AMS-IO-Agent

An AI-powered integrated circuit (IC) design automation agent specialized for Cadence Virtuoso EDA environments. This system automates IO ring generation and verification workflows using LLM-driven intelligent agents.

Overview

AMS-IO-Agent leverages large language models to automate complex analog and mixed-signal IC design tasks, specifically focusing on IO ring design. It combines domain knowledge bases, specialized tools, and intelligent agent architectures to generate schematic and layout designs for IO rings while ensuring design rule compliance.

Key Features

• Automated IO Ring Generation - Generate IO pad ring schematics and layouts automatically based on requirements
• Verification Automation - Integrated DRC, LVS, and PEX verification workflows
• Multi-Model Support - Compatible with Claude, GPT-4, DeepSeek, and other OpenAI-compatible APIs
• Knowledge-Driven Design - Structured knowledge bases for technology nodes and IO design
• Web UI - Optional Gradio-based web interface for interactive use
• Benchmark Suite - Comprehensive test cases in AMS-IO-Bench for validation
• Image Vision Analysis - Analyze screenshots, layouts, and schematics using vision models
• User Profile Management - Customizable user profiles for personalized design preferences
• Tool Statistics & Analytics - Track tool usage and performance metrics
• Task History & Query - Query and analyze past task executions
• Health Check System - Automated system diagnostics and verification

Quick Start

Prerequisites

• Python 3.8 or higher
• csh (C Shell) - usually pre-installed on Linux/macOS
• Cadence Virtuoso (for actual design execution)
• API key for supported LLM providers (Claude, OpenAI, DeepSeek, etc.)

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/AMS-IO-Agent.git
   cd AMS-IO-Agent

2. Run the quick setup script (Recommended):

   For Linux/macOS (csh):

   ./quick_start.csh

   For Windows (PowerShell - Recommended):

   .\quick_start.ps1

   For Windows (CMD):

   quick_start.bat

   This one-command script will:

   • ✅ Automatically fix executable permissions (Linux/macOS)
   • ✅ Create Python virtual environment
   • ✅ Install all dependencies
   • ✅ Generate configuration files
   • ✅ Set up Virtuoso integration

   The setup script will:

   • Create a Python virtual environment using uv
   • Install all dependencies from requirements.txt
   • Generate a .env configuration file template
   • Set up Virtuoso integration files

3. Configure environment variables:

   Edit .env with your API keys:

   # Choose your preferred LLM provider
   CLAUDE_API_BASE=https://api.anthropic.com/v1
   CLAUDE_MODEL_ID=claude-sonnet-4-20250514
   CLAUDE_API_KEY=your_api_key_here
   
   # Or use OpenAI
   GPT4_API_BASE=https://api.openai.com/v1
   GPT4_MODEL_ID=gpt-4-turbo
   GPT4_API_KEY=your_openai_key_here
   
   # Virtuoso bridge configuration (optional)
   RB_HOST=localhost
   RB_PORT=9123

4. Configure the agent:

   Edit config.yaml to set your preferences:

   model:
     active: "claude"  # or "gpt4", "deepseek", etc.
     temperature: 1.0
   
   interface:
     mode: "cli"  # or "webui" for web interface
     logging: true

Basic Usage

Run the agent in CLI mode:

python main.py

Run with web interface:

# Edit config.yaml and set interface.mode: "webui"
python main.py

Auto-load a benchmark prompt:

# Edit config.yaml:
# prompt:
#   key: "IO_28nm_3x3_single_ring_digital"
python main.py

Project Structure

AMS-IO-Agent/
├── main.py                           # Main entry point
├── api_server.py                     # FastAPI server entry point
├── config.yaml                       # Main configuration file
├── requirements.txt                  # Python dependencies
├── README.md
│
├── src/                              # Core source code
│   ├── app/
│   │   ├── intent_graph/             # Intent graph generation and validation
│   │   ├── layout/                   # IO ring layout generation
│   │   ├── schematic/                # IO ring schematic generation
│   │   └── utils/                    # App-level shared utilities
│   ├── tools/                        # Agent tools and tool config
│   │   ├── tools_config.yaml
│   │   └── python_helpers/
│   ├── scripts/                      # EDA bridge and verification scripts
│   │   ├── ramic_bridge/
│   │   ├── calibre/
│   │   └── devices/
│   ├── skill/                        # Reusable Virtuoso SKILL snippets
│   └── agent_generated/              # Agent-generated Python helpers
│
├── Knowledge_Base/                   # Structured design knowledge base
│   ├── 00_META/
│   ├── 01_CORE/
│   ├── 02_TECHNOLOGY/
│   ├── 03_DESIGN_BLOCKS/
│   └── 04_ERRORS/
│
├── AMS-IO-Bench/                     # Benchmark prompts and scenarios
│   ├── 28nm_wirebonding/
│   └── 180nm_wirebonding/
│
├── tests/                            # Unit/integration and flow tests
│   ├── run_IO_Ring_batch.py
│   ├── test_io_ring_tool.py
│   ├── test_drc_runner_tool.py
│   ├── test_lvs_runner_tool.py
│   ├── test_pex_runner_tool.py
│   └── ...
│
├── setup/                            # Cross-platform setup scripts
├── docs/                             # Additional technical docs
├── user_data/                        # User profiles and per-user artifacts
├── output/                           # Generated design artifacts
├── logs/                             # Runtime logs
└── uploads/                          # Uploaded files for API/Web flows

Core Components

Entry Point

• main.py - Agent system entry point with IO ring design capabilities

Agent Tools

Tool	Purpose
il_runner_tool.py	Execute SKILL files in Cadence Virtuoso
drc_runner_tool.py	Run Design Rule Check verification
lvs_runner_tool.py	Run Layout vs Schematic verification
pex_runner_tool.py	Run parasitic extraction
io_ring_generator_tool.py	Generate IO ring schematics and layouts
knowledge_loader_tool.py	Dynamically load knowledge from markdown files
skill_tools_manager.py	Manage and execute reusable SKILL utilities
bridge_utils.py	Virtuoso communication (RAMIC/skillbridge)
image_vision_tool.py	Analyze images (screenshots, layouts, schematics)
user_profile_tool.py	Manage user profile and preferences
health_check_tool.py	System health diagnostics and verification
task_query_tool.py	Query and analyze task history
tool_stats_tool.py	Track and analyze tool usage statistics
python_tool_creator.py	Create and manage Python helper tools

Layout Generation (src/app/layout/)

Module	Purpose
layout_generator.py	Main layout generation orchestrator
position_calculator.py	Calculate pad positions for IO rings
voltage_domain.py	Handle voltage domain logic
skill_generator.py	Generate SKILL code for layouts
auto_filler.py	Automatic filler cell generation

Schematic Generation (src/app/schematic/)

Module	Purpose
schematic_generator.py	Generate SKILL code for schematics
json_validator.py	Validate intent graph files

Workflows

IO Ring Generation Workflow

1. Generate JSON configuration from natural language requirements
2. Validate configuration against templates
3. Generate schematic SKILL code
4. Generate layout SKILL code
5. Execute in Virtuoso via bridge
6. Run DRC/LVS verification
7. Iterate if errors detected

Knowledge Bases

The system uses a unified Knowledge_Base folder with specialized modules:

01_CORE - Core Design Knowledge

• Universal design principles
• Python-SKILL integration patterns

02_TECHNOLOGY - Technology-Specific Parameters

• 28nm process parameters
• 180nm process parameters
• Technology-specific design rules

03_DESIGN_BLOCKS - Design-Specific Knowledge

• IO_RING: IO pad selection, corner devices, voltage domains

04_ERRORS

• Common error documentation
• Error patterns and solutions learned during development
• Verification errors and troubleshooting

Technologies

Technology	Usage
Python 3.x	Main programming language
smolagents	AI agent framework from Hugging Face
OpenAI-compatible APIs	LLM backend (Claude, GPT-4, DeepSeek)
Cadence Virtuoso	Target EDA environment
SKILL	Virtuoso scripting language
Gradio	Optional web UI interface

Virtuoso Bridge Options

• RAMIC Bridge (Recommended) - Socket-based communication, supports cross-server connections
• skillbridge - Direct Python-Virtuoso integration via IPC

Supported Models

The system supports any OpenAI-compatible API. Pre-configured models include:

• Claude Sonnet 4 (via Anthropic or proxies)
• GPT-4 / GPT-4o (via OpenAI or proxies)
• DeepSeek Chat v3 (via OpenRouter)
• Custom models (add to .env)

To add a new model, simply add three environment variables to .env:

MYMODEL_API_BASE=https://api.provider.com/v1
MYMODEL_MODEL_ID=model-name
MYMODEL_API_KEY=your_key_here

Then use it in config.yaml:

model:
  active: "mymodel"

Batch Experiments

IO Ring Experiments

Run batch IO ring experiments:

python tests/run_IO_Ring_batch.py

Features:

• Test cases from 3x3 to 18x18 pad configurations
• Single and double ring topologies
• Digital, analog, and mixed-signal designs
• Multi-voltage domain support

Testing

Run individual tests from the project root:

# Quick flow test
python tests/test_quick_flow.py

# IO ring tool test
python tests/test_io_ring_tool.py

# Verification tests
python tests/test_drc_runner_tool.py
python tests/test_lvs_runner_tool.py
python tests/test_pex_runner_tool.py

# System and utility tests
python tests/test_health_check.py
python tests/test_image_vision_tool.py
python tests/test_user_profile.py
python tests/test_python_helper.py

Configuration

Main Configuration (config.yaml)

All non-sensitive configuration is in config.yaml:

model:
  active: "claude"      # Model to use
  temperature: 1.0      # Creativity level (0.0-2.0)

interface:
  mode: "cli"           # "cli" or "webui"
  logging: true         # Save conversation logs
  show_code: false      # Show code execution details

prompt:
  key: null             # Auto-load benchmark (e.g., "IO_28nm_3x3_single_ring_digital")
  text: null            # Or provide direct prompt text
  config_path: "user_prompt"  # Directory for custom prompts

ramic_bridge:
  host: null            # RAMIC server host (null = use RB_HOST env var)
  port: null            # RAMIC server port (null = use RB_PORT env var)

Environment Variables (.env)

All sensitive data and API keys go in .env:

# Model API Keys
CLAUDE_API_KEY=your_key
GPT4_API_KEY=your_key
DEEPSEEK_API_KEY=your_key

# Model Endpoints
CLAUDE_API_BASE=https://api.anthropic.com/v1
CLAUDE_MODEL_ID=claude-sonnet-4-20250514

# Virtuoso Bridge
RB_HOST=localhost
RB_PORT=9123

# User Profile
USER_PROFILE_PATH=user_data/default_user_profile.md

Architecture

Agent System

The system uses a specialized agent architecture focused on IO ring design:

• IO Agent: Coordinates IO ring generation, verification, and knowledge management
• Tool Suite: Comprehensive EDA tools for Virtuoso integration and verification

Benefits:

• Focused expertise on IO ring design
• Streamlined workflow for pad ring generation
• Direct access to all necessary EDA tools

Tool Loading

Tools are configured via src/tools/tools_config.yaml and loaded dynamically based on the use case. This allows flexible tool composition for different agent types.

Dynamic Knowledge Loading

Knowledge modules are loaded on-demand from markdown files in the knowledge bases, enabling:

• Efficient memory usage
• Targeted expertise injection
• Easy knowledge base updates without code changes

Contributing

This project is under active development. Contributions are welcome in the following areas:

• Additional technology node support
• New IO pad types and configurations
• Improved error handling and recovery
• Documentation and examples
• Test coverage

Troubleshooting

Common Issues

API Connection Errors:

• Verify API keys in .env
• Check API base URLs
• Ensure network connectivity

Virtuoso Bridge Issues:

• Verify RAMIC bridge is running
• Check RB_HOST and RB_PORT settings
• Test with tests/test_il_runner_ramicbridge.py

Tool Loading Errors:

• Check src/tools/tools_config.yaml syntax
• Verify all tool files exist
• Review logs in logs/ directory

Knowledge Base Not Loading:

• Verify Knowledge_Base directory structure
• Check markdown file syntax
• Review knowledge loader tool logs

License

Proprietary - All rights reserved.

Acknowledgments

• Built with smolagents from Hugging Face
• Uses skillbridge for Virtuoso integration
• Powered by large language models from Anthropic, OpenAI, and DeepSeek

Contact

For questions, issues, or collaboration inquiries, please open an issue on GitHub.

---

Note: This tool is designed for research and development purposes in IC design automation. Always verify generated designs meet your specific requirements and design rules before fabrication.