Abstract

Process, voltage and temperature variations are on the rise with technology scaling. Nano-scale technology requires huge design margins to ensure reliable operation. Worst case design margining consumes significant amount of circuits and systems resources. In-situ error detection or correction is an alternative method for cost effective variation tolerance. However, existing in-situ error detection and correction circuits are power and area hungry since they use speculative error management, which gives less power savings at higher error rates. This paper proposes an error resilience technique utilizing available slack in the design. The proposed method uses a clock stretching circuit to relax timing margins on selected critical paths that has sufficient consecutive stage slack. We also propose a power optimization method which reshapes the critical path logic proportionate to the consecutive stage slack. Experimental results show that the proposed method achieves the power and area savings of 40% and 8% respectively compared to the worst case design approach. When compared to the TIMBER error resilience approach, the proposed method saves power more than 74% and area more than 13% at design time. Document type: Article