What question did this study set out to answer?

To improve the efficiency of high-level synthesis design space exploration by reducing prediction errors.

March 30, 2026

AI4DSE: Leveraging Dynamic Graph Neural Networks and Large Language Models for Optimizing High-Level Synthesis Design Space Exploration

Key Points

To improve the efficiency of high-level synthesis design space exploration by reducing prediction errors.
Integrated graph neural networks with large language model-enhanced meta-heuristic algorithms.
Evaluated prediction performance against traditional models such as message-passing neural networks.
Measured reductions in various metrics such as power consumption and critical path delay.
ECoGNN achieved a 57.27% reduction in prediction error for post-HLS tasks.
Results showed decreases in flip-flop usage by 17.6% and critical path delay by 33.7%.
Generated superior Pareto fronts, improving average distance from reference sets by 87.77% compared to existing methods.

Abstract

High-Level Synthesis (HLS) Design Space Exploration (DSE) is essential for generating hardware designs that balance performance, power, and area (PPA). To optimize this process, existing works often employs message-passing neural networks (MPNNs) to predict quality of results (QoR). These predictors serve as evaluators in the DSE process, effectively bypassing the time-consuming estimations traditionally required by HLS tools. However, existing models based on MPNNs struggle with over-smoothing and limited expressiveness. Additionally, while meta-heuristic algorithms are widely used in DSE, they typically require extensive domain-specific knowledge to design operators and time-consuming tuning. To address these limitations, we propose ECoGNNs-LLMMHs, a framework that integrates graph neural networks with task-adaptive message passing and large language model-enhanced meta-heuristic algorithms. Compared with state-of-the-art works, ECoGNN exhibits lower prediction error in the post-HLS prediction task, with the error reduced by 57. 27%. For post-implementation prediction tasks, ECoGNN demonstrates the lowest prediction errors, with average reductions of 17. 6% for flip-flop (FF) usage, 33. 7% for critical path (CP) delay, 26. 3% for power consumption, 38. 3% for digital signal processor (DSP) utilization, and 40. 8% for BRAM usage. LLMMH variants can generate superior Pareto fronts compared to meta-heuristic algorithms in terms of average distance from the reference set (ADRS) with average improvements of 87. 77%, respectively. Compared with the SOTA DSE approaches GNN-DSE and IRONMAN-PRO, LLMMH can reduce the ADRS by 68. 17% and 61. 53% respectively. Code and models are available at https: //github. com/wslcccc/CoGNNsLLMMH.

Bookmark

Cite This Study

Xu et al. (Sat,) studied this question.

synapsesocial.com/papers/69c9c51bf8fdd13afe0bd02e https://doi.org/https://doi.org/10.1145/3805806

Bookmark