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

Подождите немного. Документ загружается.

368 5 TEM Applications of EELS

0.4 C/cm

for a 10-μm incident beam. Radiation damage and specimen contamina-

tion are therefore potential problems. Gain variations between diode array elements

should be removed (Jonas and Schattschneider, 1993).

Electron-Compton measurements have shown that the bonding in arc-evaporated

carbon ﬁlms is predominantly graphitic (Williams et al., 1983) and revealed appar-

ent inadequacies in band structure calculations for graphite (Vasudevan et al., 1984).

By using single-crystal silicon specimens with [100] and [111] orientation and plac-

ing the collection aperture at selected points in the diffraction pattern, anisotropies in

the electron-momentum distribution have been measured and compared with γ -ray

experiments and with theory (Jonas et al., 1992; Schattschneider and Exner, 1995).

5.7 Application to Speciﬁc Materials

In this ﬁnal section, we review some applications of EELS in selected areas of chem-

istry and materials science. Answers to the relevant questions have often required

low-loss spectroscopy, core-loss spectroscopy, and ﬁne structure analysis, in addi-

tion to TEM imaging, electron or x-ray diffraction, and other analytical techniques.

The topics chosen here represent a personal choice; there have been signiﬁcant

applications in other ﬁelds such as metallurgy (Okamoto et al., 1992; Craven et al.,

2008), advanced materials (Zaluzec, 1992), and catalyst studies (Wang et al., 1987;

Bentley, 1992; Crozier et al., 2008; Zhao et al., 2010).

5.7.1 Semiconductors and Electronic Devices

As semiconductor devices have shrunk toward nanometer dimensions, TEM-EELS

has become increasingly useful for studying their material properties. For example,

it is important to know how the bandgap of a semiconductor or a dielectric changes

within a device (Gu et al., 2009). Dipole transitions in a semiconductor or insulator

do not occur unless the energy supplied exceeds the direct bandgap, typically 1 eV

or more. Therefore the energy-loss spectrum should show zero intensity, followed

by a sharp rise at the bandgap energy E

, providing a measurement of E

In practice, there are two problems. First, the zero-loss peak is very strong in thin

specimens and its tails can extend to several electron volts (Bangert et al., 1997).

These tails can be minimized by using a TEM ﬁtted with a monochromator and by

setting the parallel-recording spectrometer to a high enough energy dispersion (low

electron volt/channel). Even so, it is usually necessary to remove the high-E tail

by ﬁtting and subtraction or by Fourier ratio deconvolution (Stöger-Pollach, 2008).

Park et al. (2009) achieved consistency in measuring the bandgap of SiO

ﬁlms of

different thicknesses by performing linear ﬁts to the intensity rise and its preceding

background, choosing E

from the intersection point; see Fig. 5.51.

The second problem arises from surface plasmon and

Cerenkov modes of energy

loss, which add peaks in the region of a few electron volt. Both effects involve very

5.7 Application to Speciﬁc Materials 369

Fig. 5.51 Bandgap

measurements on silicon

oxide ﬁlms of different

thicknesses using a

monochromated TEM-EELS

system. The direct gap

(8.9 eV) was taken as the

intersection between two

linear ﬁts. From Park et al.

(2009), copyright Elsevier

small scattering angles, so they can be minimized by using an annular or off-axis

collection aperture or by subtracting spectra recorded with small and large collection

apertures (Stöger-Pollach 2008). Zhang et al. (2008) subtracted spectra recorded

with two different collection angles in order to measure the dielectric function of

diamond.

In fact, radiation losses within the bandgap region (where other loss processes

are largely absent) could provide useful information about artiﬁcial nanostructures,

since their intensity distribution is directly related to the optical densities of states

(Garcia de Abajo et al., 2003). The ability of the TEM to examine properties at the

nanometer level offers the opportunity to examine defects within such structures

(Cha et al., 2010).

Rafferty and Brown (1998) pointed out that the low-loss ﬁne structure represents

a joint density of states multiplied by a matrix element that differs in the case of

direct and indirect transitions. Assuming no excitonic states, their analysis showed

that the onset of energy-loss intensity is proportional to (E − E

)

1/2

for a direct gap

and (E −E

)

3/2

for an indirect gap.

Cubic GaN is a direct gap semiconductor whose inelastic intensity, after sub-

tracting the zero-loss tail, ﬁts well to (E − E

)

1/2

for E > E

(Lazar et al., 2003).

Residual intensity below E

might arise from indirect transitions in a surface oxide

layer, as suggested by (E −E

)

3/2

ﬁtting; see Fig. 5.52a. The anatase phase of TiO

is an indirect-gap material (E

≈ 3.05 eV), so Wang et al. (2008b) matched the

intensity between 3 and 6 eV to (E −E

)

1.5

; see Fig. 5.52b. In the case of a related

hydroxylated material H

(in the form of 8-mm-diameter multiwalled nan-

otubes), additional intensity occurred below 4 eV, suggesting transitions involving

defect states introduced by the hydroxyl groups; see Fig. 5.52c. The spectra were

recorded with an off-axis aperture (q ≈ 1nm

−1

) to minimize surface plasmon and

Cerenkov contributions.

370 5 TEM Applications of EELS

Fig. 5.52 (a) Inelastic intensity recorded from cubic GaN using a monochromated TEM-EELS

system. The curves show ﬁtted functions appropriate to direct and indirect electronic transitions;

from Lazar et al. (2003), copyright Elsevier. (b) Spectra recorded from TiO

and (c)froma

nanotube, at various scattering vectors q (in nm

–1

). Insets show ﬁts to a (E–E

)

1.5

func-

Physics

Bandgap maps can be produced by analyzing the low-loss data at each pixel

(Tsai et al., 2004). However, delocalization of the inelastic scattering limits the spa-

tial resolution to a few nanometers (Couillard et al., 2007, 2008), as discussed in

Section 5.5.3.Bosmanetal.(2009) concluded that bandgap mapping in the near-

visible region may not be feasible for embedded layers or particles less than about

10 nm in dimension.

Crystalline defects adversely affect the electronic properties of semiconductors.

High-resolution TEM can be used to investigate the atomic arrangement of indi-

vidual defects, while EELS gives information on their local electronic structure

and bonding. Combining this information can lead to an understanding of how the

atomic structure and physical properties are related. Although strong diffraction in

a crystalline specimen makes the interpretation of inelastic scattering more compli-

cated, channeling directs the electrons down atomic columns (if the beam is aligned

with a crystal axis) and improves the spatial resolution (Loane et al., 1988). The

STEM uses an annular dark-ﬁeld (ADF) detector to collect high-angle elastic scat-

tering, giving a high-resolution image with good atomic number contrast that allows

the electron beam to be positioned on an atomic column and held there long enough

to acquire useful spectra if specimen drift and radiation damage are sufﬁciently low

(Batson, 1992a, 1993c, 1995; Browning et al., 1993b; Lakner et al., 1992;Kimoto

et al., 2007; Muller et al., 2008).

An early use of STEM to investigate semiconductor defects is shown in Fig. 5.53

(Batson et al., 1986). The large tail of the zero-loss peak was subtracted from the

low-loss data and the inelastic intensity ﬁtted to (E −E

)

0.5

. With the STEM probe

located at a misﬁt dislocation, the effective energy gap E

is reduced, the additional

5.7 Application to Speciﬁc Materials 371

Fig. 5.53 Change in the intensity of inelastic scattering when a STEM probe is positioned on a

misﬁt dislocation in GaAs. From Batson et al. (1986), copyright American Physical Society. http://

link.aps.org/abstract/PRL/v57/p2729

low-loss scattering likely arising from electron excitation from ﬁlled states (at the

dislocation core) to the conduction band. Because these states depend on the struc-

ture of the dislocation, this measurement points to the possibility of using EELS to

determine the nature of individual dislocations. Takeda et al. (1994) reported that

line defect self-interstitials in silicon (identiﬁed by HREM imaging) give rise to

an energy-loss peak at 2.5 eV. Tight-binding calculations showed this peak to be

consistent with the presence of eight-membered rings.

The local dopant concentration in s ome semiconductor devices has become high

enough to be detectable by EELS. Servanton and Pantel (2010) have shown that

STEM-EELS (based on the As-L

ionization edge) can map the distribution of

arsenic dopant in silicon BiCMOS transistors and static RAM. Their sensitivity was

in the low end of the 10

–3

range, with a spatial resolution about 2 nm; see

Fig. 5.54. Line scans showed good quantitative agreement with secondary ion mass

spectrometry. The measurements were made with an incident energy of 120 keV,

below the knock-on threshold for silicon.

According to Eq. (4.33), Kramers–Kronig analysis of the low-loss spectrum

yields an electron concentration, whose value can then be used in Eq. (3.41),

together with the measured plasmon energy to give an electron effective mass

(related to bandgap and carrier mobility). In a semiconductor, the main plasmon

peak represents resonance of the valence electrons, whose effective mass may dif-

fer from that of the conduction electrons. Nevertheless, Gass et al. (2006a) reported

372 5 TEM Applications of EELS

Fig. 5.54 (a) TEM image and (b) arsenic concentration map of a 90-nm n-p-n BiCMOS transistor

(150 × 60 pixels) acquired in 2.2 h by STEM-EELS. (c) Arsenic concentration proﬁle along the

line CC



. From Servanton and Pantel (2010), copyright Elsevier

agreement with the cyclotron-resonance mass for GaAs and GaN, then used their

STEM-EELS system to produce maps of effective mass in InAs quantum dots and

GaInNAs quantum wells.

The properties of ﬁeld-effect transistors depend greatly on the interface between

the gate dielectric and the semiconductor. There have been several TEM-EELS stud-

ies of these interfaces. For example, Park and Yang (2009) used Kramers–Kronig

analysis to determine the dielectric function across a HfO

/Si interface, obtaining

good agreement with density functional calculations for HfO

away from the inter-

face. Gatts et al. (1995) applied neural pattern recognition to t he analysis of a series

of low-loss spectra recorded across a Si/SiO

interface. First-difference spectra were

ﬁtted to Si and SiO

standards, with a third component representing the interface

loss.

To lower gate voltage and power dissipation, the thickness of the FET gate insula-

tor has been reduced to values approaching 1 nm. In the case of SiO

, this thickness

corresponds to only ﬁve Si atoms and the properties of the oxide might be expected

to depart from those of the bulk solid. High-resolution STEM of cross-sectional

specimens, combined with core-loss EELS, provides a way of investigating these

effects. Figure 5.55 shows how the background-subtracted oxygen K-edge changed

when a STEM probe approached the Si interface from a native oxide. The threshold

shifted downward by 3 eV, indicating a reduced bandgap, and the sharp threshold

peak disappeared. Since this peak arises from strong backscattering of the ejected

core electron from nearest-neighbor O atoms, its absence near the interface indi-

cates a silicon-rich environment: a sub-oxide SiO

where x < 2. Similar results were

obtained for the thermally grown oxide used in device fabrication. From analysis

5.7 Application to Speciﬁc Materials 373

Fig. 5.55 Oxygen K-loss

spectra from the native oxide

on silicon (solid line b) and

from a region near the Si

interface (data points a), as

indicated in the STEM-ADF

image on the right.

Reproduced from Muller

et al. (1999), with permission

from Nature Publishing

Group

of the two ELNES components, the sub-oxide was estimated to have a width of

0.75 nm, compared to a total thermal oxide width of 1.6 nm (Muller et al., 1999).

EFTEM imaging can provide a relatively rapid way of obtaining elemental maps

of semiconductor devices, revealing their structure more clearly than a conventional

TEM image; see Fig. 5.56. Botton and Phaneuf (1999) used GIF-EFTEM imag-

ing with a narrow (4-eV) energy window to distinguish between oxide and nitride

regions in a DRAM transistor, exploiting the larger chemical shift of silicon oxide

compared to the nitride. Energy-ﬁltered TEM has also been used to display elemen-

tal distributions and composition gradients in semiconductor multilayers (Jäger and

Mayer, 1995; Liu et al., 1999).

Fig. 5.56 (a) Low-magniﬁcation zero-loss image of a trench capacitor (part of a 64-MB DRAM

chip). (b) Elemental map showing the distribution of silicon, nitrogen, and oxygen in this region.

From Botton and Phaneuf (1999), copyright Elsevier

374 5 TEM Applications of EELS

5.7.2 Ceramics and High-Temperature Superconductors

The performance of many technological materials i s dependent on the properties

of their internal surfaces (Brydson et al., 1995). In the case of Si

structural

ceramics, the mechanical and thermal properties are controlled by the atomic and

electronic structure of the crystal/amorphous interface, which may undergo inter-

atomic mixing and partial ordering involving heavy cations. Walkosz et al. (2010)

used EELS to obtain i nformation about the position of light atoms at the β-Si

SiO

interface. Figure 5.57a shows a atomic resolution bright-ﬁeld image that

reveals the presence of short-range ordering between the crystalline β-Si

and

amorphous SiO

. To identify the composition along and across the interface, atomi-

cally resolved EELS was carried out at 80 kV (avoiding radiation damage) by raster

scanning an aberration-corrected probe while collecting a spectrum (in 0.05 s) at

each pixel. Figure 5.57d, e shows the Si L

edges taken from six different posi-

tions, identiﬁed in the simultaneously acquired Z-contrast image (Fig. 5.57f) and

Fig. 5.57 (a) A 300-kV bright-ﬁeld image (contrast inverted) of a β-Si

-SiO

interface; (b)

and (c) are intensity proﬁles at positions P1 and P2, respectively; (d)and(e)showtheSiL-edge

acquired at the six different positions shown in the Z-contrast image (f), with L-edges of bulk Si

and SiO

for comparison. (g) shows O/N ratios at the six positions, normalized with respect to the

maximum (position 2). Dotted lines in (a)and(d) mark the interface between the Si

grain and

the amorphous SiO

ﬁlm; the scale bars are 0.5 nm in length. Figure adapted from Walkosz et al.

(2010), copyright American Physical Society. http://link.aps.org/abstract/PRB/v82/p081412

5.7 Application to Speciﬁc Materials 375

corresponding to the Si sites in the bulk Si

. Silicon signals at all six positions

displayed both Si-N and Si-O bonding features (peaks a, b, and c of the Si

and

SiO

reference spectra) but those at positions 1, 2, 3, and 4 reveal stronger Si-O

bonding since the peaks at 105 eV (labeled b) are more pronounced. The Si signals at

positions 4 and 6, representing the two ends of the terminating Si

structure, have

slightly different features, indicating distinct bonding characteristics. To investigate

further, O and N K-edges were acquired simultaneously with the Si L

edge, and by

integrating these signals over a 40-eV window, elemental O/N ratios were computed,

as shown in Fig. 5.57g. The ratios are different at the two ends of the terminating

structures (positions 4 and 6), suggesting a compositional asymmetry at the

surface and implying site-speciﬁc intermixing across the interface.

Klie and Zhu (2005) and Klie et al. (2008) have reviewed EELS and atomic

column STEM imaging of ceramics, including their studies of dislocations present at

◦

tilt boundaries in SrTiO

. At a dislocation core, the Ti L-edge integral was found

to be 21% higher than in the bulk; its threshold was shifted 0.8 eV lower in energy,

indicating a Ti valency of 3.6 ± 0.2; the t

peaks were suppressed, consistent with

a Ti valency below 4. The oxygen-K ELNES structure was also different at the core,

suggesting an excess negative charge.

Some ceramics have applications as thermoelectric materials. Klie and Qiao

(2010) have shown how EELS and annular bright-ﬁeld STEM can be useful for

determining the effects of structural disorder, strain, and charge transfer on the ther-

moelectric properties of the layered oxide material Ca

. STEM spectroscopic

imaging has been used to investigate atomic-scale interdiffusion at oxide interfaces

(Fitting Kourkoutis et al., 2010).

Following their discovery in 1987, cuprate superconductors were extensively

studied by TEM-EELS, including yttrium barium cuprate (YBa

7−δ

, abbre-

viated as YBCO). These ceramics contain CuO

planes perpendicular to the c

direction, with oxygen and metal atoms in between. Superconductivity involves

Cooper pairs of valence band holes and is associated with the CuO

planes, where

Cu-3d and O-2p states lie close to the Fermi level. Taking advantage of the dipole

selection rule (Section 3.7.2), the unoccupied part of these states can be investigated

by examining ﬁne structure of the copper L

and oxygen K-edges.

Figure 5.58a shows the onset of the oxygen K-edge recorded from

YBa

7−δ

. For small oxygen deﬁciency (δ ≈ 0.2), two pre-edge features are

visible: a shoulder around 535 eV, which represents transitions to unoccupied Cu-

3d-states (upper Hubbard band), and a peak around 528 eV representing transitions

to O-2p states that give rise to holes in the valence band. As the oxygen deﬁciency

δ increases, the 528-eV peak falls in intensity, indicating a decrease in hole con-

centration and a reduction in the superconducting critical temperature. For δ >0.6,

the prepeak is absent, corresponding to zero hole concentration and a material that

behaves as an electrical insulator at all temperatures.

By appropriate choice of the scattering angle and specimen orientation, it is pos-

sible to select the direction of the momentum transfer q that contributes to the

energy-loss spectrum and observe signiﬁcant differences in pre-edge structure; see

Fig. 5.58b. These observations were used to determine ﬁnal-state symmetries in

376 5 TEM Applications of EELS

Fig. 5.58 (a) Onset of the oxygen K-edge of YBCO at four values of the oxygen deﬁciency δ

(Fink, 1989). (b) Oxygen K-edge recorded at scattering angles such that the momentum transfer

was either parallel or perpendicular to the c-axis. From Fink et al. (1994), copyright Elsevier

several oxide superconductors, information that proved useful in the comparison

of different models of electrical conduction (Fink et al., 1994). The q-dependence

(dispersion) of the plasmon peak has been measured for Bi

CaCu

and inter-

preted in terms of the multilayered structure of the material (Longe and Bose, 1993).

Using a ﬁeld-emission TEM to examine Ba

1−x

BiO

, Wang et al. (1993) identi-

ﬁed a free-carrier plasmon around 2 eV that abruptly increased in strength at the

semiconductor/metal transition (x > 0.3). The dielectric function of YBCO shown

in Fig. 5.59 was derived by Kramers–Kronig analysis of EELS data and has been

discussed in terms of band structure (Yuan et al., 1988).

The superconducting properties of these ceramics depend strongly on the pres-

ence of defects, such as grain boundaries in a polycrystalline material. Making use

of the high spatial resolution of TEM-EELS, the oxygen-edge prepeak can be used

as a sensitive measure of local oxygen concentration. From EELS measurements

made using a 2-nm electron probe, Zhu et al. (1993) concluded that grain bound-

aries in fully oxygenated YBCO fall into two categories, with different structural

misorientation between the grains: those in which the 529-eV peak retains its inten-

sity across the grain boundary and those in which the intensity falls practically to

zero. The implication is t hat some boundaries are fully oxygenated and would trans-

mit a superconducting current at temperatures below 90 K, while others contain an

oxygen-depleted region (width 10 nm or less) and may act as weak links. Similar

conclusions have been reached on the basis of EELS combined with Z-contrast

STEM imaging (Browning et al., 1993a). From analysis of the oxygen K-edge struc-

ture, Browning et al. (1999) established the presence of a charge depletion layer at

the grain boundaries of YBa

7–δ

and (Bi/Pb)

. They found that

5.7 Application to Speciﬁc Materials 377

Fig. 5.59 Energy-loss function (solid curve) together with real and imaginary components of the

permittivity ε of YBCO superconductor, derived from low-loss EELS. Peaks A and B are attributed

to valence electron plasmon excitation and peak C to yttrium 4p→d transitions (N

edge). From

Yuan et al. (1988), copyright IOP Publishing. http://iopscience.iop.org/0022-3719/21/3/008

the width increased with misorientation angle and concluded that the critical current

is limited by tunneling across this depletion zone.

From measurements of the 529-eV peak, Dravid et al. (1993) showed that the

oxygen content in YBa

7−δ

may vary along a grain boundary plane (δ ﬂuc-

tuating between 0.2 and 0.4 over distances of the order of 10 nm) and proposed a

modiﬁcation to the Dayem bridge model for grain boundary structure. Browning

et al. (1991a) applied the same technique to reveal variations in oxygen content

within a grain, with an estimated accuracy of 2%. EELS has also been used to detect

the presence of carbon at grain boundaries of densiﬁed YBCO; examination of the

carbon and oxygen K-edge structure indicated the presence of BaCO

(Batson et al.,

1989).

0.5

1−x

7−δ

is another high-temperature superconductor that

has been examined by EELS (Yuan et al., 1991). Its oxygen K-edge shows a prepeak

at 529 eV, believed to represent transitions to conduction band states and not an

indication of superconductivity. However, a shoulder on the low-energy side of the

peak (attributed to transitions to hole states near the Fermi level) appeared to be

characteristic of the superconducting (x < 0.4) material. Core-loss measurements on

1−x

GaO

(Dravid and Zhang, 1992) also found two pre-edge features:

a broad peak (dependent on Ca doping) centered around 528.2 eV, associated with

normal conductivity and attributed to holes on oxygen sites that are not on Cu

planes, and a smaller peak around 527.1 eV, probably associated with holes on Cu

planes and superconductivity.

Attempts were made to fabricate Josephson junctions by depositing multilayer

structures of oxide superconductors. However, the c-axis tends to lie perpendicular

to the ﬁlm plane and the coherence length is very short (<1 nm) in this direc-

tion, placing extreme limits on the sharpness of the junction. If a ﬁlm could be

depleted in oxygen by means of an electron probe of sub-nanometer dimensions, the