Breakdown data generation and in-die deconvolution methodology to address BEOL and MOL dielectric breakdown challenges

Abstract

Both middle-of-line (MOL) gate to contact spacer dielectric and back-end-of-line (BEOL) low-k dielectric breakdown data are commonly convoluted with multiple variables induced by process steps such as lithography, etch, chemical–mechanical polish (CMP), cleaning, and thin film deposition. The traditional method of stressing one device under test (DUT) per die or multiple DUTs per die, without careful data deconvolution, is incapable of addressing current complex MOL PC-CA and BEOL low-k dielectric breakdown modeling challenges. Generally, compound Weibull distributions in various unpredictable shapes induced by various die-to-die variations would be generated and such compound distributions could lead to a wrong low-percentile failure rate projection and a non-Poisson area scaling outcome. In this paper, a generation method plus an analytics procedure to analyze die-to-die variation is proposed to soundly evaluate both MOL and BEOL dielectric time-dependent-dielectric breakdown data. Relying on such die-to-die data generation and analytics, a diagnostic reliability concept is further proposed for comprehensive process diagnostics and more accurate reliability failure rate determination.