ElastIC: An Adaptive Self-Healing Architecture for Unpredictable Silicon

  • Authors:
  • Dennis Sylvester;David Blaauw;Eric Karl

  • Affiliations:
  • University of Michigan, Ann Arbor;University of Michigan, Ann Arbor;University of Michigan, Ann Arbor

  • Venue:
  • IEEE Design & Test
  • Year:
  • 2006

Quantified Score

Hi-index 0.00

Visualization

Abstract

With continued technology scaling, silicon is becoming increasingly less predictable. Recent years have brought an acceleration of wear-out mechanisms, such as oxide breakdown and NBTI, which occur over a part's lifetime. Manufacturing device failure rates will increase significantly with decreases in device sizes, possibly reaching one in thousands or even hundreds of devices. Process variations will increase significantly in future technologies because fundamental laws of physics drive certain parametric variations, such as random dopant fluctuation (RDF) and line edge roughness, making their increased contribution to variability almost inevitable. The combination of wear-out mechanisms, RDF, and line edge roughness leads to an unpredictable silicon fabric that poses a major obstacle to reliable computing in future technologies. The authors present a broad vision of a new cohesive architecture, ElastIC, which can provide a pathway to successful design in unpredictable silicon. ElastIC is based on aggressive runtime self-diagnosis, adaptivity, and self-healing. It incorporates several novel concepts in these areas and brings together research efforts from the device, circuit, testing, and microarchitecture domains. Architectures like ElastIC will become vital in extremely scaled CMOS technologies.