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

Подождите немного. Документ загружается.

318 5 TEM Applications of EELS

Fig. 5.18 Two-beam (220) CBED patterns of silicon: (a) no energy ﬁltering and 1-s exposure,

(b) zero-loss ﬁltering and 3-s exposure. From fringe spacings, the thickness was calculated t o be

270 nm (Mayer et al., 1991). From Mayer et al. (1991), copyright San Francisco Press, Inc., with

permission

intensities using the least-squares reﬁnement technique, has made it possible to

derive structure factor amplitudes and phases with accuracies sufﬁcient to detect

small changes in crystal electron density due to chemical bonding (Zuo, 2004).

Temperature factors can be measured (Tsuda et al., 2002) and also small amounts

of lattice strain. For lattice strain measurement, zero-loss ﬁltering signiﬁcantly

improves the visibility of higher order Laue zone (HOLZ) lines in the central disk

(direct beam) of a CBED pattern. The lattice parameters of crystalline TEM spec-

imens can be obtained with 0.1 pm accuracy, sufﬁcient to measure small strains

in silicon memory devices (Kim et al., 2004). The use of a ﬁeld-emission elec-

tron source allows CBED data to be acquired from specimen areas less than 1 nm

in diameter (Xu et al., 1991). Even in the Tanaka LACBED method, in which a

small SAD aperture provides a degree of energy ﬁltering, zero-loss ﬁltering can

substantially increase the contrast in the central region of the pattern (Burgess et al.,

1994).

5.3.3 Low-Loss Images

The spatial distribution of a material having a sharp plasmon peak can be displayed

by forming an image at the corresponding energy loss. For example, Be precipitates

5.3 Energy-Filtered Images and Diffraction Patterns 319

in an Al alloy appear dark in a 15-eV image (plasmon loss of Al) but brighter than

their surroundings in an image recorded at 19 eV, the plasmon loss of beryllium

(Castaing, 1975).

Making use of the shift in plasmon energy upon alloying, Williams and Hunt

(1992) processed spectrum image data to display the distribution of Al

Li precipi-

tates in Al/Li alloys. Tremblay and L’Espérance (1994) used a similar technique to

image Al(Mn, Si, Fe) particles in aluminum alloy, deducing the volume fraction of

precipitates to be 0.81%.

Inelastic scattering by surface plasmon excitation provides intensity at energies

below the volume plasmon peak. At this energy loss, small particles show a bright

outline because the probability of surface plasmon excitation is larger for an electron

that travels at a glancing angle to the surface (Section 3.3.6). If an unidentiﬁed peak

is seen in the low-loss spectrum of an inhomogeneous specimen, forming an image

at that energy loss can help to determine if it arises from surface-mode scattering.

Some organic dyes (chromophores) have absorption peaks at energies of a few

electron volts, corresponding to visible or UV photons, and can be used as chemi-

cally speciﬁc stains in light microscopy. By forming an image at the corresponding

energy loss, their distribution can be mapped with high spatial resolution in an

energy-ﬁltering electron microscope (Jiang and Ottensmeyer, 1994).

5.3.4 Z-Ratio Images

A Z-ratio STEM image is formed by taking a ratio of the high-angle scattering

(recorded by an annular detector) and the low-angle scattering, measured through an

electron spectrometer that removes the zero-loss component; see Section 2.6.6.For

very thin specimens, the dark-ﬁeld signal represents mainly elastic scattering, while

the spectrometer signal arises from inelastic scattering. Intensity in the ratio image

is therefore a measure of the local elastic/inelastic scattering ratio, which is roughly

proportional to the local (mean) atomic number Z;seeSection 3.2.1. The aim is

usually to distinguish differences in elemental composition, while suppressing the

effects of varying specimen thickness and ﬂuctuations in incident beam current. This

technique was used by Crewe et al. (1975) to display images of single high-Z atoms

on a very thin (<10 nm) carbon substrate and subsequently investigated for imaging

small catalyst particles on a crystalline or amorphous support (Treacy et al., 1978;

Pennycook, 1981).

The Z-ratio technique has also been applied to thin sections of biological tissue

(Garavito et al., 1982); see Fig. 5.19. If the section thickness is below 50 nm, so that

plural scattering is not severe, contrast due to thickness variations (e.g., caused by a

microtome) largely cancels in the ratio image, allowing small differences in scatter-

ing power to be distinguished in unstained specimens (Carlemalm and Kellenberger,

1982).

The Z-contrast image may appear to have better spatial resolution than either

the dark-ﬁeld or inelastic image, but this effect occurs because the inelastic image

is blurred by delocalization (Section 5.5.3); upon division of the two intensities,

320 5 TEM Applications of EELS

Fig. 5.19 T4 bacteriophages adsorbed t o E. coli in a 30-nm section embedded in a 24% Sn resin

(1.1-μmﬁeldofview).(a) Annular dark-ﬁeld image; (b) inelastic/elastic Z-ratio image in which

regions of lower atomic number appear bright. From Carlemalm et al. (1982), copyright Elsevier

high-frequency components of the dark-ﬁeld image are emphasized, equivalent to

unsharp masking of photographic negatives (Ottensmeyer and Arsenault, 1983). If

the contrast in the elastic or inelastic signals is too high, the nonlinear process of

division can create image artifacts (Reichelt et al., 1984).

5.3.5 Contrast Tuning and MPL Imaging

Contrast tuning denotes the ability to choose an energy loss (typically in the range

0–200 eV) where contrast is adequate but low enough for the image to be recorded

in a single micrograph (Bauer et al., 1987; Wagner, 1990). Dynamic range is some-

times a problem in zero-loss images of thick (e.g., 0.5 μm) sections of biological

tissue because stained regions scatter very strongly relative to unstained ones.

Structure-sensitive contrast in biological tissue can sometimes be maximized

by choosing an energy loss around 250 eV, just below the carbon K-edge, so that

the contribution of carbon to the image is minimized. Structures containing ele-

ments with lower-lying edges (sulfur, phosphorus, or heavy-metal stain) then appear

bright in the image, giving a reversed “dark-ﬁeld” contrast; see Fig. 5.20. Imaging at

260 eV has been used to observe microdomain morphology in unstained polymers

(Du Chesne et al., 1992).

The most probable loss (MPL) is the energy loss at which the spectral intensity

is highest. For thin specimens (t/λ < 1), the zero-loss peak is the most intense,

but in thick specimens this peak is reduced and the MPL corresponds to the broad

maximum of the Landau distribution (Fig. 3.30), around 80 eV for 0.5-μm Epon

and 270 eV for a 1-μm section (Reimer et al., 1992). An image obtained at this

loss will have maximum intensity, a desirable property for accurate focusing and

short recording times, so that specimen drift and radiation damage are minimized.

5.3 Energy-Filtered Images and Diffraction Patterns 321

Fig. 5.20 A 30-nm section showing HIV-producing cells embedded in Epon after glutaralde-

hyde and OsO

ﬁxation and staining with uranyl acetate. (a) Unﬁltered image (the bar measures

100 nm); (b) zero-loss image showing improved contrast; (c) 110-eV image showing reduced con-

trast; (d) 250-eV image showing structure-sensitive reversed contrast. From Özel et al. (1990),

Intensity is also increased by widening the energy-selecting slit, but at the expense

of loss of spatial resolution due to higher chromatic aberration in EFTEM images.

Pearce-Percy and Cowley (1976) showed that STEM images of thick biolog-

ical objects can be obtained with near-optimum signal/noise ratio if an electron

spectrometer is used to accept all energy losses below or above the MPL (giving

bright-ﬁeld and dark-ﬁeld energy-loss images, respectively). With MPL = 150 eV,

they obtained 100-keV dark-ﬁeld images with high contrast from 1-μm-thick chick

ﬁbroblast nuclei.

5.3.6 Core-Loss Images and Elemental Mapping

The ability to display two-dimensional distributions of speciﬁc elements makes

the TEM imaging ﬁlter a powerful tool in materials analysis. As discussed in

Section 2.6.5, elemental mapping involves recording at least two images, before

and after the ionization edge. The s implest procedure is to subtract the pre-edge

and post-edge images; see Fig. 5.21 This two-window procedure works well enough

for edges with high jump ratio (obtained from very thin specimens and high con-

centrations of the analyzed element) but is unsatisfactory when quantitative results

are required (Leapman and Swyt, 1983). Negative intensities can be generated in

regions devoid of the selected element (Crozier, 1995).

Another simple procedure involves dividing the post-edge and pre-edge images

(Section 2.6.5), yielding a jump-ratio image that is largely insensitive to variations

in specimen thickness and diffracting conditions. The diffraction contrast is fur-

ther suppressed by using rocking-beam illumination (Hofer and Warbichler, 1996),

322 5 TEM Applications of EELS

Fig. 5.21 A 30-nm microtomed section of a photographic emulsion, showing silver halide micro-

crystals with a AgBr core and AgBrI shell. (a)Silver-M

map obtained by subtracting 370 and

360-eV images. (b) Iodine-M

image from subtraction of 615 and 580-eV images. From Lavergne

et al. (1994), copyright Les Editions de Physique

as illustrated in Fig. 5.22, where carbide precipitates become highly visible in a

jump-ratio image recorded at the M

-edge. For very thin specimens, the ratio-image

intensity would be proportional to elemental concentration, but in specimens of typ-

ical thickness, plural scattering background components make the ratio image only

a qualitative indication of elemental distribution (Hofer et al., 1995; Crozier, 1995).

For quantitative elemental mapping, it is desirable to record at least two pre-

edge images. If t he two energy windows are adjacent to each other, as in Fig. 1.11,

Eqs. (4.51)–(4.53) or least-squares ﬁtting can be used to evaluate the background

parameters A and r, from which the background contribution to the post-edge

Fig. 5.22 EFTEM micrographs of ferritic-martensitic 10% Cr steel containing W and Mo. (a)

Zero-loss image, showing strong diffraction contrast, (b)Fe-M

jump-ratio image recorded with

conical rocking-beam illumination, showing precipitates at grain boundaries. From Warbichler

et al. (2006), copyright Elsevier

5.3 Energy-Filtered Images and Diffraction Patterns 323

Fig. 5.23 EFTEM images (E

= 200 keV, β = 7.6 mrad) of an ion-thinned foil of ODS-niobium

alloy containing 0.3 at.% Ti and 0.3 at.% oxygen. Images (a)and(b) were recorded with 20-eV

windows below the Ti L

edge, (c) with a 20-eV window just above the Ti-edge threshold. (d)Ti

is an elemental map obtained by three-window modeling, (e) a Ti-edge jump-ratio image, and (f)

an image formed from the ratio of the two pre-edge images. From Hofer et al. (1995), copyright

Elsevier

image can be calculated by the image-acquisition computer. An example is shown

in Fig. 5.23. The two pre-edge images and the post-edge image all look similar,

but when they are combined to form an elemental map, titanium oxide precipi-

tates become visible and bend contour contrast within the foil largely disappears.

Diffraction contrast is further suppressed by dividing by a low-loss or zero-loss

image (Crozier, 1995). As further conﬁrmation that a jump-ratio or three-window

image represents elemental concentration, taking the ratio of the two pre-edge

images should yield very little contrast. This is the case i n Fig. 5.23f except along

the bend contour, where intensity ﬂuctuations may have arisen from a small change

in specimen orientation between acquisition of the two images.

Because of background extrapolation errors (Section 4.4.4), three-window mod-

eling produces a noisier image than the t wo-window methods (subtraction or

division), as seen by comparing images (d) and (e) in Fig. 5.23. To achieve ade-

quate statistics, the recording time can be increased or the number of image pixels

reduced. A good strategy is to ﬁrst acquire a jump-ratio image from the area

of interest, requiring a relatively short exposure time. If the results are encour-

aging, the three-window method can then be used to obtain a more quantitative

elemental map.

324 5 TEM Applications of EELS

Fig. 5.24 Atomic resolution images of LaB

recorded in STEM mode with 120-mrad collection

semi-angle, using a 200-kV TEM with the probe-forming and imaging lenses corrected for spheri-

cal aberration. Lanthanum atom columns appear bright in the ADF image (a)andtheLa-N

image

(b) but boron columns do not show up in the images (c and d) recorded with energy losses above

the B K-edge (188 eV). Courtesy of Sorin Lazar, FEI Company

The availability of spectrum-imaging software makes the above techniques rela-

tively easy to implement. An extended range spectrum is recorded from each image

pixel and can be analyzed in different ways later (Arenal et al., 2008).

As discussed in Section 2.6.5, elastic scattering modulation of a core-loss image

can be reduced by using a large collection angle, possible without sacriﬁcing energy

resolution if the post-specimen lenses are aberration corrected. Large collection

angle maximizes the core-loss signal but gives substantially worse edge/background

ratio, except for very thin s pecimens or high edge energies. Some impressive ele-

mental maps with atomic resolution have been obtained under these conditions (e.g.,

Muller et al., 2008) but elastic scattering may exert a more subtle effect in atomic-

scale images of crystalline specimens, as illustrated in Fig. 5.24. Here the specimen

was aligned with the electron beam parallel to the [100] direction; La atom columns

are visible in the STEM–ADF image and at the La-N

(and M

) edges, whereas

images recorded with boron-K electrons show little contrast. The proposed expla-

nation is that electrons incident on the La columns are strongly scattered and excite

the adjacent boron atoms, producing a strong B K-signal (Lazar et al., 2010). The

reverse effect would be much weaker because of the small elastic scattering power

of B atoms, as shown by their absence in the ADF image.

5.4 Elemental Analysis from Core-Loss Spectroscopy

As discussed in Chapter 1, EELS offers higher spatial resolution and elemental

sensitivity than EDX spectroscopy for some specimens, while generally requiring

more skill on the part of the operator. In this section, we discuss the data collection

strategies that have proved effective in particular cases, to complement the gen-

eral description of spectrum-processing techniques in Chapter 4. Situations speciﬁc

to particular elements are discussed in this section; results from particular materials

systems are given in Section 5.7. We begin by reviewing s ome choices of instrument

and method that are directly relevant to core-loss spectroscopy.

5.4 Elemental Analysis from Core-Loss Spectroscopy 325

Parallel recording of the energy-loss spectrum increases the speed and sensitiv-

ity of elemental analysis. Variations in sensitivity within the CCD array have been

minimized by improved design and computer processing (Section 2.5.5). For STEM

instrumentation, the use of spectrum imaging (Section 2.6.4) is attractive because

it allows extensive data processing after acquisition, including the use of principal

component and independent component analysis. Similar complete information can

be recorded by an energy-ﬁltering TEM, although the radiation dose required to

achieve the same (time-integrated) signal is higher by a factor equal to the number

of images collected (Section 2.6.6). For some inorganic specimens, this increased

dose is unimportant and is outweighed by the higher current available in a broad

beam, allowing shorter recording times and less drift of the specimen and high

voltage.

As discussed in Chapter 2, spectroscopy can be carried out with a conventional

TEM operating in either its imaging or diffraction mode. In image mode, a region of

analysis of known diameter is conveniently selected by means of the spectrometer

entrance aperture, the collection semi-angle being known if the objective aperture

has been calibrated. However, chromatic aberration of the TEM imaging lenses can

prevent precise selection of the analyzed area (see Section 2.3.2) and may result

in incorrect elemental ratios (Section 2.3.3). Chromatic effects are minimized by

changing the microscope high voltage by an amount equal to the energy loss being

analyzed, a technique previously used for serial recording. In TEM diffraction

mode, the diffraction pattern should be carefully centered about the spectrometer

entrance aperture (approximately the center of the viewing screen); the collection

angle now depends on the camera length and the aperture diameter. Provided the

incident probe diameter is small, chromatic aberration is unimportant in diffraction

mode.

A small collection angle (5–10 mrad) increases the visibility (signal/background

ratio) of an ionization edge (Section 3.5) and is appropriate for lower-energy

edges. For higher-energy edges, the problem of low intensity is reduced by choos-

ing larger β. Quantitative analysis becomes problematic if strong Bragg spots (or

rings) appear just inside or outside the aperture (Egerton, 1978a). In the case

of a small probe, the convergence angle α may easily exceed 10 mrad, and to

avoid considerable loss of signal β should exceed α (Section 4.5.3). As shown by

Fig. 4.19, quantitative core-loss analysis involves a convergence correction unless

α < β/2.

The next decision is the choice of specimen region to analyze. Sometimes this is

obvious from the TEM image (possibly aided by diffraction) but if the instrumen-

tation allows elemental mapping, a jump-ratio image (Section 5.3.6) can be useful

in selecting an area for more detailed analysis. Particularly for quantitative analysis

and ionization edges below 200 eV (where plural scattering can greatly increase the

pre-edge background), a very thin part of the specimen is needed: ideally t/λ < 0.5,

which means that the zero-loss peak contains at least 60% of the counts in the low-

loss spectrum (Section 5.1). EELS elemental analysis is often carried out using the

maximum available incident energy, since this is equivalent to using the thinnest

specimen.

326 5 TEM Applications of EELS

In the case of a crystalline specimen, its orientation relative to the incident beam

has an inﬂuence on quantitative microanalysis. It is advisable to avoid strongly

diffracting conditions, such as the Bragg condition for a low-order reﬂection or

where the incident beam is parallel to a low-order zone axis. Use of a less-diffracting

situation increases the collected signal and minimizes quantiﬁcation errors that can

arise from channeling (Section 5.6.1) and from Bragg beam contributions to the

core-loss intensity (Section 4.5.1). It may also help to improve the spatial resolution

of analysis by reducing beam spreading (Section 5.5.2).

Elements of atomic number greater than 12 allow a choice of the ionization

edge used for elemental analysis. In general, only a major edge (listed in italics in

Appendix D) is suitable, and those with a threshold energy in the range 100–2000 eV

are preferable. Edges that are sawtooth shaped or peaked at the threshold (denoted h

and w in Appendix D) are more easily identiﬁed and quantiﬁed, especially if the ele-

ment occurs in low concentration. Ionization cross sections of K-edges are mostly

known to within 10% but the situation for other edges is more variable (Egerton,

1993).

Quantiﬁcation of the core-loss signal requires its separation from the background.

The simplest procedure is to model the pre-edge background by ﬁtting within a

pre-edge region, making allowance for this background when integrating the core-

loss signal over an energy window (typically 50–100 eV) following the edge; see

Sections 4.4.1 and 4.5.1. This procedure becomes problematic when two or more

edges are close in energy or when an element is present at low concentration, or

at a low-energy edge in a specimen that is not extremely thin. In such cases, the

core-loss signal is more successfully modeled by multiple least-squares (MLS) ﬁt-

ting to the background and edge components; see Section 4.5.4. Usually standard

specimens are used to record edge shapes while calculated cross sections are used to

derive elemental ratios. Another tactic is to investigate differences in concentration

by subtracting spectra recorded from nearby regions of specimen (Section 4.5.4).

To determine elemental ratios, a choice must be made between a standardless

procedure (using calculated or parameterized cross sections; see Appendix B) and

a standards-based (k-factor) method. The standardless approach is convenient, but

requires that the collection semi-angle be known, at least approximately. The k-

factor method uses one or more standard specimens of known composition that

incorporate each analyzed element. The incident beam energy and collection angle

are not needed, provided the same values are used for the unknown and standard

specimens. Some sources of systematic error, such as poorly known cross sec-

tions and chromatic aberration effects, cancel when using the k-factor method.

Standards that have been found useful include apatite (Ca

F) and rhodizite

399

455

1135

Although core-loss spectroscopy can in principle identify any element in the peri-

odic table, some are more easily detected than others. EELS is most commonly

used for analyzing elements of low atomic number, which are difﬁcult to quan-

tify by EDX spectroscopy. In the following section, we show how EELS has been

employed to detect or quantify speciﬁc light elements.

5.4 Elemental Analysis from Core-Loss Spectroscopy 327

5.4.1 Measurement of Hydrogen and Helium

Hydrogen in its elemental form is observable from the presence of an ionization

edge. Although the ionization energy is 13.6 eV, this value corresponds to transitions

to continuum states of an isolated atom. At slightly lower energy loss, a Lyman

series of transitions to discrete levels gives peaks that may not be resolved in a

TEM spectrometer systems, the result being a structureless edge with a maximum at

about 12 eV, followed by a gradual decay on the high-loss side (Ahn and Krivanek,

1983). Energy-loss spectroscopy can be used to measure the composition of gases

(including H

) inside a TEM environmental cell, to an accuracy of 15% (Crozier and

Chenna, 2011). EELS has also detected molecular hydrogen present as bubbles in

ion-implanted SiC (Hojou et al., 1992) and in frozen hydrated biological specimens

after irradiation within the TEM (Leapman and Sun, 1995). Bubbles do not form if

the specimen is s ufﬁciently thin, suggesting t hat the hydrogen can diffuse out even

at –170

◦

C (Yakovlev et al., 2009).

Hydrogen chemically combined with other elements transfers its electrons to the

whole solid, destroying the energy levels that would give rise to a characteristic

ionization edge. Nevertheless, metallic hydrides have been detected from their low-

loss spectra; electrons donated by H atoms usually increase the valence-electron

density, shifting the plasmon peak upward by 1 or 2 eV from that of the metal;

see Section 5.2.2. In minerals, an oxygen K-edge prepeak near 530 eV, previously

thought to be indicative of hydrogen, has more recently been interpreted as due to

liberation of O

during electron irradiation (Garvie, 2010).

Hydrogen present in an organic compound inﬂuences its low-loss spectrum.

Hydrocarbon polymers have their main “plasmon” peak at a lower energy than

that of amorphous carbon (≈24 eV) because hydrogen reduces the mass density.

If hydrogen is lost, for example, during electron irradiation, the plasmon energy

increases toward that of amorphous carbon (Ditchﬁeld et al., 1973).

Hydrogen in an organic material also increases the inelastic/elastic scattering

ratio n, measurable in a conventional TEM from the total intensity I and zero-loss

intensity I

in a spectrum recorded without an angle-limiting aperture, together with

the zero-loss intensity I

recorded with a small (~2 mrad) angle-limiting aperture.

Making allowance for plural scattering,

n =

ln(I/I

)

ln(I

)

(5.15)

The specimen must be thick enough to avoid I

and I

being almost equal, otherwise

ﬂuctuations in incident beam current result in poor accuracy. This type of measure-

ment was used to monitor the loss of hydrogen from 9,10-diphenyl anthracene as a

function of electron dose (Egerton, 1976a).

Helium is produced in the form of nanometer-sized bubbles when stainless steel

(used as fuel cladding in nuclear reactors) is irradiated with neutrons. By position-

ing a STEM probe at a bubble and on the nearby metal matrix, McGibbon and