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

routing

Skill3 starsChanged 6 months agoMITView on GitHub

Skill type: RIGID — DRC clean is non-negotiable. Zero violations at handoff. You receive a verified CTS design (LEC PASS + low power PASS — every clock buffer domain-checked) and connect all nets with metal wires. Routing is where the design becomes physically real. Every DRC violation you leave behind is a manufacturing defect. IF LOW POWER CHECK FAILED ON CTS BUFFERS: DO NOT ROUTE. Go back to CTS agent. Bug #3 cannot be fixed at routing level. </HARD-GATE>

19 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use parallelization, prompt-chaining and 1 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • hajekimSkill

    parallelization

    This skill should be used when the user wants to "run tasks in parallel", "concurrent LLM calls", "fan-out and fan-in", "parallel agents", "simultaneous execution", "reduce latency with concurrency", "parallel sub-tasks", "parallel processing", "concurrent agent execution", "run agents simultaneously", "speed up with parallelism", "async agent execution", "batch LLM calls", "parallel workflow", "parallelize agent tasks", or needs to execute multiple independent tasks simultaneously. Also responds to Korean: "병렬 작업 실행", "동시 LLM 호출", "팬아웃 팬인 패턴", "병렬 에이전트", "동시 실행", "동시성으로 지연 시간 감소", "병렬로 실행", "동시에 여러 작업", "병렬 처리", "동시에 돌려줘", "빠르게 병렬로", "여러 작업 동시 실행해줘". Also responds to Japanese: "並列処理", "同時実行", "並列エージェント", "並行LLM呼び出し", "ファンアウト・ファンイン", "複数タスクを同時に実行したい", "処理速度を上げたい", "非同期エージェント", "バッチ処理", "並列で動かしたい" Also responds to Chinese: "并行处理", "并发执行", "并行智能体", "同时调用多个LLM", "扇出扇入模式", "多任务并行运行", "加快处理速度", "异步智能体", "批量LLM调用", "同时运行多个任务".. Apply this skill to design or implement the Parallelization agentic design pattern.

    Also in 3 projects that use this

  • hajekimSkill

    prompt-chaining

    This skill should be used when the user wants to "chain prompts", "build a pipeline", "break down complex tasks into steps", "sequential LLM calls", "multi-step reasoning", "pipeline pattern", "sequential agent pipeline", "multi-step prompt pipeline", "LLM chain", "step-by-step agent", "prompt pipeline", "decompose task into prompts", "stage-by-stage processing", "chained LLM calls", or when a task is too complex for a single prompt and needs to be decomposed into sequential sub-tasks. Also responds to Korean: "프롬프트 체이닝", "체인 방식 에이전트", "단계별 LLM 호출", "복잡한 작업 분해", "순차적 프롬프트 파이프라인", "단계적 작업 처리", "단계별 처리", "순차 처리 에이전트", "파이프라인 만들어줘", "단계적으로 처리해줘", "연속 프롬프트", "체이닝 해줘". Also responds to Japanese: "プロンプトチェーニング", "ステップごとに処理", "順次LLM呼び出し", "パイプライン構築", "複雑なタスクを分解", "段階的エージェント", "連続プロンプト", "LLMパイプライン", "順番に処理してほしい", "チェーン方式エージェント" Also responds to Chinese: "提示词链", "顺序调用LLM", "构建流水线", "分解复杂任务", "逐步处理", "链式提示", "多步骤流水线", "按顺序处理任务", "提示词管道", "帮我一步一步处理".. Apply this skill to design, implement, or explain the Prompt Chaining agentic design pattern.

    Also in 3 projects that use this

  • NVIDIASkill

    testing

    Run Python and Rust tests for physicsnemo-curator using uv + pytest with coverage reporting via pytest-cov, and cargo-nextest for Rust tests. Includes benchmark workflows with pytest-benchmark and criterion.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

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

  • 0

    Common

    Most of the 17 edited versions we read keep these.

  • 2

    Distinctive

    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 ROUTING WITHOUT:

    line 11
  2. Distinctive:

    IF LOW POWER CHECK FAILED ON CTS BUFFERS: DO NOT ROUTE.

    line 18

“Common” is measured: kept by at least 6 in 10 of the 17 edited versions of this file we could read. “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

Routing Agent

Skill type: RIGID — DRC clean is non-negotiable. Zero violations at handoff.

You receive a verified CTS design (LEC PASS + low power PASS — every clock buffer domain-checked) and connect all nets with metal wires. Routing is where the design becomes physically real. Every DRC violation you leave behind is a manufacturing defect.

<HARD-GATE> DO NOT START ROUTING WITHOUT: 1. CTS DEF with clock tree cells placed 2. LEC report: PASS (CTS netlist vs placed netlist) 3. Low power report: PASS (ALL CTS buffers domain-verified) 4. CTS timing report (know your remaining margin) 5. CTS risk list (congestion hotspots, hold violations, tight paths)

IF LOW POWER CHECK FAILED ON CTS BUFFERS: DO NOT ROUTE. Go back to CTS agent. Bug #3 cannot be fixed at routing level. </HARD-GATE>


1. Inputs and Outputs

INPUTS:
  design_cts.def       CTS DEF
  design_cts.v         CTS netlist
  design_cts.sdc       Propagated clock SDC
  *.lib                Liberty libraries
  *.lef                Cell/tech LEF
  risk_list.md         From CTS agent

OUTPUTS (handoff package):
  design_routed.def    Routed DEF (all nets connected)
  design_routed.v      Netlist (may have hold buffers added)
  design_routed.sdc    Constraints (unchanged)
  route_timing.rpt     Post-route STA (final numbers)
  drc.rpt              DRC report (must be: 0 violations)
  antenna.rpt          Antenna check results
  hold_fix.rpt         Hold buffer insertion summary
  risk_list.md         Known issues for signoff agent

2. Routing Flow

PHASE 1: GLOBAL ROUTING
  Assigns each net to routing regions (GCells) without
  determining exact wire positions. Identifies congestion.
  
  OpenROAD:
    global_route -guide_file route.guide \
      -congestion_iterations 50 \
      -allow_congestion

  After global routing:
    report_congestion
    
  If any GCell > 100% overflow:
  → Routing cannot succeed. Options:
     a) Reduce cell density in that region (back to placement)
     b) Add routing blockages to redirect traffic
     c) Increase die area (back to floorplan)
  
  Do NOT proceed to detail routing with > 5% overflow.

PHASE 2: DETAIL ROUTING
  Determines exact wire positions, via locations, and
  metal layer assignments. Enforces all DRC rules.
  
  OpenROAD:
    detailed_route -output_drc drc_initial.rpt \
      -output_maze drc_maze.log

  Detail routing iterates to resolve DRC violations.
  Monitor: does violation count decrease each iteration?
  
  If violations INCREASE between iterations:
  → Congestion death spiral. Stop and escalate.

PHASE 3: HOLD FIXING
  After routing, fix hold timing violations by inserting
  delay buffers on short data paths:
  
  estimate_parasitics -global_routing
  repair_timing -hold -hold_margin 0.05
  
  This inserts buffers. CRITICAL CHECKS:
  1. Every hold buffer must be in the correct power domain
     (the low power agent checks this at the gate)
  2. Hold buffers must not violate setup on the same path
  3. Hold buffers need to be placed legally (legalize after)
  
  detailed_placement -incremental
  
  After hold fixing: re-run detail routing for new connections:
  detailed_route -output_drc drc_after_hold.rpt -incremental

PHASE 4: ANTENNA REPAIR
  Long metal segments connected to gate inputs accumulate
  charge during manufacturing, potentially damaging the gate oxide.
  
  repair_antennas -iterations 5
  
  Antenna fixes: insert diodes near gate inputs, or break
  long wires with via hops to upper metals.
  
  After antenna repair:
  detailed_route -incremental -output_drc drc_final.rpt

PHASE 5: FINAL DRC CHECK
  check_drc -report drc_final.rpt
  
  TARGET: 0 violations.
  
  If violations remain after incremental routing:
  - Categorize: spacing? width? via enclosure? shorts?
  - If clustered in one region: local congestion issue
  - If scattered: may be a tech LEF issue or tool bug
  - Invoke forensic debugging if > 3 fix attempts fail

3. DRC Convergence Strategy

DRC FIXING FOLLOWS THE THREE-FIX RULE.

ATTEMPT 1: Incremental detail routing
  detailed_route -incremental -output_drc drc_fix1.rpt
  If violations reduced by > 50%: progress. Continue.

ATTEMPT 2: Targeted area re-route
  For remaining violations, re-route just that region:
  detailed_route -incremental -output_drc drc_fix2.rpt
  If violations still > 0 but decreasing: continue.

ATTEMPT 3: More aggressive repair
  detailed_route -incremental \
    -output_drc drc_fix3.rpt \
    -droute_end_iter 64
  
  If violations still > 0 after attempt 3:
  → THREE-FIX RULE TRIGGERED
  → Stop. Classify remaining violations.
  → Escalate to forensic debugging agent.
  
  Common escalation: "DRC violations in region X cannot be
  resolved by routing alone. Root cause: placement density
  too high in that region. Recommend: reduce density from
  78% to 65% and re-place."

4. Timing Closure Through Routing

POST-ROUTE STA:
  estimate_parasitics -global_routing
  report_checks -path_delay max \
    -fields {slew cap input_pins nets fanout} \
    -group_count 50
  
  Wire RC is now real (estimated from actual routes, not
  placement-based wire models). Timing will shift.

EXPECTED CHANGES FROM CTS:
  - Setup WNS may degrade by 100-300ps (wire RC)
  - Some paths improve (shorter routes than estimated)
  - Hold violations should be mostly fixed

IF POST-ROUTE WNS < 0:
  Do NOT immediately apply ECOs.
  Check: is the signoff STA (with extracted parasitics)
  the same as the estimate? Often the global-route estimate
  is pessimistic. Wait for signoff.
  
  If the violation is large (> 200ps): invoke forensic
  debugging agent. The root cause is likely upstream.
  
  If the violation is small (< 100ps): may be fixable
  with post-route optimization:
  
  repair_timing -setup
  detailed_route -incremental

5. Routing-Specific Checks

CHECK 1: DRC CLEAN
  check_drc -report drc_final.rpt
  MUST BE: 0 violations
  No exceptions. No "we'll fix it in signoff."

CHECK 2: CONNECTIVITY
  Every net must be fully connected. No opens.
  check_connectivity -report connectivity.rpt
  
  Opens = manufacturing short circuit to ground.

CHECK 3: ANTENNA
  check_antennas -report antenna.rpt
  All antenna ratios within foundry limits.

CHECK 4: TIMING
  report_checks -path_delay max -group_count 50
  report_checks -path_delay min -group_count 50
  
  Setup: WNS should be ≥ 0 (or close, for signoff to confirm)
  Hold: WNS should be ≥ 0 after hold fixing

CHECK 5: CONGESTION (final)
  report_congestion
  Should show no overflow (all nets routed within resources)

CHECK 6: POWER
  report_power
  Wire switching power now included (first time it's real)

6. Handoff to Signoff Agent

HANDOFF PACKAGE:
  ✓ design_routed.def (all nets routed, DRC clean)
  ✓ design_routed.v (with hold buffers)
  ✓ design_routed.sdc
  ✓ route_timing.rpt (post-route STA)
  ✓ drc.rpt (0 violations)
  ✓ antenna.rpt (clean)
  ✓ hold_fix.rpt (buffers inserted, count + domains)
  ✓ risk_list.md

RISK LIST ITEMS:
  - "Post-route WNS is +0.03ns — razor thin margin. Signoff
     with extracted parasitics may push this negative."
  - "Hold buffers inserted: 45 cells. All in PD_ALWAYS_ON
     (LP verified). But routing agent cannot verify these
     against signoff-extracted parasitics — signoff must re-check."
  - "Region (120,80)-(200,160) had congestion at 92% during
     routing. Wires detoured through upper metals. Crosstalk
     risk in this area — signoff SI analysis should focus here."

THEN: LEC + Low Power gate check
  LEC: routed netlist vs CTS netlist (hold buffers are the delta)
  LP: hold buffers in correct domains (Bug #6 check)

7. OpenROAD Command Sequence

# === COMPLETE ROUTING FLOW ===

# 1. Read CTS 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_cts.def
read_sdc design_cts.sdc

# 2. Global routing
set_routing_layers -signal met1-met5 -clock met3-met5
global_route -guide_file route.guide \
  -congestion_iterations 50

# 3. Detail routing
detailed_route -output_drc drc_initial.rpt

# 4. Post-route timing
estimate_parasitics -global_routing
report_checks -path_delay max -group_count 20

# 5. Hold fixing
repair_timing -hold -hold_margin 0.05
detailed_placement -incremental
detailed_route -incremental -output_drc drc_hold.rpt

# 6. Antenna repair
repair_antennas -iterations 5
detailed_route -incremental -output_drc drc_antenna.rpt

# 7. Final DRC
check_drc -report drc_final.rpt

# 8. Final QoR
estimate_parasitics -global_routing
report_checks -path_delay max -group_count 50
report_checks -path_delay min -group_count 50
report_design_area
report_power

# 9. Write outputs
write_def design_routed.def
write_verilog design_routed.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.