creates regions containing a high concentration of

heavy-metal atoms surrounded by material comprised mainly of light elements (H,

C, O), and the resulting strong variations in elastic scattering power provide usable

contrast. Because the inelastic/elastic scattering ratio is high for light elements, elec-

tron scattering in unstained regions is mainly inelastic and is therefore removed by

energy ﬁltering, leading to a further improvement in contrast. Reimer and Ross-

Messemer (1989) reported that the contrast of large-scale features in OsO

-stained

myelin was increased by a factor of 1.3 after zero-loss ﬁltering.

Because unstained biological specimens provide very low contrast, the image is

often defocused to create phase contrast. Langmore and Smith (1992) found that

zero-loss ﬁltering increased the image contrast from air-dried and frozen hydrated

TMV images by factors between 3 and 4. This improved contrast allows a reduc-

tion in the amount of defocusing, allowing better spatial resolution and increased

signal/noise ratio or reduced electron dose to the specimen (Schröder et al., 1990).

5.3 Energy-Filtered Images and Diffraction Patterns 317

In the case of crystalline specimens, defects are visible through diffraction con-

trast that arises from variations in the amount of elastic scattering, depending on

the local excitation error (deviation of lattice planes from a Bragg reﬂecting orien-

tation). The angular width of i nelastic scattering creates a spread in excitation error,

reducing the contrast of dislocations, planar defects, bend contours, and thickness

contours (Metherell, 1967), so diffraction contrast is again improved by zero-loss

ﬁltering. Higher incident energy also provides less spread in excitation error and

chromatic aberration; Bakenfelder et al. (1989) concluded that, for a 500-nm Al

ﬁlm, zero-loss ﬁltering is equivalent (in terms of image quality) to raising the micro-

scope voltage from 80 to 200 kV. Making use of zero-loss ﬁltering and of the

increased transmission (channeling) that occurs when a crystal is oriented close to a

zone axis, Lehmpfuhl et al. (1989) obtained clear 80-keV images of dislocations in

gold ﬁlms as thick as 350 nm.

At high energy loss or large scattering angle, inelastic scattering in crystals is

believed to be partly interbranch: the character of the electron Bloch wave changes

upon scattering and Bragg contrast is no longer preserved (Hirsch et al., 1977).

Although zero-loss ﬁltering removes such scattering, the overall effect appears neg-

ligible because the probability of such scattering is low. Phonon excitation also

causes interbranch scattering, but because the energy losses are below 0.1 eV, it

is not removed by energy ﬁltering.

5.3.2 Zero-Loss Diffraction Patterns

Energy ﬁltering of diffraction patterns can be accomplished with an imaging ﬁl-

ter in a conventional TEM (Section 2.6), much more efﬁciently than scanning

a diffraction pattern across the entrance aperture of a nonimaging spectrometer

(Graczyk and Moss, 1969). Zero-loss ﬁltering removes the diffuse background aris-

ing from inelastic (mainly plasmon) scattering and makes faint diffraction features

more visible (Midgley et al., 1995). Since the inelastic scattering is strongest at

small angles (Fig. 3.7), ﬁltering should be particularly advantageous for low-angle

diffraction, improving the analysis of materials with large unit cell, such as periodic

arrays of macromolecules or nonperiodic nanostructures. Filtering has also been

used to improve the visibility of reﬂection diffraction patterns recorded in a TEM

(Wang and Cowley, 1994) and to facilitate the quantitative analysis of intensities in

convergent-beam diffraction; see Fig. 5.18.

Removal of inelastic scattering also facilitates the quantitative analysis of amor-

phous materials (Cockayne et al., 1991). After subtracting a smoothly varying

atomic scattering factor, Fourier transformation of the zero-loss diffraction pattern

leads to a radial density function (RDF) whose peak positions provide interatomic

spacings to an accuracy of typically 5 pm. This technique allowed Liu et al. (1988)

to demonstrate that the ﬁrst- and second-neighbor distances in amorphous silicon

alloys decrease by up to 40 pm after doping with boron and phosphorus.

Zero-loss ﬁltering of convergent-beam diffraction (CBED) patterns, combined

with quantitative comparison between experimental and theoretical diffraction