[P] Prompt optimization for analog circuit placement — 97% of expert quality, zero training data

VizPy has demonstrated a breakthrough in applying AI to analog integrated circuit (IC) layout, a domain considered a notoriously difficult benchmark for machine learning. Unlike digital design, which has automated place-and-route (P&R) tools, analog layout requires complex spatial reasoning and multi-objective optimization to balance factors like device matching and minimizing parasitic effects. The company's novel approach uses a prompt optimizer that iteratively refines instructions to a large language model (LLM) based on analyzing pairs of failed and successful layout attempts.

This method allows the AI to learn and improve its reasoning across iterations without ever being trained on domain-specific circuit data. The system was evaluated on real-world analog circuit placement tasks, where it achieved a result quality of 97% compared to expert human designers. This represents a significant step toward automating a highly specialized, manual engineering process, potentially accelerating chip design cycles for critical components like sensors, power management circuits, and RF chips.