Personal project. I built a workflow that scores placement candidates, refines the best layout, and sends it to FreeRouting.
The challenge
Apparently poor routing results came from the test setup: unconnected ground pours inflated the unrouted count. Fixing the harness brought the same small boards to 100% routed.
Results
- One 131-component example reduced wirelength by 33.6% and crossings from 427 to 164.
- Sorting net accumulation made seeded placement reproducible across processes.
- Outputs verified by milling real boards.
