Analysis of Temporal and Spatial Temperature Gradients for IC Reliability

One of the most common causes of IC failure is interconnect electromigration (EM), which exhibits a rate that is exponentially dependent on temperature. As a result, EM rate is one of the major determinants of the maximum tolerable operating temperature for an IC and of resulting cooling costs. Previous EM models have assumed a uniform, typically worst-case, temperature. This paper presents a model that accounts for temporal and spatial variations in temperature, and shows that accounting for these variations can dramatically improve interconnect lifetime prediction accuracy. We also show that the same modeling approach applies to temperature-related gate-oxide breakdown, another common cause of IC failure. We then propose that by modeling expected lifetime as a resource that is consumed over time at a rate dependent on temperature, substantial design margin can be reclaimed. For example, for a fixed target lifetime, intermittent higher operating temperatures and performance can be tolerated if compensated by lower temperatures at other times during the product’s lifetime. This approach offers higher overall performance and/or lower cooling costs than a standard design methodology that uses a worst-case temperature criterion for reliability analysis. This report supersedes TR CS-2003-21 and TR CS-2004-07. Keywords—reliability, electromigration, gate oxide breakdown, thermal management, gradient