Physical Design Flow Orchestrator

Overview

This agent implements a comprehensive, AI-driven architecture for the entire Physical Design (PD) flow. It models industry best practices by breaking down the complex process into distinct, manageable stages: Floorplanning, Placement, detailed routing, and Sign-off verification.

The core principle is modularity: an Orchestrator directs specialized Stage Agents, which in turn rely on specific domain knowledge encapsulated in 'Skills'. This structure ensures that each step—from initial netlist ingestion to final GDS generation—is executed with rigorous state management and quality checks.

Capabilities

• Orchestration: Manages the sequence of PD stages, handling dependencies and error recovery (loop-back logic).
• State Management: Utilizes a shared JSON object contract for all inter-agent communication, ensuring data consistency across the entire flow.
• Stage Execution: Coordinates specialized agents for critical tasks like floorplanning, component placement, routing optimization, and final sign-off checks.
• Reporting: Generates structured outputs including timing, power, area reports, and the final GDSII file representation.

Example Use Cases

1. Full Chip Synthesis Simulation: Ingest a raw netlist and SDC constraints to simulate an end-to-end physical design run, identifying bottlenecks at each stage (e.g., poor timing closure after placement).
2. Design Iteration Loop: Automatically manage the loop between Placement and Routing when initial quality metrics fail, passing updated state data back to the relevant agent.
3. Technology Migration: Adapt the flow to incorporate new technology rules by updating specific 'Skill' modules without rewriting the core orchestration logic.