StanfordAHA/garnet

Tapeout: ICOVL halos and blockages (SPACE1703)

Opened this issue · 0 comments

PROBLEM:

  • halos don't seem to block routing as well as blockages
  • weird extra-big halos for vias only

Our floorplan script floorplan.tcl puts route blockages on top of each ICOVL cell, and then extends an even larger "halo" which apparently also is supposed to block routes (why? what's the difference between a blockage and a halo?). Unfortunately, the router does not seem to respect the halos as much as actual blockages, and some wires end up inside the halo region. And yes, the wires have DRC errors (in innovus anyway).

My solution is to make the blockage extend over the entire blocked-out region, with no halo. This seems to make the router behave, and the stray wires no longer appear.

A second problem: for some reason, the floorplan extends halos around ICOVL cells an extra 2 microns in each direction for VIAs only. I can't find a reason for this, and it definitely causes routing issues. So I eliminated the extra margin.

My questions:

  • Is this fix okay?
  • What's the difference between a blockage and a halo anyway?

The answer, I expect, will involve running Calibre DRC on the final result, which I hope to do soon. Meanwhile, if anyone else has insight into this problem, please chime in.