A heat pipe is a thermal transport device that uses phase-change processes and vapor diffusion to transfer large quantities of heat over substantial distances, with no moving parts and a
Trang 11 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
[1006],(60)
Lines: 1635 to 1651
———
5.7pt PgVar
———
Normal Page PgEnds: TEX [1006],(60)
In all 12 heat sinks studied, the vertical–vertical orientation, that is, a vertically oriented base with vertical fins and channels, yielded the highest heat transfer coef-ficients most often However, in a relatively large number of situations, the thermal performance of the vertical–vertical arrays was indistinguishable from that attained
by a vertical base plate with fins rotated 30° from the axis, a horizontal base plate, and
a base plate inclined 60° from the horizontal with unrotated fins On the other hand, vetical–horizontal orientation, that is, the base plate vertical and the fins rotated 90°
from the axis, led to the lowest heat dissipation rates For the unrotated fins, the low-est heat transfer coefficients were almost always found to occur at a base plate angle
of 30° from the horizontal The use of smoke revealed a relatively complex three-dimensional flow pattern around the heat sinks, with very substantial inflow from the direction of the fin tips when the base plate was strongly inclined and when the heat sinks were in the vertical base–horizontal fins orientation
The influence of the spacingz between the fins forshort and long base plates was
examined by comparing pairs of heat sinks that differed only in geometric parameters,
z (fin arrays 1 and 4, 2 and 5, 3 and 6, 7 and 10, 8 and 11, and 9 and 12 in Table
13.15) Bilitzky (1986) observed that in nearly all the configurations and operating conditions examined, the highest heat transfer coefficients were attained with the larger fin spacing However, the improvement in the heat transfer coefficient was not always sufficient to compensate for the loss of wetted fin surface area Moreover, for the horizontal base plate configurations as well as for the vertical base with horizontal fins, the total array dissipation appeared not to depend on this parameter
Among the various available cooling techniques, the use of heat pipe technology is increasing rapidly, especially in portable computers Heat pipes can provide a low-thermal-resistance path for heat transfer within electronic equipment, linking a high-powercomponent with a remotely placed heat sink orcold plate, without adding substantial weight to the system A heat pipe is a thermal transport device that uses phase-change processes and vapor diffusion to transfer large quantities of heat over substantial distances, with no moving parts and at nearly a constant temperature A heat pipe is composed of a sealed slender tube containing a wick structure, which lines the inner surface, saturated by a small amount of fluid (such as water), as shown
in Fig 13.30 It is composed of three sections: the evaporator section at one end, where heat is absorbed and the fluid is vaporized; a condenser section at the other end, where the vapor is condensed and heat is rejected; and the adiabatic section, in between, where the vapor and the liquid phases of the fluid flow in opposite directions through the core and the wick, respectively, and in which no significant heat transfer occurs between the fluid and the surrounding medium
The evaporation and condensation processes yield extremely high heat transfer coefficients, and only a modest pressure difference is required to transport the vapor
Trang 21 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
[1007],(61)
Lines: 1651 to 1657
———
0.097pt PgVar
———
Normal Page PgEnds: TEX [1007],(61)
Figure 13.30 Longitudinal cross section of a heat pipe
from the evaporator to the condenser end of the heat pipe However, very careful wick design and assembly is required to ensure that capillary forces, in the wick structure, will be successful in pumping the liquid back to the evaporator Unfortunately, even modest accumulations of noncondensable gas in the vaporspace, especially in the region adjacent to the condenser, can lead to severe deterioration in heat pipe perfor-mance, often in just a few weeks or months of operation Contamination and trapping
of airand/orothernoncondensable gases, during the fabrication process, as well as gas leakage and/or generation due to material incompatibilities, can all contribute to such a loss of heat pipe performance
Due to the small axial gradient in vapor pressure, required to propel the vapor from the evaporator to the condenser, and the strong dependence of vapor pressure on temperature embodied in the Clausius–Clapeyron equation (Dunn and Reay, 1994), even a modest internal temperature difference is sufficient to effect considerable thermal transport Moreoever, since much of the working volume of a heat pipe is occupied by vapor, such heat pipes may weigh just a few grams Indeed, a simple heat pipe with water as the working fluid may have an effective axial thermal conductivity
on the order of 100,000 W/m · K, compared with about 400 W/m · K forcopper For
example, a 0.6-cm-diameter 15-cm-long horizontal cylindrical heat pipe, with water
as the working fluid can transfer 300 W at just a 2- to 3-K temperature difference between the evaporatorend and the condenserend of the pipe
However, the resistance to heat flow in the radial direction, or perpendicular to the heat pipe surface, is often far higher than in the axial direction The relatively low thermal conductivity of the commonly used wick materials, the high porosity of the wick, and the use of low-thermal-conductivity working fluids all combine to impede the flow of heat from the heat pipe wall to the liquid–vapor interface Moreoever, the interface thermal resistance between the heat pipe and the microelectronic device, heat sink, cold plate, and/orotherelements in the thermal path frequently poses
a substantial thermal resistance that may well govern the cooling capability of the heat pipe
Although most heat pipes are cylindrical in shape, they can be manufactured in a variety of shapes, involving right-angle bends, S-turns, or spirals Heat pipes can also
Trang 31 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
[1008],(62)
Lines: 1657 to 1691
———
0.0pt PgVar
———
Normal Page PgEnds: TEX [1008],(62)
be made in a flat configuration, with a minimum thickness that currently is close to 0.3
cm (Zorbil et al., 1988) Flat heat pipes, attached directly to the back surface of a PCB, have been used successfully forcooling high-powerboards in avionic applications, in which heat must be conducted to the edges of the board, which are attached to an air-orwater-cooled, cold plate When the ultimate heat sink is the ambient air, cooling fins are usually attached to the condenser end of the heat pipe, to increase the heat transfer area and reduce the convective resistance
Recent years have seen growing interest in a special category of flat heat pipes,
popularly known as vapor chambers Vapor chambers differ from traditional
cylin-drical heat pipes in relying on two-dimensional rather than one-dimensional flow of vapor and liquid in the chamber core and in the wick, respectively Vapor chambers are used in the base of a heat sink for enhanced lateral spreading of heat from a high-heat-flux component Although the high lateral thermal conductivity of the vapor chambers
is most advantageous in this regard, the relatively low axial conductivity of the vapor chamber, often due to the low thermal conductivity of the wick, high wick porosity, and low fluid conductivity, can compromise the effectiveness of a poorly designed vaporchamber Recently, Chesseret al (2000) provided a simplified scheme to pre-dict the capillary limits of a vaporchamberand suggested ways in which this same analysis could also be used to predict the temperature variations in the flowing vapor
Alternative Designs Murthy et al (2000) presented a micromachined thermo-syphon design to be used as heat spreader The thermothermo-syphon included a central evap-orator section, with integrated fins for cooling along the edges and a microfabricated three-dimensional copper structure for enhancing boiling heat transfer Zeng et al
(2000) have proposed a two-phase cooling design based on the use of an electroki-netic (EK) pump delivering 0.05 to 10 mL/min of electroosmotic flow for the cooling
of microelectronic devices in the range 1 to 200 W Such pumps have no moving parts and are expected to develop pressure heads in excess of 1 atm, substantially above values reported for other micropumps
Thermal control of operational electronic components by direct immersion in low-boiling-point dielectric liquids dates back to the late 1940s In the mid-1980s use of immersion cooling for the Cray 2 and ETA-10 supercomputers, as well as substantial research on jet impingement and spray cooling, led to renewed interest in this technol-ogy Due to elimination of the solid-to-solid interface resistance, immersion cooling
is well suited to the cooling of advanced, high-heat-flux electronic components now underdevelopment
Table 13.16 lists the thermophysical properties for some of the most commonly used dielectric liquids It is clear from the data in this table that these fluids have higherthermal conductivities, densities, and specific heats than airand that, conse-quently, use of these fluids can be expected to offer substantial improvements relative
to conventional air cooling However, the table also indicates that due to the inferiority
of the thermophysical properties to those of water, the thermal performance afforded
Trang 41 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
[1009],(63)
Lines: 1691 to 1713
———
0.75417pt PgVar
———
Normal Page PgEnds: TEX [1009],(63)
TABLE 13.16 Thermophysical Properties of Typical Dielectric Coolants
Liquid density (kg/m3 ) 1633 1635 1592 1575 1511 Vapordensity (kg/m3 ) 11.58 11.25 12.68 13.28 7.40 Dynamic viscosity
(N· m/s2 )
4.2 × 10−4 5.2 × 10−4 4.5 × 10−4 4.2 × 10−4 5.0 × 10−4
Specific heat (J/kg · K) 1088 1059 1088 1130 979 Thermal conductivity
Heat of vaporization (J/kg)
87,927 104,675 87,927 79,553 146,824 Surface tension (N/m) 8.9 × 10−3 1.1 × 10−2 8.5 × 10−3 7.7 × 10−3 1.5 × 10−3
Source: Danielson et al (1987).
by these fluids is significantly lowerthan that expected from the use of wateras a coolant
Immersion cooling systems can be classified as passive or active Passive systems are those in which no pumping devices are used to circulate the coolant In such systems, heat transfer from the hot component to the coolant may occur due to natural convection orpool boiling In an active system, on the otherhand, the heat transfer from the immersed component is governed by forced convection and/or by flow boiling to the fluid actively circulated by a pump
As seen from Table 13.16, the Prandtl number for most commonly used dielectric fluids is very high In addition, the length scales of electronic components are usually smaller than those used to establish the free- and forced-convection correlations listed
in Table 13.4 The effect of length scale is much more pronounced on the natural convection heat transfer coefficient Park and Bergles (1988) provided a correlation accounting forthe effect of heaterlength on the natural convection heat transfer coefficient from vertically oriented heaters immersed in dielectric fluids:
Nu= a · Ra b
where
a = 0.906 1.0 + (W/W0.0111
∞)3.965
0.2745
(13.82)
b = 0.184 1.0 + (W/W2.64 × 10−5
∞)9.248
−0.0362
(13.83)
Although eq (13.81) was developed from measurements taken on vertically oriented heaters, it predicts natural convection on horizontal silicon chips quite well
The importance of substrate conduction on natural convection heat transfer from a package to a dielectric coolant has been addressed by Sathe and Joshi (1992) and by
Trang 51 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
[1010],(64)
Lines: 1713 to 1722
———
2.25099pt PgVar
———
Normal Page PgEnds: TEX [1010],(64)
Wroblewski and Joshi (1992) Both numerical studies found that a significant portion
of the heat was conducted from the power-dissipating component into the substrate and then convected to the surrounding coolant Sathe and Joshi (1992) also found that natural convection liquid cooling was preferable to natural convection air cooling only
if the ratio of the package thermal conductivity to the fluid thermal conductivity was less than 1
Due to the large liquid densities and the flow rates commonly encountered in active immersion cooling methods, the liquid flow regime is usually transitional or turbulent
The equations listed in Table 13.4 can be used to obtain fairly accurate estimates of forced-convection heat transfer coefficients on electronic components being cooled
by flowing dielectric coolants Although single-phase natural or forced convection of the dielectric coolants can be used for cooling of electronics, the high heat transfer efficiency of nucleate boiling makes this phase-change process a highly attractive mode of heat transfer Once boiling is initiated on the component very large changes
in component heat dissipation can be accommodated with a very small change in component temperature This is clear from the steep slope of the pool boiling curve forFC-72 shown in Fig 13.31
Most commonly used electronic cooling fluids have rather large air solubilities and very low surface tension Consequently, prolonged exposure of any surface to these fluids allows the fluid to dissolve air out of the microscopic surface cavities In the
Figure 13.31 Nucleate pool boiling curve for FC-72 on a horizontal silicon chip (From Watwe et al., 1997.)
Trang 61 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
[1011],(65)
Lines: 1722 to 1747
———
0.42209pt PgVar
———
Normal Page PgEnds: TEX [1011],(65)
absence of these air-filled surface cavities, boiling cannot be initiated on the surface
of the silicon Thus with increasing component heat flux the die temperature increases rapidly, while heat transfer from the component occurs primarily via natural convec-tion, as shown in Fig 13.31 When the temperature difference between the component surface and the bulk coolant is sufficiently large, boiling may be initiated, causing
a rapid decrease in the component temperature The initial increase in component temperature followed by the temperature drop subsequent to the onset of boiling is
termed incipience overshoot Such a thermal shock is rather undesirable in the design
of an immersion cooled system due to component reliability considerations
Although it is very difficult to predict incipience overshoot, You et al (1990) have presented a comprehensive investigation of the parameters that affect nucleate boiling incipience forthe highly wetting dielectric coolants Bar-Cohen and Simon (1986) have proposed the following equation to obtain and engineering estimate of the incipience temperature excursion∆Tex:
∆Tex≤ 2T sσl
h f v 4× 106
1
ρf −
1
ρl
ρ
f
ρf− ρv
− 0.003
qh2
f v
C3
pµf
σf
g(ρ f− ρv )
1/3
· Pr1.7 (13.84)
Once nucleate boiling initiates on the component, the relationship between the com-ponent heat flux and its temperature rise above that of the coolant saturation temper-ature is given by the Rohsenow (1951) correlation:
q= µf h fg
g(ρ f− ρg )
σ
C sf
a · Pr1.7
f · h fg
(T w − Tsat ) n (13.85)
Watwe et al (1997) determined that values ofa and n in eq (1.52) were 0.0075 and
7.5, respectively, for fluorocarbon FC-72 boiling on a 1-cm2silicon chip Danielson
et al (1987) have reported values ofa ranging from 0.003 to 0.0093 and values of
n ranging from 4 to 8 for nucleate boiling of FC-72 Watwe et al (1997) have also
shown that eq (13.85) can be used to predict the boiling heat transfer characteristics accurately at elevated ambient pressure or under subcooled liquid conditions
Flow boiling in channels occurs due to bubble generation at the channel walls
Depending on the fluid velocity, channel orientation with respect to gravity and hy-drodynamic effects due to vapor and liquid density differences, several different flow regimes can be identified Following initiation of boiling, the heat transfer coefficient
at the wall increases significantly while the vapor bubbles are carried away by the flowing liquid This is the bubbly flow regime With increasing vapor volume frac-tion, individual bubbles coalesce to form large vapor slugs This constitutes the slug flow regime
At very high heat flux levels, the rate of vapor generation is so high that the indi-vidual bubbles coalesce into large vapor mushrooms These large vapor mushrooms depart from the heated surface periodically due to hydrodynamic instabilities caused
Trang 71 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
[1012],(66)
Lines: 1747 to 1791
———
0.5102pt PgVar
———
Custom Page (-10.0pt) PgEnds: TEX
[1012],(66)
by density differences between the vapor and liquid flow If the surface of the heater cannot be rewetted following the cyclic departure of these vapor mushrooms, the sur-face temperature is expected to rise rapidly and cause the heater to melt or burnout
The heat flux at which such a dryout/burnout phenomenon occurs is called critical heat flux (CHF) It is extremely essential to ensure that the critical heat flux
con-dition does not occur on an immersion-cooled electronic component to ensure its functionality and reliability Watwe et al (1997) extended the classical Kutateladze (1951)–Zuber (1959) Haramura-Katto (1983) model of the pool boiling critical heat flux to propose the equation
CHF= π
24h fgρgσf g(ρ f− ρg )1/4 δ
ρs C s k s
δρs C s k s + 0.1
×1.3014 − 0.01507L)
1+ 0.03
ρ
f
ρg
0.75 C pf
h fg
(Tsat− Tbulk )
(13.86) where
L= L
g(ρ f − ρg )
andδ is the heaterthickness Equation (13.86) is a composite equation that includes effects such as heater size, heater thermophysical properties, system pressure, and bulk liquid subcooling (temperature difference between the saturation temperature and bulk liquid temperature can be used to estimate critical heat flux during pool boiling of dielectric coolants)
Lee and Simon (1989) proposed the following equation to compute the critical heat flux on heater lengths ranging from 0.25 to 3 mm for flow velocities from 1 to
17 m/s and subcooling ranging from 13 to 68°C:
CHF= ρv h f v U
0.04ρf
ρv
0.99
σf
ρf U2L
0.33
1+ 3.03ρf
ρv
0.78
C p ∆Tsub
h f v
0.42
(13.88)
Most practical immersion cooling systems operate in a closed loop, where the vapor
of the dielectric liquid is condensed and returned to the electronic enclosure Two such
systems are shown in Fig 13.32a and b In Fig 13.32a a “remote” condenser, external
to the electronic enclosure and cooled by water, air, or other fluid, condenses the vapor leaving the enclosure and directs the condensate back to the enclosure for reuse In
the configuration represented by Fig 13.32b, the condenseris located in the vapor
Trang 81 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
[1013],(67)
Lines: 1791 to 1791
———
* 43.927pt PgVar
———
Custom Page (-10.0pt) PgEnds: TEX
[1013],(67)
Figure 13.32 Two closed-loop immersion cooling systems
Trang 91 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
[1014],(68)
Lines: 1791 to 1793
———
0.097pt PgVar
———
Normal Page PgEnds: TEX [1014],(68)
Figure 13.33 Submerged condenser immersion cooling systems
space above the liquid—producing a more compact immersion module design—and the condensate drips back into the liquid
As discussed previously, due to the high solubility of air in the perfluorinated fluorocarbons, often used as immersion cooling liquids, it is not uncommon for vapor space condensers to be affected adversely by a buildup of noncondensable gas
Trang 101 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
[1015],(69)
Lines: 1793 to 1825
———
-0.50587pt PgVar
———
Normal Page
* PgEnds: Eject
[1015],(69)
Such difficulties can be avoided by submerging the condenser (i.e., heat exchanger
tubes) in the liquid, as shown in Fig 13.33a The circulating water through the tubes
absorbs the heat from the dielectric liquid, thus subcooling the liquid Any vapor bubbles generated by boiling on the component surfaces collapse and condense in the subcooled liquid
As a further modification of this approach, it is possible to use the side and top walls of the liquid-filled enclosure to serve as the submerged condenser, which can
then be externally air- or liquid-cooled, as shown in Fig 13.33b Nelson et al (1994),
Kitching et al (1995), and Geisler et al (1996) have performed empirical studies on the cooling capabilities of “passive immersion cooled multichip modules” based on the concepts discussed above Nelson et al (1994) showed that the overall thermal performance of these modules was not affected by the presence of uneven chip power densities The maximum powerthat can be dissipated from such modules depends on the nucleate pool boiling critical heat flux on the power-dissipating component as well
as the condensation limit of the condenser The thermal behavior of the module can
be plotted on performance maps where the lower bound marked is determined by natural convection heat transferon both heaterand condensersections The natural convection correlation by Park and Bergles (1988) and those listed in Table 13.4 can
be used to estimate the lowerbound
Once pool boiling is initiated on the heaters, one of the following two scenarios determines the cooling limit:
1 The maximum powerdissipation from the module can be limited by the occur-rence of the critical heat flux condition on the heater Geisler et al (1996) have shown that it may be possible to raise the critical heat flux limit by reducing the module-to-ambient thermal resistance, thereby increasing the subcooling of the coolant inside the module The Watwe et al (1997) correlation [see eq (13.86)] can be used to compute the critical heat flux limit
2 The maximum powerdissipation from the module can also be limited if the condenseris unable to condense the volume of vaporgenerated at the heater The uncondensed vapor will occupy the near wall regions of the condenser, thereby dra-matically reducing the condenser efficiency
The condensation limit on an unfinned flat plate condensercan be computed using the following correlation proposed by Gerstmann and Griffith (1966):
Nu=
0.81Ra0.193 for1010> Ra > 108 (13.89)
0.69Ra0.20 for108> Ra > 106 (13.90) where the Nusselt numberNu and the Rayleigh numberRa are defined as follows:
Nu= h
k f
σf
g(ρ f− ρv ) cos θ
1/2
(13.91)
Ra=g cos θ ρ f (ρ f− ρv )h f v
kµ ∆T
σf
g(ρ f − ρv ) cos θ
3/2
(13.92)