Applications of Multi-Variate Statistical Analysis of Spectrum Images to Microelectronic Devices

Abstract

It is not uncommon for an electrical failure in a microelectronic device to be traced back to an individual resistive contact, such as a W contact lined by TiN to the underlying metal silicide [1]. Identifying the root cause of the defective cell then requires the capability of extracting interfacial chemistry and microstructure with near atomic resolution, which is achieved by recording EELS and EDX data either along 1-d lines or from 2-d arrays. EELS data are characterized by intrinsically low signal-to-background ratios and plural inelastic-scattering effects, which presents a challenge to the reliability of the methods used to extract quantitative information from a spectrum image. Commercially available software packages by Emispec and Gatan currently provide routines for power-law or polynomial background fits and allow the net counts under peaks or edges to be determined from each pixel. in both cases each spectrum in the series is processed in the same manner.