节点文献

多芯片电子系统的内嵌式微流散热关键技术研究

Study on the Key Technology of Embedded Microfluidic Heat Dissipation for Multi-chip Electronic Systems

【作者】 张楠

【导师】 段智勇; 焦斌斌;

【作者基本信息】 郑州大学 , 物理电子学, 2022, 硕士

【摘要】 随着电子系统集成度的不断提升以及对高运算性能的追求,致使电子芯片的工作频率与热耗散功率同步增加,热流密度已经超过1000 W/cm~2,电子芯片的散热已经成为制约其性能输出的瓶颈之一。近期,随着具有优良性能多芯片组件系统的广泛应用,进一步加剧了电子系统的散热难度。例如,AMD公司的服务器芯片AMD EPYC 7002采用64核128线程,最大工作频率为3.4 GHz,热功耗达到了280 W。特斯拉的AI芯片Dojo由25个芯片组成,热功耗已经达到了10 k W。传统多芯片组件系统散热是一种基于间接式的热管理技术,如空气冷却、间接液体冷却,其散热能力被限制在了100 W/cm~2左右。这些方法都存在以下几点问题:(1)电子芯片通过热界面材料与外置散热器粘接成一体,严重限制了散热器与热源之间的热传递能力。(2)针对不同电子系统的散热需求进行定制化的散热系统设计,以至于无法采用一种通用的散热方案实现热管理,实际应用效率低。(3)在整个液冷循环系统中,若一体化模块中某段流道出现阻塞或损坏,则会对整个电子系统造成不可逆的损伤。因此,亟待开发一种针对多芯片组件系统具有高效换热能力、可更换性、可配置性的散热方法。内嵌式冷却是针对高热流密度芯片一种非常具有吸引力的散热方法,与传统的间接式散热方法相比,突破了散热器与电子芯片之间热界面材料层的热阻限制,并且具有集成化、小型化、高效率散热等优点。本文对具有多个高功率芯片电子系统的散热需求进行了分析,并针对此需求进行了相关散热研究,创新性的提出通过将单个芯片的内嵌式歧管微流道散热方法与拼图式组装方式相结合,从而实现了一种可用于多芯片组件系统并且兼具可配置性和可更换性的内嵌式冷却方法。首先,本文针对独立的高热流密度芯片开展了内嵌式歧管微流道散热研究。采用液冷循环测试平台对换热结构的热工水力性能进行实验评估。结果显示,在器件热流密度和冷却液流量分别为1200 W/cm~2和400 m L/min的条件下,集成有8×-50(歧管分液流道数量为8个;内嵌式微流道宽度为50μm)冷却结构的热测试芯片表面温度均匀性(阵列温度标准差)和最大温度能够被控制在1.88℃和69℃以内,从而验证了内嵌式歧管微流道散热方法能够实现对高热流密度芯片的有效热管理。针对多芯片组件系统的散热特性研究,采用尺寸为70mm×88 mm×9 mm的内嵌式冷却模块对3×3芯片阵列进行散热以验证多芯片内嵌式冷却方法的可行性。实验结果表明,3×3芯片阵列中每个芯片被独立冷却,彼此之间没有热耦合干扰。在400 W/cm~2和500 m L/min的条件下,温度均匀性为2.59℃,芯片阵列中最大芯片温度低于58.3℃。在相同测试条件下,探究了三种不同排列方式芯片阵列的散热特性。结果显示,冷却模块之间的压降差异和芯片温度的最大波动范围分别小于1.6 k Pa和5.58℃,验证了冷却模块可以满足不同排布方式的芯片阵列散热需求。此外,冷却模块中的单个芯片经过反复组装拆卸后散热性能保持不变,验证了该冷却方法的可更换性。因此,本论文提出的针对多芯片组件系统的冷却方法具备未来应用于高性能处理器、相控阵雷达阵列、AI运算组件以及DC/DC电源转换器等系统中的前景。

【Abstract】 With the continuous improvement of electronic system integration and the pursuit of high computing performance,the operating frequency of the chip and the heat dissipation power have increased simultaneously,and the heat flux density has exceeded 1000 W/cm~2.The heat dissipation of electronic chips has become one of the bottlenecks restricting its performance output.Recently,with the wide application of multi-chip component electronic systems with excellent system performance,the heat dissipation difficulty of the electronic system is further exacerbated.For example,AMD’s server chip AMD EPYC 7002 uses 64 cores and 128 threads,the maximum operating frequency is 3.4 GHz,and the thermal power consumption reaches 280 W.Tesla’s AI chip Dojo consists of 25 chips,and the thermal power consumption has reached 10 k W.The heat dissipation of the traditional multi-chip module system is based on an indirect thermal management technology,such as air cooling and indirect liquid cooling,and its heat dissipation capacity is limited to about 100 W/cm~2.These methods all have the following problems:(1)The electronic chip is bonded to an external heat sink through a thermal interface material to form an integrated module,which severely limits the heat transfer capability between the heat sink and the heat source.(2)Customize the heat dissipation system design according to the heat dissipation requirements of different electronic systems,so that a general heat dissipation scheme cannot be used to achieve thermal management,and the practical application efficiency is low.(3)In the entire liquid cooling circulation system,if a certain flow channel in the integrated module is blocked or damaged,it will cause irreversible damage to the entire electronic system.Therefore,it is urgent to develop a heat dissipation solution with efficient heat exchange capability,replaceability,and configurability for multi-chip component systems.Embedded cooling is a very attractive heat dissipation method for high heat flux chips.Compared with traditional indirect heat dissipation technology,it breaks through the thermal resistance limit of the thermal interface material layer between the heat sink and the electronic chip,and it has the advantages of integration,miniaturization,and high-efficiency heat dissipation.This paper analyzes the heat dissipation requirements of electronic systems with multiple high-power chips,and conducts related heat dissipation research for this requirement,and innovatively proposes a single chip embedded manifold micro-channel heat dissipation method and puzzle assembly.The concepts combine to realize an embedded cooling approach that is configurable and replaceable for multi-chip component systems.First of all,this thesis studies the heat dissipation of the embedded manifold micro-channel for the independent high heat flux density chip.The thermal-hydraulic performance of the heat exchange structure was experimentally evaluated using a liquid cooling cycle test platform.The results show that under the conditions of device heat flux density and working fluid flow rate of 1200 W/cm~2 and 400 m L/min,respectively,the surface temperature uniformity(array temperature standard deviation)and the maximum temperature of 8×50(number of manifold distribution channels:8;embedded microchannel width:50μm)cooling structure can be controlled within1.88°C and 69°C,which verifies that the embedded manifold micro-channel heat dissipation method can achieve effective thermal management of high heat flux chips.Aiming at the research on the heat dissipation characteristics of the multi-chip module system,an embedded cooling module with a size of 70 mm×88 mm×9 mm is used to dissipate heat for a 3×3 chip array to verify the feasibility of the multi-chip embedded cooling method.The experimental results show that each chip in the 3×3chip array is cooled independently,and there is no thermal coupling interference between each other.The temperature uniformity is 2.59℃at 400 W/cm~2 and 500m L/min,and the highest temperature of the chip array is less than 58.3℃.Under the same conditions,the effects of three different configurations of chip arrays on the heat dissipation characteristics of the cooling module were studied.The results show that the maximum pressure drop difference of the cooling system and the maximum temperature difference of the chip array are less than 1.6 k Pa and 5.58℃,respectively,which proves that the cooling module can meet the heat dissipation requirements of chip arrays with different configurations.In addition,the heat dissipation performance of a single chip in the cooling module remains unchanged after repeated assembly and disassembly,proving the replaceability of this cooling method.Therefore,the cooling method for multi-chip component systems proposed in this paper has the prospect of being applied to high-performance processors,AI computing components,and DC/DC power converters in the future.

  • 【网络出版投稿人】 郑州大学
  • 【网络出版年期】2024年 08期
  • 【分类号】TN405;TK124
节点文献中: 

本文链接的文献网络图示:

本文的引文网络