热的
电子工程
计算机科学
歪斜
输电线路
中间层
信号完整性
热阻
材料科学
印刷电路板
电气工程
工程类
物理
蚀刻(微加工)
图层(电子)
复合材料
电信
气象学
作者
Keeyoung Son,Seongguk Kim,HyunWook Park,Taein Shin,Keunwoo Kim,Minsu Kim,Boogyo Sim,Subin Kim,Gapyeol Park,Shinyoung Park,Seungtaek Jeong,Joungho Kim
出处
期刊:IEEE Transactions on Components, Packaging and Manufacturing Technology
[Institute of Electrical and Electronics Engineers]
日期:2022-09-01
卷期号:12 (9): 1542-1556
被引量:3
标识
DOI:10.1109/tcpmt.2022.3201673
摘要
In this article, we propose an embedded cooling structure with thermal transmission line (ECS-TTL) to improve thermal integrity (TI) and signal integrity (SI) of a high-bandwidth memory (HBM) module. The proposed hierarchical cooling scheme, ECS-TTL, consists of ECS for lowering thermal resistance and TTL for smoothing thermal distribution. The ECS cools the module’s integrated circuits by circulating fluid inside the computing module components. The TTL is a novel interconnection that focuses on transferring internal heat in a horizontal direction to achieve a uniform thermal distribution. The low and uniform thermal distribution achieved by the ECS-TTL can improve the temperature-dependent SI of an HBM module. We designed ECS-TTL by considering the structural reliability and SI of the HBM module based on its physical dimensions. We conducted TI verification of the HBM module by using a 3-D computational fluid dynamics solver; we evaluated the maximum temperature and the temperature uniformity of the HBM module with the proposed ECS-TTL and other cooling structures. Based on the TI evaluation, we conducted SI verification of the HBM by using a 3-D electromagnetic solver considering the operating thermal distribution; we evaluated the eye diagram and skew of the silicon interposer, through silicon via channels and a 3-D clock distribution network in the HBM. The evaluation verified that the proposed ECS-TTL can achieve a low and uniform thermal distribution while improving the eye diagram and skew of the HBM module compared to the conventional cooling structure.
科研通智能强力驱动
Strongly Powered by AbleSci AI