• DocumentCode
    3586229
  • Title

    Using Formal Verification of Parameterized Systems in RAW Hazard Analysis in Microprocessors

  • Author

    Charvat, Luka ; Smrcka, Ales ; Vojnar, Tomas

  • Author_Institution
    IT4Innovations Centre of Excellence, Brno Univ. of Technol., Brno, Czech Republic
  • fYear
    2014
  • Firstpage
    83
  • Lastpage
    89
  • Abstract
    Implementation of a pipeline-based execution of instructions in purpose-specific microprocessors is an error prone task, which implies a need of proper verification of the resulting design. Various techniques were proposed for this purpose, but they usually require a significant manual intervention of the developers. In this work, we propose a novel, highly automated approach for discovering RAW hazards in in-order pipelined instruction execution. Our approach combines static analysis of data paths to detect anomalies and possible hazards, followed by a transformation of detected problematic paths to a parameterised system (PS), and a subsequent formal verification to check the possibility of unhandled hazards using techniques for formal verification of PSs. We have implemented our approach and successfully applied it on multiple non-trivial microprocessors.
  • Keywords
    instruction sets; microprocessor chips; pipeline processing; program diagnostics; program verification; PS; RAW hazard analysis; anomalies detection; data paths; detected problematic paths transformation; formal verification; multiple nontrivial microprocessors; parameterised system; parameterized systems; pipeline-based instruction execution; purpose-specific microprocessors; read-after-write hazards; static analysis; Hazards; Microprocessors; Pipelines; Ports (Computers); Program processors; Radiation detectors; Registers; formal verification; microprocessor analysis; parameterised system; pipeline hazard; pipelined execution;
  • fLanguage
    English
  • Publisher
    ieee
  • Conference_Titel
    Microprocessor Test and Verification Workshop (MTV), 2014 15th International
  • ISSN
    1550-4093
  • Type

    conf

  • DOI
    10.1109/MTV.2014.21
  • Filename
    7087240