Skip to content

asap7: tech LEF still uses DATABASE MICRONS 1000, so the M4/M5 EOLKEEPOUT side extension rounds onto the adjacent track and stops working (fixed in asap7sc7p5t_28 #10; ORFS #4663 was closed) #4741

Open

Description

@Arcadia-1

Description

flow/platforms/asap7/lef/asap7_tech_1x_201209.lef on master still declares DATABASE MICRONS 1000. With that unit OpenROAD rounds several quarter-nanometre rule values to whole nanometres, and some of them stop doing what they were written for. The library repository fixed this by moving to DATABASE MICRONS 4000 (The-OpenROAD-Project/asap7sc7p5t_28 #10, merged 2026-10-13; OpenROAD #11695 added a warning for such rounding), and the same change was proposed here as #4663, which was closed without being merged on 2026-10-25. This issue is to keep track of it, with a measured effect on a routed design.

Measured effect (M4 LEF58_EOLKEEPOUT)

M4 has EOLKEEPOUT 0.025 EXTENSION 0.048 0.02425 0.048 CORNERONLY. At 1000 the side extension 0.02425 is stored as 24 nm, which is exactly the gap to the adjacent track (24 nm wire, 48 nm pitch), so a corner on the adjacent track lies on the box edge and is not caught. On a small ASAP7 block (a 32 x 256 FIFO, ~1,400 cells plus a fakeram7_256x32, 1 GHz, ORFS c06bf3c):

  • with the platform LEF: router DRC 0, but asap7.lydrc reports 16 M4.S.5 ("min. parallel run length of two M4 on adjacent tracks: 44 nm") -- each a line-end patch next to a via pad on the adjacent track, offset by 1-33 nm;
  • with the side extension changed to 0.025 in a copy of the LEF (plus one extra M4 keep-out aimed at a different rule, M4.S.3): M4.S.5 0, router DRC 0. The final run built on that LEF (with other, unrelated changes) passes timing at 1 GHz, LVS and post-route simulation.

We tested the corrected value, not the move to 4000 itself; the library PR's table lists the same M4/M5 case (and M1, M6/M7).

Suggested fix

Take the library's fix (DATABASE MICRONS 4000 in both 1x tech LEFs), as #4663 proposed, or at least the corrected values.

Activity

  1. added a commit that references this issue on Oct 28, 2026
    ee9d563
  2. Arcadia-1 commented on Oct 28, 2026

    @Arcadia-1
    Author

    Still the case at head (d4559fc): the default TECH_LEF is asap7_tech_1x_201209.lef with DATABASE MICRONS 1000 and EOLKEEPOUT 0.025 EXTENSION 0.048 0.02425 0.048 on M4/M5 (lines 303, 390; the newer asap7_tech_1x_260907.lef has the same values and is not the default). Stock designs built and checked (make ... drc) at head with the official image openroad/orfs:26Q4-231-gd4559fcdd: asap7/gcd has 2 M4.S.5 markers, asap7/aes has 1,212 M4.S.5 and 168 M5.S.5 markers, all in the top cell (routing).

  3. maliberty commented on Oct 28, 2026

    @maliberty
    Member

    The rules are problematic as the EOL side extension exceeds the spacing for the layer. I've asked for clarification. The current rounding hides it.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions