SchGen: PCB Schematic Generation with Semantic-Grounded Code Representations

Qinpei Luo, Ruichun Ma, Xinyu Zhang, Lili Qiu

Keywords: PCB schematic generation, Large language models, Generative models, Electronic Design Automation (EDA)

TL;DR: We present a semantic-grounded code representation and a new dataset to enable LLM-based PCB schematic design generation.

Abstract: Printed circuit board (PCB) design is fundamental to nearly all electronic devices and has a growing demand in embodied AI, mobile, and Internet-of-Things applications. However, the schematic design process remains expertise-intensive and time-consuming. Large language models (LLMs) have accelerated software development recently, but extending them to hardware design is challenging due to domain-specific gaps— notably the absence of large, high-quality datasets and effective representations that capture circuit semantics. This paper presents SchGen, a large language model for PCB schematic generation that lowers the barrier to creating custom, novel hardware. To address the gaps above, we curate a comprehensive dataset of PCB schematic designs and propose a novel semantic-grounded code representation, which effectively encodes spatial arrangement and wire connections of components, making schematic designs amenable to generative modeling. Experimental results demonstrate that SchGen generates high-quality schematics with superior performance on wire connectivity and spatial arrangement over baselines. Our work paves the way for transforming hardware design from a manual task into an automated process with generative AI.

Primary Area: generative models

Submission Number: 15725