prev FPGA 2002 next
Observations Summary
  1. track group segregation model for switch block design
    • simpler mathematical analysis
    • draw distinction between wire endpoints, midpoints
  2. try different endpoint models
    • disjoint, universal, Wilton: Wilton better, also most diverse
    • is Wilton better because it is diverse?
  3. different midpoint models
    • makes traditional switch blocks more diverse
  4. checkerboard layout
    • makes switch blocks even more diverse
  5. commutative switch blocks
    • order of turns unimportant, switch blocks easier to analyse
    • shifty similar to Wilton, but commutative
    • can use to optimize diversity (cdf-design)
  6. diversity measurements
    • cdf-design more diverse than shifty, disjoint
    • checkerboard increases diversity of shifty, disjoint
  7. routing results of diverse switch blocks
    • shifty consistenly uses less area
    • more diversity didn't help!
  8. in all cases, area savings small (<4%)
    • delay shows no trend, +/-4% typical
  9. is diversity worthwhile?
    • diversity: rerouting a net moves it to a new track
    • this creates conflicts with new nets, must reroute again
    • diversity is difficult for existing CAD tools to exploit