Epitaxially fabricated MS systems offer the best opportunity to understand the dependence of the electronic properties on the structure of a MS interface. The SBH at an epitaxial MS interface is frequently homogeneous and, therefore, can be directly correlated with the interface atomic structure. In addition, the atomic structure at single crystal MS interfaces can be determined by experimental means and then used to calculate the expected SBH. A comparison of the experimental and theoretical SBH allows a test of the validity of theories of the SB. Therefore, even though the formation mechanism of the SBH at ordinary MS interfaces is of the most technological interest, its deduction almost certainly has to come from the simpler, epitaxial, "model" MS systems.

Silicides are ordered metal-Si compound phases and have wide applications in Si integrated circuits as contacts and gate electrodes. Silicide thin films are usually formed by the deposition of metal on a Si substrate, followed by a high temperature treatment to induce solid-phase reactions. Because part of the substrate participates in the reaction of silicide thin films, the location of the final silicide-Si interface is usually well inside the original Si surface. This ensures intimate contact between the silicide and the Si, with transport behavior essentially independent of the initial conditions of the Si surface. A few silicides, most notably NiSi₂ and CoSi₂, have lattice structures similar to that of Si and close lattice matches with Si. As a result of these favorable conditions, these silicides can form epitaxially on silicon and the quality of these silicide-Si interfaces has been the best among MS interfaces ever fabricated. The SBH measured from single crystal silicide interfaces shows a remarkable dependence on the interface structure. The best known of these results is the dependence of the SBH on silicide orientation at the NiSi₂/Si(111) interfaces. High quality single crystals of NiSi₂ may be grown on Si(111) with either type A or type B orientation by a proper choice of template growth condition. These two orientations differ only by an azimuthal rotation: the type A silicide has the same orientation as the silicon substrate, and the type B silicide shares the surface normal <111> axis with Si, but is rotated 180o about this axis with respect to the Si. The atomic structure of both type A and type B NiSi₂/Si(111) interfaces have been studied by various experimental techniques and found to have the 7-fold structure. Intriguingly, type A and type B NiSi₂ have been found to have distinctively different SBH's, as illustrated in these current-voltage curves.

The atomic structure of the planar NiSi₂/Si(100) interface has been studied experimentally by HREM and shown to be consistent with having either 6-fold or 8-fold coordination. Theoretical calculations tend to favor the 8-fold model over the 6-fold model. Because extra streaks at the (0 1/2 1/2) related positions have been observed at flat NiSi₂/Si(100) interface by TED, it is clear that the atomic structure at this interface is much more complicated than these basic models. Its exact details notwithstanding, the flat NiSi₂/Si(100) interface obviously has an atomic structure which is entirely different from either of the two NiSi₂/Si(111) interfaces. One thus expects to find the flat NiSi₂/Si(100) interface with an SBH of its own. Indeed, uniform NiSi₂/Si(100) interfaces show a SBH which is much lower on n-type Si than either of the two NiSi₂/Si(111) interfaces, as shown in this figure. One notes that the low SBH of 0.4 eV measured from uniform NiSi₂ layers is also very different from the value of 0.6 - 0.7 eV usually observed for polycrystalline NiSi₂ on Si.

The overall picture of the formation of the SBH at single crystal MS interfaces, presented by experimental investigations, is identical to that described earlier from theoretical studies. In the few cases where theoretical calculations have enough accuracy for a meaningful comparison with experimental SBHs, most notably on the difference of SBHs between the type A and the type B NiSi₂/Si(111) interfaces, good numerical agreement was found. The message is unmistakable: the SBH depends on the atomic structure of epitaxial MS interfaces. This dependence is inconsistent with SBH models that ignore interaction between the metal and the semiconductor. It also disagrees with the basic notion of CNL-based models that the distribution of interface states is independent of the metal. It is in good agreement with the spirit of the bond polarization theory. However, one notes that the tight-binding picture of the bond polarization theory would not be able to explain the A-B dependence of the NiSi₂ SBH either. Before one rules out any SBH model based on the experimental data observed from epitaxial MS interfaces, however, one needs to appreciate the complexity of the SBH problem. Even after the entire interface electronic structures are known for a number of epitaxial MS interfaces from calculations, researchers are still at a loss as to the single most important feature or parameter that determines the SBH. There is none, other than the "charge density distribution of the entire interface". The formation of the SBH has always contained a bulk term and an interface term. Results described in this chapter give strong indication that the interface term depends on the interaction between the metal and the semiconductor. This should be regarded as the most important message from these studies. If we were to ask ourselves the question "given all the information one now knows about the electronic properties of epitaxial MS interfaces, what is the most convenient way to view the formation of the interface dipole (at any MS interface)?" The answer cannot be anything other than "bonding".