SHF: Small: Physics-Based Electromigration Assessment and Validation For Reliability-Aware Design and Management

University Of California-Riverside, Riverside CA

Investigators

Abstract

Long-term reliability has become a significant challenge for design of integrated circuit (IC) chips used in today's computers and smart phones. Among many effects, electromigration-induced reliability has become the dominant limiting factor due to aggressive transistor scaling and increasing power density. In this project, the PI proposes to address IC reliability problems by providing more accurate electromigration models and assessment techniques at both circuit level and system level so that electromigration induced reliability and aging effects can be fully accounted, leveraged and optimized at both the design stage and run time stage to improve chip reliability and their lifetime at minimum design costs. Furthermore, through the collaborations with the industry partners, the proposed techniques are expected to impact the design and tool development community, thus increasing design productivity. This project also enables the institution to hire students from underrepresented groups, thus enhancing the diversity of science and technology workforce. The project seeks to develop new physics-based electromigration models and full-chip assessment techniques, to perform model validation for accurate yet efficient electromigration verification at the design stage, and to enforce electromigration-aware reliability management at run time for nanometer IC chips. First, the research will develop new physics-accurate electromigration models to better predict mean time to failure for multi-branch interconnect trees, which are commonly seen in IC chips, for both void nucleation and void growth phases. Second, the project will develop electromigration models that can accommodate time-varying temperature and current densities, which reflect a more realistic chip working condition due to time-varying loads for both single wire and multi-branch interconnect trees. On top of this, the project team will develop resource-based electromigration models, which are more amenable to the system-level run-time reliability optimization and management.