Solving hard instances of FPGA routing with a congestion-optimal restrained-norm path search space

  • Authors:
  • Keith So

  • Affiliations:
  • University of New South Wales, Sydney, Australia

  • Venue:
  • Proceedings of the 2007 international symposium on Physical design
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

The negotiated congestion mechanism forms the basis of most published FPGA routers today, with many routers projecting congestion and any other requirements onto a scalar search space to evaluate candidate paths. In this paper, we study the numerical stability of these scalar projections as the number of iterations increase. We show that in these scalar search spaces the norm of path costs increase exponentially with the number of iterations, leading to floating-point absorption and representation problems in computer arithmetic. We propose a novel two-component totally-ordered monoid space for path candidate evaluation, that guarantees the A* search finds a path with minimum congestion cost, and has linear norm growth with respect to the number of iterations. We demonstrate the efficacy of our new algorithm by testing on hard open routing problems in the FPGA Place and Route Challenge. The router successfully found 18 new routing solutions to instances that were previously unroutable, reducing the lowest track count to route 20 standard FPGA routing benchmarks by 10.2%.