Search for: All records

Abstract The growing demand for higher computational efficiency in data centers and AI hardware is accelerating the adoption of three dimensional (3D) heterogeneous integration and immersion-cooling technologies. Oil immersion offers significantly enhanced heat dissipation compared to conventional air cooling, improving thermal stability and enabling higher power densities. However, the thermo-mechanical environment created by oil immersion also introduces new reliability challenges. In 3D packages, mismatches in material properties like Coefficient of Thermal Expansion (CTE) and Elastic Modulus (E) can induce interfacial stresses that lead to through-silicon via (TSV) cracking, printed circuit board (PCB) warpage, and solder-joint fatigue failures. This study investigates the influence of oil-immersion cooling on the thermomechanical reliability of solder joints and TSV structures in advanced 3D packages. Plastic strain accumulation in solder joints under extreme thermal cycling is analyzed to evaluate the likelihood of crack initiation and propagation. Additionally, J-integral computations are performed to quantify crack-tip driving forces in TSVs across different PCB configurations, providing a comprehensive assessment of package lifetime reliability. To further enhance TSV resilience, this work examines alternative via-fill materials—including tungsten, silver, nickel, carbon nanotubes, aluminum, molybdenum, gold, graphene, and annealed copper which are possible alternatives—and compares their crack-propagation tendencies using J-integral-based metrics. Overall, this research not only identifies reliability risks associated with oil-immersion environments but also proposes material-level mitigation strategies. 

Abstract The rapid increase in computational requirements of modern computing systems has led to exponentially increasing thermal design powers (TDPs) from the processors. Traditional air-cooling methods have reached their limits, being unable to handle the component level heat fluxes and rack level power densities of next generation data centers. A transition to single-phase immersion cooling is an effective strategy to properly manage the heat fluxes from the next generation processors, all while offering improved energy efficiency, decreased water usage, and less datacenter whitespace. To fully embrace the benefits of immersion cooling, the heatsinks used on the processors must be well designed and optimized, necessitating rapid and robust performance models. This study presents the use of neural networks to develop models that predict performance characteristics of heatsinks in single-phase immersion cooling. The networks are trained on a synthetic dataset of 15,552 points, taking inputs of immersion fluid, flow conditions, and heatsink geometry. An analysis of the various architectural and training conditions is conducted, yielding networks with enhanced validation set accuracy. When inferenced in generalization, the networks exhibit accurate predictive capabilities (4.084% MAPE) with an average inference time of 3.685 µsec. The networks are deployed in multi-variable, multi-objective optimizations, completing various 10,000-point optimization analyses in under three seconds (average 2.21s), with only a 3.9% difference from CFD optimization. A time saving analysis is performed demonstrating the extreme order of magnitude reductions in design turn around time possible. 

Abstract In the modern semiconductor industry, Silicon photonics stands at the forefront as an emerging interdisciplinary department that seeks to combine photonics and electronics on a singular silicon substrate. This combination of conventional electronic circuits and the rapid data transmission capabilities of optical communication has attracted substantial attention due to its potential to transform data communication, computing, sensing and more. Silicon photonics attempts to convert electronic signals into optical signals, facilitating data transmission at remarkable velocities. This is achieved by leveraging photonic components such as waveguides, modulators, detectors, and multiplexers all meticulously engineered using conventional silicon fabrication methodologies. The major component in silicon photonics is the silicon waveguide, a small structure that steers light, enabling its propagation with minimal attenuation. III-V materials such as Gallium Arsenide (GaAs) and Indium Phosphide (InP) are used as substrates of opto-electronic devices like vertical cavity surface emitting lasers (VCSEL), edge emitting lasers and light emitting diodes. System in Package integration of the III-V materials to silicon requires a reliable bonding technique. The present study proposes four different interconnect design namely Annular, semi-annular, flat pad and solder balls between GaAs substrate and SOI wafer. A finite element model of the package is created using Ansys and analyzed using Ansys Thermal module. Different designs are evaluated for their thermal and reliability performances in the GaAs/Si integrated circuits. The advantages and short comings of each interconnect design is discussed and suggestions are made about the affinity of each configuration for different applications 

Abstract Flushing is a crucial step performed on the cooling equipment utilized in single phase liquid cooled data centers. This process eliminates closed-loop contamination for long-term reliability. However, it is crucial to remember that total coolant removal is not feasible during the draining process because residual quantities are always left behind. Fresh coolant in data centers could be contaminated in the loop with these residuals. Bioproliferation from cross-contamination of liquid-cooled data center coolants affects reliability and efficiency. Bacterial in the coolant can cause fouling, corrosion, and inadequate heat transfer. Two tests are carried out mimicking the data center scenarios to study the formation of biogrowth. In the first test, a coupon set comprising data center TCS loop wetted materials is submerged in a known concentration of cross-contaminated solution for an extended period. Microbial growth will be analyzed by sampling every thirty days. This test replicates stagnant cooling loops and data center equipment kept in storage rooms after flushing is performed. In the second test, two separate cooling loops with a microfin channeled cold plate mounted on a heater base are utilized. These loops undergo a flushing procedure with type-II DI water and Treated water. Following the flushing procedure, a fresh coolant is filled and kept circulating within the loops. Microbial growth will be analyzed by sampling every ten days. The results highlight the need for routine bio-analysis and the adoption of efficient microbial control measures in liquid-cooled data centers. 

Abstract As the need for high-performance computing grows, data center operators have the challenge of regulating the increasing heat densities produced by current information technology equipment. Conventional air-based cooling systems are inadequate for dissipating the heat generated by high-power servers, prompting the exploration of alternative cooling techniques. The implementation of liquid cooling, particularly liquid-to-liquid cooling systems, has been identified as a potentially advantageous solution to addressing these issues. There are four distinct forms of liquid cooling technologies used in data centers. These include single-phase immersion cooling, two-phase immersion cooling, direct-to-chip liquid cooling employing single-phase microfin routed cold plates, and direct-to-chip liquid cooling employing two-phase cold plates. The coolant distribution unit (CDU) is an essential component of the liquid cooling system used in direct-to-chip liquid cooling technology. The CDU comprises a heat exchanger, circulating pumps, coolant monitoring sensors, and a programable logic controller that receives inputs and feedback from all coolant monitoring devices. The major function of the CDU is to circulate the fluid, dissipate heat, and regulate the coolant's supply temperature at a specific flowrate within the system. Alternatively, the CDU is tasked with regulating the temperature of the refrigerant supply. Defining the desired optimal efficiency level in real-world cases, in terms of percentage efficiency, would be more effective to ensure that the CDU performs at the desired level after installation. This article aims to outline several tests conducted on a CDU to evaluate its functionality and verify its proper operation. A coolant distributing unit is examined with a closed-loop cooling system. The cooling system consists of two row manifolds, eight rack manifolds, and 32 cooling loops equipped on each of the 32 Thermal test vehicles. Thermal energy may be used to power the TTVersus to a maximum of 200 KW. A suggested cleaning technique is followed to commission the CDU with the coolants at the start of the process. The tests involve operating the CDU in a variety of modes, such as pressure differential mode and flow control mode, at a range of coolant inlet temperatures that are suggested by the American Society of Heating, Refrigerating and Air-Conditioning Engineers (ASHRAE). An analysis is conducted on several real-time case scenarios, such as pump failure, low load operations, high load operations, and the response time of leak detection integrated into the CDU system. Accrediting liquid-to-liquid in-row coolant distribution systems is crucial to ensure the reliable and efficient functioning of high-power data centers. These advanced cooling technologies enhance data center performance, energy efficiency, and environmental sustainability in the current era of demanding computer workloads. They achieve this by effectively addressing thermal management-related challenges. 

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 flowrate 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. 

Abstract This paper proposes a computational fluid dynamics (CFD) simulation methodology for the multi-design variable optimization of heat sinks for natural convection single-phase immersion cooling of high power-density Data Center server electronics. Immersion cooling provides the capability to cool higher power-densities than air cooling. Due to this, retrofitting Data Center servers initially designed for air-cooling for immersion cooling is of interest. A common area of improvement is in optimizing the air-cooled component heat sinks for the fluid and thermal properties of liquid cooling dielectric fluids. Current heat sink optimization methodologies for immersion cooling demonstrated within the literature rely on a server-level optimization approach. This paper proposes a server-agnostic approach to immersion cooling heat sink optimization by developing a heat sink-level CFD to generate a dataset of optimized heat sinks for a range of variable input parameters: inlet fluid temperature, power dissipation, fin thickness, and number of fins. The objective function of optimization is minimizing heat sink thermal resistance. This research demonstrates an effective modeling and optimization approach for heat sinks. The optimized heat sink designs exhibit improved cooling performance and reduced pressure drop compared to traditional heat sink designs. This study also shows the importance of considering multiple design variables in the heat sink optimization process and extends immersion heat sink optimization beyond server-dependent solutions. The proposed approach can also be extended to other cooling techniques and applications, where optimizing the design variables of heat sinks can improve cooling performance and reduce energy consumption. 

Abstract To fulfill the increasing demands of data storage and data processing within modern data centers, a corresponding increase in server performance is necessary. This leads to a subsequent increase in power consumption and heat generation in the servers due to high performance processing units. Currently, air cooling is the most widely used thermal management technique in data centers, but it has started to reach its limitations in cooling of high-power density packaging. Therefore, industries utilizing data centers are looking to singlephase immersion cooling using various dielectric fluids to reduce the operational and cooling costs by enhancing the thermal management of servers. In this study, heat sinks with TPMS lattice structures were designed for application in singlephase immersion cooling of data center servers. These designs are made possible by Electrochemical Additive Manufacturing (ECAM) technology due to their complex topologies. The ECAM process allows for generation of complex heat sink geometries never before possible using traditional manufacturing processes. Geometric complexities including amorphous and porous structures with high surface area to volume ratio enable ECAM heat sinks to have superior heat transfer properties. Our objective is to compare various heat sink geometries by minimizing chip junction temperature in a single-phase immersion cooling setup for natural convection flow regimes. Computational fluid dynamics in ANSYS Fluent is utilized to compare the ECAM heat sink designs. The additively manufactured heat sink designs are evaluated by comparing their thermal performance under natural convection conditions. This study presents a novel approach to heat sink design and bolsters the capability of ECAM-produced heat sinks. 

Abstract Data centers have complex environments that undergo constant changes due to fluctuations in IT load, commissioning and decommissioning of IT equipment, heterogeneous rack architectures and varying environmental conditions. These dynamic factors often pose challenges in effectively provisioning cooling systems, resulting in higher energy consumption. To address this issue, it is crucial to consider data center thermal heterogeneity when allocating workloads and controlling cooling, as it can impact operational efficiency. Computational Fluid Dynamics (CFD) models are used to simulate data center heterogeneity and analyze the impact of two different cooling mechanisms on operational efficiency. This research focuses on comparing the cooling based on facility water for Rear Door Heat Exchanger (RDHx) and conventional Computer Room Air Conditioning (CRAH) systems in two different data center configurations. Efficiency is measured in terms of ΔT across facility water. Higher ΔT will result in efficient operation of chillers. The actual chiller efficiency is not calculated as it would depend on local ambient conditions in which the chiller is operated. The first data center model represents a typical enterpriselevel configuration where all servers and racks have homogeneous IT power. The second model represents a colocation facility where server/rack power configurations are randomly distributed. These models predict temperature variations at different locations based on IT workload and cooling parameters. Traditionally, CRAH configurations are selected based on total IT power consumption, rack power density, and required cooling capacity for the entire data center space. On the other hand, RDHx can be scaled based on individual rack power density, offering localized cooling advantages. Multiple workload distribution scenarios were simulated for both CRAH and RDHx-based data center models. The results showed that RDHx provides a uniform thermal profile across the data center, irrespective of server/rack power density or workload distribution. This characteristic reduces the risk of over- or under-provisioning racks when using RDHx. Operational efficiency is compared in terms of difference in supply and return temperature of facility water for CRAH and RDHx units based on spatial heat dissipation and workload distribution. RDHx demonstrated excellent cooling capabilities while maintaining a higher ΔT, resulting in reduced cooling energy consumption, operational carbon footprint (?), and water usage. 

Abstract The increasing demand for high-performance computing in applications such as the Internet of Things, Deep Learning, Big data for crypto-mining, virtual reality, healthcare research on genomic sequencing, cancer treatment, etc. have led to the growth of hyperscale data centers. To meet the cooling energy demands of HPC datacenters efficient cooling technologies must be adopted. Traditional air cooling, direct-to-chip liquid cooling, and immersion are some of those methods. Among all, Liquid cooling is superior compared to various air-cooling methods in terms of energy consumption. Direct on-chip cooling using cold plate technology is one such method used in removing heat from high-power electronic components such as CPUs and GPUs in a broader sense. Over the years Thermal Design Power (TDP) is rapidly increasing and will continue to increase in the coming years for not only CPUs and GPUs but also associated electronic components like DRAMs, Platform Control Hub (PCH), and other I/O chipsets on a typical server board. Therefore, unlike air hybrid cooling which uses liquid for cold plates and air as the secondary medium of cooling the associated electronics, we foresee using immersion-based fluids to cool the rest of the electronics in the server. The broader focus of this research is to study the effects of adopting immersion cooling, with integrated cold plates for high-performance systems. Although there are several other factors involved in the study, the focus of this paper will be the optimization of cold plate microchannels for immersion-based fluids in an immersion-cooled environment. Since immersion fluids are dielectric and the fluids used in cold plates are conductive, it exposes us to a major risk of leakage into the tank and short-circuiting the electronics. Therefore, we propose using the immersed fluid to pump into the cold plate. However, it leads to a suspicion of poor thermal performance and associated pumping power due to the difference in viscosity and other fluid properties. To address the thermal and flow performance, the objective is to optimize the cold plate microchannel fin parameters based on thermal and flow performance by evaluating thermal resistance and pressure drop across the cold plate. The detailed CFD model and optimization of the cold plate were done using Ansys Icepak and Ansys OptiSLang respectively.