上海交通大学：《探索物质微观世界》课程教学资源（文献资料）Structure and bonding at the atomic scale by scanning transmission electron microscopy

nature INSIGHTI REVIEW ARTICLES materials PUBLISHED ONLINE:24 MARCH 2009DOI:10.1038/NMAT2380 Structure and bonding at the atomic scale by scanning transmission electron microscopy David A.Muller A new generation of electron microscopes is able to explore the microscopic properties of materials and devices as diverse as transistors,turbine blades and interfacial superconductors.All of these systems are made up of dissimilar materials that,where they join at the atomic scale,display very different behaviour from what might be expected of the bulk materials.Advances in electron optics have enabled the imaging and spectroscopy of these buried interface states and other nanostructures with atomic resolution.Here I review the capabilities,prospects and ultimate limits for the measurement of physical and electronic properties of nanoscale structures with these new microscopes. roblem solving at the atomic scale has become an essential part development of multipole-based aberration correctors24-27 have of both fundamental science and modern device development enabled sub-angstrom beam sizes2-and rapid,atomic-resolution Hand in hand with the desire and ability to control materials compositional imaging3.Figure 2 shows that progress in improv at the level of atoms,comes a need to image and quantify mate- ing spatial resolution has been both rapid and ongoing,so that 0.5-A rials properties with atomic resolution'.The success or failure of beam sizes are now possible.(Note that for a limit of resolution x, semiconductor devices and their continued miniaturization relies the resolving power of the microscope"is defined as the reciprocal on critical components.These include the 5-10-atom-thick gate distance,1/6x). oxides in transistors?-,magnetoresistive read heads in hard drives Given that interatomic distances are all larger than 0.5 A,one that have layers which are only 1-2 nm thicks-7and tunnel junctions might well suspect,as did Dennis Gabor in 1948,that "resolution in magnetic memories that are also of comparable thickness-10.The will have to stop here for a lack of objects"(ref.34).Although this challenge in all these systems is not only to determine the thick- lack of suitable test objects does make it difficult to test the resolu- ness and composition of the 'bulk'portion of these ultrathin layers, tion of a microscope,there is no shortage of problems that could but also the chemistry,interdiffusion and electronic structure of the be tackled with smaller beam sizes.Atoms in amorphous materi- buried'interfaces between the layers. als or tilted crystals can appear much closer when viewed in pro- Although no single method can yet obtain complete informa- jection.Being able to resolve arbitrarily short distances at multiple tion about buried structures at the atomic scale,this goal is being projections could lead to direct structural solutions for general approached from several directions.This includes compositional amorphous materials,or atomic-resolution reconstructions of (but not bonding)information from atom-probe microscopy,struc- three-dimensional objects by electron tomography.The number of tural information from transmission electron microscopy and elec- tron tomography(see the Review by Midgley and Dunin-Borkowski elsewhere in this Insight")and scanning transmission electron microscopy (STEM),which will be discussed here.STEM has proved very effective in measuring not only compositional changes at buried interfaces,but also the electronic structure and bonding that have relevance for the mechanical4 and transport proper- ties25 of a device.Detection of single dopant atoms with STEM both on free surfacest6 and buried inside devices 7.s,has proved useful in studies ranging from the characterization of catalysts! to understanding the materials limits for transistor scaling.It is 000 also possible to detect and image the spatial distribution of single vacancies-either directly2122,or by their strain fields,or spectro- 000 scopically from their electronic fingerprints on the local densities of states23 The basic concept of a STEM is relatively simple-a high-energy electron beam is focused down to a small spot and fired through a thin sample.Signals from scattered electrons and ionized atoms are recorded as the beam is scanned across the sample to build up a two- dimensional map.Chemical and bonding information along each Figure 1|Compositional imaging at the atomic scale.a-d,An atomic- projected atomic column can be obtained by measuring the energy resolution chemical map of a (La,Sr)MnO,/SrTiO,multilayer showing the lost by transmitted electrons to core and valence excitations at each La-M(a),Ti-L (b)and Mn-L (c)edge spectroscopic signals merged into point on the map(Fig.1).The performance of the instrument is a composite map(d).The white circles show the location of the columns determined by how small a spot the electron beam can be focused of La atoms.The spectral series was recorded on an aberration-corrected to,and how much current can be maintained in that beam.Recent scanning electron microscope using electron energy-loss spectroscopy.The advances in the control of lens aberrations through the successful data took less than 1 min to acquire.(From ref.32;2008 AAAS.) School of Applied and Engineering Physics,Cornell University,Ithaca,New York 14853,USA.e-mail:dm24@cornell.edu NATURE MATERIALS VOL 8|APRIL 2009 www.nature.com/naturematerials 263 2009 Macmillan Publishers Limited.All rights reserved

nature materials | VOL 8 | APRIL 2009 | www.nature.com/naturematerials 263 insight | review articles Published online: 24 march 2009 | doi: 10.1038/nmat2380 P roblem solving at the atomic scale has become an essential part of both fundamental science and modern device development. Hand in hand with the desire and ability to control materials at the level of atoms, comes a need to image and quantify mate￾rials properties with atomic resolution1 . The success or failure of semiconductor devices and their continued miniaturization relies on critical components. These include the 5–10-atom-thick gate oxides in transistors2–4, magnetoresistive read heads in hard drives that have layers which are only 1–2 nm thick5–7 and tunnel junctions in magnetic memories that are also of comparable thickness8–10. The challenge in all these systems is not only to determine the thick￾ness and composition of the ‘bulk’ portion of these ultrathin layers, but also the chemistry, interdiffusion and electronic structure of the ‘buried’ interfaces between the layers. Although no single method can yet obtain complete informa￾tion about buried structures at the atomic scale, this goal is being approached from several directions. This includes compositional (but not bonding) information from atom-probe microscopy, struc￾tural information from transmission electron microscopy and elec￾tron tomography (see the Review by Midgley and Dunin-Borkowski elsewhere in this Insight11) and scanning transmission electron microscopy (STEM), which will be discussed here. STEM has proved very effective in measuring not only compositional changes at buried interfaces, but also the electronic structure and bonding that have relevance for the mechanical12–14 and transport proper￾ties2,15 of a device. Detection of single dopant atoms with STEM, both on free surfaces16 and buried inside devices17,18, has proved useful in studies ranging from the characterization of catalysts19,20 to understanding the materials limits for transistor scaling17. It is also possible to detect and image the spatial distribution of single vacancies — either directly21,22, or by their strain fields, or spectro￾scopically from their electronic fingerprints on the local densities of states23. The basic concept of a STEM is relatively simple — a high-energy electron beam is focused down to a small spot and fired through a thin sample. Signals from scattered electrons and ionized atoms are recorded as the beam is scanned across the sample to build up a two￾dimensional map. Chemical and bonding information along each projected atomic column can be obtained by measuring the energy lost by transmitted electrons to core and valence excitations at each point on the map (Fig. 1). The performance of the instrument is determined by how small a spot the electron beam can be focused to, and how much current can be maintained in that beam. Recent advances in the control of lens aberrations through the successful structure and bonding at the atomic scale by scanning transmission electron microscopy david a. muller1 A new generation of electron microscopes is able to explore the microscopic properties of materials and devices as diverse as transistors, turbine blades and interfacial superconductors. All of these systems are made up of dissimilar materials that, where they join at the atomic scale, display very different behaviour from what might be expected of the bulk materials. Advances in electron optics have enabled the imaging and spectroscopy of these buried interface states and other nanostructures with atomic resolution. Here I review the capabilities, prospects and ultimate limits for the measurement of physical and electronic properties of nanoscale structures with these new microscopes. development of multipole-based aberration correctors24–27 have enabled sub-ångström beam sizes28–30 and rapid, atomic-resolution compositional imaging31,32. Figure 2 shows that progress in improv￾ing spatial resolution has been both rapid and ongoing, so that 0.5-Å beam sizes are now possible. (Note that for a limit of resolution δx, the resolving power of the microscope33 is defined as the reciprocal distance, 1/δx). Given that interatomic distances are all larger than 0.5 Å, one might well suspect, as did Dennis Gabor in 1948, that “resolution will have to stop here for a lack of objects” (ref. 34). Although this lack of suitable test objects does make it difficult to test the resolu￾tion of a microscope, there is no shortage of problems that could be tackled with smaller beam sizes. Atoms in amorphous materi￾als or tilted crystals can appear much closer when viewed in pro￾jection. Being able to resolve arbitrarily short distances at multiple projections could lead to direct structural solutions for general amorphous materials, or atomic-resolution reconstructions of three-dimensional objects by electron tomography. The number of 1 School of Applied and Engineering Physics, Cornell University, Ithaca, New York 14853, USA. e-mail: dm24@cornell.edu a b c d Figure 1 | Compositional imaging at the atomic scale. a–d, An atomic￾resolution chemical map of a (La,Sr)MnO3 /SrTiO3 multilayer showing the La–M (a), Ti–L (b) and Mn–L (c) edge spectroscopic signals merged into a composite map (d). The white circles show the location of the columns of La atoms. The spectral series was recorded on an aberration-corrected scanning electron microscope using electron energy-loss spectroscopy. The data took less than 1 min to acquire. (From ref. 32; © 2008 AAAS.) nmat_2380_APR09.indd 263 11/3/09 11:12:07 © 2009 Macmillan Publishers Limited. All rights reserved

REVIEW ARTICLES INSIGHT NATURE MATERIALS DO:10.1038/NMAT2380 Aberration-corrected electron microscope TEAM/ (1.2MV) CREST 100 Dietrich 100 200kV0 Haider (200kV) 101 10 色 Electron microscope Zach,Haider (1kV SEM) 102 Borries and Ruska Marton,Ardenne (100kV) 10-3 103 L山ight Ruska (75 kV) microscope Abbe 10-4 Amici 109 Ross 101800 1850 190019502000 2050 Year Figure 2 Hardware advances in imaging microscopies.Optical microscopy reached its far-field diffraction limit roughly a century ago. Electron microscopy,exploiting the reduction in electron wavelength with increasing beam voltage,showed steady increases in resolving power for ADF over 50 years,until limited by radiation damage.Recent developments detector in correcting electron-optical aberrations have led to improvements in resolving power without having to increase the electron beam energy. Although still far from their ultimate diffraction limits,corrected electron microscopes have already demonstrated the ability to resolve sub- angstrom image features.(Adapted from ref.114:2009 Elsevier.) Electron resolvable projections increases roughly as the square of the resolu- spectrometer tion,so even improvements in resolution by a factor of two or three could make a dramatic difference.Even when such small distances do not need to be measured,a smaller probe size allows for higher Figure 3|Major elements of a scanning transmission electron contrast images and improved detectability of single dopant atoms, microscope.A high-brightness electron source produces a 100-300-keV enabling studies of a wider range of dopant species and thicker, electron beam with an energy spread of 0.3-1 eV,which can be narrowed more realistic,substrates. to below 100 meV with a monochromator.Round magnetic lenses and This article will explore the limits and opportunities for the corrective multipole optics focus the beam to a spot size of between new generation of aberration-corrected microscopes.After a brief 0.05 and -0.3 nm,which is scanned across an electron-transparent sample. description of the STEM itself,the improvements in resolution and To a first approximation,when the beam is placed on an atom column, current due to aberration-corrected optics and improved electron strong Rutherford-like scattering deflects the transmitted electrons to form sources are detailed.Improved energy resolution from mono- brighter features in an annular dark-field (ADF)image,with less scattering chromators and higher-resolution spectrometers has enabled nano- between the columns.Inelastic scattering is strongly peaked in the forward scale measurements of the photonic properties of materials and direction and is collected simultaneously with the ADF signal.The energy waveguides.Finally,with the improvements in instrumentation,it losses of the transmitted electrons reflect characteristic excitations of is the sample itself and the radiation dose it can tolerate that seems the sample in a frequency range spanning the near-infrared to hard X-ray now to be the limit to signal and resolution. regions,allowing electronic and elemental identification from a single column of atoms. The scanning transmission electron microscope The first STEM to include the necessary field-emission source and Almost a decade later,with the development of a commercially high-vacuum needed to keep samples clean during irradiation was manufactured STEM and better specimen preparation tools such as developed by Crewe in 1964(ref.35),and within a few years they ion-milling,there was a renewed interest in the instrument as an were demonstrating images of single heavy atoms's and their dif- analytical tool for metallurgy and in the semiconductor industry. fusion across thin carbon films%.The popular dark-field and high- Some of these second-generation instruments are still in operation angle dark-field imaging modes were demonstrated early during the today.In the mid-1990s,a new generation of commercial field- machine's development,including in the collection of lattice images emission TEMs with scanning attachments made STEM techniques from nanoparticles"7.As the original microscopes were intended for widely available to the materials research community and industry. biological research,these samples were used more as resolution tests The major components of a STEM are shown in Fig.3.A than for materials studies -with one exception being the study field-emission gun provides a high-coherence source of electrons of catalyst particles.Adoption of the STEM technique was slow that is accelerated to between 100 and 300 keV-energies sufficient owing to difficulties in sample preparation of bulk materials and a to penetrate samples up to 100 nm thick without significant beam lack of widespread instrument availability as a result of the vacuum spreading.A series of electron lenses and corrective optics placed and electronics requirements. before the sample focuses the beam down to a diameter smaller 264 NATURE MATERIALS|VOL 8|APRIL 2009 www.nature.com/naturematerials 2009 Macmillan Publishers Limited.All rights reserved

264 nature materials | VOL 8 | APRIL 2009 | www.nature.com/naturematerials review articles | insight NAture mAterIAls doi: 10.1038/nmat2380 resolvable projections increases roughly as the square of the resolu￾tion, so even improvements in resolution by a factor of two or three could make a dramatic difference. Even when such small distances do not need to be measured, a smaller probe size allows for higher contrast images and improved detectability of single dopant atoms, enabling studies of a wider range of dopant species and thicker, more realistic, substrates. This article will explore the limits and opportunities for the new generation of aberration-corrected microscopes. After a brief description of the STEM itself, the improvements in resolution and current due to aberration-corrected optics and improved electron sources are detailed. Improved energy resolution from mono￾chromators and higher-resolution spectrometers has enabled nano￾scale measurements of the photonic properties of materials and waveguides. Finally, with the improvements in instrumentation, it is the sample itself and the radiation dose it can tolerate that seems now to be the limit to signal and resolution. the scanning transmission electron microscope The first STEM to include the necessary field-emission source and high-vacuum needed to keep samples clean during irradiation was developed by Crewe in 1964 (ref. 35), and within a few years they were demonstrating images of single heavy atoms15 and their dif￾fusion across thin carbon films36. The popular dark-field and high￾angle dark-field imaging modes were demonstrated early during the machine’s development, including in the collection of lattice images from nanoparticles37. As the original microscopes were intended for biological research, these samples were used more as resolution tests than for materials studies — with one exception being the study of catalyst particles19. Adoption of the STEM technique was slow owing to difficulties in sample preparation of bulk materials and a lack of widespread instrument availability as a result of the vacuum and electronics requirements. Almost a decade later, with the development of a commercially manufactured STEM and better specimen preparation tools such as ion-milling, there was a renewed interest in the instrument as an analytical tool for metallurgy and in the semiconductor industry. Some of these second-generation instruments are still in operation today. In the mid-1990s, a new generation of commercial field￾emission TEMs with scanning attachments made STEM techniques widely available to the materials research community and industry. The major components of a STEM are shown in Fig. 3. A field-emission gun provides a high-coherence source of electrons that is accelerated to between 100 and 300 keV — energies sufficient to penetrate samples up to 100 nm thick without significant beam spreading. A series of electron lenses and corrective optics placed before the sample focuses the beam down to a diameter smaller Light microscope Electron microscope Aberration-corrected electron microscope Zach, Haider (1 kV SEM) Ross Amici Abbe Ruska (75 kV) Borries and Ruska Marton, Ardenne (100 kV) Dietrich (200 kV) Stuttgart (1.2 MV) TEAM/ CREST NION Haider (200 kV) Feature size (Å) 1800 10–4 10–5 10–3 10–2 10–1 100 1850 1900 Year 1950 2000 2050 104 103 102 101 100 Resolving power (Å–1) Figure 2 | Hardware advances in imaging microscopies. Optical microscopy reached its far-field diffraction limit roughly a century ago. Electron microscopy, exploiting the reduction in electron wavelength with increasing beam voltage, showed steady increases in resolving power for over 50 years, until limited by radiation damage. Recent developments in correcting electron-optical aberrations have led to improvements in resolving power without having to increase the electron beam energy. Although still far from their ultimate diffraction limits, corrected electron microscopes have already demonstrated the ability to resolve sub- ångström image features. (Adapted from ref. 114; © 2009 Elsevier.) Source Lenses Sample ADF detector Electron spectrometer Figure 3 | major elements of a scanning transmission electron microscope. A high-brightness electron source produces a 100–300- keV electron beam with an energy spread of 0.3–1 eV, which can be narrowed to below 100 meV with a monochromator. Round magnetic lenses and corrective multipole optics focus the beam to a spot size of between 0.05 and ~0.3 nm, which is scanned across an electron-transparent sample. To a first approximation, when the beam is placed on an atom column, strong Rutherford-like scattering deflects the transmitted electrons to form brighter features in an annular dark-field (ADF) image, with less scattering between the columns. Inelastic scattering is strongly peaked in the forward direction and is collected simultaneously with the ADF signal. The energy losses of the transmitted electrons reflect characteristic excitations of the sample in a frequency range spanning the near-infrared to hard X-ray regions, allowing electronic and elemental identification from a single column of atoms. nmat_2380_APR09.indd 264 11/3/09 11:12:08 © 2009 Macmillan Publishers Limited. All rights reserved

NATURE MATERIALS DOL:10.1038/NMAT2380 INSIGHT I REVIEW ARTICLES than the spacings between the projected atomic columns.The beam the aberrations corrected over a wider angular range,the numerical is then scanned across the sample,and the differences in electron- aperture of the probe-forming lens can then be increased,allowing scattering when the beam is located on and off the atomic columns more electrons to be brought to a focused spot.It has also proved nec- are recorded.The optimal sample thickness is a balance between essary to correct the collection optics so that the additional electrons being thin enough that the electron probe will not spread signifi- input at the larger angles could still be collected312 cantly in the sample,and thick enough so that bulk-like,and not A very powerful feature of EELS is that it provides both com- surface,atoms dominate the signal. positional and bonding information at very high spatial resolution. Elastically scattered electrons can undergo large angle deflections, Figure 4 shows the EELS excitation process whereby the incident and hence an annular dark-field(ADF)detector is placed below the electron transfers energy to the target atom,exciting a core electron sample to collect the scattered electrons.The high-angle electron- to empty states above the Fermi level (E).The energy loss of the scattering cross-section scales roughly as the atomic number transmitted electron is recorded by the spectrometer and provides asymptotically tending towards the Rutherford scattering limit of information equivalent to X-ray absorption spectroscopy,where the Z,and so this imaging mode is sometimes called 'Z-contrast'imag- electron's momentum transfer has the same role as the polarization ings7-.Provided the collection angles are three or more times of the X-ray beams1s2.The core-level binding energy that marks larger than the probe convergence angle,the incoherent imaging the EELS edge onset allows a unique elemental identification to approximation is satisfied for thin samples,thereby allowing rela- be made.The shape of the edge itself reflects the underlying local tively straightforward image interpretation32.To avoid contrast partial density of states modified by the presence of a core holess-57. reversals with sample thickness,the detector also needs to be suf- The core hole is relatively well screened in metals,allowing the spec- ficiently large to avoid any strongly diffracted beams from being trum to be interpreted in terms of the ground state local density of collected.When this condition is just met,diffuse scattering from phonons and lattice distortions can provide significant image contrast.This low-angle ADF signal is useful for strain mapping as well as for imaging light dopant atoms and vacancies2434.At even higher angles,these effects become less pronounced and the atomic- number contrast from nuclear scattering dominates.The resulting high-angle ADF images are easier to interpret,with the brighter features resulting from high-Z atoms,or higher densities of atoms. Electron channelling in the crystal can still provide unexpected con- trast that may confound a direct quantitative analysis,but gener- ally the contrast is still sufficient that model inputs for quantitative simulations can be directly determineds4345-7. Compositional and bonding information can be obtained from electron energy-loss spectroscopy (EELS)analysis of the inelasti- cally scattered electrons.The time-varying,electric-field pulses from the incident-beam electrons transfer energy to the sample over a range of frequencies,from the infrared to the X-ray regime, as they pass near target atoms.The resulting core-level excitations provide unique spectroscopic information about the excited atom and its bonding states(Fig.4).Fortunately,the inelastic scattering is strongly peaked in the forward direction and so passes through the hole in the centre of the ADF detector.A spectrometer placed Dipole on axis can collect upwards of 90%of the EEL signal without com- transition promising the ADF signal.This ADF/EELS geometry has been used since the earliest days of STEM,where the larger ADF signal is used to form an image and locate the probe for selected EELS measurements*537.By 1986,0.4-nm resolution composition profiles Core level were demonstrated by Scheinfein and Isaacson.Development of low-noise parallel detectors enabled the recording of spectra that showed details of chemical bonding from selected points in an atomic-resolution image4 The time required to record a single spectrum using that tech- Figure 4|Electron energy-loss spectroscopy (EELS)as a probe of local nology was~1-10 s.Even a spectral image with as few as 64 x64 pixels electronic structure and composition.EELS records the energy lost by an could take hours to record-a serious challenge when typical sample incident electron as it excites a core electron to unoccupied local states, drift rates are~5 A min-.Despite this,atomic-resolution composi- leaving behind a core hole.Analysing the resulting loss spectrum of the tion maps have been demonstrated (after significant drift correction transmitted beam provides information on the unoccupied local density and data processing)s.One of the key benefits of aberration correc- of states,partitioned by site,chemical species and angular momentum. tors is that they increase the signal,allowing collection times to be The site specificity is achieved by forming a probe small enough to only reduced from hours to under a minute(Fig.1),and also increase the excite electrons on a single atomic column.The chemical specificity is signal-to-noise ratio so that bonding features are visible22.A full two- obtained from the uniqueness of the core-level binding energy.The angular dimensional chemical map obtained by EELS that also contains bond- momentum of the final state is determined by the dipole selection rules ing information requires yet another order of magnitude increase and the angular momentum of the initial state.When the core hole is well in the beam current beyond that needed for simple compositional screened,the shape of the EELS spectrum reflects the ground state density imaging in order to improve the signal-to-noise ratio.This has been of states.In systems with strong core-hole coupling,excitonic effects can done by further correcting the electron optical aberrations to fifth- dramatically alter the shape of spectrum,but also provide useful local order,rather than only to third-order as in earlier correctors.With fingerprints of the formal charge and coordination chemistry. NATURE MATERIALS|VOL 8|APRIL 2009 www.nature.com/naturematerials 265 2009 Macmillan Publishers Limited.All rights reserved

nature materials | VOL 8 | APRIL 2009 | www.nature.com/naturematerials 265 NAture mAterIAls doi: 10.1038/nmat2380 insight | review articles than the spacings between the projected atomic columns. The beam is then scanned across the sample, and the differences in electron￾scattering when the beam is located on and off the atomic columns are recorded. The optimal sample thickness is a balance between being thin enough that the electron probe will not spread signifi￾cantly in the sample, and thick enough so that bulk-like, and not surface, atoms dominate the signal. Elastically scattered electrons can undergo large angle deflections, and hence an annular dark-field (ADF) detector is placed below the sample to collect the scattered electrons. The high-angle electron￾scattering cross-section scales roughly as the atomic number Z1.7, asymptotically tending towards the Rutherford scattering limit of Z2 , and so this imaging mode is sometimes called ‘Z-contrast’ imag￾ing18,37–39. Provided the collection angles are three or more times larger than the probe convergence angle, the incoherent imaging approximation is satisfied for thin samples, thereby allowing rela￾tively straightforward image interpretation39–42. To avoid contrast reversals with sample thickness, the detector also needs to be suf￾ficiently large to avoid any strongly diffracted beams from being collected19,38,43. When this condition is just met, diffuse scattering from phonons and lattice distortions can provide significant image contrast. This low-angle ADF signal is useful for strain mapping as well as for imaging light dopant atoms and vacancies23,43,44. At even higher angles, these effects become less pronounced and the atomic￾number contrast from nuclear scattering dominates. The resulting high-angle ADF images are easier to interpret, with the brighter features resulting from high-Z atoms, or higher densities of atoms. Electron channelling in the crystal can still provide unexpected con￾trast that may confound a direct quantitative analysis, but gener￾ally the contrast is still sufficient that model inputs for quantitative simulations can be directly determined39,43,45–47. Compositional and bonding information can be obtained from electron energy-loss spectroscopy (EELS) analysis of the inelasti￾cally scattered electrons. The time-varying, electric-field pulses from the incident-beam electrons transfer energy to the sample over a range of frequencies, from the infrared to the X-ray regime, as they pass near target atoms. The resulting core-level excitations provide unique spectroscopic information about the excited atom and its bonding states (Fig. 4). Fortunately, the inelastic scattering is strongly peaked in the forward direction and so passes through the hole in the centre of the ADF detector. A spectrometer placed on axis can collect upwards of 90% of the EEL signal without com￾promising the ADF signal. This ADF/EELS geometry has been used since the earliest days of STEM, where the larger ADF signal is used to form an image and locate the probe for selected EELS measurements35,37. By 1986, 0.4-nm resolution composition profiles were demonstrated by Scheinfein and Isaacson48. Development of low-noise parallel detectors enabled the recording of spectra that showed details of chemical bonding from selected points in an atomic-resolution image49. The time required to record a single spectrum using that tech￾nology was ~1–10 s. Even a spectral image with as few as 64 × 64 pixels could take hours to record — a serious challenge when typical sample drift rates are ~5 Å min−1. Despite this, atomic-resolution composi￾tion maps have been demonstrated (after significant drift correction and data processing)50. One of the key benefits of aberration correc￾tors is that they increase the signal, allowing collection times to be reduced from hours to under a minute (Fig. 1), and also increase the signal-to-noise ratio so that bonding features are visible32. A full two￾dimensional chemical map obtained by EELS that also contains bond￾ing information requires yet another order of magnitude increase in the beam current beyond that needed for simple compositional imaging in order to improve the signal-to-noise ratio. This has been done by further correcting the electron optical aberrations to fifth￾order, rather than only to third-order as in earlier correctors. With the aberrations corrected over a wider angular range, the numerical aperture of the probe-forming lens can then be increased, allowing more electrons to be brought to a focused spot. It has also proved nec￾essary to correct the collection optics so that the additional electrons input at the larger angles could still be collected31,32. A very powerful feature of EELS is that it provides both com￾positional and bonding information at very high spatial resolution. Figure 4 shows the EELS excitation process whereby the incident electron transfers energy to the target atom, exciting a core electron to empty states above the Fermi level (EFermi). The energy loss of the transmitted electron is recorded by the spectrometer and provides information equivalent to X-ray absorption spectroscopy, where the electron’s momentum transfer has the same role as the polarization of the X-ray beam51,52. The core-level binding energy that marks the EELS edge onset allows a unique elemental identification to be made. The shape of the edge itself reflects the underlying local partial density of states modified by the presence of a core hole53–57. The core hole is relatively well screened in metals, allowing the spec￾trum to be interpreted in terms of the ground state local density of EFermi Dipole transition Core level Figure 4 | electron energy-loss spectroscopy (eels) as a probe of local electronic structure and composition. EELS records the energy lost by an incident electron as it excites a core electron to unoccupied local states, leaving behind a core hole. Analysing the resulting loss spectrum of the transmitted beam provides information on the unoccupied local density of states, partitioned by site, chemical species and angular momentum. The site specificity is achieved by forming a probe small enough to only excite electrons on a single atomic column. The chemical specificity is obtained from the uniqueness of the core-level binding energy. The angular momentum of the final state is determined by the dipole selection rules and the angular momentum of the initial state. When the core hole is well screened, the shape of the EELS spectrum reflects the ground state density of states. In systems with strong core-hole coupling, excitonic effects can dramatically alter the shape of spectrum, but also provide useful local fingerprints of the formal charge and coordination chemistry. nmat_2380_APR09.indd 265 11/3/09 11:12:08 © 2009 Macmillan Publishers Limited. All rights reserved

REVIEW ARTICLES INSIGHT NATURE MATERIALS DOI:10.1038/NMAT2380 statess154-56.Core-hole effects can dominate in insulators,especially be possible for electron imaging.The typical electron wavelength on the cation sites,contracting the final state wavefunctions and in a TEM is on the order of 0.02 A,yet the spatial resolution of allowing a more molecular or crystal-field-like interpretation of the such a microscope is 1-2 A,between 50 and 100 times'worse'than local bonding environment to be mades.s6.58-40. the wavelength.Wavelength for wavelength,conventional elec- The microscopic insight into electronic structure at buried inter- tron lenses have similar optical properties to the bottom of a beer faces provided by EELS has proved useful in the study of transistor bottle with respect to spherical and chromatic aberrations.This is miniaturization.The narrowest feature in modern transistors is the not the result of imperfect engineering,but rather a consequence of gate oxide.This is a dielectric layer roughly 5-6 atoms thick and tra- Laplace's equation,where the electric and magnetic potentials used ditionally made from SiO,-which was just recently replaced with to focus the electron beam cannot take on arbitrary shapes in free Hf-based dielectric materials that for the moment are about twice as space.The result is that positive spherical and chromatic aberrations thick3.As device-scaling continues,these layers will also have to are unavoidable for static round lenses.This fundamental limit was shrink.However,their continued scaling is limited by the interfacial first identified by Scherzer7 in 1936,who also suggested methods reaction layers and altered electronic structure present at the inter- to correct these aberrations by considering non-static,non-round face22.For example,in the five-Si-atom-thick gate oxide shown in lenses or placing free charges in the path of the beam.Of these, Fig.5,at least two of those five atoms will be at the silicon/oxide the most practical approach has proved to be abandoning cylindri- interfaces,and EELS shows a very different local electronic struc- cal symmetry for some of the lenses and instead adding a series of ture for the interfacial atoms.These atoms are expected to have very multipole optical elements that correct the leading terms in an ever- different electrical and optical properties from the desired bulk SiO, increasing power-series expansion of the wavefront distortions9 yet they comprise a significant fraction of the dielectric layer257. It took nearly 50 years before the precision,stability and software In more-ionic materials,the local EELS fingerprint can be used control needed to implement these corrector designs could be real- to map the spatial distribution of formal charges at interfaces and ized24257.Figure 2 shows that since the proof of concept was estab defects3-65.The 3d transition metal edges are well suited to this task lished,corrector development has been rapid (and is still continuing), and atomic-scale analysis of the Ti-L edge has provided the spatial with probe sizes close to 0.5 A now reached2930.Aberration correctors distribution of conduction electrons and their screening lengths in have now removed spherical aberration(and higher-order geometric LaTiO,/SrTiO,multilayers,where a highly conducting sheet of elec- aberrations)as limiting factorsInstead,chromatic aberration, trons is formed at the interface between a Mott insulator and a band combined with the finite energy spread of the electron source,are the insulator4.Corrected microscopes should prove especially useful current limits on spatial resolution The resolution and aperture for identifying single dopant atoms inside a specific device,where size can be increased by building a chromatic aberration corrector?1 the 'tyranny of small numbers'requires two-dimensional imaging increasing the beam voltage or reducing energy spread of the source. instead of simply line profiles to obtain sufficient counting statistics Improvements of a factor of two appear possible,but not without from volumes that contain only a few atoms47.66. cost:increasing the beam voltage also increases the risk of radiation damage-so beam voltage is more likely to be decreased in future Advances in electron optics instruments,only exacerbating the chromatic problems.Reducing Although the ability to focus an electron beam down to a spot size the source energy spread by adding a monochromator reduces the of around 2 A with a simple,round magnetic lens is remarkable, available beam current.Cryogenically cooled sources can narrow the 2 A is still far from the fundamental performance limit that could source energy spread and improve gun brightness"2,but have proved difficult to implement owing to the required mechanical stability. b Corrective optics are also needed to retain and improve the 1,400 collection efficiency of the EELS spectrometer32 Without these corrections,much of the additional current provided by the probe a-Si 1.200 corrector will not enter the spectrometer,yet will add to the radia- 9 tion dose of the sample.The effects can be dramatic-an uncor- 1.000 rected spectrometer on a corrected microscope can lose as much as 90%of the inelastic signal and provide no more signal than a 800 completely uncorrected microscope.These very low collection effi- ciencies introduce additional problems in image formation and S102 interpretation,because the incoherent imaging approximation no 600 longer holds?-75.Diffraction contrast artefacts,such as bend con- tours and thickness fringes,can appear and elastic scattering arte- 400 facts can dominate the EELS spectrum image76-7.Strong elastic scattering on heavy-atom columns can prevent inelastic electrons 200 from entering the spectrometer,resulting in contrast reversals and inelastic images that reflect mostly the elastic scattering and not the 4 上www0 elemental distribution in the sample. STEM image 520530540550560570 A common symptom of this elastic scattering artefact is that Energy loss (eV) all inelastic maps peak at the same spatial positions,independent of the actual location of the elements.This problem was identified Figure 5|Electron energy-loss spectrum recorded point by point across a and demonstrated by quantitative theory and experimental work gate stack containing a thin gate oxide.a,Circles in the annular dark-field for BiosSrosMnO,-where the 'inelastic'images are dominated by image indicate where each spectrum was taken.The [100]silicon substrate the strong elastic scattering from the Bi sites that largely prevents is in the lower half,the gate oxide is in the middle and the deposited a-Si gate electrons on the Bi column from entering the detector?.The result electrode is in the top half.b,The background-corrected oxygen K edges for the (111)zone axis is an apparent map of Mn,where the Mn for the various positions across the gate stack.The centre of the oxide layer signal is smallest on the only columns that actually contain Mn,and shows bulk-like oxygen bonding (blue dots)and altered oxygen bonding(red largest where there are no Mn atoms?6.The effect can be suppressed dots)is present at atoms near the interfaces.(From ref.2:1999 NPG.) by increasing the collection angle so that almost all of the elastically 266 NATURE MATERIALS VOL 8|APRIL 2009 www.nature.com/naturematerials 2009 Macmillan Publishers Limited.All rights reserved

266 nature materials | VOL 8 | APRIL 2009 | www.nature.com/naturematerials review articles | insight NAture mAterIAls doi: 10.1038/nmat2380 states51,54–56. Core-hole effects can dominate in insulators, especially on the cation sites, contracting the final state wavefunctions and allowing a more molecular or crystal-field-like interpretation of the local bonding environment to be made51,56,58–60. The microscopic insight into electronic structure at buried inter￾faces provided by EELS has proved useful in the study of transistor miniaturization. The narrowest feature in modern transistors is the gate oxide. This is a dielectric layer roughly 5–6 atoms thick and tra￾ditionally made from SiO2 — which was just recently replaced with Hf-based dielectric materials that for the moment are about twice as thick3,61. As device-scaling continues, these layers will also have to shrink. However, their continued scaling is limited by the interfacial reaction layers and altered electronic structure present at the inter￾face2,62. For example, in the five-Si-atom-thick gate oxide shown in Fig. 5, at least two of those five atoms will be at the silicon/oxide interfaces, and EELS shows a very different local electronic struc￾ture for the interfacial atoms. These atoms are expected to have very different electrical and optical properties from the desired bulk SiO2, yet they comprise a significant fraction of the dielectric layer2,57. In more-ionic materials, the local EELS fingerprint can be used to map the spatial distribution of formal charges at interfaces and defects63–65. The 3d transition metal edges are well suited to this task and atomic-scale analysis of the Ti–L edge has provided the spatial distribution of conduction electrons and their screening lengths in LaTiO3/SrTiO3 multilayers, where a highly conducting sheet of elec￾trons is formed at the interface between a Mott insulator and a band insulator64. Corrected microscopes should prove especially useful for identifying single dopant atoms inside a specific device, where the ‘tyranny of small numbers’ requires two-dimensional imaging instead of simply line profiles to obtain sufficient counting statistics from volumes that contain only a few atoms47,66. advances in electron optics Although the ability to focus an electron beam down to a spot size of around 2 Å with a simple, round magnetic lens is remarkable, 2 Å is still far from the fundamental performance limit that could be possible for electron imaging. The typical electron wavelength in a TEM is on the order of 0.02 Å, yet the spatial resolution of such a microscope is 1–2 Å, between 50 and 100 times ‘worse’ than the wavelength. Wavelength for wavelength, conventional elec￾tron lenses have similar optical properties to the bottom of a beer bottle with respect to spherical and chromatic aberrations. This is not the result of imperfect engineering, but rather a consequence of Laplace’s equation, where the electric and magnetic potentials used to focus the electron beam cannot take on arbitrary shapes in free space. The result is that positive spherical and chromatic aberrations are unavoidable for static round lenses. This fundamental limit was first identified by Scherzer67 in 1936, who also suggested methods to correct these aberrations by considering non-static, non-round lenses or placing free charges in the path of the beam68. Of these, the most practical approach has proved to be abandoning cylindri￾cal symmetry for some of the lenses and instead adding a series of multipole optical elements that correct the leading terms in an ever￾increasing power-series expansion of the wavefront distortions69. It took nearly 50 years before the precision, stability and software control needed to implement these corrector designs could be real￾ized24,25,70. Figure 2 shows that since the proof of concept was estab￾lished, corrector development has been rapid (and is still continuing), with probe sizes close to 0.5 Å now reached29,30. Aberration correctors have now removed spherical aberration (and higher-order geometric aberrations) as limiting factors24,26–28,31. Instead, chromatic aberration, combined with the finite energy spread of the electron source, are the current limits on spatial resolution27,31. The resolution and aperture size can be increased by building a chromatic aberration corrector71, increasing the beam voltage or reducing energy spread of the source. Improvements of a factor of two appear possible, but not without cost: increasing the beam voltage also increases the risk of radiation damage — so beam voltage is more likely to be decreased in future instruments, only exacerbating the chromatic problems. Reducing the source energy spread by adding a monochromator reduces the available beam current. Cryogenically cooled sources can narrow the source energy spread and improve gun brightness72, but have proved difficult to implement owing to the required mechanical stability. Corrective optics are also needed to retain and improve the collection efficiency of the EELS spectrometer31,32. Without these corrections, much of the additional current provided by the probe corrector will not enter the spectrometer, yet will add to the radia￾tion dose of the sample. The effects can be dramatic — an uncor￾rected spectrometer on a corrected microscope can lose as much as 90% of the inelastic signal and provide no more signal than a completely uncorrected microscope. These very low collection effi￾ciencies introduce additional problems in image formation and interpretation, because the incoherent imaging approximation no longer holds73–75. Diffraction contrast artefacts, such as bend con￾tours and thickness fringes, can appear and elastic scattering arte￾facts can dominate the EELS spectrum image76–78. Strong elastic scattering on heavy-atom columns can prevent inelastic electrons from entering the spectrometer, resulting in contrast reversals and inelastic images that reflect mostly the elastic scattering and not the elemental distribution in the sample. A common symptom of this elastic scattering artefact is that all inelastic maps peak at the same spatial positions, independent of the actual location of the elements. This problem was identified and demonstrated by quantitative theory and experimental work for Bi0.5Sr0.5MnO3 — where the ‘inelastic’ images are dominated by the strong elastic scattering from the Bi sites that largely prevents electrons on the Bi column from entering the detector76. The result for the 〈111〉 zone axis is an apparent map of Mn, where the Mn signal is smallest on the only columns that actually contain Mn, and largest where there are no Mn atoms76. The effect can be suppressed by increasing the collection angle so that almost all of the elastically 520 0 200 400 600 800 1,000 1,200 1,400 530 540 Energy loss (eV) Oxygen K edge intensity (arbitary units) STEM image 550 560 570 a-Si SiO2 Si 5.42 Å a b Figure 5 | electron energy-loss spectrum recorded point by point across a gate stack containing a thin gate oxide . a, Circles in the annular dark-field image indicate where each spectrum was taken. The [100] silicon substrate is in the lower half, the gate oxide is in the middle and the deposited a-Si gate electrode is in the top half. b, The background-corrected oxygen K edges for the various positions across the gate stack. The centre of the oxide layer shows bulk-like oxygen bonding (blue dots) and altered oxygen bonding (red dots) is present at atoms near the interfaces. (From ref. 2; © 1999 NPG.) nmat_2380_APR09.indd 266 11/3/09 11:12:09 © 2009 Macmillan Publishers Limited. All rights reserved

NATURE MATERIALS DOL:10.1038/NMAT2380 INSIGHT I REVIEW ARTICLES 40 mrad 25 mrad 10m 1.000 100 10 Probe size (A) Figure 6|Phase space for operation of a scanning electron microscope at a fixed source brightness.The goal of the largest probe current in Ga 63 pm the smallest-diameter probe can be reached by correcting the electron optics to allow a larger optimal probe-forming aperture.Red data points Figure 7 Direct demonstration of sub-angstrom resolution ADF-STEM represent typical operating conditions for a 100-keV cold field emission imaging.The sample is [211]-oriented wurtzite GaN,showing a 0.63-nm gun on various generations of corrected microscopes with different probe spacing between the projected Ga columns (yellow).The lighter N atoms semi-angles-uncorrected (10 mrad),third-order corrected (25 mrad) (blue circles)are not visible in the image.The elastic signal is displayed and fifth-order corrected(40 mrad).At high beam currents,the instrument n false colour to emphasize the Ga columns.(From ref.29;2008 performance is brightness-limited,where spatial resolution is determined by Cambridge Univ.Press.) the source size.Only by changing the physical nature of the electron source can the beam current be increased without increasing the probe size or or a 16-fold increase in beam current.Another factor of two increase numerical aperture.In the limit of zero beam current,the spatial resolution in numerical aperture may still be technically feasible-allowing is determined by the diffraction limit from the largest usable aperture.The electron probes as small as 0.25 A,or a 64-fold increase in beam cur- aperture size can be increased by correcting optical aberrations. rent.Fundamentally,there is no physical law preventing a smaller probe from being designed,but the technical and engineering chal- scattered electrons enter the spectrometer.However,at such large lenges become very severe.Although aberration correctors correct angles the energy resolution of the spectrum can be degraded.With aberrations,they do not correct instabilities.Mechanical stability proper design and corrective optics,the energy resolution can be and environmental sensitivity often limit the usability of instruments preserved and incoherent EELS maps,free of diffraction and other today.Without careful room design,the microscope is also likely to elastic artefacts,can be obtained".As with elastic scattering,chan- function as an expensive barometer,magnetometer,accelerometer, nelling inside the crystal can further complicate a quantitative inter- thermoanemometer,microphone and truck detector283. pretation of the EELS signal. One additional constraint on the spatial resolution of the EELS Electron sources method is the delocalization oftheinelasticscatteringprocess47 One future area of improvement is that of the electron sources Even if the electron beam could be made arbitrarily small,the elec- themselves.For a given aperture size (determined by the electron tric field surrounding the electron is over a longer range.Even optics),the source size and usable beam current is set by the source though the time-varying electric fields of an electron travelling in brightness,which is a physical property of the source itself.Source a straight line must be evanescent,the near-field terms fall off over size can be traded for beam current at a fixed aperture size and a small but finite distance.The spatial decay of the field is faster for brightness.Figure 6 shows the phase space of beam current versus larger energy losses,but it is not until a few hundred volts energy resolution for a typical STEM.There is a hundredfold difference loss that the localization reaches atomic dimensions.A localiza- between state of the art and conventional instruments.At low beam tion of 1-2 A is expected for typical energy losses of 300-1,000 eV currents the diffraction-limited spot size set by the corrected optics (refs 73,74,81).At higher energies the signal becomes weaker,and dominates,and at high beam currents the finite source size limits noise limits the spatial resolution.At lower energies,the charac- the resolution.Figure 7 shows the 0.63-A spacing in GaN imaged teristic delocalization length is likely to be larger than the spacing by aberration-corrected STEM (these results have been reported between atoms,and image resolution and direct interpretability are by both Japanese%and US groups2).The same probe size could lost?.For EELS work,the main benefit of correctors is likely to be be reached with a lower-brightness source,but the images would the very large increases in signal at a resolution of 1-2 A rather than appear noisier. deep,sub-angstrom spatial resolution.The increased signal allows Most commercial microscopes use thermally assisted or Schottky for faster data acquisition,making it practical to record full spectral field emitters,but cold field emitters that rely on direct tunnelling images rather than only line profiles. out of the tip have a tenfold higher beam brightness in practice. At present,correctors allow up to a factor of four increase in The intrinsic energy spread of both sources is about 0.3 eV at room the numerical aperture(from 10 to 40 mrad),leading to a fourfold temperature and low beam current4.Increasing the extraction cur- improvementin spatialresolution(to2 A-;that is,aspacingof0.5A),rents increases electron-electron scattering in the beam,which NATURE MATERIALS VOL 8 APRIL 2009 www.nature.com/naturematerials 267 2009 Macmillan Publishers Limited.All rights reserved

nature materials | VOL 8 | APRIL 2009 | www.nature.com/naturematerials 267 NAture mAterIAls doi: 10.1038/nmat2380 insight | review articles scattered electrons enter the spectrometer. However, at such large angles the energy resolution of the spectrum can be degraded. With proper design and corrective optics, the energy resolution can be preserved and incoherent EELS maps, free of diffraction and other elastic artefacts, can be obtained31. As with elastic scattering, chan￾nelling inside the crystal can further complicate a quantitative inter￾pretation of the EELS signal. One additional constraint on the spatial resolution of the EELS method is the delocalization of the inelastic scattering process73,74,79,80. Even if the electron beam could be made arbitrarily small, the elec￾tric field surrounding the electron is over a longer range. Even though the time-varying electric fields of an electron travelling in a straight line must be evanescent, the near-field terms fall off over a small but finite distance. The spatial decay of the field is faster for larger energy losses, but it is not until a few hundred volts energy loss that the localization reaches atomic dimensions. A localiza￾tion of 1–2 Å is expected for typical energy losses of 300–1,000 eV (refs 73,74,81). At higher energies the signal becomes weaker, and noise limits the spatial resolution. At lower energies, the charac￾teristic delocalization length is likely to be larger than the spacing between atoms, and image resolution and direct interpretability are lost79. For EELS work, the main benefit of correctors is likely to be the very large increases in signal at a resolution of 1–2 Å rather than deep, sub-ångström spatial resolution. The increased signal allows for faster data acquisition, making it practical to record full spectral images rather than only line profiles. At present, correctors allow up to a factor of four increase in the numerical aperture (from 10 to 40 mrad), leading to a fourfold improvement in spatial resolution (to 2 Å–1; that is, a spacing of 0.5Å), or a 16-fold increase in beam current. Another factor of two increase in numerical aperture may still be technically feasible — allowing electron probes as small as 0.25 Å, or a 64-fold increase in beam cur￾rent. Fundamentally, there is no physical law preventing a smaller probe from being designed, but the technical and engineering chal￾lenges become very severe. Although aberration correctors correct aberrations, they do not correct instabilities. Mechanical stability and environmental sensitivity often limit the usability of instruments today. Without careful room design, the microscope is also likely to function as an expensive barometer, magnetometer, accelerometer, thermoanemometer, microphone and truck detector82,83. electron sources One future area of improvement is that of the electron sources themselves. For a given aperture size (determined by the electron optics), the source size and usable beam current is set by the source brightness, which is a physical property of the source itself. Source size can be traded for beam current at a fixed aperture size and brightness. Figure 6 shows the phase space of beam current versus resolution for a typical STEM. There is a hundredfold difference between state of the art and conventional instruments. At low beam currents the diffraction-limited spot size set by the corrected optics dominates, and at high beam currents the finite source size limits the resolution. Figure 7 shows the 0.63-Å spacing in GaN imaged by aberration-corrected STEM (these results have been reported by both Japanese30 and US groups29). The same probe size could be reached with a lower-brightness source, but the images would appear noisier. Most commercial microscopes use thermally assisted or Schottky field emitters, but cold field emitters that rely on direct tunnelling out of the tip have a tenfold higher beam brightness in practice. The intrinsic energy spread of both sources is about 0.3 eV at room temperature and low beam current84. Increasing the extraction cur￾rents increases electron–electron scattering in the beam, which 1 Probe size (Å) 1 10 10 100 Probe current (pA) 1,000 40 mrad 25 mrad 10 mrad Figure 6 | Phase space for operation of a scanning electron microscope at a fixed source brightness. The goal of the largest probe current in the smallest-diameter probe can be reached by correcting the electron optics to allow a larger optimal probe-forming aperture. Red data points represent typical operating conditions for a 100-keV cold field emission gun on various generations of corrected microscopes with different probe semi-angles — uncorrected (10 mrad), third-order corrected (25 mrad) and fifth-order corrected (40 mrad). At high beam currents, the instrument performance is brightness-limited, where spatial resolution is determined by the source size. Only by changing the physical nature of the electron source can the beam current be increased without increasing the probe size or numerical aperture. In the limit of zero beam current, the spatial resolution is determined by the diffraction limit from the largest usable aperture. The aperture size can be increased by correcting optical aberrations. Ga 63 pm Figure 7 | Direct demonstration of sub-ångström resolution ADF–stem imaging. The sample is [211]-oriented wurtzite GaN, showing a 0.63-nm spacing between the projected Ga columns (yellow). The lighter N atoms (blue circles) are not visible in the image. The elastic signal is displayed in false colour to emphasize the Ga columns. (From ref. 29; © 2008 Cambridge Univ. Press.) nmat_2380_APR09.indd 267 11/3/09 11:12:10 © 2009 Macmillan Publishers Limited. All rights reserved