On the Resilience of RTL NN Accelerators: Fault Characterization and Mitigation

Abstract: Machine Learning (ML) is making a strong resurgence in tune with the massive generation of unstructured data which in turn requires massive computational resources. Due to the inherently compute- and power-intensive structure of Neural Networks (NNs), hardware accelerators emerge as a promising solution. However, with technology node scaling below 10nm, hardware accelerators become more susceptible to faults, which in turn can impact the NN accuracy. In this paper, we study the resilience aspects of Register-Transfer Level (RTL) model of NN accelerators, in particular, fault characterization and mitigation. By following a High-Level Synthesis (HLS) approach, first, we characterize the vulnerability of various components of RTL NN. We observed that the severity of faults depends on both i) application-level specifications, i.e., NN data (inputs, weights, or intermediate), NN layers, and NN activation functions, and ii) architectural-level specifications, i.e., data representation model and the parallelism degree of the underlying accelerator. Second, motivated by characterization results, we present a low-overhead fault mitigation technique that can efficiently correct bit flips, by 47.3% better than state-of-the-art methods.

• The document serves as a demonstrative template and guide for using the IEEEtran.cls LaTeX class file to structure articles for IEEE Computer Society journals.
• It aims to streamline manuscript preparation by providing a pre-structured framework that helps authors conform to complex IEEE formatting requirements more efficiently.
• The guide details essential elements including document class options, handling peer review formats, and incorporating sections like appendices, acknowledgments, and biographies.

Overview of "Bare Advanced Demo of IEEEtran.cls for IEEE Computer Society Journals"

This document serves as a demonstrative template rather than a traditional research paper, focusing on the usage of the IEEEtran.cls class file for authoring journal articles specifically within IEEE Computer Society publications using \LaTeX. Authored by Michael Shell, alongside John Doe and Jane Doe, the paper functions primarily as a guide to facilitate the initial stages of drafting and structuring documents according to IEEE standards.

Structure and Focus

The document does not explore original research; instead, it systematically introduces key aspects of formatting and typographic standards prescribed by IEEE. The purpose is to provide a preliminary architecture for authors, streamlining the creation of articles with appropriate style and structure. This includes instructions on organizing sections such as the title, authorship details, abstract, and bibliography.

Implications for Authors

While this demonstrative file might not contain empirical results or theoretical advancements, its practical utility is significant for new and seasoned authors in the IEEE community. By offering a solid foundation, it assists in reducing the time and effort involved in conforming to complex formatting requirements. The clarity and user-friendliness of this template may lead to more efficient manuscript preparation and submission processes, enhancing the overall quality and consistency of articles within IEEE journals.

Detail Orientation

The paper highlights essential elements such as the utilization of document class options to switch between draft and journal modes, handling of peer review-specific formatting, and guidelines for incorporating various structural components like appendices, acknowledgments, and author biographies. The template also encapsulates intra-document navigation using labels and sections pivotal for comprehensive document architecture.

Future Directions

The development of such templates could benefit from feedback and iterative adjustments informed by authors’ experiences. Exploring automated mechanisms or interactive tutorials embedded within templates might further assist users in intuitively navigating intricate formatting nuances. As \LaTeX\ evolves and integrates with contemporary document preparation systems, refining such templates remains a practical focus for the IEEE community to ensure accessibility and usability across diverse technological skill levels.

Conclusion

Although this document is primarily instructional, its impact lies in its facilitation of scholarly communication through standardized documentation. By simplifying the task of manuscript preparation, it allows researchers to focus more intently on the content of their work rather than on the intricacies of formatting requirements. It exemplifies the pivotal role of well-structured templates in the publication ecosystem, underscoring the ongoing need for resources that bridge the gap between effective technical writing and stringent publication standards.