Output remapping technique for critical paths soft-error rate reduction

Qian Ding, Yu Wang, Hui Wang, Rong Luo, Huazhong YangIET Comput. Digit. Tech 2010 · DOI

TL;DR

This paper introduces an "output remapping" technique to reduce the soft-error rate (SER) in critical paths of combinational logic circuits. By strategically replacing certain gates with others that have longer propagation delays, the method effectively filters out glitches caused by particle strikes, achieving significant SER reduction (59.2% to 89.8%) with minimal delay or area/power overhead.

What Problem Does This Paper Address?

As semiconductor technology scales down, circuits become more susceptible to soft errors – transient failures caused by particle strikes that flip bits. While memory elements have traditionally been the primary concern, combinational logic is increasingly vulnerable due to reduced node capacitances and lower supply voltages. Existing methods to mitigate soft errors in combinational circuits often introduce substantial penalties in terms of delay, area, or power consumption, making them impractical for critical paths where performance is paramount. This paper aims to address this gap by proposing a technique that effectively reduces SER in critical paths without significant performance degradation.

Methods

The core of the proposed solution is an "output remapping" technique. This method leverages the understanding that glitches need to propagate to the output to cause a soft error. The authors identified that critical output nodes are particularly important for SER reduction. Their approach involves:

  • Identifying Critical Outputs: Focusing on outputs whose delay is close to the overall circuit delay.
  • Gate Replacement: Replacing existing gates driving these critical outputs with more complex logic gates that have longer propagation delays. This increased delay acts as a filter, attenuating or eliminating glitches that are too short to propagate through the longer delay path.
  • Trade-off Analysis: The technique is based on a trade-off between propagation delay and SER. By increasing the delay of the last stage in a critical path, the authors exploit the electrical masking effect to reduce the likelihood of a soft error.
  • SER Analysis Tool: The authors developed a C++-based SER analysis tool to evaluate their technique. This tool models glitch generation and propagation, using HSPICE simulations to characterize standard cell libraries.

Datasets

The paper evaluates its proposed technique using benchmark circuits from the ISCAS85 suite. These are standard, publicly available circuit designs used for testing and comparing digital circuit design methodologies. The authors specifically mention using PTM 45 nm models and the Nangate Open Cell Library for their simulations.

Key Insights and Ideas

  • Critical Output Importance: The research highlights that output nodes, especially those on critical paths, are the most crucial for SER reduction. Focusing mitigation efforts here yields the greatest impact.
  • Delay as a Glitch Filter: Increasing the propagation delay of gates, particularly at the output stage of critical paths, is an effective way to filter out short-duration glitches caused by particle strikes.
  • Technology Scaling Benefits: The authors observed that the width of particle-strike-induced glitches scales down with technology scaling. This suggests that their output remapping technique will continue to be effective as technology advances.
  • Minimal Overhead: The proposed method achieves significant SER reduction with negligible delay, area, and power overhead, making it a practical solution for critical paths.

Key Results

The output remapping technique demonstrated significant SER reduction across various ISCAS85 benchmark circuits, ranging from 59.2% to 89.8%. This is comparable to or better than existing methods, such as those focusing on electrical masking effects, which achieved reductions from 53.1% to 95.2%.

Crucially, the delay penalty introduced by the output remapping technique was minimal, with most circuits experiencing no delay penalty. Only one circuit (C432) showed a 4% delay penalty, attributed to its small size and the prevalence of critical paths. In contrast, a comparative method [9] had an average delay penalty of 4.9% to 7.2%. The area and power overhead were also deemed negligible due to the focused modification of only a small percentage of gates (critical output nodes).

Limitations and Future Work

  • Delay Penalty in Specific Cases: While generally minimal, a delay penalty can occur in very small circuits or those with a high density of critical paths. The authors suggest that further optimization might be needed for such cases.
  • Technology Scaling: The paper posits that the technique scales well with technology, but this is an expectation based on their analysis of glitch width scaling. Further experimental validation across a wider range of technology nodes would be beneficial.
  • Complex Glitch Interactions: The analysis primarily focuses on individual glitches. Future work could explore the impact of multiple simultaneous glitches and their interactions.
  • Optimization for Specific Architectures: While effective for general combinational logic, tailoring the output remapping strategy for specific circuit architectures or applications could yield even greater benefits.

Summary updated

BibTeX
@article{ding2010output,
  author = {Qian Ding and Yu Wang and Hui Wang and Rong Luo and Huazhong Yang},
  title = {{Output remapping technique for critical paths soft-error rate reduction}},
  journal = {IET Comput. Digit. Tech.},
  year = {2010},
  doi = {10.1049/iet-cdt.2009.0038},
}
Figure