Frontiers in Heat and Mass Transfer (FHMT), 1, 013002 (2010) Global Digital Central

DOI: 10.5098/hmt.v1.1.3002 ISSN: 2151-8629

Frontiers in Heat and Mass Transfer

Available at www.ThermalFluidsCentral.org

VOLUME OF FLUID SIMULATION OF BOILING TWO-PHASE FLOW IN

A VAPOR-VENTING MICROCHANNEL
Chen Fang*, Milnes David, Anita Rogacs, Kenneth Goodson
Mechanical Engineering Department, Stanford University, Stanford, CA, 94305

ABSTRACT
Vapor-venting microchannel heat exchangers are promising because they address the problems of high pressure drop, flow instability, and local dry-
out that are common in conventional two-phase microchannel heat sinks. We present a 3D numerical simulation of the vapor-venting process in a
rectangular microchannel bounded on one side by a hydrophobic porous membrane for phase-separation. The simulation is based on the volume of
fluid (VOF) method together with models for interphase mass transfer and capillary force. Simulation shows the vapor-venting mechanism can
effectively mitigate the vapor accumulation issue, reduce the pressure drop, and suppress the local dry-out in the microchannel. Pressure surge is
observed in the vapor-venting channel. The simulation provides some insight into the design and optimization of vapor-venting heat exchangers.

Keywords: Volume of Fluid, Porous Membrane, Capillary-induced Phase Separation

microchannels sandwiching a porous, hydrophobic membrane in-

1. INTRODUCTION between. A similar approach was applied to microdirect methanol fuel
cells for the reaction byproduct removal (Meng et al., 2004; Meng et
Continuing increases in component density and reductions in VLSI al., 2007). An adiabatic test using an air-water mixture showed an air-
chip dimensions demand the development of high-performance and mass venting efficiency exceeding 95% and a significant pressure drop
low-profile thermal packaging solutions. According to the International compared to the non-venting model. However, the opaque vapor-
Technology Roadmap for Semiconductors high performance venting membrane and a silicon substrate made the flow pattern
microprocessors require heat removal rates exceeding 200W with a visualization in the microchannel unavailable. A solution to that issue is
thermal resistance of up to 0.1 K/W (David et al., 2007). Conventional using a quartz substrate and/or a semi-transparent membrane. As an
cooling techniques based on fans and heat pipes do not scale well with alternative to the silicon-based vapor venting heat exchanger, David et
diminished device dimensions. Several studies (Tuckerman et al.; Qu et al. (2008) developed vapor-venting copper heat exchangers with
al., 2002; Kandlikar et al., 2004; Garimella et al., 2004) showed that integrated hydrophobic PTFE membranes. However, no overall
single-phase liquid convection in microchannels can dissipate large reduction in thermal resistance or pressure-drop was observed in the
heat fluxes only at the expense of high pumping power. Two-phase open venting device. The analysis attributed this to the water saturation
boiling flow in microchannels promises low thermal resistances with of the carbon-fiber membrane that drastically reduced the venting area.
low pumping power requirements. However several drawbacks make Similar to silicon-based devices, the copper heat exchanger lacked the
two-phase heat sinks less attractive at present. These include flow optical access, making it hard to study the vapor-venting mechanism.
instabilities (Kennedy et al., 2000), high peak pressure drops (Zhang et More recently, visualization of the gas-venting process was achieved
al., 2005; Garimella et al., 2002), and dry-out, which causes hotspots. by David et al. (2009) and Alexander et al. (2008) using a porous
These issues are at least partially attributed to the excessive vapor hydrophobic Teflon membrane and microfabricated hydrophobic
phase accumulation in the microchannel during the boiling process. trenches for phase separation, respectively. Both studies were based on
One proposed solution to the vapor phase accumulation, put forth by an adiabatic system involving the air-water mixture and a single air
Zhou et al. (2006), is to locally vent the vapor phase generated in the injection site emulating the boiling nucleation. Although those studies
system into a separate channel, leaving the liquid channel primarily yielded valuable information about the flow pattern and vapor-venting
occupied by the liquid phase. The vapor-escape mechanism reduces the effectiveness, the coupling between heat transfer and flow were not
two-phase pressure drop to pressures seen in systems of single-phase considered.
microchannels while maintaining the high heat fluxes associated with Transient 3D numerical simulations based on the Volume Of Fluid
the phase change process. The removal of the vapor phase may also (VOF) method provide an alternative way to explore the bubble
minimize the vapor-expansion-induced flow instability and the dry-out dynamics and venting mechanism, which is essential for both
caused by the local liquid film rupture. understanding the underlying physics of the vapor-venting process and
The core technological challenge of the vapor-venting heat device optimization purposes. Unlike many other numerical methods
exchanger lies in achieving phase separation that is reliable and leakage treating liquid and vapor phase separately, the VOF method
free for the liquid phase. The first experimental realiziation of this dynamically tracks the interface and requires no knowledge of the
approach was provide by David et al. (2007). The design, as illustrated interface location (e.g., Fang et al., 2008). Yang et al. (2008)
in Fig.1, incorporates a pair of parallel liquid and vapor silicon performed VOF simulations of boiling flow of R141B in a horizontal

* Corresponding author, Email: chenfang@stanford.edu

1
Frontiers in Heat and Mass Transfer (FHMT), 1, 013002 (2010) Global Digital Central
DOI: 10.5098/hmt.v1.1.3002 ISSN: 2151-8629

coiled tube. The calculations treated the phase change as a local 1 ∂u ∂u

Sij = ( j + i ) (2)
temperature dictated process based on a quasi thermo-equilibrium
assumption, which yielded a flow pattern in good agreement with the 2 ∂xi ∂x j
experimental observation. Zhang et al. (2001) relaxed the thermo- The velocity field u is subject to the incompressibility constraint
equilibrium assumption and treated the condensation as a process ∇ u = 0 (3)
dictated by the heat transfer rate across the interface. However, the VOF method defines the interface separating immiscible liquids of
model requested an artificial source term to be added to the enthalpy different density and viscosity based on a phase indicator – the volume
equation to keep the liquid-vapor interface temperature at the saturation fraction function α . To allow for mass conservation of each phase, an
point. The VOF method was also used to model other types of phase additional equation is introduced:
change process, such as boiling (Welch et al., 2000), solidification ∂α
(Takata et al., 1999), and laser-induced vaporization (Ganesh et al., + u ∇α = Src (4)
1997). The phase change process in the porous structure poses a new ∂t
challenge to the modeling due to the interaction between porous media where the volumetric source term Src dictates the phase change rate.
and fluid. Most previous studies utilized either the pore network model We use (4) to represent and track the interface which is transported by
(Figus et al., 1999) or the finite element model which treats the liquid the velocity field u . The energy equation is written as:
and vapor zone separately while coupling two phases by imposing a ∂( ρ E ) / ∂t + ∇[u ( ρ E + p)] = ∇( K ∇T ) + Q (5)
pressure jump on the interface (Kaya and Goldak, 2006; Ren et al., where Q represents the phase change induced energy transfer, which is
2007; Figus et al., 1999). Alternatively, VOF method dramatically computed from the evaporation mass rate Src and the latent heat h:
facilitates the interface tracking. However, the capillary force Q = −hSrc (6)
associated with the multiphase flow in porous media needs to be
The density, viscosity, thermal conductivity, and energy at the interface
properly modeled within the VOF framework.
are calculated using the volume fraction α :
We use the VOF method to study 3D boiling flow in a rectangular
microchannel with a hydrophobic vapor-venting membrane on one ρ = αρl + (1 − α ) ρv (7)
boundary. The phase separation is modeled using a capillary force μ = αμl + (1 − α ) μv (8)
model. The simulation illustrates the impact of applied heat flux,
membrane hydrophobicity, and channel geometry on the flow pattern, K = α K l + (1 − α ) K v (9)
pressure drop, and temperature profile in the vapor-venting (αρl Cv ,l + (1 − α ) ρv Cv ,v )(T − Tsat )
E= (10)
microchannel. A comparison between the conventional channel and the
vapor-venting channel clearly shows the pros and cons of the new
ρ
design. The simulation result provides a tool for designing and The current VOF model employs the Marker and Cell (MAC) finite
optimizing the next generation vapor-venting micro heat exchanger. difference method based on rectangular grids. The volume fraction C
and the pressure p are defined at the center of each cell, while the
horizontal and vertical velocity components are defined at the
midpoints of the vertical and horizontal sides, respectively. At each
time step the volume fraction α , which indicates the two-phase
interface position, is first updated followed by the evaluation of the
surface tension force. Finally, the velocity and pressure fields are
solved using Eq.(1).
The interface location is updated using the Young’s piecewise
linear interface construction (PLIC) scheme (Youngs, 1982), in which
(a) Schematic of a vapor-venting microchannel the normal vector n of the interface is first related to the gradient of
volume fraction C by n = ∇C / | ∇C | . Then, a flat plane having a
normal vector n is used to approximate the real liquid-gas interface
within each cell. With the knowledge of the interface position change,
the volume of fluid transferring into the neighboring cells is evaluated
and the volume fraction α updated. Due to the small Reynolds number
(b) Liquid containment enabled by a hydrophobic membrane which is typical for microscale flows, a laminar flow model is used.
The equations are solved by an algebraic multigrid (AMG) solver (Issa
Fig. 1 Schematic of the vapor-venting microchannel design and the et al., 1991), and the pressure-implicit with splitting of operators
liquid-vapor transport through a hydrophobic porous membrane (PISO) scheme for pressure-velocity coupling (Brackbill et al., 1992).
2.2 Surface Tension in Channel Region
2. PHYSICAL MODEL
Interfacial surface tension arises as the result of intermolecular
attractive forces. The Young-Laplace equation expresses the surface
2.1 Governing Equations tension in terms of the surface curvature k , surface tension
The VOF method uses one set of governing equations to model both the coefficient σ , and the surface normal n :
liquid and vapor region, which substantially facilitates the interface f = σ kn (11)
tracking and interfacial force treatment. The flow is modeled with the In our simulation, surface tension is modeled using the Continuous
Navier-Stokes equation: Surface Force (CSF) scheme (Brackbill et al., 1992). The surface
∂u (1) tension is modeled by a source term f in the momentum equation:
ρ( + (u ∇)u ) = −∇p + ∇( μ S ) + f
∂t 2ρ
where ρ , μ and f are the density, viscosity, and body force term, f =σk ∇α (12)
ρ1 + ρ 2
respectively. The rate of strain tensor S can be expressed as:
where ρ is given by (7). The density ρ in Eq.(12) inhibits the
numerical smearing of the interface and is elaborated by Brackbill et

2
Frontiers in Heat and Mass Transfer (FHMT), 1, 013002 (2010) Global Digital Central
DOI: 10.5098/hmt.v1.1.3002 ISSN: 2151-8629

al.(1992). In the present study, Maragoni effects are not considered, Equation (20) assumes an isotropic and homogeneous effective thermal
therefore σ is treated as constant. The curvature k is defined in terms of conductivity keff , which can be obtained based on the membrane
∧
the divergence of the unit normal n : porosity ϕ :
 1  keff = ϕ k f + (1 − ϕ )ks (21)
k = −(∇ • n ) = − ∇ • ( ∇α )  (13)
In Eqs. (20) and (21), the fluid properties ρ f , k f , and energy E f are
 ∇α 
where determined by Eqs. (7), (9), (10), respectively. The energy source term
∧ n Q is calculated by Eq. (6).
n= (14)
n 2.4 Phase Change Model
Substituting (13) into (12) yields: Yang et al. (2008) used a mass transfer model proposed by Lee (1980)
∇α 1  ∇α  (15) to simulate the two-phase boiling flow in a coiled tube. By assuming
f = 2σρ ( • ∇) ∇α − (∇ 2α )  that phase change occurs at a quasi thermo-equilibrium state at a
ρ l + ρ g ∇α  ∇α  constant pressure and that the phase change rate S is mainly dictated
In applying formulation (15), a Marker and Cell (MAC) like by the local temperature Tl , the saturation temperature Tsat , and the
discretization scheme is used. A unit vector n is calculated from
discretizing the volume fraction α of the immediate 26 neighboring volumetric fraction of the liquid/vapor phase α l and α v , we have:
cells, based on the relation n = ∇α . Initially, a cell corner value of the Src = rlα l ρl (Tl − Tsat ) / Tsat Tl ≥ Tsat (22)
normal vector is computed. Then required cell-centered values are Src = rvα v ρv (Tl − Tsat ) / Tsat Tl < Tsat
obtained by averaging the corner values. Also, ∇ 2α is calculated by Yang et al. (2008) recommends setting the empirical coefficient rl
discretizing the normal vector values obtained at the eight cell corners.
and rv to be 100 s −1 in order to numerically maintain the interface
2.3 Flow in Porous Membrane
temperature at Tsat , and the choice was justified by a good agreement
Unlike in the channel region, flow field in the porous region is between the model prediction and the experiment. Too large rl and
described by the superficial velocity us defined based on the volumetric
rv induces a numerical oscillation, whereas excessively small values
flow rate instead of the real physical velocity. The incompressibility
constraint yields: cause the interfacial temperature to substantially deviate from Tsat . In
∇us = 0 (16) the present study, the source term Src in Eq. (4) is evaluated using Eq.
For a low flow velocity, we have a laminar flow with a negligible (22), where α l and α v can be obtained within the framework of VOF
inertial force. The momentum equation reads: model. An alternative method of regulating the interface temperature
μ (17) based on artificial parameters was discussed by Zhang et al. (2001).
−∇P = us − f capillary
K 2.5 Geometric Configuration and Boundary Conditions
which assumes homogeneous, isotropic porous media giving a constant
permeability K . f capillary represents the capillary force acting on the In the present study, a straight channel with a rectangular cross-section
is considered. A vapor-venting porous membrane covers the top of the
liquid-vapor interface. Unlike the surface tension force on the liquid-
channel. A uniform heat flux is applied to the bottom and two
vapor interface in the channel region, the capillary force in the porous
sidewalls. All other walls are set to be adiabatic. A velocity and a
media represents the interaction between the porous structure and a set
pressure boundary condition are assigned to the channel inlet and
of small meniscuses on the order of the pore size, and is therefore
outlet, respectively. Pure liquid water is pumped into the channel at a
independent of the macroscopic curvature of the interface. By replacing
temperature around 3K below the saturation point, creating a sub-
the σ k term in (12) with the capillary pressure jump Pcapillary across the
cooling boiling environment. The membrane-channel interface has a
interface, we have: no-slip boundary condition and an effective contact angle, while the
2ρ (18) other side of the membrane has a pressure boundary condition. The
f capillary = Pcapillary ∇α geometry of the simulation domain is shown in Fig.2 and the
ρ1 + ρ 2
parameters are listed in Table 1.
The capillary pressure Pcapillary depends on the pore size and the
wettability of the porous structure. A simplified model treating the
porous structure as a bundle of capillary tubes gives the equation:
4σ f cos θ
Pcapillary = (19)
D
where D is the average pore size and σ f is the surface energy. The
equilibrium contact angle θ is a measure of the porous media
wettability. A hydrophobic porous membrane with large θ leads to a
capillary force pointing toward the liquid phase, preventing the liquid
leakage through the porous media. In this study, the venting membrane
is treated as a homogeneous porous media, therefore Pcapillary is Fig. 2 Computational domain and boundary conditions
independent of the orientation and the location of the meniscus.
The energy equation in the porous region takes the form: Models discussed in Section 2.1-2.4 are applied to the channel and
∂[ϕρ f E f + (1 − ϕ ) ρ s Es ] membrane region to account for a variety of physical processes. The
+ ∇[us ( ρ f E f + P)] = ∇(keff ∇T ) + Q (20) phase change model and surface tension model are assigned to the
∂t channel domain, while the capillary force model is employed in the

3
Frontiers in Heat and Mass Transfer (FHMT), 1, 013002 (2010) Global Digital Central
DOI: 10.5098/hmt.v1.1.3002 ISSN: 2151-8629

membrane region to simulate the capillary-induced phase separation. where h is the liquid column height, ρ , μ , K , ε are the liquid density,
The VOF method has been employed in commercial package FLUENT dynamic viscosity, permeability, and porosity, respectively. Equation
to simulate multiphase flow with a free surface and is used in the (23) can be readily solved using the 4th order Runge-Kutta method and
present study. However, the modeling of surface tension acting on the the solution is compared with the current VOF model prediction, as
liquid-vapor interface in the microchannel (Section 2.2), the capillary shown in Fig.3(b). A perfect match for liquid rise velocity is obtained
force model (Section 2.3), and the phase change model (Section 2.4) for the whole imbibition process, indicating that the current model can
are implemented using a customized user defined code, which is correctly capture the dynamics of capillary transport in porous media
hooked up with the main VOF code to create a closed form model. x 10
-5
1.8
In the present study, a 3D structured orthogonal grid with about
191,400 computation nodes was employed. The grid independency was 1.6 Liquid Filling in Porous Tube
validated by performing different numerical simulations for one 1.4

Liquid Front Velocity (m/s)

particular combination of heat flux, inlet velocity, and outlet pressure
1.2
with varying number of nodes. This original grid was refined twice by
1
increasing the number of cells by 20% and 40% in every direction
leading to a 95% and 200% finer grid, respectively. Simulation results 0.8 Analytical Solution
showed that the difference between them is minor. Hence, the original 0.6
VOF Solution

computational grid with 191,400 was adopted for simulation cases. The
0.4
mesh statistics are listed in Table 1.
0.2

Table 1 Input parameters for the present vapor-venting modeling 0

0 1 2 3 4 5
Time (s) 4
x 10
Parameter Value
(a) Pressure profile at t = 3.5e4s (b) Liquid rise velocity
Channel Length ( mm ) 5 Fig. 3 Calculated pressure profile and liquid rise velocity for the
Channel Width ( μ m ) 300 simulation of water rise in a vertical porous tube
Channel Height H c ( μ m ) 150
Membrane Thickness H m ( μ m ) 50
4. RESULTS AND DISCUSSION
Membrane Permeability-Liquid ( m )
2 4e-22
4.1 Flow Pattern Overview
Membrane Permeability-Vapor ( m )
2 8.8e-15
The vapor venting mechanism dramatically changes the mixture quality
Membrane Contact Angle (deg) 140
distribution along the channel, which in turn exerts a significant impact
Membrane Pore Size ( μ m ) 0.24 on the flow pattern. Figure 4 shows the comparison of flow patterns
Capillary Pressure (Pa)* 8.9e5 between a vapor-venting channel and a conventional channel for heat
Membrane Porosity 0.6 fluxes ranging from 10 W / cm2 to 23 W / cm2 , in which the left and
Surface Heat Flux ( Wm−2 )** Varying right column correspond to the vapor-venting and conventional
channel, respectively. To make the two channels comparable, identical
Liquid Inlet Velocity ( m / s ) 0.27
initial condition and boundary condition are assigned to both channels
Outlet Gauge Pressure (Pa) Varying except that the vapor-venting channel has its top solid wall replaced by
Membrane Outer Surface Gauge Pressure (Pa) 0 a hydrophobic porous membrane.
Mesh Size-Channel Space (L x W x H) 330 x 20 x 25 For a low heat flux of 10 W / cm2 , as shown in Fig. 4(a), the
Mesh Size-Membrane Space (L x W x H) 330 x 20 x 4 upstream of both channels is dominated by a bubbly flow. Simulation
shows that the bubble size increases in the streamwise direction with
Inlet Liquid Temperature (K) 370
the liquid phase temperature. For a vapor bubble attached to a
superheated surface, the evaporation occurs near the bubble base (Fig.5
region A) which supplies the vapor to the bubble, while the bubble top
3. VALIDATION OF CAPILLARY FORCE MODEL interface contacting the subcooled liquid (Fig.5 region B) is dominated
The phase change and the surface tension models used in the present by the condensation consuming the vapor phase. The bubble size at
study have been validated in previous literatures (Brackbill et al., 1992; equilibrium is dictated by the balance between boiling and
Yang et al., 2008). The newly proposed capillary force model in Eq. condensation, and increases with the bulk liquid temperature. A
(18) is validated with a study of the capillary inhibition process into a relevant modeling work for this process was done by Li et al. (2002).
straight porous cylinder. The water inhibition into the initially dry The occurrence of vapor venting requires that the bubble grows to a
porous media is driven by the assigned capillary pressure Pcapillary of certain size and touch the channel top wall. Hence, the small
equilibrium bubble size in the upstream suppresses the vapor-venting,
700Pa. Figure 3(a) illustrates a snapshot of the pressure profile in the making upstream flow patterns in both channels similar. Small bubbles
porous cylinder at time 3.5e4s. The modeling prediction of capillary- reaching the departure size may detach from their nucleation sites and
induced pressure jump across the air-liquid interface agrees well with be convected to the downstream. Those convected bubbles,
the specified Pcapillary . For comparison, a simple 1-D analytical equation nevertheless, tend to move towards the center of the channel due to the
can be used to describe the capillary transport in the domain: Fahraeus-Lindquist effect (Lee et al. 2004), and therefore never touch
the membrane surface for venting. In the downstream region, however,
a large equilibrium size tends to induce the coalescence of neighboring
d 2h dh μ dh (23)
ρε h 2
+ ρε ( )2 + h + ρ gε h − Pcap = 0 bubbles, resulting in a vapor venting process (Fig.4(a1)~4(a4)). In
dt dt K dt contrast, over-sized bubbles in the ordinary channel coalesce and form
elongated slugs (Fig.4(a6)~4(a8)), which significantly reduces the
cross-sectional area available for the liquid flow.

4
Frontiers in Heat and Mass Transfer (FHMT), 1, 013002 (2010) Global Digital Central
DOI: 10.5098/hmt.v1.1.3002 ISSN: 2151-8629

(a1) (a5)
t = 2.3532e -2

(a2) (a6)
t = 2.3582e -2

(a3) (a7)
t = 2.3632e -2

(a4) (a8)
t = 2.3957e -2
(a) Case 1: 10 W/cm^2

(b1) (b4)
t = 2.6059e -2

(b2) (b5) Dry-out

t = 2.6334e -2 Region

(b3) (b6)
t = 2.6484e -2
(b) Case 2: 18 W/cm^2

(c1) (c4)
t = 2.0557e -2

(c2) (c5)
t = 2.1332e -2

(c3) (c6)
t = 2.2932e -2
(c)Case 3: 23 W/cm^2
Fig. 4 Comparison of two-phase flow patterns in vapor-venting and conventional channels for various heat fluxes (perspective view)

5
Frontiers in Heat and Mass Transfer (FHMT), 1, 013002 (2010) Global Digital Central
DOI: 10.5098/hmt.v1.1.3002 ISSN: 2151-8629

dampened by the viscous resistance (Fig.6(h)). The simulated venting

process is similar to what was observed by David et al. (2009) in a
adiabatic water-air system, as shown in Fig.7. The bubble venting
duration is around 0.5ms in both experiment and simulation.

Membrane

liquid

vapor

(a) 23.482ms (b) 23.545ms

Fig. 5 Equilibrium between condensation and evaporation on a
bubble interface during the nucleate boiling

The flow pattern for an intermediate heat flux of 18 W / cm2 is

shown in Fig.4(b). A raised heat flux causes the bulk liquid
temperature to grow at a faster rate, yielding a larger equilibrium
bubble size as opposed to that in case 1. The simulation clearly
shows that the vapor venting process (Fig.4(b1)~4(b3)) can
effectively limit the buildup of the two-phase mixture quality in the
downstream and hinder the formation of the elongated liquid slug
and the annular flow, which appears in the downstream region of the (c) 23.570ms (d) 23.632ms
conventional channel (Fig.4 (b4)~4(b6)) leaving only a small fraction
of cross section for liquid transfer. A similar flow pattern was also
observed in a microchannel condensation (Fang et al., 2010). The
long slug in the conventional channel is initially lubricated by thin
liquid films separating the vapor core with the channel wall. The
excessively long slug eventually leads to the liquid film rupture,
yielding dry-out regions (Fig.4(b5)~4(b6)). The dry-out issue was
also addressed by Chang and Pan (2007), who stated that the
dominant effect of the surface tension force over the buoyancy and
lift forces at small scale exacerbates the dry-out condition. In the
vapor-venting channel, however, the elongated slug is absent and the
(e) 23.682ms (f) 23.732ms
dry-out may never occur. The simulated flow pattern here agrees
with the experimental observation reported previously (Zhang et
al.,2005; Alexander and Wang, 2008).
Figure 4(c) shows the flow pattern for a high heat flux of
23 W / cm2 . Such high heat flux generates the vapor at a rate
exceeding the venting capacity of the channel. Consequently, the
excessive vapor accumulation and the slug/annular flow pattern
appear in both channels. Unlike those in case 1 and case 2, the flow
patterns shown in Fig.4c do not indicate a quasi-static state. Instead,
the quality of mixture continuously grows with the time and the dry-
out region expands across the whole channel. (g) 23.782ms (h) 23.832ms
A single bubble venting process is illustrated in Fig.6 for heat Fig. 6 Velocity field of a typical bubble-venting process. The cross
flux of 10 W / cm2 . The bubble initially grows to the channel size section of the membrane-channel assembly is shown and blue curves
and comes into contact with the membrane surface (Fig.6(a)). The indicate the vapor-liquid interface.
venting process starts with the rupture of the liquid film separating
the vapor and membrane, and the surface-tension-induced high
pressure in the bubble drives the vapor into the membrane region
through a neck (Fig.6(b)). The vapor flow experiences a high
resistance in the membrane, causing the vapor flow to divert to the
sidewalls and expanding the venting neck sidewards, thus forming a 0 ms 0.15ms 0.3ms
cylindrical vapor column (Fig.6(c)~6(d)). A pair of vortex was also
induced by the expansion process (Fig.6(c)). The sideward expansion
eventually transforms the vapor column into a semi-spherical bubble
attached to the membrane surface, and the venting area and venting 0.45ms 0.6ms
rate reach a peak value, as is evidenced by the velocity field in the Fig.7 Experimental observation of a venting process in a water-
membrane region in Fig.6(e). The continuous venting process causes
air adiabatic system at a liquid flowrate of 0.26ml/min and an air
the bubble to shrink (Fig.6(f)~6(g)) and finally disappear, leaving the
flow rate of 0.23ml/min (David et al., 2009).
channel with a big circulation of liquid flow, which is eventually

6
Frontiers in Heat and Mass Transfer (FHMT), 1, 013002 (2010) Global Digital Central
DOI: 10.5098/hmt.v1.1.3002 ISSN: 2151-8629

4.2 Pressure Distribution

Figure 8 shows a snapshot of the pressure contour when the channel
and the membrane are occupied by the single liquid and the single
vapor phase, respectively. A capillary pressure jump across the
membrane-channel interface is evident, which prevents the liquid
phase from leaking into the membrane. Serving as a no-slip solid
wall to the liquid phase, the hydrophobic membrane allows the
development of the standard Poiseuille flow in the channel, leading
to a linear pressure drop in the longitudinal direction.

Fig.9 Comparison of relationship between the mean pressure

drop and the heat flux for two types of microchannels
Fig.8 Calculated pressure contour in the membrane and the
microchannel prior to onset of boiling
4.2.2 Pressure Fluctuation
Figure 10 illustrates the temporal pressure evolution for both types of
4.2.1 Reduction of Average Pressure Drop
channels at a heat flux of 11 W / cm2 . The pressure fluctuation
Figure 9 shows the comparison of the pressure drop between the appears in both conventional and vapor-venting channels. These
conventional channel and the vapor-venting channel for various heat pressure fluctuations, which are observed mainly in microchannels,
fluxes. For a heat flux lower than 6 W / cm2 , the channel is occupied are primarily due to bubble formation, expansion and departure
by the liquid single phase and both types of channel yield identical (Zhang et al., 2002; Zhang et al., 2005) The present simulation also
pressure drops. As the heat flux exceeds 6 W / cm2 , the boiling predicts pressure fluctuation on the order of a few hundred Pa in a 5
occurs, starting near the channel outlet and propagating upstream mm long channel, though their reported frequency is limited by the
with an increase in heat flux. As discussed by Zhang et al. (2002), computational time step size and thus lacks physical meaning.
after the onset of boiling, the confined geometry of the microchannel
without vapor-venting causes the generated vapor phase to rapidly
fill and clog the channel, leaving only a small fraction of the cross
section available for liquid transfer. This added friction dramatically
lifts the pressure drop. Our simulation shows that the pressure drop
increases with the heat flux at an approximate rate of 160 N / W
within the range of 6 W / cm2 to 25 W / cm2 . The result is consistent
with the prediction of the steady-state model proposed by Koo et al.
(2001), which was also used by Zhang et al. (2002), Zhang et al.
(2005) and David et al. (2007) for the comparison with experimental
measurement.
For the vapor-venting microchannel, the pressure drop initially
increases with the heat flux after the onset of boiling at a rate similar
Fig.10 Temporal pressure evolutions in two types of
to that of the conventional channel, as shown in the region A of
Fig.9. However, the curve starts to flatten out with a further increase microchannels (heat flux: 11W/cm^2). Significant pressure
in the heat flux beyond around 10 W / cm2 (Fig.9 region B). The
surges are present in the vapor-venting channel.
pressure drop becomes insensitive to the rise of heat flux in the high
In the vapor-venting channel, bubbles still experience an initial
heat flux regime, which clearly shows the advantage of the vapor-
expansion. However, the expansion process is halted when the
venting channel over the conventional channel in terms of the
bubble reaches the channel size. Hence, the high frequency
pressure drop reduction. The initial increase of the pressure drop in
component of the pressure fluctuation also exists in the vapor-venting
region A is attributed to the presence of moving small bubbles which
channel, although the average amplitude is lower than that in the
are not big enough to contact the membrane surface and escape.
conventional channel. In addition to the high frequency component,
Those bubbles are convected downstream, partially blocking the
the simulation also reveals a new pressure fluctuation mode with a
channel and increasing the overall pressure drop. An optimization of
low frequency of around 1kHz and a high amplitude. The peak
the channel structure would facilitate the escape of those small
amplitude of this low frequency fluctuation can be up to several
bubbles and may relieve the pressure buildup in region A. Instead of
thousand Pa, as shown by spikes on the red curve in Fig.10. A
the slug and annular flow in the conventional channel, the high heat
synchronized examination of the flow pattern and the pressure time
flux in the vapor-venting channel only induces a massive vapor-
series implies that the pressure surge is induced by the final stage of
escaping process and the pressure stops further increasing due to the
the bubble venting, as shown in Fig.11. During the venting process,
limited mixture quality buildup.
the liquid cavity caused by the bubble shrinkage is filled by upstream
liquid, which leads to a discontinuity in the liquid flowrate at the
venting site (Fig.11 (A)~11(B)). The velocity field shown in Fig.12

7
Frontiers in Heat and Mass Transfer (FHMT), 1, 013002 (2010) Global Digital Central
DOI: 10.5098/hmt.v1.1.3002 ISSN: 2151-8629

clearly shows the abrupt flowrate drop at the venting site. The from the inlet liquid temperature 370K to the saturation point where
flowrate immediately downstream of the venting site gets boiling occurs.
vanishingly small and in some extreme cases may even result in For a low heat flux of 10 W / cm2 , the temperature profile in both
reverse flow. Due to the incompressibility of the liquid however, the channels remains flat over the whole boiling region, and the supplied
completion of the venting process dictates an immediate removal of heat is purely absorbed by the phase change process. Similar
such flowrate discontinuity, leading to an abrupt deceleration of the temperature profile was also observed by Zhang et al. (2002) in the
upstream liquid and an acceleration of the downstream liquid. silicon microchannel heat sink. Although the conventional channel
Consequently, the induced inertial force causes a pressure surge in and the vapor-venting channel are dominated by the bubbly-slug
the channel (Fig.11(C)). The amplitude of the pressure surge is flow (Fig.13(A)) and pure bubbly flow (Fig.13(B)), respectively, the
negatively correlated with the distance between the channel inlet and existence of the liquid film on the channel wall in both channels
the venting site, and the fluctuation frequency is dictated by the sustains a continuous evaporation and a stable wall temperature is
number of bubbles undergoing the venting process per unit time. The maintained for both flow patterns.
overall pressure fluctuation in the vapor-venting channel is the As the heat flux increases to 18 W / cm2 , the vapor venting
superposition of the bubble-expansion-induced high frequency
process starts to play a critical role in preventing the formation of the
component and the flowrate-discontinuity-induced low frequency
elongated liquid slug and maintaining the liquid film evaporation
component.
(Fig.13(C)). In contrast, the excessive vapor accumulation in the
10000 conventional channel rapidly creates an elongated vapor slug in the
8000
upstream and an annular flow in the downstream (Fig.13(D)), where
the liquid film rupture and the local dry-out appears on the channel
Pressure Drop (Pa)

6000
bottom. When dry-out occurs, the thermal resistance between the
4000
channel wall and the liquid absorbing the heat surges drastically, and
2000 the heat removal capacity in those corresponding areas is reduced by
0 several orders of magnitude (Chang and Pan 2007). Consequently,
0.01 0.0102 0.0104 0.0106 0.0108 0.011 0.0112 0.0114 0.0116 0.0118 0.012
the bottom temperature in the conventional channel surges to around
Time (s)
420K, which is in sharp contrast to the almost unchanged
A: temperature in the vapor-venting channel.
t=20.007ms By further ramping the heat flux up to 23 W / cm2 , we may make
B: the vapor generation rate significantly exceed the vapor removal
t=20.132ms capacity of the membrane. Consequently, an annular/stratified flow
accompanied by the dry-out occurs in both channels, leading to a
C:
drastic temperature surge (Fig.13(E)~13(F)).
t=20.232ms
The simulation shows that the vapor-venting mechanism can
Fig.11 Pressure surge process versus vapor-venting flow patterns
dramatically suppress the local dry-out, and hence substantially
(heat flux: 10 W/cm^2)
enhance the critical heat flux that a microchannel heat sink can
operate at.
Although the simulation shows that the vapor-venting mechanism
may exacerbate rather than mitigate the pressure fluctuation issue in 4.4 Impact of Membrane Surface Hydrophobicity
a single channel, it may still suppress the pressure fluctuation in a
multi-channel system, considering the vapor-venting mechanism can Recently, a number of experimental studies were conducted to study
effectively alter the demand curve which is responsible for the the air-venting process in the microchannel, which bears some
pressure fluctuation caused by Ledinegg instability in an ordinary resemblance with the present vapor-venting microchannel. However,
multi-channel system without vapor-venting. different flow patterns were reported by those studies. David et al.
(2009) reported the stratified flow at a moderate to high air flow rate
in the copper microchannel with a porous hydrophobic Teflon
1.5m/s membrane wall. The contact angle at the membrane wall is 140o. In
contrast, Alexander et al. (2008) tested the air-venting process in the
microchannel with parallel breather ports located at the sidewall. The
breather ports are coated with a Silane SAM rendering a contact
angle of 116.3o, while the channel wall is kept moderately
hydrophilic with a contact angle around 90o. Instead of the stratified
flow, the study reported the slug flow pattern during the venting
Low flow rate region process. Similar flow pattern was also reported by the Meng et al.
High flow rate region
(2004) using a Teflon membrane breather with the surface contact
Fig.12 Velocity discontinuity during venting. A transition from the angle of 60 o to 67 o for the methanol. The discrepancy is possibly
high flow rate region to the low flow rate region occurs during the attributed to the variance in the wettability of the breather’s surface,
bubble venting process. which is hydrophobic (David et al., 2009) and semi-hydrophilic
(Meng et al., 2004; Alexander et al., 2008).
4.3 Temperature Distribution
To verify this argument, the vapor-venting process in the channel
In addition to the pressure drop reduction, another major advantage with various membrane surface contact angles is simulated. The
of the vapor-venting microchannel, as suggested by Zhou et al. resulting flow patterns are shown in Fig.14. The bubbly flow
(2006), lies in its mitigation of the local dry-out which leads to (Fig.14(a)) and elongated slug flow (Fig.14(c)) are observed in
hotspots on the chip. Our simulation firmly supports this argument. channels with moderately hydrophilic membrane and hydrophilic
Figure 13 shows the temperature distribution along the channel membrane, respectively, which is similar to those reported previously
bottom subject to a variety of heat fluxes, with the corresponding (Alexander et al., 2008). In contrast, the bubble in the
flow patterns illustrated. In all cases, the temperature quickly rises superhydrophobic case (Fig.14(b)) tends to spread on the membrane
surface and form a vapor buffer layer.

8
Frontiers in Heat and Mass Transfer (FHMT), 1, 013002 (2010) Global Digital Central
DOI: 10.5098/hmt.v1.1.3002 ISSN: 2151-8629

E
450 F
10W/cm2 Venting
10W/cm2 No Venting
440 18W/cm2 Venting
18W/cm2 No Venting
Channel Bottom Temperature (K) 430 23W/cm2 Venting D
23W/cm2 No Venting
420

410

400 C

390

380

370
0 0.5 1 1.5 2 2.5
Location(mm)
B
A

Fig.13 Temperature distribution along the bottom of the channel for two types of microchannels and various heat fluxes.

The simulation shows that the formation of the vapor buffer layer on the flow pattern is illustrated in Fig.15, which clearly shows that
significantly increases the contact area between the vapor and the condensed liquid in the membrane can degenerate the vapor-
membrane, therefore enhancing the vapor-venting rate, whereas the venting channel into a conventional channel with a hydrophobic top
hydrophilic membrane surface in Fig.14(c) minimizes the vapor- surface (Fig.15(a)). Due to the low permeability of the membrane to
membrane contact area and almost completely inhibits vapor venting. the liquid water (4e-22 m2 in the present study) the condensed water
is retained in the membrane and blocks the passage in the porous
structure available for vapor venting. The membrane condensation
may be mitigated by vaporizing and removing the water from the
(a) Moderately hydrophobic membrane surface (114.5 deg) membrane.

(b) Hydrophobic membrane surface (contact angle: 179 deg) (a) Channel with membrane condensation

(c) Hydrophilic membrane surface (contact angle: 28.3 deg) (a) Channel with dry membrane
Fig.14 Impact of membrane surface hydrophobicity on the vapor- Fig.15 Impact of membrane condensation on the vapor-venting
venting flow pattern (10 W / m 2 ) process. Note how membrane condensation prevents vapor from
escaping

4.5 Impact of Membrane Condensation 5. CONCLUSIONS

David et al. (2008) tested the operation of the vapor-venting copper The present study numerically simulates the 3D transient vapor-
heat exchangers with built-in PTFE membranes and found that in venting process in a microchannel with a hydrophobic porous
certain conditions the device fails to show the improvement in the membrane using the VOF method along with a capillary force model.
pressure drop, implying that “the membrane acts like a solid wall”. A comparison is provided between the vapor-venting channel and the
Due to the lack of the visualization capability, such hypothesis has conventional channel for the two-phase flow patterns, the pressure
yet to be confirmed. drop, and the temperature profile. The effects of the membrane
surface wettability and the membrane condensation on the heat
To verify this argument, we artificially induce the condensation exchanger performance are also discussed. We conclude:
in the membrane by applying a heat sink of 0.5 W / cm2 to the top
surface of the membrane. The impact of the membrane condensation

9
Frontiers in Heat and Mass Transfer (FHMT), 1, 013002 (2010) Global Digital Central
DOI: 10.5098/hmt.v1.1.3002 ISSN: 2151-8629

1. Over a full range of the heat flux below the critical value, the Fang, C., Steinbrenner, J., Wang, F., and Goodson, K., 2010, “Impact
vapor-venting channel yields a significantly smaller pressure drop as of Wall Hydrophobicity on Condensation Flow and Heat Transfer in
opposed to the conventional microchannel. As the heat flux exceeds Silicon Microchannels,” J. Micromech. Microeng., 20, 045018
the critical value, both channels are occupied by the annular flow and doi: 10.1088/0960-1317/20/4/045018
yield similar pressure drop.
2. The temporal pressure fluctuation in the vapor-venting channel is Fang, C., David, M., Wang, F., and Goodson, K., 2010, “Influence of
the superposition of a high-frequency-low-amplitude component Film Thickness and Cross-sectional Geometry on Hydrophilic
induced by the bubble expansion and a low-frequency-high- Microchannel Condensation,” Int. J. Multiphase Flow, in press
amplitude component caused by the bubble venting, the latter being doi:10.1016/ j.ijmultiphaseflow.2010.04.005
absent in the conventional channel. Ganesh, R.K., Faghri, A., and Hahn, Y., 1997, “A Generalized
3. Vapor-venting mechanism can significantly inhibit the liquid film Thermal Modeling for Laser Drilling Process: Part I- Mathematical
rupture and local dry-out, a phenomenon commonly appearing in the Modeling and Numerical Methodology,” Int. J. Heat Mass Transfer,
conventional microchannel when heat flux is high. The temperature 40, 3351-3360
profile remains stable in the vapor-venting channel for a full range of doi:10.1016/S0017-9310(96)00368-7)00368-7
heat fluxes below the critical value.
4. Superhydrophobic membrane surface induces the formation of a Garimella, S., 2002, “An Experimentally Validated Model for Two-
buffer vapor layer, which enhances the vapor-venting efficiency due phase Pressure Drop in the Intermittent Flow Regime for Circular
to the increased vapor-venting area. The hydrophilic membrane Microchannels,” J. Fluids Eng., 124, 205-214
surface significantl1y compromises the device performance due to doi:10.1115/1.1601258
the diminished contact area between vapor and membrane.
Garimella, S., and Singhal, V., 2004, “Single Phase Flow and Heat
5. The vapor-venting process suffers from the condensation in the
Transport and Pumping Considerations in Microchannel Heat Sinks,”
membrane. Possible solutions include maintaining the temperature of
Heat Transfer Engineering, 25, 15-25
the membrane and finding a highly permeable membrane with a fine-
pore, superhydrophobic surface to create a high capillary force.
doi:10.1080/01457630490248241
The present numerical model may assist in the design and Issa, R.I., Gosman, A.D., and Watkins, A.P., 1991, “The Computation
optimization of the next generation vapor-venting heat exchanger of Compressible and Incompressible Recirculating Flows by a Non-
with an enhanced heat removal capability. iterative Implicit Scheme,” J. Comput. Phys., 93, 388-410.
doi:10.1016/0021-9991(86)90100-2
REFERENCES
Kennedy, J.E., Roach, G.M., Dowling, M.F., Abdel-Khalik, S.I.,
Alexander, B.R., and Wang, E.N., 2008. “Design of a Microscale Ghjaasiaan, S.M., and Jeter, S.M., 2000, “The Onset of Flow
Breather for Two-phase Thermal Management Devices”, IMECE Instability in Uniformly Heated Horizontal Microchannels,” J. Heat
2008-68898, Proceedings of 2008 ASME International Mechanical Transfer, 122, 118-125
Engineering Congress and Exposition, Boston MA. doi:10.1115/1.521442521442
Brackbill, J.U., Kothe, D.B., and Zemach, C., 1992, “A Continuum Koo, J., Jiang, L., Zhang, L., Zhou, P., Banerjee, S.S., Kenny, T.W.,
Method for Modeling Surface Tension,” J. Comput. Phys., 100, 335- Santiago, J.G., and Goodson, K.E., 2001, “Modeling of Two-phase
354. Microchannel Heat Sinks for VLSI Chips,” Proceedings of Int.
doi:10.1016/0021-9991(92)90240-Y MEMS Workshop, 422-426
Chang, K., and Pan, C., 2007, “Two-phase Flow Instability for Kandlikar, S., and Grande, W., 2004, “Evaluation of Single Phase
Boiling in a Microchannel Heat Sink,” International Journal of Heat Flow in Microchannels for High Heat Flux Chip Cooling-
and Mass Transfer, 50, 2078-2088 Thermohydraulic Performance Enhancement and Fabrication
doi:10.1016/j.ijheatmasstransfer.2006.11.014 Technology,” Heat Transfer Engineering, 25, 5-16
doi:10.1080/0145763049051977201457630490519772
David, M.P., Khurana, T., Hidrovo, C., Pruitt, B.L., and Goodson,
K.E., 2007, “Vapor-venting, Micromachined Heat Exchanger for Kaya, T., and Goldak, J., 2006, “Numerical Analysis of Heat and
Electronics Cooling,” Proceedings of IMECE, Seattle, WA. Mass Transfer in the Capillary Structure of a Loop Heat Pipe,” Int. J.
Heat and Mass Transfer, 49, 3211-3220
David, M.P., Marconnet, A., and Goodson, K.E., 2008,
doi:10.1016/j.ijheatmasstransfer.2006.01.028
“Hydrodynamic and Thermal Performance of a Vapor-venting,
Micromachined Copper Heat Exhcnager,” Proceedings of ICNMM, Lee, W.H., 1980, “A Pressure Iteration Scheme for Two-phase Flow
Darmstadt, Germany. Modeling,” Multiphase Transport Fundamentals, Reactor Safety,
Applications, Hemisphere Publishing, Washington, DC
David, M.P., Steinbrenner, J., Miler, J., and Goodson, K.E., 2009,
“Visualization and Analysis of Venting from a Single Microchannel Li, Z.M., 2002, “Mass Transfer Behavior at Bubble Surface During
Two-phase Copper Heat Exchanger,” Proceedings of InterPack, San Nucleate Boiling,” Heat and Mass Transfer, 38, 433-439
Francisco, California. doi:10.1007/s002310100253
Figus, C., Bray, Y., Bories, S., and Prat, M., 1999, “Heat and Mass Lee, P., Tseng, and F., Pan, C., 2004, “Bubble Dynamics in
Transfer with Phase Change in a Porous Structure Partially Heated: Microchannels. Part I: Single Microchanel,” Int. J. Heat and Mass
Continuum Model and Pore Network Simulations,” Int. J. Heat and Transfer, 47, 5575-5589
Mass Transfer, 42, 2557-2569 doi:10.1016/j.ijheatmasstransfer.2004.02.031
doi: 10.1016/S0017-9310(98)00342-1
Meng, D.S., Cubaud, T., Ho, C., Kim, C., 2004, “A Membrane
Fang, C., Carlos, C., Wang, F., Eaton, J., and Goodson, K., 2008, “3- Breather for Micro Fuel Cells with High Concentration Methanol,”
D Numerical Simulation of Contact Angle Hysteresis for Microscale Proc. Tech. Dig. Solid-state Sens., Actuator, Microsyst. Workshop,
Two-phase Flow,” Int. J. Multiphase Flow, 34, 690-705 Hilton Heat Island, SC, 141-144
doi:10.1016/j.ijmultiphaseflow.2007.08.008

10
Frontiers in Heat and Mass Transfer (FHMT), 1, 013002 (2010) Global Digital Central
DOI: 10.5098/hmt.v1.1.3002 ISSN: 2151-8629

Meng, D.S., Cubaud, T., Ho, C., and Kim, C., 2007, “A Methanol- Youngs, D.L., 1982, “Time-dependent Multi-material Flow with
tolerant Gas-venting Microchannel for a Microdirect Methanol Fuel Large Fluid Distortion,” in: Numerical Methods for Fluid Dynamics,
Cell,” Journal of Microelectromechanical Systems, 16, 1403-1410 K.W. Morton and M.J. Baines (eds.), New York Academic Press,
273–285.
Qu, W., and Mudawar, I., 2002, “Experimental and Numerical Study
of Pressure Drop and Heat Transfer in a Single-phase Microchannel Yang, Z., Peng, X.F., and Ye, P., 2008, “Numerical and Experimental
Heat Sink,” Int. J. Heat and Mass Transfer, 45, 2549-2565 Investigation of Two Phase Flow During Boiling in a Coiled Tube,”
doi:10.1016/S0017-9310(01)00337-4 International Journal of Heat and Mass Transfer, 51, 1003-1016
doi:10.1016/j.ijheatmasstransfer.2007.05.025
Ren, C., Wu, Q., and Hu, M., 2007, “Heat Transfer with Flow and
Evaporation in Loop Heat Pipe’s Wick at Low or Moderate Heat Zhang, Y., Faghri, A., and Shafii, M.B., 2001, “Capillary Blocking in
Fluxes,” Int. J. Heat and Mass Transfer, 50, 2296-2308 Forced Convective Condensation in Horizontal Miniature Channel,”
doi:10.1016/j.ijheatmasstransfer.2006.10.029 Journal of Heat Transfer, 123, 501-511
doi:10.1115/1.13518081.1351808
Tuckerman, D.B., and Pease, R.F.W., 1981, “High-performance Heat
Sinking for VLSI,” IEEE Electron Device Letters, Vol.EDL2, No.5, Zhang, L., Koo, J., Jiang, L., Asheghi, M., Goodson, K.E., Santiago,
126 and J.G., Kenny, T.W., 2002, “Measurements and Modeling of Two-
phase Flow in Microchannels with Nearly Constant Heat Flux
Takata, Y., Shirakawa, H., Sasaki, H., Kuroki, T., and Ito, T., 1999, Boundary Conditions,” Journal of Microelectromechanical Systems,
“Numerical Analysis of Rapid Solidification in a Single Roller
11, 12-19
Process,” Heat Transfer – Asian Research, 28, 34-49
doi:10.1109/84.982858
doi:10.1002/(SICI)1523-1496(1999)28:1<34::AID-
HTJ4>3.0.CO;2-0 Zhang, L., Wang, E.N., Goodson, K., and Kenney, T., 2005, “Phase
Change Phenomena in Silicon Microchannels,” International Journal
Welch, S.W.J., and Wilson, J., 2000, “A Volume of Fluid Based of Heat Mass Transfer, 48, 1572-1582
Method for Fluid Flows with Phase Change,” Journal of
doi:10.1016/j.ijheatmasstransfer.2004.09.048
Computational Physics, 160, 662-682
doi:10.1006/jcph.2000.6481 Zhou, P., Goodson, K.E., and Santiago, J., 2006, U.S. Patent 6, 994,
151 B2

11