This paper reviews our recent investigations of compound semiconductors and heterovalent interfaces using the technique of aberration-corrected scanning transmission electron microscopy. Bright-field imaging of compound semiconductors with a collection angle that is comparable in size to the incident-beam convergence angle is demonstrated to provide better atomic-column visibility for lighter elements in comparison with the more traditional high-angle annular-dark-field approach. Several pairs of Group II–VI/Group III–V compound semiconductors with zincblende structure have been studied in detail. These combinations are all valence-mismatched (i.e., heterovalent), and include CdTe/InSb (Δa/a ≤ 0.05%), ZnTe/InP (Δa/a = 3.8%), and ZnTe/GaAs (Δa/a = 7.4%). CdTe/InSb (001) interfaces are observed to be defect-free with a slight lattice contraction at the interface plane. For interfaces with larger lattice-parameter mismatch, the primary interfacial defects are identified as Lomer edge dislocations and perfect 60° dislocations. However, the atomic structure of the dislocation cores has not yet been unambiguously determined.

The integration of dissimilar materials is highly desirable for many different types of device applications but often challenging to achieve in practice. The unrivalled imaging capabilities of the aberration-corrected electron microscope enable enhanced insights to be gained into the atomic arrangements across heterostructured interfaces. This paper provides an overview of our recent observations of oxide-semiconductor heterostructures using aberration-corrected high-angle annular-dark-field and large-angle bright-field imaging modes. The perovskite oxides studied include strontium titanate, barium titanate, and strontium hafnate, which were grown on Si(001) and/or Ge(001) substrates using the techniques of molecular-beam epitaxy or atomic-layer deposition. The oxide layers displayed excellent crystallinity and sharp, abrupt interfaces were observed with no sign of any amorphous interfacial layers. The Ge(001) substrate surfaces invariably showed both 1× and 2× periodicity consistent with preservation of the 2 × 1 surface reconstruction following oxide growth. Overall, the results augur well for the future development of functional oxide-based devices integrated on semiconductor substrates.

The mean inner potential (MIP) and inelastic mean free path (IMFP) of undoped ZnTe are determined using a combination of off-axis electron holography and convergent beam electron diffraction. The ZnTe MIP is measured to be 13.7±0.6 V, agreeing with previously reported simulations, and the IMFP at 200 keV is determined to be 46±2 nm for a collection angle of 0.75 mrad. Dynamical effects affecting holographic phase imaging as a function of incident beam direction for several common semiconductors are systematically studied and compared using Bloch wave simulations. These simulation results emphasize the need for careful choice of specimen orientation when carrying out quantitative electron holography studies in order to avoid erroneous phase measurements.

The mean-free-paths for inelastic scattering of high-energy electrons (200 keV) for AlAs and GaAs have been determined based on a comparison of thicknesses as measured by electron holography and convergent-beam electron diffraction. The measured values are 77 ± 4 nm and 67 ± 4 nm for AlAs and GaAs, respectively. Using these values, the mean inner potentials of AlAs and GaAs were then determined, from a total of 15 separate experimental measurements, to be 12.1 ± 0.7 V and 14.0 ± 0.6 V, respectively. These latter measurements show good agreement with recent theoretical calculations within experimental error.

Wedge polishing was used to prepare one-dimensional Si n-p junction and Si p-channel metal-oxide-silicon field effect transistor (pMOSFET) samples for precise and quantitative electrostatic potential analysis using off-axis electron holography. To avoid artifacts associated with ion milling, cloth polishing with 0.02-μm colloidal silica suspension was used for final thinning. Uniform thickness and no significant charging were observed by electron holography analysis for samples prepared entirely by this method. The effect of sample thickness was investigated and the minimum thickness for reliable results was found to be ∼160 nm. Below this thickness, measured phase changes were smaller than expected. For the pMOSFET sample, quantitative analysis of two-dimensional electrostatic potential distribution showed that the metallurgical gate length (separation between two extension junctions) was ∼54 nm, whereas the actual gate length was measured to be ∼70 nm by conventional transmission electron microscopy. Thus, source and drain junction encroachment under the gate was 16 nm.

Extended abstract of a paper presented at Microscopy and Microanalysis 2004 in Savannah, Georgia, USA, August 1–5, 2004.

Extended abstract of a paper presented at Microscopy and Microanalysis 2004 in Savannah, Georgia, USA, August 1–5, 2004.

It has long been realized that imaging with hollow cone illumination (HCI) should, in theory, improve the directly interpretable resolution of TEM by as much as 100% (albeit at the expense of contrast). The principle of HCI was first proposed by Scherzer in 1949 and then reinvented by Hanssen and Trepte in 1971. As opposed to axial illumination, HCI effectively eliminates zeroes and reversals of the transfer function providing direct interpretability of the resulting images. In addition to the substantial resolution enhancement, HCI should reduce significantly the phase-contrast noise inherent in axial HRTEM images. However, there are experimental obstacles for high resolution HCI which make its practical application very difficult to implement. To our knowledge, all observations using HCI so far have not shown all of the expected improvement predicted theoretically. This is believed to be due to the fact that accurate coma-free alignment is required to substantially improve the resolution.

This paper extends the concepts that were developed to explain the structural rearrangement of the wurtzite AlN lattice due to incorporation of small amounts of oxygen, and to directly use them to assist in understanding the polytypoid structures. Conventional and high-resolution transmission electron microscopy, specific electron diffraction experiments, and atomistic computer simulations have been used to investigate the structural nature of the polytypoids. The experimental observations provide compelling evidence that polytypoid structures are not arrays of stacking faults, but are rather arrays of inversion domain boundaries (IDB's). A new model for the polytypoid structure is proposed with the basic repeat structural unit consisting of a planar IDB-P and a corrugated IDB. This model shares common structural elements with the model proposed by Thompson, even though in his model the polytypoids were described as consisting of stacking faults. Small additions (≃ 1000 ppm) of silicon were observed to have a dramatic effect on the polytypoid structure. First, it appears that the addition of Si causes the creation of a new variant of the planar IDB (termed IDB-P'), different from the IDB-P defect observed in the AlN-Al2O3 polytypoids; second, the addition of Si influences the structure of the corrugated IDB, such that it appears to become planar.

The model proposed by Harris et al. [J. Mater. Res. 5, 1763–1773 (1990)], describing planar inversion domain boundaries in aluminum nitride, consists of a basal plane of aluminum atoms octahedrally coordinated with respect to oxygen, and with a translation of R = 1/3〈1011〉. This thin sandwich is inserted onto the basal plane of the wurtzite structure of aluminum nitride. This model does not take into consideration any interfacial relaxation phenomena, and is arguably electrically unstable. Therefore, this paper presents a refinement of the model of Harris et al., by incorporating the structural relaxations arising from modifications in local chemistry. The interfacial structure was investigated through the use of conventional transmission electron microscopy, convergent electron diffraction, high resolution transmission electron microscopy, analytical electron microscopy, and atomistic computer simulations. The refined planar inversion domain boundary model is closely based on the original model of Harris et al.; however, the local chemistry is changed, with every fourth oxygen being replaced by a nitrogen. Atomistic computer simulation of these defects, using a classical Born model of ionic solids, verified the stability of these defects as arising from the adjustment in the local chemistry. The resulting structural relaxations take the form of a 0.3 mrad twist parallel to the interface, a contraction of the basal planes adjacent to the planar inversion domain boundary, and an expansion of the c-axis component of the displacement vector; the new displacement vector across the interface is R = 1.3〈1010〉 + ∊〈0001〉, where ∊meas = 0.387 and ∊calc = 0.394.

Three distinct morphologies of curved (curved, facetted, and corrugated) inversion domain boundaries (IDB's), observed in aluminum nitride, have been investigated using conventional transmission electron microscopy, convergent beam electron diffraction, high-resolution transmission electron microscopy, analytical electron microscopy, and atomistic computer simulations. The interfacial structure and chemistry of the curved and facetted defects have been studied, and based upon the experimental evidence, a single model has been proposed for the curved IDB which is consistent with all three observed morphologies. The interface model comprises a continuous nitrogen sublattice, with the aluminum sublattice being displaced across a {1011} plane, and having a displacement vector R = 0.23〈0001〉. This displacement translates the aluminum sublattice from upwardly pointing to downwardly pointing tetrahedral sites, or vice versa, in the wurtzite structure. The measured value of the displacement vector is between 0.05〈0001〉 and 0.43〈0001〉; the variation is believed to be due to local changes in chemistry. This is supported by atomistic calculations which indicate that the interface is most stable when both aluminum vacancies and oxygen ions are present at the interface, and that the interface energy is independent of displacement vector in the range of 0.05〈0001〉 to 0.35〈0001〉. The curved IDB's form as a result of nonstoichiometry within the crystal. The choice of curved IDB morphology is believed to be controlled by local changes in chemistry, nonstoichiometry at the interface, and proximity to other planar IDB's (the last reason is explained in Part III). A number of possible formation mechanisms are discussed for both planar and curved IDB's. The Burgers vector for the dislocation present at the intersection of the planar and curved IDB's was determined to be b = 1/3〈1010〉 + t〈0001〉, where tmeas = 0.157 and tcalc = 0.164.