Leveraging Large Language Models for the Automated Documentation of Hardware Designs

Abstract

The current interest in using Large Language Models (LLMs) to increase work productivity has also affected the Electronic Design Automation (EDA) industry. The main focus here lies on generating and automating the hardware design itself. We propose an alternative application of LLMs, using them to create human-readable design documentation from formal hardware models. We integrate LLMs into a documentation generation framework, which reads a formal specification of a hardware system and pre-processes it into suitable prompts. These prompts guide an LLM in articulating a natural language description of the system. Our findings affirm the potential of LLMs as practical tools to alleviate the documentation burden on engineers. However, they are not a standalone solution yet. A manual review of the generated documents is still required to ensure adherence to specific style and structure guidelines. Nonetheless, we show that this process demands significantly less effort than writing documentations manually.