International Journal of Heat and Mass Transfer 160 (2020) 120203
`
`Contents lists available at ScienceDirect
`
`International Journal of Heat and Mass Transfer
`
`journal homepage: www.elsevier.com/locate/hmt
`
`Design of a single-phase immersion cooling system through
`experimental and numerical analysis
`a , ∗
`
`Chin-Chi Cheng
`
`b, Po-Chun Chang a , Hsing-Chieh Li a , Fu-I Hsu
`
`
`
`
`
`a Department of Energy and Refrigerating Air-Conditioning Engineering, National Taipei University of Technology, 1, Section 3, Zhongxiao East Road, Taipei
`10608, Taiwan R.O.C
`b VICI Holdings, No. 2-7, Sec. 2, Ren’ai Rd., Zhongzheng Dist., Taipei City 100, Taiwan (R.O.C.)
`
`a r t i c l e
`
`i n f o
`
`a b s t r a c t
`
`Article history:
`Received 31 March 2020
`Revised 8 June 2020
`Accepted 12 July 2020
`Available online 31 July 2020
`
`Keywords:
`Immersion cooling
`Single-phase
`Heat sink
`Synthetic oil
`Circulation
`Computer simulation software
`Heat distribution
`
`Cooling electronics systems more effectively could enable computer systems to utilize power more effi-
`ciently. Immersion cooling is a potential method to cool computers and other electronics by submerging
`them into a thermal conductive dielectric liquid or coolant. In this study, an innovative cooling struc-
`ture and procedure of a single-phase immersion cooling system using heat sink and forced circulation
`is presented. The straight finned heat sink is attached on the CPU surface, and the entire mainboard is
`submerged in an engineered fluid, 3 M Novec 7100, which is able to dissipate heat and is used in im-
`mersion cooling applications. An experimental test is utilized to verify a simulation model built using the
`computer simulation software. Three circulating speeds of the liquid coolant along with the use of two
`different materials for the heat sink are chosen to explore the cooling effects of a single-phase immer-
`sion cooling system. The heat distribution of the designed model at various flow rates of liquid coolant
`and materials was observed through the cross-sectional viewpoint and 3D isothermal surface model. The
`results of the simulations show that the faster flowing speed of liquid coolant would remove more heat,
`and cause lower temperature of CPU. However, the flow of coolant was encumbered due to slower circu-
`lating speed. It also caused the higher temperature values and unbalanced heat distribution. The highest
`temperatures were measured, and the unbalanced heat distribution of the model was observed. It is
`noted that the material of the heat sink do not significantly affect the results. These outcomes are able to
`provide the system designer with useful information to increase power densification and guarantee the
`safe operation of the related computer, server and communication systems.
`
`© 2020 Elsevier Ltd. All rights reserved.
`
`1. Introduction
`
`Cooling electronics systems efficiently could drive the power
`utilization more concentrated. The International Electronics Manu-
`facturing Initiative (iNEMI) Technology Roadmap reports that how
`to draw the heat flux generated by chips within an area of po-
`
`tentially only 300 W/cm 2 while keeping the operating tempera-
`tures below 85 °C will be the main challenge for the micro and
`power-electronics industry [1] . Therefore, the cooling technology
`to dissipate the generated heat by chips and maintain the safe
`operation of electronic devices is a critical issue. Several meth-
`ods, including thermoelectric cooler (TEC), heat pipe, vapor cham-
`
`∗
`
`Correspondence to: Chin-Chi Cheng, Dept of Energy and Refrigerating Air-
`Conditioning Engineering Taipei, 1, Sec. 3, Zhongxiao E. Rd., Taipei 10608, Taiwan,
`R.O.C.
`E-mail addresses: newmanch@ntut.edu.tw (C.-C. Cheng), qazxd5522@gmail.com
`(P.-C. Chang), shingjay@nsrrc.org.tw (H.-C. Li), louis.hsu@viciholdings.com (F.-I. Hsu).
`
`https://doi.org/10.1016/j.ijheatmasstransfer.2020.120203
`0017-9310/© 2020 Elsevier Ltd. All rights reserved.
`
`ber, nanofluids, single-phase cooling, two-phase cooling, immer-
`sion cooling and vapor compression refrigeration (VCR) system are
`options for this purpose. The TEC coupling with thermosiphone
`loop [2] or heat sink [3] has higher efficiency and temperature uni-
`formity comparing with other methods, and its performance in-
`crease with thermoelectric operating voltage and the dimension
`closeness between thermoelectric module and central processing
`unit (CPU). Heat pipes are able to dissipate substantial amount of
`heat with a relative small temperature drop along the heat pipe
`while providing a self-pumping ability due to an embedded porous
`material in their structure. The limiting factor for the heat transfer
`capability of a heat pipe is related to the working fluid transport
`capability [4] . Based on phase change heat transfer phenomena,
`vapor chamber draws high heat flux from a small heating source,
`and spreads to larger area through vaporization and condensation
`of working fluid inside it. The vapor chamber can keep the system
`at constant temperature nearly to avoid overheating or other dam-
`age. The thermal performance of vapor chamber is related to the
`
`Immersion Systems LLC – Ex. 1016
`PGR 2021-00104 (U.S. 10,820,446 B2)
`1 of 12
`
`

`

`2
`
`C.-C. Cheng, P.-C. Chang and H.-C. Li et al. / International Journal of Heat and Mass Transfer 160 (2020) 120203
`
`Nomenclatures
`
`Symbols
`g
`Cp
`J
`k
`K
`m
`P
`q
`Q
`T
`V
`W
`
`•K) specific heat capacity (J/kg
`
`weight
`
`joule
`
`heat transfer coefficient (W/m 2 K)
`Kelvin temperature
`meter
`pressure (Pa)
`heat flux (W/m 2 )
`
`heating source
`temperature (K)
`average velocity (m/s)
`watt (J/s)
`
`•s) dynamic viscosity (Pa
`
`Greek letters
`μ
`
`ρ
`
`density (kg/m 3 )
`(cid:3)
`cross product
`
`Acronyms
`3D
`three dimensions
`Al
`aluminum
`BDF
`back difference in phase
`CFD
`computational fluid dynamics
`CPU
`central processing unit
`Cu
`copper
`LHPs
`loop heat pipes
`PC
`personal computer
`TDP
`thermal design power
`TEC
`thermoelectric cooler
`VCR
`vapor compression refrigeration
`
`property of working fluid, material and geometric structure of wick
`structure [5] . The vapor chamber embedded with heat sinks with
`various fin configurations draw much attention in thermal man-
`agement of electronics devices recently [6] . Nanofluids, the mixture
`of base fluid with nanoparticles, are able to improve the thermal
`conductivity and convection coefficients of fluid [7] . It is one of the
`candidates for further miniaturization of electronic devices. But a
`balance among volumetric concentration of nanoparticles, the flow
`rate and other variables to satisfy the economy and power con-
`sumption of cooling the system is still unconcluded [8] . Two-phase
`system of the micro-evaporator cooling cycles, composed by the
`pump [9] , the compressor, the loop heat pipes (LHPs) [10] or the
`hybrid [11] , could not only improve the heat removal efficiency
`but also permit the reuse of waste heat. It is suitable for a com-
`puter blade server, green data center and supercomputers, etc. VCR
`system, consisting compressor and heat exchanger (evaporator and
`condenser), can maintain low junction temperature while dissipate
`high heat flux [12] . Miniature VCR system can fit the requirement
`of limited space of today’s electronic devices by combining the
`miniature compressor and microchannel heat exchanger. The main
`concern for VCR system is the cost [13] .
`Among them, immersion cooling is a potential method to cool
`computer and other electronics, such as complete servers and
`mainboard, by submerging them into a thermal conductive dielec-
`tric liquid or coolant [14] . This method removes the generated heat
`from the system by circulating the liquid throughout the system in
`order to continually dissipate heat that is generated by the compo-
`nents. The immersion cooling market is projected to increase from
`USD 177 million in 2019 to USD 501 million by 2024 [15] due to
`increased advances and uses in immersion edge computing, arti-
`ficial intelligence, and cloud computing [16] . The requirement for
`
`the working fluid of an immersion cooling system is that it must
`be non-conductive with the electronics. Several types of fluids are
`suitable for cooling, however, the use of liquids is one of the most
`prominent and efficient candidates [14] . Four types of fluids, in-
`cluding deionized water, mineral oil, fluorocarbon-based fluids and
`synthetic oil, are suitable for this application. Depending on the
`properties of the fluid, the fluids used in the immersion cooling
`technology can be classified into a single-phase or two-phase sys-
`tems. The difference between these two systems is whether the
`fluid that is used changes its state or not during the process of
`cooling the electronics. Matsuoka, et al. [17] evaluated the cool-
`ing performance of a liquid immersion cooling system with natu-
`ral convection for high power server boards used in data center
`by computational fluid dynamics (CFD) software packages simu-
`lation and actual experiments. They tested several fluids, includ-
`ing silicone oil, soybean oil, and perfluorocarbon structured liq-
`uids, and found that the smoother refrigerant was better for cool-
`ing the high power CPU. Wagner, et al. [18] compared the cooling
`performance of three different liquid cooling technologies, includ-
`ing single-phase cold plate, two-phase immersion cooling with 3 M
`R (cid:4)
`Novec
`649 and single-phase mineral oil immersion. The experi-
`mental results indicated that the single-phase cold plate achieve
`the highest power dissipation with the lowest thermal resistance
`of 0.048 K/W, the two-phase immersion cooling performed sec-
`ond best, then was the single-phase mineral oil immersion cool-
`ing. Sathyanarayana, et al. [19] presented the mixture formula-
`tions of Novec fluid (pure HFE 7200) with alcohols and ethers
`(HFE 7200 and methanol; HFE 7200 and ethoxybutane), which
`equipped low thermal conductivity and specific heat, to improve
`heat transfer properties and applicability of heat transfer fluids.
`The thermal performance of these new fluid mixtures was tested
`on a 1 cm × 1 cm silicon (Si) substrate having copper nanowire
`arrays at saturation conditions. Based on previous research results,
`fluorocarbon-based fluids evaporated in the largest amounts during
`the cooling process, and the mineral oil removed the least amount
`of heat from a system compared with other fluids, whereas the
`synthetic oil showed the best cooling performance.
`Chen, et al. [20] installed radiation fins to the surface of a CPU
`to see if this could enhance heat-removing capabilities. They found
`that the novel three-dimensional (3D) stacked package structure
`with horizontal fins did in fact lead to better heat radiation ca-
`pacity than the conventional design. Qiu, et al. [21] investigated
`the thermal behavior of 3D stacked dies using the cooling tech-
`nologies of natural convection, forced air cooling and water im-
`mersion cooling methods. These models were built by using one
`of ANSYS analysis modules for electronic system thermal manage-
`ment, which is ICEPAK software. Simulation models were built for
`experimental validation and further thermal analysis. The results
`indicated that the thermal resistances of these three methods were
`26, 7.6 and 0.6 °C/W, respectively. The immersion cooling method
`had the best cooling capability and the lowest temperature at the
`hotspot, which was the area of highest current density. An, et al.
`[22] compared a two-phase immersion cooling solution for high-
`powered processor designs with various heating sources by us-
`ing ANSYS Fluent for a 3D numerical analysis. Two arrangements,
`including two vertically mounted heat sources to achieve higher
`packing density of the server and a single heat source, were im-
`mersed in a bath of 3 M Novec70 0 0 as the phase change coolant.
`They found that Novec70 0 0 was able to cool a 5 cm × 5 cm heat
`source in a vertical orientation with power as high as 225 W (heat
`
`flux 9 W/cm 2 ). In addition to ANSYS Fluent, the generally utilized
`CFD software packages for geophysical modeling also include COM-
`SOL Multiphysics. Wang, et al. [23] investigated the electrical field
`distribution rule for direct current method forward modeling and
`point electric source by using COMSOL Multiphysics. Through com-
`paring and analyzing COMSOL Multiphysics modeling results and
`
`Immersion Systems LLC – Ex. 1016
`PGR 2021-00104 (U.S. 10,820,446 B2)
`2 of 12
`
`

`

`C.-C. Cheng, P.-C. Chang and H.-C. Li et al. / International Journal of Heat and Mass Transfer 160 (2020) 120203
`
`3
`
`Fig. 1. Model structure of an immersion cooling system for the CPU of a personal
`computer.
`
`theoretic values of typical models, they indicated that the valid-
`ity and feasibility of direct current method forward modeling is
`proved and the COMSOL Multiphysics was suitable in geophysi-
`cal modeling. By using COMSOL Multiphysics software, Kocman,
`et al. [24] also carried out a simulation of three-phase squirrel cage
`induction motor. They indicated that the simulation results from
`COMSOL match the catalog values rather well in the values of sta-
`tor current and motor torque.
`Based on the previous researches, an innovative cooling struc-
`ture and procedure of a single-phase immersion cooling system
`with heat sink and forced circulation of liquid are presented in this
`study. The straight finned heat sink is attached on the CPU surface.
`The entire mainboard is immersed in the engineered 3 M Novec
`7100 fluid for forced circulation cooling. An experimental test is
`utilized to verify the simulation model built by using computer
`simulation software. Three circulation speeds of the liquid coolant
`and two materials for the heat sink are chosen to explore the cool-
`ing effects of the designed single-phase immersion cooling system.
`
`2. Model
`
`2.1. Model configuration
`
`In order to simulate the effect of the immersion cooling sys-
`tem with circulating liquid and heat sink placed on the CPU sur-
`face of a personal computer (PC), a three dimensional model was
`built using the computer simulation software, as shown in Fig. 1 .
`The model includes a CPU, which is the main heat source of a PC,
`with a straight finned heat sink on its surface. The size of CPU is
`40 mm-in-square. The chosen materials of the heat sink were alu-
`minum (Al) and copper (Cu). The liquid coolant flowed into the PC
`box from the bottom of heat sink, passed the heat sink and flowed
`out through the top left of the box. The model was assumed as a
`steady, incompressible and laminar flow. The momentum equation
`was expressed as [ 25 , 26 ]:
`
`ρ( DV / Dt ) = −∇ P + ρg + μ∇ 2 V
`
`where ρ(DV/Dt) is the force, ∇P is the pressure, ρg is the weight-
`ing force, and μ∇ 2 V is the viscosity force. All the forces are related
`
`to unit volume. The heat transfer function is expressed as [27–29] :
`Q = ρC p (∂ T /∂ t) + ρC p u · ∇T + ∇ · q
`
`(cid:4) q = −k ∇T
`
`(1)
`
`(2)
`
`(3)
`
`Fig. 2. Schematic drawing of the immersion cooling system for a CPU of a PC. The
`entire system is immersed in the coolant.
`
`where Q is the heating source, C p is the heat capacity, T is the tem-
`perature, q is the heat flux, and k is the heat transfer coefficient.
`The dimensions of the designed model is 320 mm in length with
`a height of 240 mm and a width of 180 mm.
`
`2.2. Model validation
`
`In order to verify the simulation model, an experimental struc-
`ture is shown in Fig. 2 . The structure includes a CPU (I9–9900 K,
`Intel, USA) on a mainboard (ROG MAXMIMUS X HERO, Asus, Tai-
`wan), and a power supply unit (PSU) (CMPSU-10 0 0HX/PHP, Corsair,
`USA). The I9–9900 K 8 core, 16 thread CPU with a 3.60 GHz pro-
`cess base frequency, and 5.00 GHz maximum turbo frequency, as
`well as a 95 W thermal design power (TDP), is the main source
`of heat within the PC. The mainboard is equipped with 4 × DIMM
`(maximum 64 GB), DDR4 SDRAM memory, two Aura 4-pin RGB-
`strip headers, intuitive Sonic Studio III and Sonic Radar III, and LGA
`1151 processor socket. The PSU is equipped with a cooling fan and
`provides the output power of 10 0 0 W under the operating condi-
`tions of between 100 and 240 V, 6.5–13 A, 47–63 Hz and a tem-
`perature of 50 °C.
`The entire structure with the mainboard and its peripherals are
`immersed in the engineered fluid coolant (3 M Novec 7100, USA).
`The properties of engineered fluid are: coolant itself has a boil-
`ing point of 61 °C, a vaporization heat of 112 kJ/kg, a specific heat
`of 1.183 kJ/kg-K and a thermal conductivity of 0.069 W/m-K. Af-
`ter removing the heat from the CPU with the heat sink, the liq-
`uid coolant flowed out from the outlet at the top left of the PC
`box, passed through a water tank, pump and a radiator for cool-
`ing where it. Finally, it flowed back to the PC box through the in-
`let at bottom side and continued the next circulation cycle. Dur-
`ing the experiment, the consumed power of the PC was measured
`by a power meter (SPG-26MS, Panasonic, Japan) within an accu-
`racy of 0.5% when measuring power consumption, and an accu-
`racy within 0.2% for voltage and 0.3% for current, respectively. The
`temperatures of liquid coolant contacting with the CPU with heat
`sink and under the outlet location were measured by a digital ther-
`mometer (DE-3004 Type-k, DER EE, Taiwan). The resulting temper-
`atures were within an accuracy of ±(0.3% + 1 °C) under the range of
`0~10 0 0 °C. For verifying the experimental and simulation results,
`the CPU with heat sink made by Al was run at a power of 50, 60,
`and 70 W sequentially while the circulated liquid coolant flowed
`
`Immersion Systems LLC – Ex. 1016
`PGR 2021-00104 (U.S. 10,820,446 B2)
`3 of 12
`
`

`

`4
`
`C.-C. Cheng, P.-C. Chang and H.-C. Li et al. / International Journal of Heat and Mass Transfer 160 (2020) 120203
`
`Table 1
`Testing results of the four mesh types.
`
`Mesh types
`
`Coarser
`
`Coarse
`
`Normal
`
`Fine
`
`Mesh points
`Average quality of elements
`
`26,458
`0.5518
`
`44,782
`0.5872
`
`80,282
`0.6128
`
`355,967
`0.6599
`
`Fig. 4. Temperature comparisons between experimental and simulated results mea-
`sured (a) in front of the CPU with heat sink and (b) below the outlet, as shown in
`Fig. 1 . (For interpretation of the references to colour in this figure legend, the reader
`is referred to the web version of this article.)
`
`3. Results
`
`To enhance the accuracy and resolution of the simulated results
`and reduce the time necessary to make calculations and lighten
`the demands on the PC made by running simulations, proper mesh
`sizes and time intervals are very important. In this study four mesh
`sizes of Coarser, Coarse, normal and fine, were utilized to compare
`the mesh points of the model and the time it takes for the PC to
`make calculations. The testing results of these four mesh types are
`listed in Table 1 . The larger amount of mesh points produces better
`simulation results, but increases the period for PC to make calcu-
`lations. The range of the average quality of elements is between 0
`and 1, and larger values are better.
`In Table 1 , the normal mesh type illustrated has 80,282 mesh
`points and an average quality of 0.6128. Among these four mesh
`
`Fig. 3. Temperatures measured at front of heat sink by using the time interval of
`free and exact levels of 0.1, 0.2, 0.4 and 0.8 second.
`
`at a rate of 0.8 m/s for 10 min. The measured temperatures of liq-
`uid coolant in front of CPU with heat sink and under the outlet
`location, as shown in Fig. 2 (a) and (b), compared with the simula-
`tion results. After analysis, the heat radiation effects on the models,
`which the CPU with Al/Cu heat sink was run at a power of 60 W
`and circulated by the liquid coolant at flow rates of 0.4, 0.8 and
`1.2 m/s, would be discussed further.
`
`Table 2
`Mean and standard deviation of temperatures at five points around heat sink.
`
`Front
`
`Top
`
`Bottom
`
`Right
`
`Left
`
`Mean temperature ( °C)
`−4 °C)
`Standard deviation (10
`
`300.09855
`7.83454
`
`300.18024
`2.90511
`
`300.02221
`0.849125
`
`300.16738
`0.542748
`
`300.15226
`302.2
`
`Immersion Systems LLC – Ex. 1016
`PGR 2021-00104 (U.S. 10,820,446 B2)
`4 of 12
`
`

`

`C.-C. Cheng, P.-C. Chang and H.-C. Li et al. / International Journal of Heat and Mass Transfer 160 (2020) 120203
`
`5
`
`Fig. 5. A cross-sectional view showing the simulation results in the temperature
`variations measured in the immersion cooling system when viewed from the XZ
`axis shown in (a) ~ (e) from 2 to 10 min mark. (For interpretation of the references
`to colour in this figure legend, the reader is referred to the web version of this
`article.)
`
`Fig. 6. A cross-sectional view of the simulation results in the temperature varia-
`tions measured in the immersion cooling system when viewed from the YZ axis
`shown in (a) ~ (e) from 2 ~ 10 min mark. (For interpretation of the references to
`colour in this figure legend, the reader is referred to the web version of this article.)
`
`Immersion Systems LLC – Ex. 1016
`PGR 2021-00104 (U.S. 10,820,446 B2)
`5 of 12
`
`

`

`6
`
`C.-C. Cheng, P.-C. Chang and H.-C. Li et al. / International Journal of Heat and Mass Transfer 160 (2020) 120203
`
`types, the normal type is equipped with enough of a mesh quality
`to make calculations in a reasonable period of time based on the
`capacity of the PC. After choosing the mesh type, the backward
`differentiation formula (BDF) of free and exact levels are used to
`verify the proper time intervals, so called time steps. The BDF is a
`linear multistep method that, for a given function and time, calcu-
`late approximately the derivative of the given function using the
`computed results of the previous time points for increasing the
`accuracy of the approximation. This method is particularly suit-
`able for stiff differential equations and Differential Algebraic Equa-
`tions (DAEs). By illustrating the temperatures measured from the
`front, left, right, top and bottom of the heat sink center, the simu-
`lation results utilizing the time intervals of the free and exact lev-
`els were compared. In this study, four exact time intervals of 0.1,
`0.2, 0.4 and 0.8 second and free level were utilized. The distance
`between the heat sink center and the five points was 50 mm, as
`shown in Fig. 1 . The temperature measured at front of heat sink by
`using the time interval of free and exact levels of 0.1, 0.2, 0.4 and
`0.8 second, as shown in Fig. 3 , were typified in symbols of square,
`circle, top-triangle, bottom-triangle and diamond, respectively. In
`Fig. 3 , these temperatures simulated by using these time intervals
`were very consistent, and their mean value and standard devia-
`tion at the 10 min mark were 300.09885 and 7.83454 × 10
`−4 °C,
`respectively. The mean and standard deviation of temperature by
`using these five time intervals at these five points were listed in
`Table 2 . In Table 2 , all standard deviations of these temperatures
`measured at five points were less than 0.01 °C, except that at the
`left side of heat sink, indicating that the temperatures simulated
`by using these five time intervals were in consistence. Therefore,
`the time step at free level would be utilized for simulation in this
`study to reduce the calculation period.
`In order to understand if the simulated results close to the ex-
`perimental ones and if the model settings were proper or not, the
`temperatures in front of the CPU with heat sink and under the out-
`let location were measured, as shown in Fig. 4 (a) and (b), respec-
`tively. The flow rate of the liquid coolant was 0.8 m/s, and the ma-
`terial of heat sink was Al. The experimental and simulated results
`are shown as hollow and filled symbols, respectively. The varying
`power levels of 50, 60, and 70 W which were run through the CPU
`are represented as circles, squares and triangles, respectively. In
`Fig. 4 (a), the temperature measured in front of the CPU with heat
`sink rose from 38 to 59, 63 and 68 °C within the 30 second mark
`while the CPU was running at 50, 60 and 70 W, respectively. These
`temperatures reached their peak values of 60, 65 and 71 °C after
`4 min and 45 second mark. The simulated results also expressed
`similar tendencies. The temperature differences between the simu-
`lated and experimental results at end of the experiment increased
`along with the power that was run through the CPU. They were 1,
`1.2 and 3 °C (or 1.7, 1.5 and 4%) for the CPU running at 50, 60 and
`70 W, respectively.
`In Fig. 4 (b), the temperatures measured under the outlet loca-
`tion rose from 28 to 30 °C within the 30 second mark, and reached
`their final values of 30, 31 and 32 °C with the CPU running at 50,
`60, and 70 W respectively. The temperature variations at this loca-
`tion were smaller than those measured in front of CPU with heat
`sink, due to the circulation of the liquid coolant. This will be dis-
`cussed further in the following sections. The simulated results ex-
`pressed similar tendencies. The temperature differences between
`the simulated and experimental results at end of experiment in-
`creased as the power to the CPU was increased. The temperature
`differences measured were 0.29, 0.75 and 1.23 °C (or 1, 2.5 and
`3.9%) when the CPU was running at 50, 60, and 70 W, respec-
`tively. It is noted that the simulated model neglected the heat pro-
`duced from the mainboard and the several capacitors that are on
`the memory unit. This is the reason why the simulated temper-
`atures under the outlet location were less than the experimental
`
`Fig. 7. A cross-sectional view of the simulation results of the temperature varia-
`tions measured in the immersion cooling system when viewed from the XY axis
`shown in (a) ~ (e) from 2 ~ 10 min mark. (For interpretation of the references to
`colour in this figure legend, the reader is referred to the web version of this article.)
`
`ones. Even though, the differences between the simulated and ex-
`perimental temperatures were still within the acceptable ranges.
`Therefore, this model can be utilized to simulate the heat radiation
`
`Immersion Systems LLC – Ex. 1016
`PGR 2021-00104 (U.S. 10,820,446 B2)
`6 of 12
`
`

`

`C.-C. Cheng, P.-C. Chang and H.-C. Li et al. / International Journal of Heat and Mass Transfer 160 (2020) 120203
`
`7
`
`Fig. 8. Front and side-views of the simulation results in temperature variations within the immersion cooling system at the 10 min mark at flow rates of (a) ~ (b) 0.4 m/s,
`(c) ~ (d) 0.8 m/s, and (e) ~ (f) 1.2 m/s, respectively.
`
`effects of a CPU with various materials of heat sink and circulation
`speeds of the liquid coolant.
`
`4. Discussions
`
`4.1. Heat distribution from various angles
`
`Based on the designed model, showing the distribution of the
`heat radiation around the CPU from various angles can guide the
`readers to better understand how the liquid coolant and heat sink
`remove heat. In this section, an example of CPU with heat sink
`running at 60 W and cooled at a flow rate of 0.4 m/s was ob-
`served. The various angles of heat distribution are shown in cross-
`
`sections of the XZ, YZ, and XY axis of the designed model in Figs. 5 ,
`6 , and 7 respectively. It is noted that the temperature mark was
`from 30 0.0 0 to 30 0.30 K for these three figures. Fig. 5 (a) ~ (e) il-
`lustrates the flow state and temperature variations of the liquid
`coolant during a process time from the 2 to 10 min marks, respec-
`tively. There were three cross-sections on the XZ axis, as indicated
`in Fig. 5 (a). The 3rd cross-section shows a location that is past the
`inlet, outlet and heat sink surfaces. The liquid coolant flowed in
`the model through the inlet, passed the heat sink and increased
`the temperature of the area located at the top of the heat sink.
`The heat also expanded into the 2nd cross-section. In Fig. 5 (b),
`the heat expanded further into the surrounding area, and even to
`the left side of CPU and into the 1st cross-section. In Fig. 5 (c) ~
`(e), the heat distribution reached a stable state. The temperature
`
`Immersion Systems LLC – Ex. 1016
`PGR 2021-00104 (U.S. 10,820,446 B2)
`7 of 12
`
`

`

`8
`
`C.-C. Cheng, P.-C. Chang and H.-C. Li et al. / International Journal of Heat and Mass Transfer 160 (2020) 120203
`
`Fig. 9. Cross-sectional views of the simulation results showing the temperature variations in the immersion cooling system operating at the 10 min mark with a coolant
`flow rates of (a) ~ (c) 0.4 m/s, (d) ~ (f) 0.8 m/s, and (g) ~ (i) 1.2 m/s, respectively.
`
`Fig. 10. Cross-sectional views of the simulation results of temperature variation of immersion cooling system operated at the 10 min mark with a coolant flow rates of
`0.8 m/s. Heat sinks of CPU are made by (a) ~ (c) Al and (d) ~ (f) Cu, respectively.
`
`measured at the top of CPU was the highest. The temperature on
`the right side of the CPU was higher than that of the left side. The
`liquid coolant that flowed from the inlet had the lowest tempera-
`ture.
`Fig. 6 (a) ~ (e) illustrates the flow state and temperature varia-
`tions of the liquid coolant on the YZ axis during a process time
`from the 2 to 10 min mark, respectively. There are four cross-
`sections on the YZ axis from left to right. The 3rd cross-section
`
`shows a location that is past the heat sink and the inlet. The
`outlet location is located between 1st and 2nd cross-sections. In
`Fig. 6 (a), the 3rd cross-section shows the liquid coolant flowing
`from the inlet, passing the heat sink and removing the heat to the
`top of the heat sink in a bar shape. The removed heat also ex-
`panded into the 4th cross-section. In Fig. 6 (b), the removed heat
`expanded further to the 1st and 2nd cross-sections. In Fig. 6 (c) ~
`(e), the removed heat reached a stable state. The top of the 3rd
`
`Immersion Systems LLC – Ex. 1016
`PGR 2021-00104 (U.S. 10,820,446 B2)
`8 of 12
`
`

`

`C.-C. Cheng, P.-C. Chang and H.-C. Li et al. / International Journal of Heat and Mass Transfer 160 (2020) 120203
`
`9
`
`cross-section had the highest temperature. The liquid coolant flow-
`ing from the inlet had the lowest temperature. The temperature in
`the 4th cross-section was higher than those in the 1st and 2nd
`cross-sections.
`Fig. 7 (a) ~ (e) shows the flow state and temperature distribution
`of the liquid coolant on the XY axis during a process time from the
`2 to 10 min mark, respectively. There are five cross-sections on the
`XY axis, and the 4th cross-section is located past the heat sink.
`In Fig. 7 (a), the first 4 cross-sections show that the liquid coolant
`passed through the heat sink and removed the heat towards the
`right side of 1st cross-section. In Fig. 7 (b), the removed heat ex-
`panded towards the direction of the left side of 1st cross-section
`and towards the direction of the right side of the 5th cross-section.
`In Fig. 7 (c) ~ (e), the removed heat reached a stable state. The high-
`est temperature appeared at the top and right side of the heat sink.
`The area from the heat sink to the outlet also had a higher tem-
`perature. The temperature at the right side of the heat sink was
`higher than that of the left side. The liquid coolant that flowed
`from the inlet to the heat sink had the lowest temperature.
`
`4.2. The heat distribution under various flow rates of the liquid
`coolant
`
`After attaining a clear picture of the heat distribution from the
`XZ, YZ, and YX axis cross-sectional views of the designed model,
`it would be quite valuable to also investigate the effect the flow
`rate of the liquid coolant has on the heat radiation of the CPU and
`heat dispersion throughout the liquid coolant. Fig. 8 shows a front-
`side view of the temperature variations around the CPU with heat
`sink during different flow rates of the liquid coolant of 0.4, 0.8 and
`1.2 m/s at the 10 min mark of the process time. The material of
`heat sink was Al, and the CPU was running at 60 W. Fig. 8 (a), (c)
`and (e) illustrates a side-view of the temperature variations around
`the CPU at flow rates of 0.4, 0.8 and 1.2 m/s, respectively. The high-
`est temperatures of CPU at the 3 different flow rates were 348, 340
`and 334 K, respectively. The faster flowing speed of liquid coolant
`would remove more heat, and cause the lower temperature of CPU.
`The temperature measured on the heat sink surface was ranged
`from 300 to 305 K. Fig. 8 (b), (d) and (f) illustrates a front-view of
`the temperature variations around the heat sink at flow rates of
`0.4, 0.8 and 1.2 m/s, respectively. The area at the bottom of the
`heat sink surface showed the lowest temperatures, due to being in
`direct contact with the liquid coolant. The heat of the heat sink
`that originated from the CPU was removed by the liquid coolant
`in a flow direction from the bottom towards the top. The slower
`flow rate of the liquid coolant caused the more concentrated heat
`around the heat sink. The temperatures measured at