## Abstract

This paper investigated the effect of heat spreading on the boiling of the Novec 649™ for two-phase immersion cooling of electronics. Reference pool boiling tests were performed by attaching a 25.4 mm by 25.4 mm square copper plate to a same-sized heater, thus minimizing lateral heat spreading. Experimental measurements showed that the critical heat flux (CHF) happened at a heat flux of 17.4±0.8 W/cm2. Then, lateral heat spreading through the heat spreader was studied by attaching larger (47 mm by 47 mm) spreaders with four different thicknesses to the copper plate. With an increase in the integrated heat spreader (IHS) thickness from 1 mm to 6 mm, the CHF increased by more than 60% at the saturation condition. One plate was a 1 mm-thick IHS removed from a commercial microprocessor. In this case, the CHF happens at 8.6 W/cm2 (50% lower compared to the reference case) in the saturation condition. At CHF, the boiling can be observed on the whole surface, with columns and slugs regime at the center and the fully developed nucleate boiling regime at the edges. This nonuniform boiling was more pronounced in subcooled conditions, in which the CHF occurred at the center while there were regions at the edges that had no boiling. Finally, the performance of a microporous-coated IHS (with 3.15 mm thickness) was compared to the 6 mm thick IHS. The thermal resistance was almost equal for powers above 200 W. This indicates that lateral heat spreading is a critical parameter for the thermal design of immersion cooling along with microporous coating.

## 1 Introduction

In a typical data center, the thermal management of server electronics approximately consumes a third of the electricity used by the facility [1,2]. As a consequence of global demand for computing, an increase from about 152 billion kWh/year in 2005 to 238 billion kWh/year in 2010 of global electricity usage in data centers has been observed, corresponding to roughly 1.3% of global electricity use [3]. In 2013, the reported overall data center energy usage only in the United States was 91 billion kWh/year and it is anticipated to reach 140 billion kWh/year in 2020, resulting in operating electricity rates of \$13B/year and yearly carbon emissions of 150 million metric tons [4]. Additionally, the ever-increasing power density demand (more than 300 W per socket for advanced microprocessors in year 2020 [5,6]) poses stringent limitations on the thermal performance of conventional air-cooled systems [6,7]. Despite many attempts to develop compact novel air-cooled technologies, the limitations on air-cooled are still present [6,8,9]. Due to higher heat handling capacity, liquid cooling has been used as an alternative to air-cooled electronics. As an example, Schmidt [10] has demonstrated a 3500 times higher cooling capacity with single-phase pumped liquid cooling compared to air cooling. Besides, liquid cooling can lessen noise levels due to a reduced number of fans in the server room. However, single-phase liquid cooling is generally considered to be an expensive solution with potentially hazardous failures [11], due to numerous technical problems, including biological growth, galvanic corrosion, and the electrical conductivity of water that could create damage in case of leakage. Also, as the liquid cooling components are often connected in series and parallel loops to provide energy-efficient solutions and manageable tubing layouts, additional heat could be gained by the cooling fluid while moving from a processor to the next in series, causing a higher junction temperature in the second microprocessor [5,12]. Two-phase liquid cooling with dielectric liquids has recently seen a resurgence of interest. Its advantages include better thermal performances due to the release of latent heat in boiling, and since the liquid is dielectric, a reduced risk of damage to electronic equipment in case of leakage [1318]. The two-phase liquid cooling could be passive (immersion cooling [19]) or active (pumped liquid multiphase cooling [5]). Although pumped two-phase liquid cooling has high heat handling capacity, challenges linked to flow boiling instabilities in the systems limit its use when multiple evaporators are connected to one coolant distribution unit. These evaporators could generate different heat fluxes by introducing different liquid to vapor volume ratio, and this adds more challenges for the implementation of two-phase pumped systems in the near future [16]. In addition, using pumps potentially adds cost, reliability issues and increases energy consumption in the system, relative to passive two-phase liquid cooling.

Two-phase dielectric liquid immersion cooling presents a high potential for the energy-efficient cooling of data centers [6,2023]. In an extensive review by Kheirabadi and Groulx [6] on the existing technologies for cooling data centers, it was concluded that the most attractive liquid cooling solution was open bath two-phase passive immersion cooling (also known as pool boiling). This technology provides an adaptable passive and all-liquid cooling solution with direct contact of liquid with electronics. Two issues associated with immersion cooling, coolant degassing and vapor loss, can be managed by localized pool boiling and condensation only on microprocessor in small close chamber [21]. In addition, compatibility issues of electronics material with dielectric liquids have been observed in field experiments [18,20]. Finally, additional examination of lateral heat spreading and other surface effects on boiling heat transfer on the casing of high-power electronic components appear to be essential to minimize the thermal resistance from the electronics to the liquid in pool boiling [6].

As a result, the two main obstacles to the adoption of pool boiling are material compatibility issues of the immersed electronic components with the dielectric liquids, as well as suboptimal thermal performances, both resulting from the submerged parts not having been thermally designed to work under boiling conditions [20,22,23].

To overcome these challenges, a possible approach is localized heat removal on the elements with high-density power dissipation, like central processing units and graphics processing units [24]. In this approach, the boiling heat transfer happens only on the components that generate most of the heat, and the liquid is not in contact with other parts of the electronic system that could react with the cooling liquid in the long term. As a result, the reliability risk could be much smaller compared to total immersion, but it still needs to be validated experimentally. However, it might still be necessary to redesign the integrated heat spreaders (IHSs) or casings used in air/liquid-cooled electronics for pool boiling applications with dielectric liquids, to achieve optimal thermal performances.

Heat removal by boiling over the heated surfaces comes with three specific technological challenges. The first is the boiling crisis, a phenomenon that can be described as the apparition of a vapor film between the immersed surface and the immersion liquid when reaching the critical heat flux (CHF) [23]. This limits the heat flux passing from the submerged parts to the liquid, resulting in degraded thermal performance that could lead to overheating. The second challenge is caused by the sudden increase of the heat transfer coefficient (HTC) when the liquid undergoes phase change (from the natural convection in the liquid to the boiling). The HTC change creates a temperature overshoot before the initiation of boiling, followed by a temperature drop. The third is the low pool boiling HTC for dielectric liquids on conventional, untreated electronic component surfaces that needs to be increased to keep the microprocessor junction temperature at acceptable value in high-power applications.

Some research work has addressed the aforementioned challenges by changing the boiling surface properties in order to: (1) promote the initiation of nucleate boiling at low wall superheat (the difference between the temperature of a surface and the saturation temperature of an adjacent liquid that is heated by the surface); (2) increase the HTC; and (3) delay the occurrence of CHF [25]. It has been shown that the best performance can often be obtained using microporous coatings, in which up to a 60% increase in CHF has been reported [25,26]. Arik et al. [27] found that the maximum CHF value on a diamond-base microporous-coated silicon heater reached 47 W/cm2, at an ambient pressure of 3 atm and nearly 50 K of subcooling (when a liquid existing at a temperature below its normal boiling point) with the FC-72 dielectric liquid, corresponding to a 60% CHF enhancement compared to a bare silicon surfaces. A more recent study on a surface modification for pool boiling enhancement showed that electrophoretic-deposited surfaces on copper could enhance the HTC to acetone and the HFE-7200 liquid by up to 70% and 190%, respectively, but without much improving the CHF [28]. Considering all of the surface improvement still for a high power electronic chip, having a component with an extended area to enhance boiling heat transfer is essential.

While many studies have focused on surface enhancements [25], the detailed modeling of heat fluxes in pool boiling electronic cooling has not received as much attention. The effect of heat spreading on the cooling performance of an immersed computer chip cooled by pool boiling was studied numerically by Ali and El-Genk [26] for uniform values of the nucleate boiling. These values were extracted from their experimental work on pool boiling of the PF-650 dielectric liquid on Cu microporous surfaces. In their modeling, they did not consider HTC values that might vary across the boiling surface due to heat fluxes in the plane of the boiling surface (the lateral heat spreading effect). No other literature is known to study this effect under pool boiling cooling systems. Without a detailed study of the impact of the heat spreader plate thickness supporting boiling on heat spreading, we cannot propose an optimal solution to minimize its effect on the junction-to-liquid thermal resistance.

A nickel-coated copper IHS is frequently used in microprocessor packaging to laterally spread the heat from the silicon chip to the base plate of the air-cooled heatsink. The typical thickness of the IHS (for air-cooled application) is around 1 mm. Based on the numerical simulation of a microprocessor with immersion cooling, it has been reported that by increasing the thickness of the IHS from 1 mm to 4 mm, a 30% reduction on the junction-to-liquid thermal resistance at an uniform HTC of 4000 W/m2-K could be achieved [23] This improvement is mainly attributed to the better lateral heat spreading for a thicker IHS. It is thus clear that specific thermal design guidelines might have to be adopted to optimize the performance of components operated under localized liquid immersion cooling. However, simple numerical models such as those from references [23,24] cannot precisely compute the nonuniform temperature distribution due to nonuniform HTC on boiling surface. This is first due to the complexity of the boiling phenomena and spatial variations of the HTC on the boiling surface, and second due to the coupled nature of boiling with heat spreading (will be discussed later in this paper). Therefore, the objective of this study is to understand the effects of IHS thickness and surface characteristics through heat spreading on the HTC and CHF under pool boiling in a dielectric liquid, in order to improve heat transfer models and design principles for microprocessor immersion cooling applications. To the best of our knowledge, the performance of the IHS and the role played by heat spreading under different pool boiling conditions have not been reported in the literature. We anticipate that the results of this study will allow the design of IHS optimized for immersion cooling under various boiling conditions and thermal loads.

This study will focus on describing and quantifying the different observed phenomena. Details about the physical modeling of these phenomena will be discussed in another study (numerical study), where the nonuniform HTCs and temperature distributions will be coupled in numerical simulations. In this paper, Sec. 2, we detail the experimental facility, including the test bench, heater assembly, surface characterization, and contact angle measurements, including experimental uncertainties. In Sec. 3.1, we first define the reference saturation pool boiling condition without spreading by attaching a 25.4 mm by 25.4 mm copper plate to a same-sized heater and report the measured boiling curves, HTC, CHF, and a comparison with the correlations reported in the literature. Lateral spreading is then studied in Sec. 3.2, using copper plates of four different thicknesses. The boiling performance of an IHS with a microporous coating is reported in Sec. 3.3, and the junction-to-liquid thermal resistances are compared with different bare copper IHS to demonstrate the importance of the lateral heat spreading as key design parameter even if surface of plates would be modified with surface enhancement coating. Section 4 concludes the paper.

## 2 Experimental Facilities

### 2.1 Test Bench.

This section describes the equipment used for the experiments. The setup has been built to study the pool boiling experiments of dielectric liquids by considering all the elements' compatibility with the liquid. The main components of the setup comprised two chillers, a 5 kW power supply, a data acquisition and control system, a pump, a customized boiling tank, and a cold plate on top of the tank. The boiling tank contained the low-boiling point (49 °C) dielectric fluid (3M Novec 649). The chemical properties of the dielectric liquid are presented in Table 1. Its vapor density is more than 11 times higher than air density, which enables its degassing without a vacuum system.

Table 1

Chemical properties of Novec 649 [29]

Fluoroketone—Novec 649
Molecular weight (g/mol)316
Boiling point/saturation temperature (°C)49
Freeze point (°C)–108
Vapor pressure (kPa)40
Heat of vaporization (kJ/kg)88
Liquid density (kg/m3)1600
Vapor density (Ref Std air = 1)11.6
Specific heat (J/kg K)1103
Thermal conductivity (W/m-K)0.059
Surface tension (mN/m)10.8
Dielectric constant at 1 kHz1.8
Kinematic viscosity (cSt)0.4
Fluoroketone—Novec 649
Molecular weight (g/mol)316
Boiling point/saturation temperature (°C)49
Freeze point (°C)–108
Vapor pressure (kPa)40
Heat of vaporization (kJ/kg)88
Liquid density (kg/m3)1600
Vapor density (Ref Std air = 1)11.6
Specific heat (J/kg K)1103
Thermal conductivity (W/m-K)0.059
Surface tension (mN/m)10.8
Dielectric constant at 1 kHz1.8
Kinematic viscosity (cSt)0.4

Figure 1 shows the general configuration of the test bench setup.

Fig. 1
Fig. 1

The boiling experiments were performed in a cylindrical stainless steel tank (diameter: 30 cm; height: 24 cm). The liquid level for the experiments was set at 15 cm over the boiling surface. The tank was thermally insulated from the environment using polyurethane (thermal conductivity at 20 °C: 0.03 W/m-K). To facilitate observation, five windows were installed, four circular windows on the sides and one elongated window on the front. The tank was sealed with high temperature silicone gaskets and stainless steel flanges. K-type thermocouples (Omega, M12KIN) were installed inside the tank to monitor the vapor and liquid temperatures during experiments. Figure 2 shows the different parts of the boiling tank.

Fig. 2
Fig. 2

The tank was designed to operate at atmospheric pressure, with more cooling capacity than heating capacity. A pressure gauge (OMEGA, PX409) was installed to monitor the pressure. A heat sink/condenser with cylindrical fins (Coolinovation, 3-615020 M) was installed through the top cover. Over the heat sink, a cold plate (Lytron, CP15G01) was attached with thermal paste (Arctic, MX-4, 8.5 W/m-K). The cold plate was connected to a 1 kW water-cooled chiller (Lytron, RC011J03). There was a stainless steel plate over the cold plate to apply pressure on the gasket and ensure a good thermal contact between the heat sink and the cold plate. The heat sink and the cold plate were used to condense the vapor inside the boiling tank.

A micropump (Ismatec, MCP-Z) and a liquid–liquid plate type heat exchanger (Lytron, LL510G12) were used to circulate and maintain the dielectric liquid temperature during the tests. In all the tests, the liquid temperature was controlled with an accuracy of ±1 °C. The liquid-liquid heat exchanger was connected to a 4 kW water-cooled chiller (Lytron, RC045). A cylindrical heater (Omega, CIR-2046) was immersed in the dielectric liquid to preheat the liquid to the designated temperature before each test. This heater was also used prior to each test for degassing the liquid by keeping the liquid temperature close to its saturation temperature for 1 hour. Since the vapor density of the liquid is larger than air, the tank degassing was performed by opening and closing a vent valve placed on top of the tank. After degassing, the vent was kept open during all the experiments to ensure the atmospheric pressure in the tank. The flow meters are used to ensure the repeatability of the subcooled experiment (bulk liquid temperatures less than saturation temperature) at the same operating conditions.

### 2.2 Heater Assembly.

The heater housing (diameter: 16.51 cm; height: 5 cm) comprised a ceramic heater, a ceramic holder, a spring, a copper disk, and a PTFE support. The PTFE support (0.25 W/m-K) was used as a thermal insulator and to hold the heater assembly. The heater (Watlow, CER-1-01), had an integrated K-type thermocouple. The ceramic holder was used to maintain the heater in the center of the copper disk. Thermal paste was applied between the heater and the copper disk to obtain a good thermal contact. A spring was employed below the heater to supply moderate pressure between the heater and the PTFE support. The copper disk (385 W/m-K at 20 °C) had three sections, with a total thickness of 8 mm. The bottom section was in contact with the heater and had a diameter of 5 cm. The top section, which could be in contact with the dielectric liquid, had a square shape and a surface area of 6.45 cm2 (side length: 25.4 mm). There was a high temperature silicone gasket on the middle section of the copper disk to prevent any dielectric liquid from leaking into the heater housing. Two K-type thermocouples (Omega, TJC36 sheath diameter of 0.25 mm) were installed inside the copper disk. One was in the center of the bottom section inside a groove just over the heater, and the other was placed inside the top section, 1 mm below the surface. The wall temperature values (Twall) that are reported in this study were read from the thermocouple in the top section (Fig. 3).

Fig. 3
Fig. 3

The copper disk second section was pressed against the PTFE cover. The top section of the copper disk passed through the PTFE cover. Over this cover, an IHS could be installed (uniform pressure applied in all corners via a bolted plate to have similar contact conditions in all the experiments (the amount of torque applied to the corners of the IHS was 0.8 (N-m)) as a boiling surface, or the copper surface could be directly in contact with the dielectric liquid. The heater was powered by a DC power supply (Keysight, N8921A). The heat losses through the sides, top, and bottom surfaces were calculated based on the guard heater concept in a natural convection regime and extrapolated for a higher power. It was found that this heat loss was <1% for natural convection and at most 4% close to CHF for the subcooled scenarios. The amount of heat loss was estimated to be <2% for all boiling tests performed at saturation temperature. The heater is also modeled in numerical study and confirmed the low amount of heat loss. Therefore, the electrical power applied to the heating element has been used directly for calculations for the boiling curves and HTCs.

All the monitored data (temperature, pressure, power, flowrate) were recorded using a digital acquisition system (National Instruments, PXIe-1078). In all subcooled scenarios, the flowrate was low enough (<200 mL/min) to not disturb the pool boiling studies. The steady-state was considered reached when the peak-to-peak variation of temperature was less than 0.2 °C over a period of 120 s. Temperature was then obtained by averaging over a period of 30 s. Also, the locations of the tank inlet and outlet were designed to have a minimal effect on pool boiling experiments.

### 2.3 Surface Properties and Contact Angle Measurements.

A nickel-coated copper IHS (used for microprocessor cooling, 47 mm by 47 mm), 2000-grit sand paper polished copper plates (2 mm, a 4 mm and a 6 mm thick) and a 3M porous coated copper IHS from BOYD corporation (3.15 mm thick copper with 0.7 mm porous coating) were used in this study. Contact angle measurements were performed with a KRUSS drop shape analyzer (DSA30) on the nickel-coated IHS (Fig. 4) and on a polished copper surface (Fig. 5). Due to the high rate of evaporation of the Novec liquid, the test duration was 1 s at room temperature (20 °C) at atmospheric condition. The camera speed was 50 frames per second. A small decrease in the contact angle from the first and the last measured data points could be attributed to the evaporation (see Figs. 4(b) and 4(c)); the uncertainty of the contact angle measurements was estimated to be ±3 deg. It could be concluded that the average contact angle on the bare copper is 20 deg, and for the nickel-coated copper (Fig. 5) is 19 deg. We used the contact angles (20 deg) for the calculation of the CHF in correlation equations.

Fig. 4
Fig. 4
Fig. 5
Fig. 5

Figure 6 shows a scanning electron microscope (SEM) image of the 3M porous coating. The nonuniform porous coating had an average pore size between 75 μm and was fused on a bare copper disk. More information on the coating process can be found in reference [19]. Due to the highly wetting and fast-wicking properties of Novec 649 on the porous coating, it was not possible to measure a contact angle at a speed of 50 frames per second.

Fig. 6
Fig. 6

### 2.4 Experimental Uncertainties.

The experimental uncertainties are calculated using the combined root-sum-square method presented by Moffat [30]. First, the combined standard deviation on the measurements were computed using
$δS= ∑i=1i=K (∂S∂xi δxi)2$
(1)

where $δS$ is the standard deviation on a value S that is calculated from dependent measurements of X, with standard deviation $δxi$.

All uncertainties are presented as 95% confidence intervals, obtained by multiplying the standard deviations by a coverage factor of 2.

As an example, the heat flux ($q″$) was calculated as
$q′′=V·IW·L$
(2)
where V and I are the power supply voltage and current, while W and L are the width and the length of the IHS. The standard uncertainty on the heat flux thus is
$δq′′=( ∂q″∂VδV)2+( ∂q″∂IδI)2+( ∂q″∂WδW)2+( ∂q″∂LδL)2$
(3)

where we assumed that the standard deviations on the length and the width of the IHS were both equal to $δp=0.1 mm$, corresponding to machining precision. The relative standard deviations on the voltage and the current measurements ($δv /V$ and $δI /I$) were 2% and 1%, respectively. The relative uncertainty on the heat flux $δq′′/q″$ was thus calculated to be 4.5%.

The HTC is obtained by dividing the associated heat flux with the temperature difference between the boiling surface (Twall) and the saturation temperature of the liquid (Tsat) as follow:
$HTC=q″(Twall−Tsat)$
(4)
The junction-to-liquid thermal resistance (Rj−l) is calculated using the electric power consumption and the temperature difference ($Twall−Tsat$). The expression for Rj−l is as follow:
$Rj−l=Twall−TsatV·I$
(5)

The standard deviation on the temperature measurements with thermocouples was estimated at 0.1 K. As a result, the relative uncertainty was 12% on measured HTC values and 4.5% on measured thermal resistance values.

## 3 Results and Discussion

In this section, we first present the pool boiling curve of Novec 649 at its saturation temperature for a Lp= 25.4 mm square copper plate without lateral heat spreading. The effect of lateral heat spreading was then investigated by attaching four square copper plates of larger lateral dimensions to the copper plate (L = 47 mm, L/Lp = 1.85), with each plate having a different thickness (tIHS = 1 mm, 2 mm, 4 mm, and 6 mm). Finally, the performance of a boiling enhancement coating (BEC) copper plate (same area as other IHS) was evaluated at different subcooled temperatures.

### 3.1 Reference Saturation Boiling Test.

The size of the heater was matched with the size of the copper surface exposed to the boiling liquid, thus minimizing lateral heat spreading. Figure 7(a) shows the pool boiling curve, i.e., the heat flux $q″$ transferred from the copper surface at temperature Twall to the liquid at the saturation temperature Tsat. The boiling initiated at a wall superheat temperature (TwallTsat) of 12 °C. The boiling process could be divided into three regimes: the distinct-bubble nucleate boiling regime (I), the fully developed nucleate boiling regime (II), and the columns and slugs regime (III).

Fig. 7
Fig. 7

In regime I (identified with the dotted circle in Fig. 7(a)), the boiling initiated from the corner (I.I) then moved toward the center of the copper disk (I.II), and finally, it covered almost the whole surface (I.III). It was observed that by increasing the heat flux, the wall superheat temperature remained almost constant in regime I (circled region in Fig. 7(a)). In the fully developed nucleate regime, all the surface was covered by bubbles (II). The CHF happened at (17.4±0.8) W/cm2 (regime III).

Figure 7(b) shows a comparison of the measured HTC and correlations published in the literature, calculated using the properties of Novec 649. Both the Cooper [31] and Stephan and Abdelsalam [32] correlations compared well to our measurements, especially at lower heat fluxes. A larger deviation from the Ribatski and Jabardo [33] correlation is likely due to the fact that they developed this correlation for cylindrical surfaces.

We studied the literature to find CHF correlations to compare to our measurements; see Table 2. It was found that Liao et al. [34] present the closest agreement (relative discrepancy <3%).They experimented on a smooth surface for three kinds of heat transfer surfaces, resulting in different solid–liquid contact angles: a normal surface with a contact angle of 55 deg, a hydrophilic surface with a contact angle of 30 deg, and a super hydrophilic surface with a contact angle of 0 deg. Since the liquid used in our study has a low contact angle, we have the closest agreement with that correlation. In the theoretical formula provided by Kandlikar [35], the effect of contact angle was also considered, but they were not in the two other correlation from Zuber et al. [36] and Yagov [37]. The experimental value of CHF (explained in detail later in Sec. 3.2.1) was compared to the aforementioned correlations noted in Refs. [3134].

Table 2

CHF correlations

ReferenceCorrelation formulaCalculated CHF (W/cm2)RemarksRelative discrepancy (%)
Zuber et al. [36]$qCHF″=0.131ρghfg[σg(ρf−ρg)/ρg2]1/4$14.1418.7
Yagov [37]$qCHF,l″=0.5hfg81/55σ9/11ρg13/110kf7/110g21/55f(Prf)vf1/2cp,f3/10Ri79/110Tsat21/22$ for $P/Pc<0.001$ where $f(Prf)=(Prf9/81+2Prf1/4+0.6Prf19/24)4/11$$qCHF,h″=0.06hfgρg3/5σ2/5[g(ρf−ρg)/μf]1/5$ for $P/Pc>0.003$$qCHF″=(qCHF,h″3+qCHF,l″3)1/3$ for $0.00112.09Calculated using equation for $P/Pc>0.003$30.5
Kandlikar [35]$qCHF″=1+cosα16[2π+π4(1+cosα)cosθ]1/2×ρghfg[σg(ρf−ρg)/ρg2]1/4$19.23Contact angle $α=20 deg$; surface orientation $θ= 0 deg$9.9
Liao et al. [34]$qCHF″=0.131[−0.73+1.731+10−0.021×(185.4−θ)][1+55−α100(0.56−0.0013θ)]×ρghfg[σg(ρf−ρg)/ρg2]1/4$16.9Contact angle $α=20 deg$; surface orientation $θ=0 deg$2.9
ReferenceCorrelation formulaCalculated CHF (W/cm2)RemarksRelative discrepancy (%)
Zuber et al. [36]$qCHF″=0.131ρghfg[σg(ρf−ρg)/ρg2]1/4$14.1418.7
Yagov [37]$qCHF,l″=0.5hfg81/55σ9/11ρg13/110kf7/110g21/55f(Prf)vf1/2cp,f3/10Ri79/110Tsat21/22$ for $P/Pc<0.001$ where $f(Prf)=(Prf9/81+2Prf1/4+0.6Prf19/24)4/11$$qCHF,h″=0.06hfgρg3/5σ2/5[g(ρf−ρg)/μf]1/5$ for $P/Pc>0.003$$qCHF″=(qCHF,h″3+qCHF,l″3)1/3$ for $0.00112.09Calculated using equation for $P/Pc>0.003$30.5
Kandlikar [35]$qCHF″=1+cosα16[2π+π4(1+cosα)cosθ]1/2×ρghfg[σg(ρf−ρg)/ρg2]1/4$19.23Contact angle $α=20 deg$; surface orientation $θ= 0 deg$9.9
Liao et al. [34]$qCHF″=0.131[−0.73+1.731+10−0.021×(185.4−θ)][1+55−α100(0.56−0.0013θ)]×ρghfg[σg(ρf−ρg)/ρg2]1/4$16.9Contact angle $α=20 deg$; surface orientation $θ=0 deg$2.9

σ (N/m) is surface tension of Novec 649, ρg (kg/m3) is the density of Novec 649 vapor, ρl (kg/m3) is the density of the dielectric liquid. and hfg (J/kg) is its latent heat of evaporation.

### 3.2 Lateral Spreading Scenario.

In this section, the effect of heat spreading is first investigated on a 1 mm thick nickel-coated IHS removed from a commercial microprocessor, at different subcooled temperatures. Then, the effect of thickness on the boiling heat transfer of dielectric liquid is discussed.

#### 3.2.1 Boiling Tests on a 1 mm Thick Integrated Heat Spreader.

The boiling tests were performed on the 1 mm-thick nickel-coated IHS at four subcooled temperatures, i.e., ΔTsub =(TsatTf) equal to 0 °C, 14 °C, 24 °C, and 34 °C.

The resulting boiling curves are shown in Fig. 8(a). Boiling initiated at a wall superheat temperature of 13 °C, 28 °C, 38 °C, and 48 °C, respectively, for the subcooled temperatures listed earlier. While bubbles nucleated at random locations on the surface in the saturation condition (see Fig. 8(b) (i)), in all the subcooled conditions the bubbles originated from the center of the IHS, where the heater was located (see Fig. 8(c) (i)). In the saturation condition, at the heat flux of 7 W/cm2, a region with almost no bubbles was observed at the edges while at the center, a fully developed nucleate boiling was formed (see Fig. 8(b) (ii)). This phenomenon was clearer when ΔTsub was 34 °C, where a larger area was not covered with bubbles at the edges (see Fig. 8(c) (ii)). It was observed that the CHF happened at 8.6 W/cm2 in the saturation condition (see Fig. 8(b) (iii)). At CHF, the boiling could be seen on the whole surface, in the columns and slugs regime at the center and in the fully developed nucleate boiling regime at the edges. By increasing the subcooled temperature, the CHF was delayed and occurred at higher heat fluxes. For instance, the CHF happened at 11.5 W/cm2 when ΔTsub was 34 °C. Even in that case, there were regions at the edges where there was no boiling. The CHF was delayed by more than 33% when ΔTsub was 24 °C compared to when ΔTsub was 0 °C. However, there was not much gain on the CHF when increasing the subcooled temperature from 24 °C to 34 °C. This indicates that the effect of reducing the liquid temperature on delaying the CHF reaches a plateau as further decreasing the liquid temperature does not improve the CHF.

Fig. 8
Fig. 8

Based on these observations in saturation and subcooled experiments, it is suggested that the boiling phenomena are highly coupled with the spreading of the heat in the plane of the IHS, and the physical mechanism significantly differs with no lateral heat spreading scenario by having more than one regime of boiling on the IHS surface.

#### 3.2.2 Effect of the Thickness of the Integrated Heat Spreader.

The boiling curves with the liquid at its saturation temperature for copper plates with different thicknesses are shown in Fig. 9(a). Boiling started almost at the same wall superheat, around 13 °C. At low heat fluxes, all the plates show the same heat dissipation capacity. The performance is first seen to deviate for the 6 mm thickness at a heat flux of 2 W/cm2, with a higher heat dissipation capacity compared to the thinner plates. The boiling curves for the 1 mm, 2 mm, and 4 mm plates start diverging at a wall superheat temperature around 30 °C and at the heat flux of 5 W/cm2. The CHF is increased by more than 60% by increasing the thickness from 1 mm to 6 mm. This suggests that the effects of lateral spreading are significant on both the heat dissipation (higher slope of the boiling curve corresponding to higher HTC) and the CHF occurrence.

Fig. 9
Fig. 9

Figure 9(b) shows the typical patterns of the central bubble columns in the fully developed nucleate boiling regime, at 7 W/cm2. For the 1 mm-thick plate, the central bubble column that forms on the top surface has an area that is similar to the area of the heater. The dashed lines in the figures delineate the central column from no boiling or distinct-bubble nucleate boiling regions. With increasing thickness, this boiling region moves toward the edges of the copper plates and more of their surface is covered with boiling from the fully developed nucleate boiling regime. This increased coverage shows that increasing the thickness improves lateral heat spreading.

Plates with different thicknesses have also been tested under different subcooled temperatures (see Fig. 10). The boiling incipience is delayed compared to the saturation cases on different thick ces but almost happens at a similar wall superheat. The superior performance of the thicker plates remains at all subcooled levels. For example, when ΔTsub is 14 °C, the CHF is 18 W/cm2 on the 6 mm plate compared to 10 W/cm2 on the 1 mm plate. The results show that there is no significant gain in the CHF by further increasing the subcooled temperature from 24 °C to 34 °C, for all different plate thicknesses.

Fig. 10
Fig. 10

### 3.3 Junction-to-Liquid Thermal Resistance.

Junction-to-liquid thermal resistance is a key parameter for electronics cooling applications and it can be a proper metric to compare all the different scenarios. Figure 11 presents the junction-to-liquid thermal resistance for all the plates and IHSes tested in this study, at the saturation condition. BEC shows the lowest thermal resistance at low power due to earlier boiling incipience. Apart from the BEC, all the other plates show almost the same thermal resistance when the power is below 50 W. At higher power, the 6 mm-thick plate has the lowest thermal resistance for uncoated surfaces. When increasing the power to more than 200 W, its thermal resistance reaches similar values to the BEC. By comparing the thermal resistances, it is made clear that the lateral heat spreading is a critical phenomenon for immersion cooling applications in which a 6 mm-thick copper can perform like a 4 mm porous coated surface at the thermal resistance equal to 0.14 °C/W.

Fig. 11
Fig. 11

The results also show little change in thermal resistance for the BEC for a power over 200 W. Using a BEC could nevertheless lower the risk of reaching the CHF, due to its higher operation range (more than 480 W).This reveal in addition to all the efforts toward the boiling surface enhancement we need to consider heat spreading as a key parameter to be optimized the performance of future heat spreader designed (size and thickness) considering the chip size for immersion cooling of micro-electronics.

## 4 Conclusion

This paper investigated the effect of the heat spreading on the boiling of a highly wetting dielectric liquid for electronic cooling applications. The reference test was performed by attaching a 25.4 mm by 25.4 mm bare copper plate to a same-sized heater, thus minimizing lateral heat spreading. The pool boiling tests were performed at the saturation state of Novec 649 on the bare copper. It has been found that the CHF happened at (17.4±0.8) W/cm2 while the maximum HTC was (0.8±0.1) W/cm2-K. Both HTC and CHF were compared with available correlations in the literature, and a very close match was found with some of the published data. Then, the role of heat spreading under different pool boiling conditions was studied by measuring the thermal performance of plates that were larger than the heater area. Saturated and subcooled boiling studies on a 1 mm-thick plate revealed that two distinct boiling regimes appear on its surface due to lateral heat spreading and bubble generation modulated by the temperature of the surface. At the CHF with the liquid at its saturation temperature, boiling occurred on the whole surface, in the columns and slugs regime at the center, and in the fully developed nucleate boiling regime at the edges. Due to nonuniformity on boiling heat transfers, the CHF's was 8.6 W/cm2, 50% less than the reference nonlateral-spreading test. By reducing the liquid temperature to a subcooled state, the CHF was delayed to higher heat fluxes, and a region without boiling could be observed at the edges. The effect of thickness on improved lateral heat spreading was investigated on copper plates with different thicknesses. The data revealed a heat dissipation capacity increasing with the plate thickness. Also, the CHF was increased by more than 60% by increasing the thickness from 1 mm to 6 mm. A microporous BEC coated copper plate was tested under different boiling conditions, and early boiling incipience on the porous-coated surface (almost 10 °C lower compared to that on bare copper) was observed.

It was found that the lateral heat spreading was a critical design parameter for immersion cooling applications. Nonoptimized IHS (1 mm thick) could reduce the CHF to 50% less value than the reference no lateral spreading scenario. Calculated junction-to-liquid thermal resistance revealed, a 6 mm thick bare copper plate could perform like a 4 mm porous coated surface at a power higher than 200 W. Therefore, in addition to all the efforts toward the boiling surface enhancement, we need to consider heat spreading as a critical parameter for future heat spreader design (size and thickness) for immersion cooling of micro-electronics. Calculations with HTC and temperature fields varying across the boiling surface due to heat fluxes in the surface plane will be presented in a separate paper.

## Acknowledgment

The authors would also like to thank Hubert Pelletier, Jeremy Roussel-Francoeur, Gabriel Parent, and Maxime Desmarais-Laporte for their technical assistance during the test bench development.

## Funding Data

• Natural Sciences and Engineering Research Council of Canada (NSERC) under the Collaborative Research and Development (CRD) Program with Varitron Technologies Inc. and Systemex Energies Inc. (Funder ID: 10.13039/501100000038).

## Nomenclature

• cp =

heat capacity

•
• g =

gravitational acceleration

•
• I =

electrical current (A)

•
• k =

thermal conductivity

•
• L =

length of the IHS m)

•
• Pc =

critical pressure (Pa)

•
• Pr =

Prandtl number

•
• q″ =

heat flux (W/m2)

•
• R =

thermal resistance (oC/W)

•
• S =

standard deviation

•
• T =

temperature (oC)

•
• V =

voltage (V)

•
• W =

width of the IHS (m)

### Greek Symbols

Greek Symbols

• α =

contact angle

•
• θ =

surface orientation

•
• σ =

surface tension (N/m)

•
• ν =

kinematic viscosity

### Subscripts

Subscripts

• CHF =

critical heat flux

•
• f =

fluid/liquid

•
• g =

gas/vapor

•
• j–l =

junction to liquid

•
• p =

plate

•
• sat =

saturation

•
• Sub =

subcooling

•
• wall =

top of the heater

## References

1.
Garimella
,
S. V.
,
Persoons
,
T.
,
Weibel
,
J.
, and
Yeh
,
L. T.
,
2013
, “
Technological Drivers in Data Centers and Telecom Systems: Multiscale Thermal, Electrical, and Energy Management
,”
Appl. Energy
,
107
, pp.
66
80
.10.1016/j.apenergy.2013.02.047
2.
Masanet
,
E.
,
Shehabi
,
A.
,
Lei
,
N.
,
Smith
,
S.
, and
Koomey
,
J.
,
2020
, “
Recalibrating Global Data Center Energy-Use Estimates
,”
Science
,
367
(
6481
), pp.
984
986
.10.1126/science.aba3758
3.
Garimella
,
S. V.
,
Yeh
,
L.-T.
, and
Persoons
,
T.
,
2012
, “
Thermal Management Challenges in Telecommunication Systems and Data Centers
,”
IEEE Trans. Compon., Packag. Manuf. Technol.
,
2
(
8
), pp.
1307
1316
.10.1109/TCPMT.2012.2185797
4.
Whitney, J., and Delforge, P.,
2021
, “
Data Center Efficiency Assessment
,” NRDC, accessed Apr. 28, 2021, www.nrdc.org/sites/default/files/data-center-efficiency-assessment-IP.pdf
5.
Kulkarni
,
D.
,
Tang
,
X.
,
Ahuja
,
S.
,
Dischler
,
R.
, and
Mahajan
,
R.
,
2018
, “
Experimental Study of Two-Phase Cooling to Enable Large-Scale System Computing Performance
,”
Proceedings of the 17th InterSociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems, ITherm 2018
, San Diego, CA, May 29–June 1.
6.
,
A. C.
, and
Groulx
,
D.
,
2016
, “
Cooling of Server Electronics A Design Review of Existing Technology
,”
Appl. Therm. Eng.
,
105
, pp.
622
638
.10.1016/j.applthermaleng.2016.03.056
7.
,
S. T.
, and
Kumar
,
R.
,
2014
, “
Twenty First Century Cooling Solution: Microchannel Heat Sinks
,”
Int. J. Therm. Sci.
,
85
, pp.
73
92
.10.1016/j.ijthermalsci.2014.06.013
8.
Ghaffari
,
O.
,
Solovitz
,
S. A.
,
Ikhlaq
,
M.
, and
Arik
,
M.
,
2016
, “
An Investigation Into Flow and Heat Transfer of an Ultrasonic Micro-Blower Device for Electronics Cooling Applications
,”
Appl. Therm. Eng.
,
106
, pp.
881
889
.10.1016/j.applthermaleng.2016.06.094
9.
Ghaffari
,
O.
,
Ikhlaq
,
M.
, and
Arik
,
M.
,
2015
, “
An Experimental Study of Impinging Synthetic Jets for Heat Transfer Augmentation
,”
Int. J. Air-Cond. Refrig
.
,
23
(
3
), p. 1550024.10.1142/S2010132515500248
10.
Schmidt
,
R.
,
2021
, “
Packaging of New Servers-Energy Efficiency Aspects
,” IBM Corporation, Armonk, New York, accessed Apr. 28, 2021, https://e3s-center.berkeley.edu/wp-content/uploads/2017/07/RogerSchmidt.pdf
11.
Alkharabsheh
,
S.
,
,
U. L. N.
,
Ramakrishnan
,
B.
,
Ghose
,
K.
, and
Sammakia
,
B.
,
2018
, “
Failure Analysis of Direct Liquid Cooling System in Data Centers
,”
ASME J. Electron. Packag
.,
140
(
2
), p.
020902
.10.1115/1.4039137
12.
Yuksel
,
A.
,
Mahaney
,
V.
,
Marroquin
,
C.
,
Tian
,
S.
,
Hoffmeyer
,
M.
,
Schultz
,
M.
, and
Takken
,
T.
,
2021
, “
An Overview of Thermal and Mechanical Design, Control, and Testing of the World's Most Powerful and Fastest Supercomputer
,”
ASME J. Electron. Packag.
,
143
(
1
), p.
011005
.10.1115/1.4046847
13.
Bar-Cohen
,
A.
,
1983
, “
Thermal Design of Immersion Cooling Modules for Electronic Components
,”
Heat Transfer Eng.
,
4
(
3–4
), pp.
35
50
.10.1080/01457638108939607
14.
Abreu
,
V.
,
Harrison
,
M.
,
Gess
,
J.
, and
Moita
,
A. S.
,
2018
, “
Two-Phase Thermosiphon Cooling Using Integrated Heat Spreaders With Copper Microstructures
,”
Proceedings of the 17th InterSociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems, ITherm 2018
,
San Diego, CA, May 29–June 1.
10.1109/ITHERM.2018.8419644
15.
Gess
,
J. L.
,
Bhavnani
,
S. H.
, and
Johnson
,
R. W.
,
2015
, “
Experimental Investigation of a Direct Liquid Immersion Cooled Prototype for High Performance Electronic Systems
,”
IEEE Trans. Compon., Packag. Manuf. Technol.
,
5
(
10
), pp.
1451
1464
.10.1109/TCPMT.2015.2453273
16.
Khalili
,
S.
,
Rangarajan
,
S.
,
Gektin
,
V.
,
Alissa
,
H.
, and
Sammakia
,
B.
,
2020
, “
An Experimental Investigation on the Fluid Distribution in a Two-Phase Cooled Rack Under Steady and Transient Information Technology Loads
,”
ASME J. Electron. Packag.
,
142
(
4
), p.
041002
.10.1115/1.4048180
17.
Pelletier
,
M.
,
Normandin
,
I.
, and
Jasmin
,
S.
,
2017
, “
Apparatus and Methods for Cooling a CPU Using a Liquid Bath
,” Patent No. 9655279.
18.
Campbell
,
L. A.
,
Chu
,
R. C.
,
David
,
M. P.
,
Ellsworth Jr
,
M. J.
,
Iyengar
,
M. K.
, and
Simons
,
R. E.
,
2013
, “
Multi-Fluid, Two-Phase Immersion-Cooling of Electronic Component (S)
,” Patent No.
US8619425B2
19.
Tuma
,
P. E.
,
2010
, “
The Merits of Open Bath Immersion Cooling of Datacom Equipment
,” 26th Annual IEEE Semiconductor Thermal Measurement and Management Symposium (
SEMI-THERM
), Santa Clara, CA, Feb. 21–25, pp.
123
131
.10.1109/STHERM.2010.5444305
20.
Coles
,
H.
,
Herrlin
,
M.
, and
Org
,
E.
,
2016
,
Lawrence Berkeley National Laboratory Recent Work Title Immersion Cooling of Electronics in DoD Installations
.
21.
Al Sayed
,
C.
,
Ghaffari
,
O.
,
Grenier
,
F.
,
Tong
,
W.
,
Bolduc
,
M.
,
Morissette
,
J. F.
,
Jasmin
,
S.
, and
Sylvestre
,
J.
,
2020
, “
Localized Pool Boiling and Condensation Experiments Over Functional CPU: Optimizing the Overall Thermal Resistance Via Different Heat Transfer Scenarios
,”
InterSociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems, ITHERM
,
Orlando, FL, July 21–23, pp. 410–416.
10.1109/ITherm45881.2020.9190371
22.
Tuma
,
P.
,
Schulz
,
J.
,
Bonk
,
J.
,
Roy
,
R.
,
DeLorme
,
M.
, and
Logan
,
P.
,
2018
, “
An Exposure Assessment Approach for Potential Thermal Degradation Compounds in 2-Phase Immersion Cooling Operations
,”
Process Saf. Prog.
,
37
(
3
), pp.
369
375
.10.1002/prs.11956
23.
Ghaffari
,
O.
,
Grenier
,
F.
,
Morissette
,
J. F.
,
Bolduc
,
M.
,
Jasmin
,
S.
, and
Sylvestre
,
J.
,
2019
, “
Pool Boiling Experiment of Dielectric Liquids and Numerical Study for Cooling a Microprocessor
,”
InterSociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems, ITHERM
,
IEEE Computer Society
, Las Vegas, NV, May 28–31, pp.
540
545
.10.1109/ITHERM.2019.8757380
24.
Sylvestre
,
J.
, and
Jasmin
,
S.
,
2019
, “
Apparatus and Methods for Cooling of an Integrated Circuit
,” U.S. Patent 10,390,460.
25.
Leong
,
K. C.
,
Ho
,
J. Y.
, and
Wong
,
K. K.
,
2017
, “
A Critical Review of Pool and Flow Boiling Heat Transfer of Dielectric Fluids on Enhanced Surfaces
,”
Appl. Therm. Eng.
,
112
, pp.
999
1019
.10.1016/j.applthermaleng.2016.10.138
26.
Ali
,
A. F.
, and
El-Genk
,
M. S.
,
2012
, “
Spreaders for Immersion Nucleate Boiling Cooling of a Computer Chip With a Central Hot Spot
,”
Energy Convers. Manag.
,
53
(
1
), pp.
259
267
.10.1016/j.enconman.2011.09.007
27.
Arik
,
M.
,
Bar-Cohen
,
A.
, and
You
,
S. M.
,
2007
, “
Enhancement of Pool Boiling Critical Heat Flux in Dielectric Liquids by Microporous Coatings
,”
Int. J. Heat Mass Transfer
,
50
(
5–6
), pp.
997
1009
.10.1016/j.ijheatmasstransfer.2006.08.005
28.
Wu
,
Z.
,
Cao
,
Z.
, and
Sundén
,
B.
,
2019
, “
Saturated Pool Boiling Heat Transfer of Acetone and HFE-7200 on Modified Surfaces by Electrophoretic and Electrochemical Deposition
,”
Appl. Energy
,
249
, pp.
286
299
.10.1016/j.apenergy.2019.04.160
29.
3M Science. Applied to Life.TM,
2021
, “
3MTM NovecTM Engineered Fluid, 649
,” 3M Science. Applied to Life.TM, accessed Apr. 29, 2021, www.3mcanada.ca/3M/en_CA/p/d/b5005005025/
30.
Moffat
,
R. J.
,
1988
, “
Describing the Uncertainties in Experimental Results
,”
Exp. Therm. Fluid Sci.
,
1
(
1
), pp.
3
17
.10.1016/0894-1777(88)90043-X
31.
Cooper
,
M. G.
,
1984
, “
Heat Flow Rates in Saturated Nucleate Pool Boiling-A Wide-Ranging Examination Using Reduced Properties
,”
,
16
(
C
), pp.
157
239
.10.1016/S0065-2717(08)70205-3
32.
Stephan
,
K.
, and
Abdelsalam
,
M.
,
1980
, “
Heat-Transfer Correlations for Natural Convection Boiling
,”
Int. J. Heat Mass Transfer
,
23
(
1
), pp.
73
87
.10.1016/0017-9310(80)90140-4
33.
Ribatski
,
G.
, and
Jabardo
,
J. M. S.
,
2003
, “
Experimental Study of Nucleate Boiling of Halocarbon Refrigerants on Cylindrical Surfaces
,”
Int. J. Heat Mass Transfer
,
46
(
23
), pp.
4439
4451
.10.1016/S0017-9310(03)00252-7
34.
Liao
,
L.
,
Bao
,
R.
, and
Liu
,
Z.
,
2008
, “
Compositive Effects of Orientation and Contact Angle on Critical Heat Flux in Pool Boiling of Water
,”
Heat Mass Transfer Stoffuebertragung
,
44
(
12
), pp.
1447
1453
.10.1007/s00231-008-0384-6
35.
Kandlikar
,
S. G.
,
2001
, “
A Theoretical Model to Predict Pool Boiling CHF Incorporating Effects of Contact Angle and Orientation
,”
ASME J. Heat Transfer-Trans. ASME
,
123
(
6
), pp.
1071
1079
.10.1115/1.1409265
36.
Zuber
,
N.
,
1959
, “Hydrodynamic Aspects of Boiling Heat Transfer (Thesis), Report No.
AECU-4439
. 10.2172/4175511
37.
Yagov
,
V. V.
,
2014
, “
Is a Crisis in Pool Boiling Actually a Hydrodynamic Phenomenon?
,”
Int. J. Heat Mass Transfer
,
73
, pp.
265
273
.10.1016/j.ijheatmasstransfer.2014.01.076