agentleFS
RAMA-L7RAMA-L7/vlsi-pd-staff-engineer/skills/placement/SKILL.md

placement

Skill3 starsChanged 6 months agoMITView on GitHub

Skill type: FLEXIBLE — principles are firm, tactics adapt to design characteristics. You receive a verified floorplan (LEC PASS + low power PASS) and produce a legal placement where every standard cell has a physical location. Placement quality determines 80% of the final timing result. If any item is missing, request it from the floorplan agent. </HARD-GATE>

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    DO NOT START PLACEMENT WITHOUT:

    line 10
  2. Distinctive:

    If any item is missing, request it from the floorplan agent.

    line 17

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Placement Agent

Skill type: FLEXIBLE — principles are firm, tactics adapt to design characteristics.

You receive a verified floorplan (LEC PASS + low power PASS) and produce a legal placement where every standard cell has a physical location. Placement quality determines 80% of the final timing result.

<HARD-GATE> DO NOT START PLACEMENT WITHOUT: 1. Verified floorplan DEF with rows, macros placed, PG defined 2. LEC report: PASS (floorplan netlist == synthesis netlist) 3. Low power report: PASS (domain boundaries correct) 4. Congestion estimate from floorplan (to know problem areas) 5. SDC + liberty files

If any item is missing, request it from the floorplan agent. </HARD-GATE>


1. Inputs and Outputs

INPUTS:
  design_fp.def        Floorplanned DEF
  design_fp.v          Gate-level netlist
  design_fp.sdc        Timing constraints
  *.lib                Liberty timing libraries
  *.lef                Cell/tech LEF
  risk_list.md         From floorplan agent

OUTPUTS (handoff package):
  design_placed.def    Placed DEF (all std cells legalized)
  design_placed.v      Netlist (may have buffer/inverter additions
                       from placement optimization)
  design_placed.sdc    Constraints (unchanged)
  placement_qor.rpt    WNS, TNS, cell count, utilization
  congestion.rpt       Post-placement congestion map
  timing_histogram.rpt Slack distribution
  risk_list.md         Known issues for CTS agent

2. Placement Flow

PHASE 1: GLOBAL PLACEMENT
  Finds approximate locations minimizing total wire length.
  Cells overlap — that's expected at this stage.
  
  OpenROAD:
    global_placement -density <target> \
      -pad_left 2 -pad_right 2 \
      -timing_driven

  DENSITY PARAMETER:
    Matches floorplan utilization target:
    - 0.65-0.75 for standard designs
    - 0.55-0.65 for designs with macros
    - Lower density = easier routing but larger area
    
    Start with the floorplan utilization and adjust if
    congestion exceeds 85% in any region.

PHASE 2: TIMING-DRIVEN OPTIMIZATION
  Repairs timing violations found by global placement STA:
  
  estimate_parasitics -placement
  report_checks -path_delay max -group_count 10
  
  repair_timing -setup
  
  This may insert buffers and resize cells.
  Record: how many buffers added? Cell count change?

PHASE 3: DETAILED PLACEMENT (LEGALIZATION)
  Snaps every cell to a legal row position with no overlaps:
  
  detailed_placement
  
  After legalization:
  - Every cell sits on a row
  - No cell overlaps another
  - All cells align to the site grid (0.46μm for Sky130)
  
  check_placement -verbose
  
  If violations: cells didn't fit. Options:
  - Increase die area (back to floorplan agent)
  - Reduce utilization target
  - Review macro placement for wasted space

PHASE 4: POST-PLACEMENT OPTIMIZATION
  Second timing pass after legalization may have shifted cells:
  
  estimate_parasitics -placement
  repair_timing -setup
  repair_timing -hold -hold_margin 0.05
  
  report_checks -path_delay max -group_count 20
  report_checks -path_delay min -group_count 10
  report_design_area

3. Congestion Management

THE #1 PLACEMENT FAILURE MODE: congestion hotspots that make
routing impossible. Detect early, fix during placement, not routing.

CONGESTION CHECK:
  estimate_parasitics -placement
  report_congestion
  
  TARGETS:
    H-congestion < 80% everywhere
    V-congestion < 80% everywhere
    No single GCell > 90% in either direction

IF CONGESTION > 85% IN A REGION:
  Option 1: Increase cell padding
    global_placement -density 0.60 -pad_left 4 -pad_right 4
    
  Option 2: Add placement blockages in hot region
    create_blockage -type placement -area {x1 y1 x2 y2}
    Re-run global placement
    
  Option 3: Density screens (reduce density in congested area)
    set_placement_density -area {x1 y1 x2 y2} -density 0.50
    
  Option 4: Escalate to floorplan agent
    "Congestion > 90% in region (x1,y1)-(x2,y2) even with
     reduced density. Root cause: macro channel too narrow.
     Recommend: widen channel from 20μm to 40μm."

CONGESTION ANTI-PATTERNS:
  ❌ Ignoring congestion "because routing will fix it"
     → Routing cannot add resources. It can only detour.
  ❌ Reducing density globally when the hotspot is local
     → Wastes area everywhere, barely helps the hotspot
  ❌ Adding blockages without checking timing impact
     → Pushing cells away increases wire length on critical paths

4. Timing-Driven Placement

CRITICAL PATH AWARENESS:
  After global placement, identify the 20 worst paths:
  
  estimate_parasitics -placement
  report_checks -path_delay max -fields {slew cap nets fanout} \
    -group_count 20
  
  For each failing path:
  - Is the delay dominated by wire delay (placement issue)?
  - Or cell delay (synthesis issue)?
  
  If wire delay > 40% of path: cells are too far apart.
  The placement optimizer should handle this, but if it doesn't:
  
  Manual intervention:
    # Move a critical cell closer to its timing neighbor
    place_cell <instance> <x> <y> -fixed
    legalize_placement

USEFUL SKEW PREPARATION:
  Post-placement is when you identify candidates for useful skew
  (to be applied during CTS). Flag in the risk list:
  
  "Path group <name> has -0.3ns slack but the preceding path
   has +1.2ns slack. CTS agent: consider 200ps useful skew
   borrowing from the preceding path."

FANOUT OPTIMIZATION:
  repair_design -max_fanout 30
  
  This inserts buffer trees for high-fanout nets.
  Record: how many buffers inserted? These all need LEC verification.

5. Placement QoR Checks

CHECK 1: TIMING
  report_checks -path_delay max -group_count 50
  Target: WNS should be within 200ps of synthesis WNS
  (placement adds wire delay but shouldn't catastrophically degrade)
  
  If WNS degraded by > 500ps from synthesis:
  → Check if a critical path routes through a congested region
  → Check if cells were pushed far from optimal by macro blockage
  → This may be a floorplan-level issue (five-level trace)

CHECK 2: CONGESTION
  report_congestion
  Target: no region > 85%
  If exceeded: fix before proceeding (see section 3)

CHECK 3: UTILIZATION AND CELL COUNT
  report_design_area
  Cell count should match synthesis + buffer insertions
  If cell count dropped: cells were removed (investigate!)
  If cell count increased significantly: too many buffers

CHECK 4: PLACEMENT LEGALITY
  check_placement -verbose
  Must be clean: 0 overlaps, 0 site violations
  If violations exist: do NOT proceed to CTS

CHECK 5: HOLD SLACK ESTIMATE
  report_checks -path_delay min -group_count 20
  Hold violations are expected (CTS skew not applied yet)
  but extremely negative hold slack (< -1ns) suggests a
  placement problem (cells too close together on hold paths)

CHECK 6: POWER ESTIMATE
  report_power
  Significant increase from synthesis → investigate
  Could indicate unnecessary buffer insertion

6. Handoff to CTS Agent

HANDOFF PACKAGE:
  ✓ design_placed.def
  ✓ design_placed.v (with any inserted buffers)
  ✓ design_placed.sdc
  ✓ placement_qor.rpt (WNS, TNS, area, cells)
  ✓ congestion.rpt
  ✓ timing_histogram.rpt
  ✓ risk_list.md

RISK LIST EXAMPLES:
  - "WNS degraded from +1.99ns (synth) to +0.85ns (placement).
     0.85ns margin remains for CTS insertion delay + route wire RC."
  - "Congestion at 78% in region (120,80)-(200,160) near SRAM macro.
     Routing agent: expect detours in this area."
  - "3 paths with slack < 200ps — CTS must minimize insertion delay
     on these path groups. Useful skew candidates flagged."
  - "12 buffers inserted for fanout repair. LEC must verify."

THEN: LEC + Low Power gate check before CTS proceeds.
  LEC: placed netlist vs floorplan netlist (buffers are the delta)
  LP: isolation/LS cells not moved out of domain boundaries

7. OpenROAD Command Sequence

# === COMPLETE PLACEMENT FLOW ===

# 1. Read floorplanned design
read_lef sky130_fd_sc_hd.tlef
read_lef sky130_fd_sc_hd_merged.lef
read_liberty sky130_fd_sc_hd__tt_025C_1v80.lib
read_def design_fp.def
read_sdc design_fp.sdc

# 2. Global placement
global_placement -density 0.70 \
  -pad_left 2 -pad_right 2 \
  -timing_driven

# 3. Post-GP timing repair
estimate_parasitics -placement
repair_timing -setup
repair_design -max_fanout 30

# 4. Detailed placement (legalization)
detailed_placement
check_placement -verbose

# 5. Post-legalization timing repair
estimate_parasitics -placement
repair_timing -setup
repair_timing -hold -hold_margin 0.05

# 6. QoR reports
report_checks -path_delay max -group_count 20
report_checks -path_delay min -group_count 10
report_design_area
report_power
report_congestion

# 7. Write outputs
write_def design_placed.def
write_verilog design_placed.v

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.