Modeling and Analyzing the Implementation of Latency-Insensitive Protocols Using the Polychrony Framework

  • Authors:
  • Bin Xue;Sandeep K. Shukla

  • Affiliations:
  • FERMAT Lab, Virginia Polytechnic Institute and State University, Blacksburg, VA, USA;FERMAT Lab, Virginia Polytechnic Institute and State University, Blacksburg, VA, USA

  • Venue:
  • Electronic Notes in Theoretical Computer Science (ENTCS)
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

As Globally Asynchronous and Locally Synchronous (GALS) based System-on-chip (SoC) are gaining importance, a special case of GALS when the global clocking is preserved, but the interconnect delays of multiple clock cycles are to be tolerated has also been proposed, and used. In some cases, such designs, known as Latency-Insensitive Protocol (LIP) based SoC integration are also general enough to work when the global clocking is not present. In either case, the protocols are complex, and many optimized implementations of such protocols need verification that indeed they work correctly with respect to the specification of the system. Usually the specification of the system is fully globally clocked with negligible interconnect delays, so that the specification can be first implemented as synchronous design with traditional tools. GALS or LIP refinements are then applied to tolerate the multi-cycle interconnect delays, or fully asynchronous interconnect communication, as the case may be. Verifying that such refinements are correctness preserving, researchers have used model checking in the past. In this paper, we present static analysis based framework for such verification. Our framework makes use of the Polychrony framework and associated semantic analysis techniques, in the form of endo-isochrony. We show a number of LIP protocols to preserve the correctness with respect to their fully synchronous specifications using our framework. We believe, designers can verify LIP implementations with clever optimizations using our framework much more readily than when using model checking.