# Physical Design Extension Plan — Floorplan Exploration on openMSP430 **Status**: baseline OpenROAD RTL-to-GDSII closure complete; clock-gated post-route comparison remains future work. ## Goal Extend this study from synthesis-level PPA (Design Compiler + PrimeTime PX, 32 nm) down to **full RTL-to-GDSII physical design** with the open-source **OpenROAD** flow, and run a systematic **floorplan exploration**: how do core utilization, aspect ratio, and placement density change timing (WNS/TNS), routing congestion, wirelength, area, and power? Two things make this a natural extension rather than a new project: 1. The clock-gating variants studied at synthesis level here get re-measured **after placement and routing**, closing the loop the original study left open (synthesis-level power numbers vs. post-route reality). 2. The sweep is driven by a small Python tool built for this repo — config generation, batch flow execution, report parsing, and trade-off plots — i.e. flow/tool infrastructure work, not just tool usage. ## Flow and platform - **Flow**: [OpenROAD-flow-scripts (ORFS)](https://github.com/The-OpenROAD-Project/OpenROAD-flow-scripts), run via the official Docker image (no EDA licenses needed). - **Platform**: `sky130hd` (SkyWater 130 nm HD standard cells) as primary; `nangate45` as a sanity cross-check if time allows. - **Design**: the same openMSP430 RTL in `rtl/`, top module `openMSP430`, baseline configuration first, clock-gated variant (`opt1`) second. ## Phases ### Phase 1 — Baseline RTL-to-GDSII - [x] Install ORFS via Docker and verify the toolchain in the container - [x] Write `physical-design/orfs/config.mk` + `constraint.sdc` for openMSP430 on sky130hd - [x] Run the full flow: synth → floorplan → place → CTS → route → finish - [x] Record baseline metrics (WNS/TNS, area, utilization, wirelength, power) and final-route images ### Phase 2 — Floorplan sweep tool - [x] `run_sweep.py`: generate config variants with a distinct `FLOW_VARIANT` - [x] `parse_reports.py`: scrape ORFS JSON logs into one CSV (one row per variant) - [x] `plot_results.py`: area-vs-WNS Pareto plot, congestion heatmap by utilization - [x] Targeted sweep: baseline, 45% util, 35% util, 35% util with met2 minimum routing, and the final 30% util / 0.40 density / 1-site padding closure run - [x] Written findings in `physical-design/results/summary.md` ### Phase 3 — Floorplan/routability closure exercise - [x] Identify the initial routing bottleneck at high utilization - [x] Recover detailed-route closure using floorplan/placement knobs only: utilization, placement density, core margin, and one-site cell padding - [x] Document the closure loop and final metrics ### Phase 4 — Clock-gating closes the loop (future work) - [ ] Run the `opt1` clock-gated variant through the same flow - [ ] Compare post-route power vs. the synthesis-level numbers reported in the original study - [ ] One-page summary table: baseline vs. gated, synthesis vs. post-route ## Directory layout ``` physical-design/ README.md quickstart + status orfs/ config.mk ORFS design config (sky130hd, openMSP430) constraint.sdc clock constraint sweep/ run_sweep.py config generation + batch flow execution parse_reports.py ORFS reports -> results.csv plot_results.py Pareto / heatmap plots results/ checked-in CSVs, summary, images, and compact final artifacts ``` ## Metrics collected per run | Metric | Source | |---|---| | WNS / TNS (setup) | OpenSTA final timing report | | Core / die area, utilization | floorplan + finish reports | | Total wirelength | detailed route report | | DRC violation count | detailed route log | | Total / leakage power | final power report | | Runtime per stage | flow logs | ## Ground rules - Every commit is authored and committed by the repository owner; no AI co-author trailers. - Results in the README must be reproducible from the checked-in configs alone. - No result is reported until the corresponding phase is actually done.