Large language models (LLMs) are emerging as powerful tools for hardware design, with recent work exploring their ability to generate register-transfer level (RTL) code directly from natural-language specifications. This paper provides a survey and evaluation of LLM-based RTL generation. We review twenty-six published efforts, covering techniques such as fine-tuning, reinforcement learning, retrieval-augmented prompting, and multi-agent orchestration, and we analyze their contributions across eight methodological dimensions including debugging support, post-RTL metrics, and benchmark development. Building on this review, we experimentally evaluate frontier commercial models---GPT-4.1, GPT-4.1-mini, and Claude Sonnet 4---on the VerilogEval and RTLLM benchmarks under both single-shot and lightweight agentic settings. Results show that these models achieve up to 89.74% on VerilogEval and 96.08% on RTLLM, matching or exceeding prior domain-specific pipelines without specialized fine-tuning. Detailed failure analysis reveals systematic error modes, including FSM mis-sequencing, handshake drift, blocking vs. non-blocking misuse, and state-space oversimplification. Finally, we outline a forward-looking research roadmap toward natural-language-to-SoC design, emphasizing controlled specification schemas, open benchmarks and flows, PPA-in-the-loop feedback, and modular assurance frameworks. Together, this work provides both a critical synthesis of recent advances and a baseline evaluation of frontier LLMs, highlighting opportunities and challenges in moving toward AI-native electronic design automation.
Ravindran et al. (2025) studied this question.