Design Reuse via Automated Requirements-Driven Retrieval: A Framework for Electronic Hardware

Abstract Design reuse accelerates hardware development by improving efficiency, reducing costs, and shortening time-to-market. However, retrieving reusable circuit blocks directly from natural-language requirements remains challenging, particularly in confidentiality-driven low-resource domains with heterogeneous artifacts and nonstandard representations. We introduce a structured framework for requirement-driven retrieval tailored to these constraints, combining dataset curation, sparse lexical search, semantic reranking, and a large language model (LLM) for final candidate block selection. To support this, we introduce R3SET, a structured dataset that links 338 real-world hardware requirements to 92 reusable circuit blocks extracted from eight open-source projects. Each block is described using expert-reviewed, LLM-assisted metadata authoring. To simulate realistic retrieval conditions, R3SET also includes 100 synthetic distractor blocks derived from vendor artifacts. To our knowledge, it is the first dataset linking requirements to reusable electronics at this level of granularity. Evaluated on R3SET, the hybrid retriever achieves 92% Hit@5 and 89% Hit@10 on the one-to-one subset (n=271), where each requirement corresponds to a single circuit block. On the full dataset including cases without a matching block (n = 328), the LLM slightly outperformed the retriever's top-1 accuracy (60.1% vs 58.5%) while generating decision rationales intended to support explainability. This study contributes a structured, low-resource-compatible framework for modeling requirement-to-block reuse and outlines future directions in automated input elicitation and schematic segmentation.