AUTOGATE: Automated Clock Gating via Toggling-Aware LLM-based RTL Rewriting

2026-06-16 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Hardware Architecture · Depth: Expert, quick

Summary

AUTOGATE is the first agentic framework designed for industry-grade Register-Transfer Level (RTL) power optimization, specifically targeting fine-grain clock gating (FGCG). It addresses the limitations of current manual FGCG optimization and prior LLM-based approaches, which struggle with long waveform traces and scaling to large hierarchical codebases while maintaining correctness. AUTOGATE employs an ML-LLM co-design, utilizing an ML-based clustering algorithm to distill raw toggling traces into compact representations that guide LLM-based RTL rewriting, avoiding direct LLM processing of raw waveform data. For scalability, it uses a hierarchical multi-agent architecture to decompose designs into optimizable modules. Evaluations show AUTOGATE reduces dynamic power by 49.31% on average for small designs, 19.34% on NVDLA, 7.96% on BlackParrot, and up to 6.86% on highly optimized proprietary production designs.

Key takeaway

For AI Hardware Engineers optimizing power in complex RTL designs, AUTOGATE offers a significant advancement over manual or prior LLM-based methods. You should consider integrating this ML-LLM co-design approach to overcome challenges with long waveform traces and hierarchical scalability. This framework demonstrates substantial dynamic power reductions, making it a viable solution for improving energy efficiency in your next-generation silicon.

Key insights

AUTOGATE automates fine-grain clock gating for RTL power reduction using an ML-LLM co-design and hierarchical multi-agent architecture.

Principles

• Distill complex data for LLM guidance.
• Decompose large designs hierarchically.
• Combine ML and LLMs for specialized tasks.

Method

AUTOGATE uses ML clustering to condense toggling traces, guiding LLM-based RTL rewriting. A hierarchical multi-agent system then decomposes and coordinates optimization across design hierarchies.

In practice

• Apply ML clustering to waveform data.
• Implement multi-agent design decomposition.
• Integrate LLMs for RTL rewriting tasks.

Topics

• Fine-Grain Clock Gating
• RTL Power Optimization
• LLM-based Design
• Multi-Agent Systems
• Hardware Architecture
• Machine Learning Clustering

Best for: Research Scientist, AI Scientist, AI Hardware Engineer, Machine Learning Engineer

Related on AIssential

• Startup Wants to Speed Up Chip Design With AI in RTL to GDSII Path — AI can dramatically accelerate chip physical design by understanding circuits from first principles and automating complex reasoning tasks.
• Exploring LLM-based Verilog Code Generation with Data-Efficient Fine-Tuning and Testbench Automation — Automated testbench generation enables data-efficient LLM fine-tuning for high-quality Verilog code generation.
• Embedded Arena: Iterative Optimization via Hardware Feedback — Hardware-in-the-loop LLM agents can autonomously optimize embedded AI models, surpassing manual and feedback-free methods.
• Google's TurboQuant Crashed the AI Chip Market — Google's Turbo Quant algorithm offers 6x memory reduction and 8x speedup for LLM inference with zero accuracy loss.
• ChatNeuroSim: An LLM Agent Framework for Automated Compute-in-Memory Accelerator Deployment and Optimization — ChatNeuroSim automates CIM accelerator design and optimization using LLM agents and design space pruning for faster, more efficient exploration.
• AutoVecCoder: Teaching LLMs to Generate Explicitly Vectorized Code — AutoVecCoder enables LLMs to generate high-performance, explicitly vectorized code by combining data synthesis and reinforcement learning.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.