Physical feature-based machine learning of BEOL thermal conductivity in 3D ICs

doi:10.1016/j.mejo.2025.107024

CSpace

	Physical feature-based machine learning of BEOL thermal conductivity in 3D ICs
	Liu, Yunting 1,2,3; Fu, Rong 1,2,3; Zhu, Jixiang 4; Zhang, Kun 4; Chen, Chuan 1,2,3; Li, Jun 1,2,3; Cao, Liqiang 1,2,3
	2026-02-01
发表期刊	MICROELECTRONICS JOURNAL
ISSN	0959-8324
卷号	168 页码:11
摘要	Accurate prediction of effective thermal conductivity in back-end-of-line (BEOL) stacks is crucial for 3D memory-on-logic design. We propose a physics-informed machine-learning framework that converts GDSII layouts into representative volume elements (RVEs), labels their directional conductivities using a matrix-based finite-volume solver with preconditioned conjugate gradients (PCG), and learns topology-conductivity mappings via a 50-dimensional hybrid embedding (HYB) combining 25 physics-informed (PH) descriptors and 25 principal components (PCs). The PCG solver reproduces finite-element results within 4 % deviation while reducing labeling time by over 90 %. Across independent validation regions, LS-Boost achieves approximate to 5 % mean directional MAPE and Ridge <10 %, with PH features showing dominant importance over PCs. The trained models generalize well to unseen layouts and modules, maintaining anisotropic trends and spatial fidelity. Full-window inference over a 540 x 540 mu m(2) BEOL block completes within minutes, yielding interpretable, direction-resolved conductivity maps for fast and physically consistent thermal analysis of stacked systems.
关键词	3D integrated circuits Advanced back-end-of-line interconnects Anisotropic thermal conductivity Thermal management
DOI	10.1016/j.mejo.2025.107024
收录类别	SCI
语种	英语
WOS研究方向	Engineering ; Science & Technology - Other Topics
WOS类目	Engineering, Electrical & Electronic ; Nanoscience & Nanotechnology
WOS记录号	WOS:001647992700002
出版者	ELSEVIER SCI LTD
引用统计
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/42964
专题	中国科学院计算技术研究所
通讯作者	Chen, Chuan; Li, Jun
作者单位	1.Univ Chinese Acad Sci, Sch Integrated Circuits, Beijing 100049, Peoples R China 2.Chinese Acad Sci, Key Lab Fabricat Technol Integrated Circuits, Beijing 100029, Peoples R China 3.Chinese Acad Sci, Inst Microelect, Microsyst Packaging Res Ctr, Beijing 100029, Peoples R China 4.Chinese Acad Sci, Inst Comp Technol, Res Ctr Intelligent Comp Syst, SKLP, Beijing 100029, Peoples R China
推荐引用方式 GB/T 7714	Liu, Yunting,Fu, Rong,Zhu, Jixiang,et al. Physical feature-based machine learning of BEOL thermal conductivity in 3D ICs[J]. MICROELECTRONICS JOURNAL,2026,168:11.
APA	Liu, Yunting.,Fu, Rong.,Zhu, Jixiang.,Zhang, Kun.,Chen, Chuan.,...&Cao, Liqiang.(2026).Physical feature-based machine learning of BEOL thermal conductivity in 3D ICs.MICROELECTRONICS JOURNAL,168,11.
MLA	Liu, Yunting,et al."Physical feature-based machine learning of BEOL thermal conductivity in 3D ICs".MICROELECTRONICS JOURNAL 168(2026):11.