The Role of Defects and Metal States at the Metal-Semiconductor Interface

Abstract

Previously proposed Schottky barrier models that attempt a microscopic description of the observed behavior fall basically into the categories of metal induced gap states (MIGS) models 1-5 and defect models.6-9 These two types of models are mutually exclusive, as the MIGS models generally ignore the influence of defects and the defect models, for their part, ignore the importance of screening by the metal. Historically the MIGS models preceded the others and were an outgrowth of Bardeen’s10 suggestion that a large density of interface states was responsible for the generally weak dependence of Schottky barrier heights on the metal (workfunction). Bardeen’s notion that these states derive from the semiconductor surface states (dangling bonds) was incorporated into the MIGS picture by Mele, et al. 3 and recently by Lefebvre, et al. 11, who considered the interaction of metallic states with the empty, cation-derived dangling bond states. The consequence of this interaction is the formation of broadened resonances which tail into the semiconductor bandgap. The rehybridization of the dangling bonds with the metallic states at the interface was not considered. In general the MIGS models encounter difficulties in the alignment of the metal-semiconductor bandstructures, a task which is model dependent.1,4 This problem was circumvented by aligning the MIGS with the charge neutrality level (CNL) of the semiconductor.5’12 The MIGS models are often more successful at estimating the index of interface behavior S\(\left( {S \equiv d\Phi _b^n/d{x_m}} \right)\) where \(\Phi _b^n\) represents the Schottky barrier height for an n-type semiconductor and Xm the Pauling electronegativity of the metal13) since this quantity is insensitive to the magnitude of \(\Phi {}_b^n\), and strongly dependent only on the inverse of the interface density of states Ds.14 Various approaches to calculate Ds have been reported, ranging from self-consistent pseudopotential methods,2 which gave the best agreement with experiment, scaling approaches based on the mean optical gap,3 which give poorer agreement, to one-dimensional model calculations,4 which agree even less with experimental results.