Low-Energy Electron Microscopy: Imaging Surface Dynamics

In 1985, Ernst Bauer and his student Wolfgang Telieps published a stunning set of images that abruptly solved a long-debated question in surface science: What is the nature of the phase transition that occurs on the (111) surface of silicon? ¹ Determining the complex ordered arrangement of atoms, or “reconstruction,” that occurs there had been one of the hottest problems in surface science for nearly 25 years. ² One of their images (see figure 1) shows a sharply defined coexistence between two structural phases and demonstrates a first-order—rather than a continuous second-order—transition between an ordered (bright) and disordered (dark) arrangement of atoms at the surface. Bauer and Telieps’s unambiguous answer about the nature of that disordering transition dramatically introduced a very powerful probe of solid surfaces: low-energy electron microscopy (LEEM).

Since publication of those early images, an increasing number of investigators have used LEEM to gain insights into a variety of dynamical processes on surfaces. After all, whether the interest is crystal growth, catalysis, or thin-film or multilayer materials engineering, surface science is based on understanding surfaces not as static objects, but as dynamic systems. Key questions include, How do structural transitions nucleate and propagate across a surface? What are the limiting processes in atom transport across a surface during temperature treatments or material deposition?

At roughly the same time that successful LEEM research began, scanning tunneling microscopy burst on the scene as an important technique in surface science. Although STM, with its subnanometer resolution, was well suited for examining surface structures and atomic arrangements, LEEM, with its large field of view and fast imaging rate, was well poised for studying dynamics.

LEEM combines a number of spectroscopic techniques and uses the fact—already exploited in low-energy electron diffraction—that electrons with energies of a few to a few tens of eV are extremely sensitive to near-surface atomic order. Depending on the energy of the incident electron beam, subtle differences in local atomic structure or composition can give rise to dramatic contrast in LEEM images. An example is the contrast, evident in figure 1, that results from a strong difference in the specularly diffracted intensity between ordered and disordered structures. What distinguishes LEEM from LEED and scattering techniques is its real-time surface imaging capability (although the diffraction pattern is still available in the back focal plane of the objective lens). Image formation with electrons is analogous to light-microscope imaging. The backscattered electron beam diverges and refocuses through electrostatic and magnetic lenses on its way to a phosphor screen (see the box on page 52). LEEM is the fruit of more than 20 years of work by Bauer, who recognized by the early 1960s that the high reflectivity of low-energy electrons from surfaces would make surface-sensitive imaging practical. The first published images caused great excitement in the surface science community because they were acquired with high spatial resolution, typically several nanometers, and at video rates. LEEM can image in situ samples over a wide range of temperatures and in the presence of incident fluxes of atoms, ions, or light. That flexibility allows researchers to more easily control the evolution of nanostructures.

Cobalt on silicon

Bauer and Telieps’s initial observation of phase coexistence became the starting point for a growing number of discoveries using LEEM. One crucial problem in modifying surfaces is determining how the deposition of different types of atoms influences the structures that form and coexist. For example, one of us (Phaneuf) and Peter Bennett at Arizona State University used LEEM to investigate the structures that evolve when a very small flux of cobalt adsorbs onto the Si(111) surface at elevated temperatures. ³ Our results demonstrated an interesting consequence of certain impurities on the transition: a stabilization of the disordered phase below the clean-surface transition temperature. Starting with a single-phase (7 × 7) ordered surface (figure 2(a)), Co segregated to steps and domain boundaries of the ordered structure (figure 2(b)). Further deposition—up to a tenth of a monolayer—caused the disordered Co-containing structure to spread outward, leaving an almost entirely dark image. Previous STM measurements suggest that the disordered, so-called “(1 × 1),” regions contain a random arrangement of Co-containing clusters and individual Si adatoms decorating a bulklike surface termination. ⁴ The disordered structure visibly grew outward from steps and domain boundaries at the expense of the ordered regions when either the Co coverage was increased at a fixed temperature or when the temperature was raised at a fixed Co coverage. Images recorded while heating or cooling the surface revealed that the transition was reversible, so that a local equilibrium existed between phases.

That insight about the transition would have been almost impossible to deduce from diffraction studies alone because the disordered structure produces no extra reflections in the diffraction pattern. Increasing the Co coverage produces a large and continuous change of the transition temperature, with a depression of 170°C resulting from only a tenth of a monolayer. The driving force is well known from bulk alloy systems: Adding an insoluble impurity to a solid phase lowers its melting point and allows the impurity to segregate to the liquid phase, in which its solubility is higher. In our case, lowering the disordering temperature allowed Co-containing clusters to segregate out of the ordered surface regions into the disordered phase. The shape of the phase boundary in figure 2(c) indicates the presence of a small repulsion between the Co-containing particles in the disordered phase.

Even for atomically clean Si(111) surfaces, the fact that there are coexisting ordered and disordered regions at all is surprising. At first, such coexistence seems at odds with the Gibbs phase rule, which predicts that only a single phase should exist at a given temperature and pressure, with sharp transitions between phases. Using LEEM, Jim Hannon and his collaborators at IBM ⁵ recently discovered that long-range interactions between phase boundaries preclude a sharp transition between phases, even when the defect-free regions of the surface are arbitrarily large—the so-called thermodynamic limit. The nature of the domain boundary interactions can stem either from the two phases having different surface stresses or from their having different work functions. The temperature dependence of the relative sizes of the two types of regions on the surface are consistent with a model that includes a term in the free energy corresponding to the elastic interaction between “force monopoles” at the boundaries between phases.

Step dynamics

The use of LEEM has progressed gradually, from qualitative imaging to fully quantitative analysis. The almost unique ability of the LEEM to form real-time images over a wide range of temperatures and conditions allows researchers to extract thermodynamic quantities that would be difficult, if not impossible, to obtain using other techniques.

Consider the work of Ruud Tromp and his collaborators at IBM. ⁶ Those researchers observed the Brownian motion of monolayer surface steps and studied the evolution of monolayer islands on the (001) surface of Si. Both phenomena are mediated by the attachment and detachment of Si atoms from the edges of the steps or islands and by their diffusion across the surface. Which of the activated atomic mechanisms is rate limiting, attachment/detachment or diffusion? Thermodynamics distinguishes the two cases based on a statistical analysis of the Fourier components of the spatial and temporal correlations along a fluctuating step edge. In particular, the amplitude A(q) of the Fourier component corresponding to the wavevector q = 2π/λ should be inversely proportional to λ² if attachment/detachment is rate limiting, but to λ³ if, instead, fluctuations result from a diffusion-limited process. (See discussions in the articles by Zoltán Toroczkai and Ellen D. Williams, Physics Today, December 1999, page 24 , and by Harold Zandvliet, Bene Poelsema and Brian Swartzentruber, Physics Today, July 2001, page 40 .)

Island hopping

Wolfgang Theis, Bartelt and Tromp ⁶ also studied the “ripening” that occurs on Si at sufficiently high temperature as larger islands grow at the expense of smaller islands on the surface. The well-known Gibbs–Thomson effect, in which the chemical potential of an island is proportional to the inverse of its radius of curvature, explains the ripening. Figures 3(a) and 3(b) show LEEM snapshots of how a configuration of islands coexisting on the surface has evolved. One might guess that the time dependence of the island sizes could be predicted from a mean-field model, in which a uniform chemical potential describes the sea of atoms diffusing from island to island. As it turned out, that idea was nearly correct: Starting with a refined approximation—allowing the chemical potential in different cells around each island to take on different values, depending on island sizes—the model becomes essentially perfect. Compare the experimental measurement in figure 3(b) with the results from computations in figure 3(c). What’s striking is that the model takes kinetic and energy parameters obtained from thermal step fluctuations and successfully predicts the histories of individual islands.

Those observations highlight the importance of the sea of detached atoms, always present on surfaces at finite temperature. Although this “adatom lattice gas” is too dilute and too mobile to be observed directly, Tromp and Marian Mankos ⁶ successfully measured its density. They analyzed the appearance of islands on extremely large step-free regions of a lithographically altered Si(001) surface subsequent to an abrupt drop in temperature. Presumably, those islands were made up of atoms in the adatom lattice gas phase before the temperature was dropped. Measuring the sizes of the islands, the researchers were able to extract the concentration of Si atoms on the surface as a function of temperature. A simple Arrhenius model indicates that the diffusing species on the surface are not isolated atoms at all, but pairs of atoms, or dimers. These same dimers are known from STM images to line up in rows and form the anisotropic reconstruction mentioned above.

Self-assembly

Many of the early experiments—and those showcased in this article so far—concentrated on Si surfaces because of Si’s technological importance and the ease with which flat, atomically clean Si surfaces can be prepared. However, a number of groups have investigated the evolution of metal surfaces during deposition, alloying, or relaxation toward equilibrium. A striking example of the rich variety of phenomena that can occur as an alloy system orders comes from recent work by Gary Kellogg’s group at Sandia National Laboratories. ⁷ Using LEEM, those researchers observed self-assembly on the nanometer scale during the deposition of lead vapor onto an atomically clean (111) surface of copper at 400°C. Initially, the deposition of lead at that temperature forms a disordered Pb–Cu alloy. Once the alloy covers the surface, additional Pb beads up into compact, nearly circular 70-nm islands. As the coverage increases, the islands begin to order into a hexagonal pattern that Richard Plass describes as a droplet phase (figure 4(a)). After still more coverage, the islands coalesce into a configuration of stripes (figure 4(b)) and then into an arrangement of holes in the lead matrix; that arrangement constitutes the so-called negative droplet phase (figure 4(c)).

The still images, acquired over several minutes, do not convey the richness of the structures’ evolution: the nucleation of droplets containing tens of thousands of atoms occurring in a single 30-ms frame. Observation of individual droplets reveals that the islands move within a force field corresponding to dipole–dipole interactions. That motion is consistent with the idea that the interaction forces between islands are mediated by strain fields generated within the substrate. It is remarkable, and perhaps unexpected, that this behavior depends on a high mobility of solid, crystalline islands that contain hundreds of thousands of Pb atoms. The discovery of self-assembly on a mesoscopic-length scale highlights the advantages of LEEM with its unique combination of high data rate and control of the sample environment.

Magnetic phenomena

The examples discussed so far illustrate what sort of surface features—atomic steps and domains of different atomic structure, for instance—produce strong contrast using conventional LEEM sources. A spin-polarized electron source extends LEEM’s sensitivity to include magnetic phenomena. When a spin-polarized beam of low-energy electrons illuminates a magnetized surface, the reflectivity of that surface can vary dramatically, depending on the relative orientation between the spin of the scattering electrons and the magnetization vector. Bauer’s group pioneered the use of this exchange-scattering asymmetry for imaging by developing the first spin-polarized LEEM, or SPLEEM. ⁸ In these microscopes, the spin-polarization orientation of the illuminating electron beam is adjustable. So, comparing a set of images that were recorded using different beam polarizations permits the determination of the local orientation of the magnetization vector in magnetic domains. Usually, one first uses opposite polarization alignments of the illuminating beam to record two LEEM images, and then forms a pixel-by-pixel difference image. This method enhances magnetic contrast, while simultaneously suppressing other LEEM contrast mechanisms. Since the instrument records pairs of images, the SPLEEM technique is marginally slower than conventional LEEM, but shares the advantages of an easily variable sample environment.

In situ sample processing is particularly interesting for magnetic materials, because magnetic properties depend very sensitively on microstructure. And SPLEEM offers the capability to track the evolution of both structural and magnetic properties simultaneously. For example, Thomas Duden summarized in a recent review ⁹ how magnetic thin films and multilayers can be grown in situ while one simultaneously records LEEM images of the sample morphology and SPLEEM images of the exchange-scattering asymmetry.

How magnetic microstructure responds to changing conditions of temperature or externally applied fields is an important question for both applied and basic physics. Using a SPLEEM at Lawrence Berkeley National Laboratory, Helmut Poppa and one of us (Schmid) recently demonstrated the feasibility of operating the instrument while simultaneously applying a controllable magnetic field to the sample. ¹⁰ A particularly interesting case of field-dependent magnetic domain patterns occurs in thin-film samples having a magnetic anisotropy perpendicular to the film. Figure 5(a) shows a typical SPLEEM image of the zero-field domain structure of a 2.5-monolayer-thick film of iron on Cu(100) at approximately 42°C. Under those conditions, the sample magnetization alternately pointed up and down in a regular pattern of dark and bright stripe domains. When a weak magnetic field was applied perpendicular to the film plane—parallel and antiparallel to magnetizations of the stripes—stripe domains whose magnetization aligned with the applied field simply grew in width at the expense of antialigned stripes. One might imagine that, as a function of increasing field, the aligned stripes would continue growing and eventually form one single large domain. However, that did not happen. In sufficiently large fields, typically at around 1 mT for the temperatures and film thickness in the experiment, the stripe domain patterns became unstable and we found transitions to magnetic droplet and inverse droplet phases.

It is no coincidence that images of the magnetic domain patterns in figure 5 look similar to the compositional domain patterns found in the Pb–Cu surface alloy of figure 4. The compositional and magnetic domain patterns are both stabilized by long-range, dipolar repulsive forces that compete with short-range ordering forces. The distance dependence of magnetic dipolar interactions has the same form as the distance dependence of the strain fields that cause repulsions in the surface alloy phase. In both systems, the competing forces drive the system toward minimum energy configurations consisting of the ordered, equilibrium domain patterns. In fact, general models for two-phase, two-dimensional systems with dipolar interactions explain exactly the type of pattern formation behavior observed in these two rather different systems. ¹¹ Such models predict stable striped domain patterns when the area fraction of the two domain types is near 1/2, while for relative domain area fractions around 1/3 and 2/3, droplet and inverse droplet phases, respectively, should be observed.

Future directions

Remaining challenges in LEEM involve improving the spatial resolution and pushing the time resolution from milliseconds to nanoseconds or better. With spatial resolution presently better than 50 Å, the LEEM built by Tromp at IBM ¹² approaches the limit imposed by the accelerating field of the objective lens. The SMART project at Bessy II (Berlin Electron Storage Ring Society for Synchrotron Radiation) uses an electron mirror to correct for the aberrations of the objective lens; researchers predict resolution approaching 10 Å or better with such corrections. ¹³ A second approach, investigated at the University of Mainz in Germany by Gerd Schönhense and his group, aims to achieve a similar result by using carefully tuned time-varying electrostatic fields. ¹⁴ Finally, researchers are exploring stroboscopic techniques to track very fast repetitive processes at surfaces. The Mainz group, for instance, is using a pulsed x-ray light source to image the switching of magnetic domains.

In addition to the applications from examples provided in this article, a number of other exciting studies are filling the literature: temperature- or impurity-driven faceting of initially uniform surfaces, structural phase formation and separation during the growth of organic semiconductor thin films, and phase separation during surface chemical reactions, among others. For a more extensive list, interested readers may consult the reviews in references and , the Web site at http://www.leem-user.com , and the proceedings of three international workshops held on LEEM and the related technique of photoemission electron microscopy ¹⁶ In PEEM, a UV or soft x-ray light source replaces the incident electron beam, and photoelectrons provide the chemical and surface potential contrast. ¹⁷ There are now working microscopes in several laboratories in the US, Europe, Hong Kong, and Japan, and the list of applications continues to grow. With the capability of probing surfaces with high spatial, temporal, spin, and energy resolution, LEEM, SPLEEM, and PEEM each promise to vastly expand our understanding of dynamics at solid surfaces.