AI PCB Routing Without the Epic Fails: Inside DeepPCB
Summary
Host Zach Peterson interviews Elaine Sam Cohen and Nabil Chaa from Deep PCB to discuss their AI routing engine and the new "Cooper" natural language assistant. The episode explores how using reinforcement learning and LLMs to capture design intent prevents common hardware errors that traditional optimizers miss. The guests demonstrate how Cooper orchestrates routing tasks, identifies missing constraints like reference planes, and integrates with other code-based design tools. They also outline future developments, including improved support for differential pairs and a more immersive user i…
Topics discussed
Intro: Communicating design intent to AI tools
Welcome and guest introductions
Guest backgrounds: Elaine Sam Cohen and Nabil Chaa
Deep PCB origins and shift to PCB design
Reinforcement learning approach to routing
Sequential decision making and reward signals
Two-level system: Routing engine and Cooper orchestrator
Task-based AI vs. one-prompt LLM approaches
Context awareness and user control over design rules
AI opportunities in placement, schematics, and manufacturing
Focus on DRC-clean routing and detailed execution
Future potential for BOM optimization and design reuse
Code-based PCB design and LLM limitations in geometry
Deep PCB as a compiler for schematics and routing
Aggregating multiple AI tools and workflow integration
Preventing hardware fails via natural language intent
AI detecting and fixing design errors automatically
Demo: Setting layer types and trace widths
Demo: Using Cooper to configure and start routing
Future roadmap: UI enhancements and feature expansion
Conclusion: AI augmenting designers and where to learn more
Listen ad-free on Castria