Cold Plate Assembly with Thermal Barrier for Stacked HBM Cooling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional cooling solutions for computing devices, particularly those with vertically stacked high-bandwidth memory (HBM), struggle to efficiently cool each layer of the stack due to the stacked configuration, leading to inadequate cooling and potential overheating.
Innovation Solution
A cold plate assembly with separate cooling structures and coolant loops for different temperature sections of a computing device, where a first cooling structure provides cooling to a processor at a higher temperature, and a second cooling structure, connected via a thermal barrier, provides cooling to memory components at a lower temperature.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a single cold plate is used to cool all components, then the processing components can be cooled, but the memory components cannot be adequately cooled due to the stacked configuration
Solution Approach 1:
The cold plate is divided into multiple independent cooling zones (first cooling zone, second cooling zone, third cooling zone) that can be independently controlled. Each zone has its own temperature control capability, allowing different temperatures to be applied to different components (processor, HBM memory layers) simultaneously. This segmentation resolves the contradiction by enabling both adequate cooling of all components and flexible temperature control.
Solution Approach 2:
Different regions of the cold plate are designed with different thermal characteristics to match the specific cooling requirements of underlying components. The first cooling zone contacts the processor with specific thermal properties, while the second and third cooling zones contact HBM memory layers with different thermal properties. This local differentiation allows each component to receive optimized cooling appropriate to its thermal requirements.
2Reliability
If the cold plate is cooled to the restrictive temperature of HBM components, then HBM can be cooled adequately, but processing components are overcooled resulting in unnecessary energy expenditure
Solution Approach 1:
The temperature control system is segmented into independent zones with separate temperature control. The first cooling zone can maintain a higher temperature suitable for the processor, while the second and third cooling zones maintain lower temperatures for HBM components. This prevents unnecessary overcooling of the processor and reduces overall energy consumption while still providing adequate cooling to memory components.
Solution Approach 2:
Each cooling zone is optimized for its specific component's thermal requirements. The processor-contacting zone operates at a temperature appropriate for CPU/GPU cooling, while the memory-contacting zones operate at lower temperatures. This localized temperature optimization eliminates the waste of cooling high-power components to temperatures far below their requirements.
3Reliability
If coolant flows over memory components first before flowing to the processor, then memory cooling is prioritized, but memory components still experience inadequate cooling leading to premature wear
Solution Approach 1:
The cooling system uses separate cooling zones with independent temperature control rather than a single sequential cooling path. The second and third cooling zones are dedicated to HBM memory layers and can maintain lower temperatures independently of the processor cooling zone. This ensures adequate cooling to memory components throughout their operational life, preventing premature wear and extending lifespan.
Solution Approach 2:
The cold plate design provides localized optimized cooling contact with each HBM layer through separate cooling zones. Each zone is thermally optimized for memory component cooling, ensuring consistent and adequate heat removal from all memory layers. This prevents thermal stress and degradation that would otherwise shorten memory component lifespan.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This solution enables tailored cooling for different components of a computing device, optimizing temperature control and reducing energy expenditure by ensuring each component is cooled to its specific thermal requirements.
Implementation Method 1
The cold plate assembly further includes a thermal barrier between the first cooling structure and the second cooling structure
Implementation Method 2
The first cooling structures is configured to provide cooling from a flow of first coolant to a first temperature section of the computing device
Data Source
AI summary
A cold plate assembly for cooling a computing device includes a first cooling structure and a second cooling structure. The first cooling structure is configured to provide cooling from a flow of first coolant to a first temperature section of the computing device. The second cooling structure is connected to the first cooling structure and is configured to provide cooling from a flow of second coolant to a second temperature section of the computing device. The second temperature section has a lower temperature threshold than the first temperature section. The cold plate assembly includes a thermal barrier between the first cooling structure and the second cooling structure.