Composition and chemical width of ultrathin amorphous films at grain boundaries in silicon nitride

Abstract

Two different electron energy loss spectroscopy (EELS) quantitative analytical methods for obtaining complete compositions from interface regions are applied to ultrathin oxide-based amorphous grain boundary (GB) films of ∼ 1 nm thickness in high-purity HIPed Si3N4 ceramics. The first method, 1, is a quantification of the segregation excess at interfaces for all the elements, including the bulk constituents such as silicon and nitrogen; this yields a GB film composition of SiN0.49±1.4O1.02±0.42 when combined with the average film thickness from high resolution electron microscopy (HREM). The second method, II, is based on an EELS near-edge structure (ELNES) analysis of the Si–L2,3 edge of thin GB films which permits a subtraction procedure that yields a completeEELS spectrum, e.g., that also includes the O–K and N–K edges, explicitly for the GB film. From analysis of these spectra, the film composition is directly obtained as SiN0.63±0.19O1.44±0.33, close to the one obtained by the first method but with much better statistical quality. The improved quality results from the fewer assumptions made in method II; while in method I uniform thickness and illumination condition have to beassumed, and correction of such effects yields an extra systematic error. Method II is convenient as it does not depend on the film thickness detected by HREM, nor suffer from material lost by preferential thinning at the GB. In addition, a chemical width for these films can be deduced as 1.33 ± 0.25 nm, that depends on an estimation of film density based on its composition. Such a chemical width is in good agreement with the structural thickness determined by HREM, with a small difference that is probably due to the different way in which these techniques probe the GB film. The GB film compositions are both nonstoichiometric, but in an opposite sense, this discrepancy is probably due to different ways of treating the surface oxidation layers in both methods.