Data Center Cooling Research Articles

Experimental study of the performance of liquid cooling tank used for single-phase immersion cooling data center

High efficiency cooling method in data center (DC), such as single-phase immersion cooling system, has received more attention. For practical application, serval servers would be placed in a liquid cooling tank where the coolant should be evenly transferred to every server. Therefore, the performance of liquid cooling tank is curial to the operation of DC, especially for the temperature uniformity among GPUs. In this study, the liquid cooling tank was designed and the impact of inlet temperature, coolant flowrate, and the load ratio of server on the performance was experimentally conducted. Results showed that, decreasing inlet temperature of coolant and the load ratio of GPUs, and increasing the flowrate are benefit to decrease the average GPU temperature. The decrease of dynamic viscosity and the increase of thermal conductivity is also helpful to the heat exchange. When the inlet temperature increased from 20 °C to 50 °C, the outlet temperature was increased by 30 °C; however, the average GPU temperature was only increased by 21 °C. The designed liquid cooling tank showed a good temperature uniformity, where the temperature deviation was within 5 °C and the square deviation was in the range of 1.0–1.5.

Multi-objective cooling control optimization for air-liquid cooled data centers using TCN-BiGRU-Attention-based thermal prediction models

Multi-objective cooling control optimization for air-liquid cooled data centers using TCN-BiGRU-Attention-based thermal prediction models

Study on the operation strategy of the water cooling system in the single-phase immersion cooling data center considering cooling tower performance

Energy analysis from an entire systems perspective can further improve the energy efficiency of single-phase immersion cooling technology. In this study, a water cooling system for the single-phase immersion data center was developed using a combination of FLOWNEX SE and FLUENT software. Firstly, the energy consumption analysis was carried out from the perspective of cooling system, and an optimal operation strategy was identified to minimize energy consumption. Furthermore, the influence of ambient temperature and chip power on the operation strategy and energy consumption were analyzed, and the optimal operation strategy under different conditions was given. Finally, a comparison was made between the two operating modes of the water cooling system. The analysis results showed that there was an optimal operation strategy which should be adapted based on ambient temperature and chip power to reduce the power consumption. Moreover, it was noted that when the environmental temperature rises above 22 °C, the PUE of the water cooling system surpasses 1.2, indicating that the water cooling system becomes unsuitable for single-phase immersion data centers in such conditions. In addition, the excessive cooling of the chip can be avoided through the optimal operation strategy. And when the chip power was 600 W, the optimal operation strategy can reduce the PUE by 16 % and save 5.71 kW energy.

Real-time particle visualization in an electric-cooled cloud chamber using machine learning

This paper presents the development and implementation of an electric-cooled cloud chamber designed to visualize alpha and beta particle tracks. Traditional cloud chambers rely on dry ice for cooling, which can be cumbersome and impractical for extended use. Our approach employs a Peltier device coupled with a commercial CPU cooler, providing a stable, large field of view, and long-lasting cooling environment. Additionally, we have integrated a real-time visualization system utilizing OpenCV and machine learning with TensorFlow. This system accurately identifies and labels particle tracks, enhancing the educational experience. The cloud chamber operates seamlessly with consumer laptops, such as the MacBook Air with an M1 chip, making it accessible and convenient for educational purposes. Initial qualitative results demonstrate the system’s effectiveness. Future work will focus on refining detection accuracy and exploring additional applications. This advancement not only improves the usability of cloud chambers but also offers new possibilities for interactive scientific education.

Methodology to Characterize Row Manifolds for High Power Direct to Chip Liquid Cooling Data Centers

Abstract Demand is growing for the dense and high-performing IT computing capacity to support artificial intelligence, deep learning, machine learning, autonomous cars, the Internet of things, etc. This led to an unprecedented growth in transistor density for high-end CPUs and GPUs, creating thermal design power (TDP) of even more than 700 watts for some of the NVIDIA existing GPUs. Cooling these high TDP chips with air cooling comes with a cost of the higher form factor of servers and noise produced by server fans close to the permissible limit. Direct-to-chip cold plate-based liquid cooling is highly efficient and becoming more reliable as the advancement in technology is taking place. Several components are used in the liquid-cooled data centers for the deployment of cold plate based direct to chip liquid cooling like cooling loops, rack manifolds, CDUs, row manifolds, quick disconnects, flow control valves, etc. Row manifolds used in liquid cooling are used to distribute secondary coolant to the rack manifolds. Characterizing these row manifolds to understand the pressure drops and flow distribution for better data center design and energy efficiency is important. In this paper, the methodology is developed to characterize the row manifolds. Water-based coolant Propylene glycol 25% was used as the coolant for the experiments and experiments were conducted at 21 °C coolant supply temperature. Two, six-port row manifolds' P-Q curves were generated, and the value of supply pressure and the flow rate were measured at each port. The results obtained from the experiments were validated by a technique called Flow Network Modeling (FNM). FNM technique uses the overall flow and thermal characteristics to represent the behavior of individual components.

Gross Error Detection and Data Correction in IIoT-Based Data Center Cooling System

A data center always require a proper cooling system. This research study a data center with water based cooling system that consists of two chillers and two in-rack coolers. To control the system, an Industrial Internet of Things (IIoT) infrastructures has been deployed. It able to monitors real-time data from various sensors such as temperature (T), pressure (P), water flow (Q). The data were supposed to be used for optimization. However, early assessment showed that there were discrepancies between the sensors. Therefore, data reconciliation method is essential to get the valid data from the sensors. This paper discusses the implementation of gross error detection and correction by using least square method and bias compensation.

Model-based control strategy with linear parameter-varying state-space model for rack-based cooling data centers

Cooling system energy consumption occupies approximately 40% of the total energy consumption in data centers, and efficient control is crucial for improving cooling efficiency and reducing the operational costs of data centers. This paper introduced a novel economic model predictive control (EMPC) strategy based on a linear parameter-varying state-space model. The primary objective was to maintain the thermal environment of data centers while minimizing the energy consumption of the cooling system. Therein, a linear parameter-varying state-space model was developed based on the parameter identification method to precisely predict the temperature field of the rack-based cooling data center in real-time. The energy management and thermal regulation performance of the EMPC and the proportion integration differentiation (PID) controller were comparatively analyzed through simulation experiments. Moreover, the impacts of the server workload level and optimization time domain on the performance of EMPC were comprehensively investigated. The results show that, the proposed EMPC was able to minimize the energy consumption by optimizing the supplied cold air temperature and airflow rate simultaneously. In comparison with the PID controller, EMPC not only achieves a 9.7% energy saving by preventing server over-cooling but also diminishes server temperature fluctuations, promoting the safe and stable operation of the server. The research shows that the EMPC reveals excellent performance along with all server workload levels. In addition, the length of the prediction horizon has a notable impact and three minutes is suggested for the rack-based cooling system.

Life-cycle optimal design and energy benefits of centralized cooling systems for data centers concerning progressive loading

Cooling of data centers requires a significant amount of energy, comparable to the energy consumption of the servers themselves. The current design of the centralized cooling systems for data centers is based on ideal IT loading conditions (i.e., 100 % loading). However, such conventional design often results in significant oversized cooling systems and leads to substantial energy waste, since most data centers operate at part load in their lifespan. To address this issue, this study proposes an optimal design for centralized cooling systems with multiple chillers under progressive loading. The optimization problem, aimed at minimizing life-cycle cost, is formulated adopting SLSQP (Sequential Least Squares Programming) algorithm. A cooling system model is developed using the manufacturer's performance data of cooling equipment. The optimal designs in different climate zones are identified according to energy performance under full-range loads and ambient temperatures. Furthermore, this study comprehensively analyzes and compares free cooling hours, cooling energy, and life-cycle cost of the optimized designs with conventional designs. The results show that the optimized cooling systems could operate more energy-efficiently, despite decreased free cooling hours (13–860). Significant cooling energy savings over the lifespan could be achieved, i.e., 4–22 %, corresponding to the PUE reductions of 0.02–0.11, depending on climate conditions and control strategies.

Unveiling the magic of twist angle: Thermal transport in Two-dimensional Graphene/MoS2/Graphene Heterostructures

Thermal transport properties of two-dimensional vdW heterostructures have attracted considerable research interests in electron-related devices due to their powerful capability of multi-functional integration. Herein, we investigated the in-plane and out-of-plane thermal conductivity of graphene/MoS2 heterostructures under various twist angles utilizing molecular dynamics simulations. It was found that the in-plane thermal conductivity of MoS2 layer in heterostructures decreases with twist angle. Notably, the lowest thermal conductivity occurs at a twist angle of 10.893°, just 47.3 % of that in untwisted MoS2 layers in heterostructure. In addition, the interlayer interfacial thermal resistance between graphene and MoS2 layer increases with the twist angle but decreases with the layer numbers of graphene. The atomic position deviation, potential energy surface, phonon state of density are calculated to reveal the regulation mechanism of thermal conductivity by twist angle. Interestingly, it is found that the twist angles and the layer numbers of graphene can control the atomic vibration arising from potential energy surfaces and phonon transport through the overlap of phonon density of states between graphene and MoS2 layers. This study can provide valuable insights into the thermal conductivity modulation and phonon transport mechanism of heterostructures.

Wide-range thermal conductivity modulation based on protonated nickelate perovskite oxides

The perovskite oxides ReNiO3 (Re = rare-earth elements) are promising functional materials due to their strongly correlated electrons. Except for the well-known intrinsic metal-insulating transition in these materials, recent progresses have proved that protonation of ReNiO3 can bring about interesting Mott transition in this series. To date, in these protonated species (H-ReNiO3), huge resistance switching, fast ionic diffusion, and their applications as an iontronic transistor, memristor, and fuel cell are reported. In this work, the thermal conductivities of H-ReNiO3 (Re = La, Nd, Sm, and Eu) epitaxial thin films are investigated. The protonation-induced Mott transition can effectively modulate the electronic thermal conductivity while the lattice thermal conductance is less affected. Hence, at room temperature, the metallic LaNiO3 and NdNiO3 exhibit reversible wide thermal conductivity modulation, in ranges of 2.6–12.0 and 1.6–8.0 W m−1 K−1, respectively. These values are much larger than other thermal regulation materials based on transition metal oxides. Thus, our work reveals the great potential of ReNiO3 being applied as a thermal-regulating material. The fast ionic diffusion in H-ReNiO3 also guarantees that a fast response and wide-range thermal transistor can be realized by H-LaNiO3 and H-NdNiO3 in the future.

Integrating project management with advanced cooling solutions for data centers: A path to enhanced energy efficiency

Data centers play a crucial role in modern digital infrastructure, but their energy consumption and associated carbon footprint are significant concerns. Among the various energy-consuming components of data centers, cooling systems stand out as major contributors. This paper explores the integration of project management principles with advanced cooling solutions to enhance energy efficiency in data centers. By effectively managing projects and implementing innovative cooling technologies, data center operators can achieve significant reductions in energy consumption and operational costs. The integration of project management practices with advanced cooling solutions involves several key steps. First, project management methodologies such as Agile or Waterfall are applied to plan and execute the implementation of advanced cooling technologies. This includes assessing current cooling systems, identifying suitable advanced solutions, and developing a timeline for implementation. Next, advanced cooling solutions such as liquid cooling, containment systems, or free cooling are deployed to reduce the energy required for cooling data center equipment. One of the primary benefits of integrating project management with advanced cooling solutions is enhanced energy efficiency. Advanced cooling technologies can significantly reduce the energy consumed by data center cooling systems, leading to lower operational costs and a smaller carbon footprint. Additionally, the use of project management methodologies ensures that the implementation of these technologies is carried out efficiently and effectively. Furthermore, this integration can lead to improved data center performance and reliability. By optimizing cooling systems, data center operators can ensure that IT equipment operates within optimal temperature ranges, reducing the risk of overheating and downtime. Additionally, the implementation of advanced cooling solutions can future-proof data centers against increasing heat loads from high-density computing equipment. In conclusion, integrating project management with advanced cooling solutions offers a promising path to enhance energy efficiency in data centers. By adopting this approach, data center operators can achieve significant energy savings, reduce their environmental impact, and improve overall operational efficiency.

Green Data Center Cooling Control via Physics-guided Safe Reinforcement Learning

Deep reinforcement learning (DRL) has shown good performance in tackling Markov decision process (MDP) problems. As DRL optimizes a long-term reward, it is a promising approach to improving the energy efficiency of data-center cooling. However, enforcement of thermal safety constraints during DRL’s state exploration is a main challenge. The widely adopted reward-shaping approach adds negative reward when the exploratory action results in unsafety. Thus, it needs to experience sufficient unsafe states before it learns how to prevent unsafety. In this article, we propose a safety-aware DRL framework for data-center cooling control. It applies offline imitation learning and online post-hoc rectification to holistically prevent thermal unsafety during online DRL. In particular, the post-hoc rectification searches for the minimum modification to the DRL-recommended action such that the rectified action will not result in unsafety. The rectification is designed based on a thermal state transition model that is fitted using historical safe operation traces and able to extrapolate the transitions to unsafe states explored by DRL. Extensive evaluation for chilled water and direct expansion-cooled data centers in two climate conditions show that our approach saves 18% to 26.6% of total data-center power compared with conventional control and reduces safety violations by 94.5% to 99% compared with reward shaping. We also extend the proposed framework to address data centers with non-uniform temperature distributions for detailed safety considerations. The evaluation shows that our approach saves 14% power usage compared with the PID control while addressing safety compliance during the training.

Electrically regulated thermal conductivity of aramid polymer systems

Aramid polymers, renowned for their electronic insulation and thermal conductive properties, are widely adopted as thermal management materials in power electronics. However, the thermal conductivity of aramid polymers under electric field has not been thoroughly understood. In this study, we investigated the thermal conductivity of amorphous and aligned aramid polymer systems under electrical field utilizing equilibrium molecular dynamics simulations. Simulation results showed that the alignment of polymer chain can significantly enhance the thermal conductivity of aramid polymer systems, achieving up to 10.13 W/m-K. Moreover, polarization of aligned aramid polymer was observed when the applied electric field exceeded 14 V/nm. Interestingly, the thermal conductivity of aligned aramid polymer was selectively modulated by the applied electric field. To unravel the underlying phonon mechanism, the molecular orientation of polymer chains and phonon spectral information were analyzed. Our study provides guidance into understanding thermal transport mechanism and thermal conductivity modulation in polymers.

Multi-objective optimization of temperature uniformity in the immersion liquid cooling cabinet with Taguchi-based grey relational analysis

Single-phase immersion liquid cooling technology can meet the cooling needs of high-density Data Centers. To improve the temperature uniformity in an immersion liquid cooling cabinet and prevent local hot spots that reduce the efficiency of servers, the structural parameters of the cabinet were optimized in this study. The fluid path is mainly affected by the position of inlet and outlet, length, Angle and number of baffles. To evaluate the effects of these factors on the temperature uniformity in the cabinet, Taguchi-based grey relational analysis method and computational fluid dynamics were used to optimize the system, the standard deviation and maximum temperature difference are determined as evaluation indices, and L16 (44) orthogonal array was used for experimental design. The baffle Angle is judged to be the most important structural parameter by the Main Effect Analysis, whose contribution rate is 49.23%. The multi-objective optimization is transformed into a single objective by the grey relational analysis, and the influence rate of each structural parameter is shown. The optimal combination of structural parameters is determined. The results show that the temperature uniformity of the optimized cabinet is improved by 46%, which provides a reference for the design of immersion liquid cooling Data Centers.

Thermal Characterization of Subcooled Flow Boiling in a Pin-Fin Coldplate With Non-Uniform Heating

Abstract Coldplates are a crucial component in various cooling applications, such as cooling data center servers and power electronics. The unprecedented growth in electronics power density, along with the resulting ultrahigh heat fluxes, demands a transition from single-phase forced convection to two-phase flow boiling heat transfer. The majority of studies in the literature have focused on flow boiling in fin-enhanced silicon microgaps and microchannels, with only a few addressing flow boiling in millimeter-scale heat sinks. In the present study, flow boiling of HFE-7200 dielectric fluid in a millimeter-scale pin-fin coldplate is experimentally investigated under nonuniform heating conditions. Four background heaters represent the low-dissipating-power devices. On the other hand, five hotspot heaters mimic the high-heat-flux devices and generate heat fluxes ranging from 50 W/cm2 to 1000 W/cm2, corresponding to hotspot heat inputs ranging from 62.5 W to 1.25 kW, respectively. The coldplate's thermohydraulic performance is investigated for various flow rates and inlet temperature ranging from 0.5 L/min to 1.5 L/min and from 25 °C to 60 °C, respectively. A high-speed camera is utilized for a narrow field of view (FOV) flow visualization at a frame rate of 2229 fps while a digital camera is used for a wider FOV at 60 fps. Flow visualization demonstrated the transition between bubbly, slug/churn, and stratified two-phase flow regimes.

Optimization and comprehensive evaluation of liquid cooling tank for single-phase immersion cooling data center

Single-phase immersion cooling has emerged as a highly promising solution for addressing energy challenges and the substantial heat dissipation demands in data centers (DCs). The liquid cooling tank is a curial component in such system where numerous servers are immersed and the coolant are evenly flowed to each server. However, its performance and the comprehensive evaluation for the tank have not received much attention. In this study, by using the validated numerical model, the different tank structures, with or without consideration of baffle and porous panel, were firstly discussed. Then, four key parameters of tank, including inlet flowrate (Parameter A), the ratio of separation chamber height to server height (Parameter B), the ratio of pressure chamber height to server height (Parameter C), and the porosity of the porous panel (Parameter D) were explored in detail. Finally, the tank performance is evaluated and optimized by the orthogonal analysis and the entropy weight method (EWM). Results showed that, flow uniformity within the tank can be enhanced by incorporating a porous panel, and it was advisable to minimize the gaps between servers, despite the potential increase in pressure loss. Parameter A significantly impacted both the maximum temperature and pressure loss. Parameters B and C demonstrated effects on coolant weight and pressure loss, while Parameter D influenced pressure loss. By using orthogonal analysis and the EWM, maximum temperature carried the highest weight (0.362), followed by temperature difference (0.245), coolant weight (0.216), and pressure loss (0.177). The recommended values for Parameters A-D were 4 m3/h, 1.35 %, 4.05 %, and 11.33 %, respectively, which could be also considered as universality, since the comprehensive evaluation index only changed by 8 % when the server number increased from 10 to 40.

Effect of van der Waals homogeneous interface on lattice thermal conductivity of Janus WSSe bilayer

Vertically stacked 2D materials have provided an unprecedented platform to identify various physical properties and discover novel interfacial emergent phenomena. In this study, the lattice thermal conductivity of Janus WSSe bilayer, involving diverse interfacial stacking configurations, has been clarified by solving the phonon Boltzmann transport equation based on first-principles calculations. The effect of homogeneously stacking on thermal conductivity and potential modulation approaches has also been revealed for Janus WSSe. Our results indicate that the thermal transport in Janus WSSe bilayer can be suppressed by van der Waals interface, and depends on both the stacking patterns and interfacial compositions (S-S, Se-Se and Se-S interfaces). Moreover, the interface effect can be further modulated by an applied vertical pressure. A further reduction in thermal conductivity can be achieved for Janus WSSe bilayer under vertical pressure, implying great potential for efficient thermal management and sensing applications.

Energy savings in direct air-side free cooling data centers: A cross-system modeling and optimization framework

The advancement of intelligent technology has led to a significant increase in the electricity consumption of data centers. Efficient free cooling and cross-system optimization are major means to improve data center energy efficiency. This work aims to develop a cross-system modeling and optimization methodology to reduce the energy consumption within data centers equipped with direct air-side economizer. A series of physics-based models are built, accounting for free cooling costs, supply air conditions (including both temperature and humidity), and outdoor weather fluctuations. A holistic energy optimization framework is proposed to determine optimal system setpoints that accommodate changes in computing workload arrival rates and outdoor environments. Experimental results demonstrate that the optimal cost characteristic curve for free cooling can be represented as a piecewise function. The advantages of the proposed method can save up to more than 5% of overall energy, which occurs during summer weeks, and more than 46% of cooling energy during winter weeks. The highest share of cooling energy consumption is the supply air fan, reaching up to more than 92% in some situations. The operation of a direct air-side economizer is found to contain four optimal modes.

Does artificial intelligence promote energy transition and curb carbon emissions? The role of trade openness

A more comprehensive understanding of the impact of artificial intelligence (AI) on energy transition and carbon emissions could help to use AI to achieve carbon neutrality. To this end, the STIRPAT approach, the mediation effect technique and the panel threshold technique are developed using the panel data in 69 countries from 1993 to 2019. The results show that: (i) AI promote energy transition and carbon emission reduction, and trade openness (indicated by imports, exports and total trade volume) has the mediating effect. (ii) There is a single-threshold of trade openness in the impact of AI on carbon emissions. When trade openness is below the threshold, AI has an insignificant impact on carbon emissions; when trade openness crosses the threshold, AI has a significant negative impact on carbon emissions. There is a double-threshold of trade openness in the impact of AI on energy transition. When trade openness is lower than the first threshold, the impact of AI on energy transition is not significant; When trade openness is higher than the second threshold, the positive impact of AI on energy transition is increased. (iii) When considering the heterogeneity of income levels and AI levels, the trade threshold for achieving carbon emission reductions in the high-income group is lower than that of the global group, and the trade threshold for achieving carbon emission reductions in the low-AI level group is higher than that of the global group. While this study unequivocally delineates the affirmative role of artificial intelligence in carbon emission reduction and energy transformation, particularly in the context of trade openness, we concurrently acknowledge that this viewpoint is not devoid of contention. Amidst the rapid advancement of technology and the landscape of open trade, we discern the presence of counterarguments. The efficacy of artificial intelligence is susceptible to the influence of multifaceted factors. It is imperative to consider associated factors, such as the significant energy consumption required for storing and cooling data centers and servers. The study's conclusions aid policymakers in devising nuanced emission reduction policies tailored to specific needs.

Design and Optimization of Heatsink for an Active CPU Cooler using Numerical Simulations

Heat generation in CPUs with high computing performance has been a very serious issue that deteriorates their performance. To ensure that CPUs operate at their maximum potential, it is essential to maintain their temperature below 80 °C. Forced convection coolers comprising of a heatsink and fan are considered the most effective means of satisfying operating temperature requirements for a CPU in order to ensure its maximum performance. A heatsink design paired with a fan with airflow speed of 80 cubic feet per minute (CFM) for a CPU Cooler targeted at CPUs with maximum heat generation of 380 watts operating in ambient temperature of 25 °C, is developed using numerical methods of Computational Fluid Dynamics (CFD) and topology optimization using ANSYS Mechanical and ANSYS Fluent. Various fin profiles, fin arrangements, fin numbers and heatsink materials are comparatively analyzed. The best results from the comparative analyses are combined to present a base design capable of keeping the CPU temperature below 80 °C, which is the requirement for ensuring maximum computing performance. A 30 fin heatsink with covered rectangular plate fins in an arc-shaped placement configuration is determined to provide the maximum cooling performance. In materials, Silicon Carbide resulted in the lowest CPU temperature of 78°C, followed by copper at 84 °C. Silicon Carbide heatsink successfully managed to satisfy the requirements for maximum CPU performance. Copper heatsink is unlikely to cause CPU failure, but it fails to meet conditions for maximum CPU performance. In addition, this base design is then optimized for material cost reduction using topology optimization which resulted in 13% reduction in material cost with a negligible cooling performance reduction of only 0.32%. The overall design of the cooler can be improved by incorporating fan design and various CPU load conditions into the design parameters in future research.