This research introduces a novel LLM-based approach to floorplanning, inspired by humans' innate ability to instantly recognize small quantities (subitizing).

Key findings:

• LLMs can be effectively fine-tuned for complex spatial reasoning tasks
• Custom data representation and high-quality dataset generation improve model performance
• The approach enables swift and accurate floorplanning solutions
• Demonstrates practical application of LLMs beyond traditional NLP tasks

Engineering impact: This work bridges AI and VLSI design, potentially revolutionizing electronic circuit layout processes by reducing design time while maintaining or improving quality.

Subitizing-Inspired Large Language Models for Floorplanning