FLAG: Formula-LLM-Based Auto-Generator for Baseband Hardware

Traditional hardware auto-generators efficiently assist designers in circuits generation, but still demand high programming skills. Large Language Models (LLMs) allow natural language programming, however, struggle with complex concepts in hardware design. To address this, we propose a formula-LLM-based auto-generator for baseband hardware, called FLAG. FLAG enhances LLMs’ understanding of complex formulas, and decreases the demands for designers’ skills. We showcase the ability of FLAG by generating a polar encoder with code length 8 and parallelism 4. Additionally, a polar encoder with code length 1024 and parallelism 4 is generated by FLAG, achieving comparable performance to state-of-the-art approaches. To the best of our knowledge, this is the first paper on LLM-assisted baseband hardware generation, which demonstrates the potential for large-scale baseband hardware generation using LLMs.