Heteroepitaxy is an important area of materials research because of its applicability to a variety of technological applications. One area presently attracting a great deal of research is heteroepitaxy of electronic and optoelectronic materials. Here the word heteroepitaxy is emphasized to distinguish it from the much more common epitaxial growth of essentially the same material where only variations in the concentration of dopants is allowed. When materials scientists and engineers from the electronics industry refer to heteroepitaxy, they mean the oriented-crystal growth of some interesting (at least, to them!) material on a single-crystal substrate of some other material. By using heteroepitaxy, it is possible to tailor the electronic or optical properties in a way that is just not possible with doped, epitaxial growth of the same material. There are some rare examples of pairs of materials that have the same crystal structure, are very closely lattice-matched, and are dissimilar enough in their optical and/or electronic properties to make heteroepitaxy interesting. One such couple, AlxGa1−xAs/GaAs (lattice mismatch of 0.1% when x = 1), has been extensively exploited over the last 30 years.

The growth of semiconductor heteroepitaxial layers is assuming ever greater importance due to the demands of modern electronic-device fabrication. Furthermore although low-strain heterosystems such as AlGaAs/GaAs remain the basis of many device structures, there is an increasing trend also to use more highly strained combinations such as InGaAs/GaAs and SiGe/Si. However the growth of the latter epitaxial systems must be approached with great care in order to achieve the optimum layer structural quality. Of course for any given alloy-layer composition, interfacial misfit defects in general will be introduced when the layer thickness exceeds a critical value, as originally described by Frank and van der Merwe, and Jesser and Matthews. (See the article by F.K. LeGoues in this issue for more general considerations of misfit-defect introduction.) In addition the morphology of such strained layers must be given very careful attention. It is the purpose of this article to examine our current understanding in this latter area.

When any epitaxial layer is grown, initially it might be expected that a flat surface would result under ideal growth conditions when internal defects have been eliminated. This would be expected to minimize the surface step density and hence the surface energy. However nature has a way of confounding our simplest expectations and while for homoepitaxial layers in the absence of kinetic effects this expectation virtually can be realized, the presence of strain in a heteroepitaxial system can severely affect the observed layer morphology.

Strain relaxation in lattice-mismatched, heteroepitaxial systems is one of the classic problems in materials physics, which has gained new urgency with the increased applications of strained layers in microelectronic systems. In general both the structure and the integrity of the thin films are strongly influenced by strain. For instance it has long been known that under strain, the growth changes from an initial layer-by-layer growth mode to one with three-dimensional islanding. In the seminal works of van der Merwe, and Matthews and Blakeslee, this change in growth mode is explained in terms of the introduction of strain-relieving misfit dislocations, which appear when the film has reached some critical thickness. Recently it has become clear that this change in growth mode can take place even without the introduction of misfit dislocations. Such dislocation-free coherent islanding, or “roughening,” has been observed experimentally both in Ge/Si and in InGaAs/GaAs systems. Furthermore recent experiments show that in Ge/Si(100) systems, the thin films display a curious asymmetry with respect to the sign of the strain: Films under compression roughen by forming coherent islands while those under tension remain relatively smooth. A possible mechanism behind this strain-induced type of roughening is the subject of this article.

The strain-induced transition of a planar film to a three-dimensional island morphology is presently a significant issue in the growth of semiconductor thin films. Strain-induced roughening can be problematic in the fabrication of coherently strained device structures where it is important to understand the early stages of the transition to avoid or suppress three-dimensional (3D) growth. On the other hand, the strain-driven transition is beneficial for the self-assembly of quantum dots where it is necessary to control the size distribution and self-organizing behavior of the islands. In both cases, it is clearly important to identify and understand the kinetic pathways to island formation. From a more basic perspective, the strain-induced transition of epitaxial films allows us to study in detail the interplay between elastic stresses and surface energy in a carefully controlled experimental environment. One would therefore hope that the lessons learned from model semiconductor systems will be of relevance to understanding related phenomena in other areas of materials science and physical metallurgy.

Ihe strain-induced two-dimensional (2D)-to-3D transition in the Si-Ge system is manifested by a rich variety of observed surface morphologies. In the case of pure Ge on Si(OO1), the 4% misfit strain induces the formation of so-called hut clusters with curious elongated shapes. Such islands form almost immediately after the deposition of a wetting layer. In the case of lower misfit alloys, a more gentle ripple morphology can result that develops far from the interface. A general trend in all of the experiments is the decreasing size of typical morphological features with increasing misfit stress. In this article, largely guided by our experimental results, we adopt a nucleation and growth description of the 2D-to-3D transition. This approach appears particularly well-suited to explaining the wide spectrum of morphological development present in the Si-Ge system.

Recently much interest has been devoted to Si-based heteroepitaxy, and in particular, to the SiGe/Si system. This is mostly for economical reasons: Si-based technology is much more advanced, is widely available, and is cheaper than GaAs-based technology. SiGe opens the door to the exciting (and lucrative) area of Si-based high-performance devices, although optical applications are still limited to GaAs-based technology. Strained SiGe layers form the base of heterojunction bipolar transistors (HBTs), which are currently used in commercial high-speed analogue applications. They promise to be low-cost compared to their GaAs counterparts and give comparable performance in the 2-20-GHz regime. More recently we have started to investigate the use of relaxed SiGe layers, which opens the door to a wider range of application and to the use of SiGe in complementary metal oxide semiconductor (CMOS) devices, which comprise strained Si and SiGe layers. Some recent successes include record-breaking low-temperature electron mobility in modulation-doped layers where the mobility was found to be up to 50 times better than standard Si-based metal-oxide-semiconductor field-effect transistors (MOSFETs). Even more recently, SiGe-based p-type MOSFETS were built with oscillation frequency of up to 50 GHz, which is a new record, in any p-type material for the same design rule.

The advantage that epitaxy offers the electronics and optoelectronics industries is that it allows the possibility of producing precisely controlled layers of very high crystal quality. Heteroepitaxy of different materials offers the promise of tailoring device layers in clever ways that nature did not intend. However unlike fruit juices, nature has made it difficult to epitaxially combine different materials. As the preceding articles have clearly pointed out, it is very difficult to obtain smooth epitaxial layers that are free both of defects and strain when there is a lattice mismatch between the layers and their substrates.

As already discussed in this issue, a uniform network of dislocations at the interface between a flat, uniform epitaxial layer and its substrate can completely relieve strain in the majority of the epitaxial layer. This would be a satisfactory situation for many devices so long as the active region of the device could be kept away from the interface. The problem is how to introduce the dislocations in an appropriate way. When an epitaxial layer has a larger lattice parameter than the underlying substrate, a misfit dislocation running along the interface represents a plane of atoms that has been removed from the epitaxial layer. (One would insert a plane of atoms if the epitaxial lattice parameter was smaller. For simplicity however we will continue to assume that the epitaxial layer has a larger lattice parameter.) It is not possible for a whole half plane of atoms, bounded by the dislocation at the interface and the substrate edges along the two sides, to be removed at once. The boundary between where the extra plane of atoms has been removed and where the epitaxial layer has not relaxed yet will represent a threading dislocation. This threading dislocation would continue to move as the size of the misfit dislocation along the interface grows. Ideally it moves all the way out to the substrate edge and vanishes there while the misfit dislocation along the interface would end up extending from one side of the substrate to the other. However other dislocations and other kinds of defects can effectively pin the threading dislocation resulting in an epitaxial layer with many threading dislocations. Unfortunately these threading dislocations are generally detrimental to most kinds of devices. It is precisely this high density of threading dislocations that limits applications of many heteroepitaxial layers.

Recent technological and materials advances in semiconductors have brought about the possibility of producing heterostructures within which carriers are confined to an ultrasmall region of space (a few thousand atoms) by a potential barrier. When the dimensions of the confining potential are smaller than the electron wavelength (a few tens of nanometers), the semiconductor electronic and optical properties are drastically altered. In these so-called quantum structures, carrier energy levels are quantized and their energy depends on the confining-potential dimensions and magnitude.

Some of these quantum structures have already found technological applications. For example the quantum-well (QW) semiconductor laser is part of every CD player. It is also widely used as the light source for intercontinental optical communications. The carrier confining potential in this case is provided by two wider bandgap semiconductor layers sandwiching a thin (3–20 nm) smaller bandgap semiconductor film. The carriers have two degrees of freedom within the QW. The QWs are grown by epitaxial deposition on a crystalline substrate. The substrate may or may not be lattice-matched with the epitaxial film. In some instances, a small lattice mismatch may be required to obtain the desired band-gap value for the QW material. These are the so-called pseudomorphically strained QW structures and devices.