seveibar/am3352-ram-dogbone-and-single-layer-route-test

AM3352BZCZ100 processor connected to a W631GG6MB-12 DDR3 memory chip through a routed, length-matched 16-bit DDR interface with clock/data-strobe pairs and control/address signals.

Version
0.0.9
License
unset
Stars
0

design/rounded-criteria.md

# Rounded-routing criteria, version 2

The user's correction changes the objective: fewer vertices must not justify sharper corners, and length tuning must be curvy rather than square-wave-shaped. The v1 simplification score remains a record of the earlier stage; it is not a score for the final curved paths.

## Hard checks

- Preserve all 47 connections, endpoints, signal layers, top dogbones, and 94 vias. Add no routing vias.
- Pass native DRC and independent copper-clearance/connectivity checks; keep simple, non-self-intersecting centerlines.
- Replace turns with tangent circular fillets. Approximate arcs with heading increments at most 5°; independently reject any final heading jump above 5.001° (numerical allowance).
- Retain 0.635 mm byte-bus skew and 0.127 mm pair skew limits.

## Radius and timing selection

Start with corner trims up to 0.5 mm and 45% of either adjacent segment, then reduce the trim if foreign traces or through vias obstruct the arc. Clearance comparisons use a 0.00005 mm numerical tolerance; nominal trace width and clearance remain 0.1 mm, and native DRC is still required.

Rounding reduces centerline length. Select each bus's permitted reductions with a 20% skew-limit margin, then tighten its differential pairs with a 40% margin. Scale fillets on each net to achieve its assigned reduction. The result is a real change in the copper geometry and timing lengths.

A uniform minimum radius is **not** guaranteed: some inherited short segments and crowded tuning regions permit only tight curves. The angular check establishes the absence of acute vertices, not a minimum-curvature or signal-integrity signoff. Radius-constrained re-routing would be a separate, stronger requirement.

## Evidence

`validation/rounded-score.json` compares phases 3 and 4: 194 acute vertices before, zero after; maximum rendered heading change 178.25° before, 5.00° after. `validation/summary.json` and `validation/copper-audit.json` record the independent timing and copper checks. `validation/rounded-comparison.png` shows the previous and rounded routes.