Therm-FM datasets
收藏资源简介:
Therm-FM数据集由东方理工学院等研究机构联合创建,旨在支持三维集成电路热模拟的深度学习研究。该数据集包含公开HotSpot基准测试和工业级3D-IC封装案例,涵盖不同芯片结构、空间分辨率及稳态/瞬态热分析任务,数据通过高保真有限元仿真和热等效模型生成,有效降低了数据获取成本。数据集通过多保真度训练策略构建,融合了低成本近似仿真与稀疏高保真样本,为热场预测提供了多尺度物理特征。该数据集主要应用于芯片热管理优化领域,解决了传统热模拟方法计算成本高、跨设计迁移能力弱的核心问题,为基于预训练物理先验的神经网络算子提供了标准化评估基准。
Therm-FM dataset was co-developed by the Oriental Institute of Technology and other research institutions, aiming to support deep learning research for thermal simulation of three-dimensional integrated circuits (3D ICs). This dataset includes publicly available HotSpot benchmark tests and industrial-grade 3D-IC packaging cases, covering diverse chip architectures, spatial resolutions, and steady-state/transient thermal analysis tasks. The data is generated through high-fidelity finite element simulation and thermal equivalent models, effectively lowering the cost of data acquisition. Constructed via a multi-fidelity training strategy, the dataset integrates low-cost approximate simulations and sparse high-fidelity samples, providing multi-scale physical features for thermal field prediction. Primarily applied in the field of chip thermal management optimization, this dataset addresses the core problems of traditional thermal simulation methods, including high computational costs and poor cross-design transferability, and serves as a standardized evaluation benchmark for neural network operators based on pre-trained physical priors.
Therm-FM 数据集概述
基本信息
- 数据集地址:https://github.com/haiyangxin/Therm-FM
数据集描述
该数据集目前仅提供了名称“Therm-FM”,暂无其他详细信息。具体内容有待仓库作者进一步补充。

- 1Therm-FM: Foundation Model is ALL YOU NEED for 3D-ICs Thermal Simulation中国科学技术大学·计算机科学与技术学院; 东方理工学院; 清华大学; 谢菲尔德大学; 东方理工学院·宁波数字孪生研究院 · 2026年



