iPL · placement engine

# Core capabilities

# Global placement

iPL uses a Nesterov gradient-descent solver with an e-Density electrostatic density model to spread standard cells evenly in continuous space while minimizing weighted wirelength (WA model). When Agents call iPL.place with effort="estimate", they get a fast HPWL estimate without moving cells; with effort="standard", the full three-stage flow runs. After each step, intermediate state exports automatically to iDB as vectorized features for AiEDA Library.

# Legalization

The Abacus legalization engine aligns post-global-placement cells to rows and sites, removes overlaps, and satisfies process constraints. Agents can pass a frozen_objects list to lock placed macros or critical-path standard cells; legality checks apply only to unfrozen regions. After legalization, dirty_set returns the set of moved cells—Agents can pass these downstream to iSTA for incremental timing analysis.

# Detailed placement

Detailed placement swaps adjacent cell positions within local windows to optimize wirelength, timing, and routability. A max_displacement constraint prevents critical cells from drifting too far from their legalized positions. After each swap, Agents immediately get updated HPWL, timing slack, and congestion density, forming a closed optimization loop—something that traditionally requires Tcl scripts calling multiple tools repeatedly.

# Incremental legalization

When Agents add, remove, or move cells locally at ECO scale, they need not rerun full-chip placement. Calling runIncrLG legalizes only changed cells (via inst_list) and their neighborhoods. Incremental runs typically take 1–5% of full legalization time, sharply reducing Agent decision-loop latency.

# Agent calling patterns

iPL exposes three equivalent calling protocols. Agents pick the one best suited to their runtime—Python SDK for native integration, MCP for remote tool services, Tcl for existing script ecosystems.

from ieda import Client

client = Client("http://localhost:9099")

# Standard full-flow placement
result = client.call(
    "iPL.place",
    design_ref="snap_7f3a",
    effort="standard",
    region="core",
    intent_ref="intent_v2",
)

# Access structured results
print(result.hpwl)            # 1823400.5
print(result.density)         # 0.738
print(result.congestion_map)  # <binary grid>
print(result.design_ref)      # "snap_7f3a_v2"
print(result.dirty_set)       # ["u_alu_0", "u_reg_5", ...]

# Incremental legalization: only fix local area
incr = client.call(
    "iPL.legalize",
    design_ref="snap_7f3a_v2",
    frozen_objects=["macro_0", "macro_1"],
)

// JSON-RPC over MCP
{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "name": "iPL.place",
    "arguments": {
      "design_ref": "snap_7f3a",
      "effort": "standard",
      "region": "core",
      "intent_ref": "intent_v2"
    }
  }
}

// Response
{
  "hpwl": 1823400.5,
  "density": 0.738,
  "congestion_map": "base64://...",
  "design_ref": "snap_7f3a_v2",
  "dirty_set": ["u_alu_0", "u_reg_5"],
  "warnings": []
}

# Full placement flow
iPL::place -design snap_7f3a -effort standard

# Individual stage control
iPL::run_gp -design snap_7f3a -target_density 0.8
iPL::run_lg -design snap_7f3a -max_displacement 50000
iPL::run_dp -design snap_7f3a

# Incremental legalization
iPL::run_incr_lg -inst_list {u_alu_0 u_reg_5}

# Query results
iPL::report_hpwl
iPL::report_density
iPL::check_legality

# Input / output contract

Request Schema — iPL.place

Result Schema

# Role in the flow

RTL
 │
 ▼
┌──────────┐
│   iMap   │  Logic synthesis & tech mapping
└────┬─────┘
     │
     ▼
┌──────────┐
│   iFP    │  floorplanning (Floorplan)
└────┬─────┘
     │
     ▼
┌──────────┐
│  iPDN    │  power grid planning
└────┬─────┘
     │
     ▼
┌──────────────────────────────────────┐
│              iPL  ◆                  │
│  global placement → legalization → detailed placement          │
│  ↑ iSTA (pre-placement timing annotation)    │
│  ↓ iSTA (incremental timing analysis, dirty_set) │
└──────────────────┬───────────────────┘
                   │
                   ▼
          ┌──────────────┐
          │     iCTS     │  Clock tree synthesis
          └──────┬───────┘
                 │
                 ▼
          ┌──────────────┐
          │     iTO      │  timing optimization
          └──────┬───────┘
                 │
                 ▼
          ┌──────────────┐
          │     iRT      │  Global & detailed routing
          └──────────────┘

# Related resources