Skip to content
AGENT-NATIVE EDA PLATFORM

EDA tooling built for
IC design Agents

40+ headless EDA tools, one unified database, and a full Agent API—so your LLM can drive chip design end to end. From RTL to GDS, every step can be called, observed, and decided by an Agent.

agent.call
# Agent calls EDA tools directly
$ agent.call("iPL.place", {
    design: "aes_core",
    effort: "standard",
    intent: intent_ref
  })
→ PlacementResult { hpwl: 1245678, density: 0.72, ... }
iMapsynthesisiFPfloorplaniPDNpoweriPLplacementiCTSclockiTOoptimizationiRTroutingiSTAtimingAiEDAdesign dataiPCLlayout modeliMapsynthesisiFPfloorplaniPDNpoweriPLplacementiCTSclockiTOoptimizationiRTroutingiSTAtimingAiEDAdesign dataiPCLlayout model
40+Agent-native tools
1Unified DB iDB
5Silicon-proven tape-outs
48+Papers
400K+Codebase size
GPL-3.0License
ARCHITECTURE

Four-layer architecture, Agent-Native by Design

From the unified data base to Agent services—every layer is designed to be called by LLMs.

L1 · Unified base

iDB · Platform · Interface · Evaluation

iDB is the hub—every tool shares one design state. A standard Interface makes each capability a first-class API, with evaluation measuring tools and Agents.

For tool authors: one interface joins the ecosystem. For Agent authors: one shared semantic space.

iDB · Platform · Interface · Evaluation
┌──────────────────────────────────────┐
│              Interface               │
│   (Python / MCP / Tcl standard interface)      │
├──────────────────────────────────────┤
│                iDB                   │
│   Unified design DB · shared in-memory state       │
├──────────────────────────────────────┤
│   Platform   ·   Evaluation          │
│   Flow orchestration   Quality metrics             │
└──────────────────────────────────────┘
    Shared design state · API first-class
L2 · Agent-native EDA tools

22 RTL→GDS point tools

Synthesis, floorplan, power, place, clock, optimize, route, timing, power, DRC, extraction—each tool is headless with a unified Agent API. Call them alone or orchestrate with iFlow.

Every tool is directly callable by Agents and returns structured results.

22 headless tools
RTL → iMap → iNO → iFP → iPDN
         → iPL → iCTS → iTO → iRT
         → iSTA → iPA → iDRC → iRCX
              ↓
         GDSII + reports
   Each tool is directly callable by Agents
L3 · Agent infrastructure

Runtime · Observer · Router · Oracle · Factory

Runtime manages experiments and state; Observer watches tool runs; Router picks models by task; Oracle supplies ground truth; Factory produces training samples.

Make Agent experiments manageable, state observable, and model routing smart.

Runtime · Observer · Router · Oracle · Factory
┌──────────────────────────────────────┐
│  Runtime        Observer             │
│  Experiments        Observe             │
├──────────────────────────────────────┤
│  Router         Oracle               │
│  Route models        Reference             │
├──────────────────────────────────────┤
│  Factory                             │
│  Samples                             │
└──────────────────────────────────────┘
  Experiments · observation · model routing
L4 · Agent services

Planner · Intent · Verify · Tech · Memory · Bridge

Planner turns natural-language intent into tool sequences; Intent captures constraints; Verify checks intermediates; Tech injects process knowledge; Memory keeps context; Bridge connects external models and knowledge bases.

Plan, intent, verify, knowledge—so Agents truly understand chip design.

Planner · Intent · Verify · Tech · Memory · Bridge
┌──────────────────────────────────────┐
│  Planner    Intent      Verify       │
│  Plan    Intent    Verify      │
├──────────────────────────────────────┤
│  Tech       Memory      Bridge       │
│  Tech    Memory   Bridge     │
└──────────────────────────────────────┘
  Plan · intent · verify · knowledge
DEVELOPER JOURNEY

Three steps to let your Agent drive chip design

From picking tools to wiring your LLM—one Agent API keeps the path smooth.

STEP 1

Pick tools

Browse 40+ Agent-native tools—22 EDA point tools covering RTL→GDS, plus Agent infra and services. Each tool ships with clear API docs and call examples.

40+ Agent-native tools
EDA tools (22)    Infra (8)    Services (10+)
─────────────────────────────────────────────
iMap  iNO  iFP   Runtime         Planner
iPDN  iPL  iCTS  Observer        Intent
iTO   iRT  iSTA  Router          Verify
iPA   iDRC iRCX  Oracle          Tech
iFV   iEval ...  Factory  ...    Memory  ...
STEP 2

Call the API

One Agent API across Python, MCP, and Tcl. Same semantics, same return shape—results are ready for Agents to consume.

Python · MCP · Tcl
from ieda import Agent

agent = Agent("aes_core")
result = agent.call("iPL.place",
  effort="standard")
# → PlacementResult { hpwl: 1245678,
#     density: 0.72, ... }
{
  "tool": "iPL.place",
  "params": {
    "design": "aes_core",
    "effort": "standard",
    "intent": "intent_ref"
  }
}
→ PlacementResult { ... }
iPL::place -design aes_core \
  -effort standard \
  -intent $intent_ref
# → PlacementResult { hpwl 1245678
#     density 0.72 ... }
STEP 3

Integrate your Agent

Connect your LLM to the iEDA Agent API. Planner decomposes natural-language goals into tool calls; Runtime manages experiments; Observer feeds intermediate results—so your Agent can drive the flow like a seasoned engineer.

LLM + iEDA Agent API
User: "Place aes_core with standard constraints"
        ↓
   ┌─────────────────────┐
   │  Your LLM    │
   └────────┬────────────┘
            ↓
   ┌─────────────────────┐
   │  iEDA Agent API     │
   │  Planner  →  iPL    │
   │  Observer →  Verify │
   └─────────────────────┘
            ↓
   PlacementResult + feedback
SILICON-PROVEN

Silicon-proven tape-outs

The toolchain has taped out chips across process nodes; silicon data keeps feeding engineering and datasets.

01

110nm validation chip

Full-flow bring-up; first silicon loop for the iEDA toolchain from RTL to GDS.

02

28nm SoC tape-out

Advanced-node validation; timing closure and power opts confirmed on silicon.

03

28nm accelerator chip

Domain-specific accelerator full flow; silicon data feeds tool algorithms.

04

110nm mixed-signal chip

Digital+analog mixed-signal flow validation, expanding toolchain coverage.

05

28nm research tape-out

Continuous iteration—silicon data drives tool improvements and datasets.

Talent training & knowledge → Academy

Courses and hands-on tutorials are no longer part of the product narrative. Enter via Academy when you need systematic learning (hosted on the same origin for now; may move later).

Ready to let your Agent drive chip design?

40+ headless EDA tools, a unified database, and a full Agent API—from RTL to GDS, your LLM can call every step.