P34 is the cleanup pass after P33 closed setup timing.
The target was narrow: keep the RV32IM runtime and ACT4/Sail coverage from P33,
then attack the remaining electrical design-rule violations. P33 ended with
setup and hold clean, but still had 2241 max-slew violations and 291
max-cap violations.
The useful RTL change was to stop pushing a large asynchronous CPU reset tree through the whole core. P34 registers the loader-to-CPU release and makes the main core reset synchronous. The backend config then leans harder on slow-corner repair, post-global-route design repair, post-global-route timing repair, stronger slew/cap margins, and 250 um wire repair.
Command:
make harden PROJECT=34_rv32im_drv_cleanupResult: PASS, with residual DRV warnings.
| check | result |
|---|---|
| Directed RV32IM runtime | PASS |
ACT4/Sail rv32i/I | PASS=39 FAIL=0 NOT RUN=0 |
ACT4/Sail rv32i/M | PASS=8 FAIL=0 NOT RUN=0 |
| Full RISC-V compliance | NOT RUN |
| LibreLane flow | PASS |
| Run directory | projects/34_rv32im_drv_cleanup/librelane/runs/RUN_2026-05-01_14-58-10 |
| Final GDS | projects/34_rv32im_drv_cleanup/librelane/runs/RUN_2026-05-01_14-58-10/final/gds/top.gds |
| Metrics | projects/34_rv32im_drv_cleanup/librelane/runs/RUN_2026-05-01_14-58-10/final/metrics.json |
| Magic DRC | PASS (0 errors) |
| KLayout DRC | PASS (0 errors) |
| LVS | PASS (0 errors) |
| Antenna | PASS (0 violations) |
| Setup timing | PASS (0 violations, worst setup slack 7.475 ns) |
| Hold timing | PASS (0 violations, worst hold slack 0.106 ns) |
| Max slew | FAIL (12 violations) |
| Max cap | FAIL (3 violations) |
| Max fanout | FAIL (9 violations) |
Two harden attempts were useful. The first P34 run, RUN_2026-05-01_14-36-37,
proved the reset cleanup helped: slew dropped to 1688 and cap dropped to
196. The second run, RUN_2026-05-01_14-58-10, used stronger repair settings
and pushed the residual violations down to 12 slew and 3 cap.
That second run was not free. It inserted thousands of timing-repair buffers, needed antenna jumpers and diodes, and made detailed routing work through a large temporary violation count before converging to zero detailed-route DRC. But it did converge, and Magic DRC, KLayout DRC, LVS, and antenna all passed.
What remains: inspect the final handful of DRV/fanout violations directly instead of treating this as a broad architecture problem. P34 turns the issue from “the backend is shouting everywhere” into “there are specific nets to look at.”