See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/231978974

Mass flow and tangential momentum accomodation in silicon micromachined

channels

Article  in  Journal of Fluid Mechanics · June 2001

DOI: 10.1017/S0022112001004128

CITATIONS READS

251 611

3 authors, including:

Kenneth S Breuer
Brown University
296 PUBLICATIONS   7,206 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

dynamic gliding in flying squirrels View project

comparing the energetic cost of increasing aerodynamic force through different kinematic mechanisms (robotic bat) View project

All content following this page was uploaded by Kenneth S Breuer on 18 September 2014.

The user has requested enhancement of the downloaded file.

J. Fluid Mech. (2001), vol. 437, pp. 29–43. Printed in the United Kingdom 29
c 2001 Cambridge University Press

Mass flow and tangential momentum

accommodation in silicon micromachined
channels
By E R R O L B. A R K I L I C1 , K E N N E T H S. B R E U E R2
A N D M A R T I N A. S C H M I D T3
1
Phasyn, San Carlos, CA 94070, USA
2
Division of Engineering, Brown University, Providence RI 02912, USA
3
Microsystems Technology Laboratories, Massachusetts Institute of Technology,
Cambridge, MA 02139, USA

(Received 4 October 1999 and in revised form 18 December 2000)

High-precision experimental results are reported showing the tangential momentum

accommodation coefficient (TMAC) for several gases in contact with single-crystal
silicon to be less than unity. A precise and robust experimental platform is demon-
strated for measurement of mass flows through silicon micromachined channels due
to an imposed pressure gradient. Analytic expressions for isothermal Maxwellian slip
flows through long channels are used to determine the TMAC at a variety of Knud-
sen numbers. Results from experiments using nitrogen, argon and carbon dioxide are
presented. For all three gases the TMAC is found to be lower than one, ranging from
0.75 to 0.85.

1. Introduction
The emergence of microelectromechanical systems (MEMS) as an important tech-
nology has raised a variety of new scientific questions related to the application of
conventional tools of analysis to the unfamiliar world of micron-sized fluidic sys-
tems. Due to their small size, the surface area to volume ratio in MEMS can be
large and thus surface effects, such as viscous interactions and surface chemistry,
become relatively more important than is typically true in a larger-scale device. As
an example of this scaling one can consider micromachined accelerometers, which
are typically composed of a vibrating proof mass suspended a few microns from a
stationary substrate. With such small clearances, viscous (squeeze-film) damping in
the small gap is typically quite high, dominating the performance of the device. In
order to minimize the damping, the gap is often vented using perforations in the
proof mass, or packaged in vacuum. Despite these steps, viscous damping still limits
device performance and is often poorly predicted using conventional continuum and
crude non-continuum analysis (such as a pressure-dependent effective viscosity). One
explanation for the lack of agreement between experiment and prediction is often that
some kind of ‘new physics’ might be required to explain experimental observations
for such micron-sized systems.
One source of such ‘new physics’ is the possibility that, at such small scales, non-
continuum effects become important to the fluid–surface interactions, including the
presence of slip flow and incomplete momentum and energy accommodation between
30 E. B. Arkilic, K. S. Breuer and M. A. Schmidt
the gas and the substrate. The latter possibility is a reasonable suspicion given that
the surface of MEMS is often a highly polished crystal face of the silicon substrate
and such ordered surfaces are known to exhibit sub-unity accommodation (e.g. Seidl
& Steinheil 1974). The purpose of this paper is to explore the basis for incomplete
accommodation in MEMS devices.
1.1. Models for slip flow
Slip flow, which represents a rarefied gas phenomenon in which the common no-slip
boundary condition breaks down at scales commensurate with the mean free path
of the gas, is commonly described by the Maxwellian (Maxwell 1879) boundary
condition for the fluid velocity, u, at a solid surface (y = 0):

2 − σ ∂u 
u(y = 0) = λ , (1.1)
σ ∂y y=0
where λ is the mean free path of the gas. The constant σ is the tangential mo-
mentum accommodation coefficient (TMAC) which can vary from zero (specular
accommodation) to one (complete, or diffuse accommodation). Although Maxwell’s
original formulation used a slightly different argument, the most intuitive description
of the TMAC is that it represents the average streamwise (or tangential) momentum
exchange between impinging gas molecules and the solid surface:
ur − ui
σ= , (1.2)
Uw − ui
where ui is the average incident streamwise velocity, ur is the average reflected
streamwise velocity and Uw is the streamwise velocity of the solid surface. For most
gas–surface interactions at macroscopic (engineering) conditions, σ = 1, is the most
appropriate choice.
Many researchers have measured the TMAC under a variety of conditions (for
example Seidl & Steinheil 1974; Thomas & Lord 1974; Nakarjakov et al. 1995).
For the most part, the test conditions have been at high vacuum, and have used
molecular beams impinging on carefully prepared substrates. Very few, if any results
are available at atmospheric conditions and using ‘as-prepared’ materials. Although
the TMAC should not be affected by operating pressure or incidence angle, the
surface conditions (level of absorbed materials and surface roughness, etc.) can affect
the TMAC and for this reason, a study of MEMS surfaces at ‘standard’ conditions
is of interest.
1.2. Microchannel flows
In recent years, the high-Knudsen-number regime has become accessible to experi-
ments at atmospheric conditions by using microfabricated flow channels. Pfahler et al.
(1991) and subsequently Harley et al. (1995) presented some of the first analytic and
experimental work on rarefied microchannel flows and demonstrated the existence
of slip flows in microchannels by observing an increase in the mass flow through
the channel over what would be predicted using continuum (non-slip) calculations.
However, the fabricated channels did not possess well-controlled surface structure
(the channel was made by the bonding of dry-etched silicon and glass). In addition,
they used a single-tank accumulation technique to measure the mass flow which is
susceptible to thermal fluctuations and thus restricted their measurements to relatively
large mass flows (this is discussed in more detail in § 3.1). Pong et al. (1994) presented
results from experiments on microchannels, fabricated using surface micromachining
 Tangential momentum accommodation in microchannels 31
techniques with integrated pressure transducers, and reported pressure distributions
measured along the length of the channel. Subsequent experiments by Shih et al.
(1996) presented mass flow measurements and pressure distributions which were not
in agreement with any theory. However, some questions remain about the dimensional
tolerances of their microchannels (which are fabricated by surface micromachining
techniques) and the accuracy of the mass flow measurement technique, which, like
that of Harley et al. (1995), was a single-tank accumulation technique and is thus very
sensitive to thermal fluctuations for these low mass flow regimes. Arkilic & Breuer
(1993) and Arkilic, Schmidt & Breuer (1997b) explored, using a simple analysis, the
effects of rarefaction and variable density in long microchannels, matching both the
experimental data of Harley et al. (1995) and the measured pressure distributions of
Pong et al. (1994).
Experimental results were also presented by Arkilic, Schmidt & Breuer (1994,
1997b). In this work, the flow rate of helium through carefully fabricated microchan-
nels was measured and it was demonstrated that, as predicted, rarefaction was present
in the microchannel flow. In addition, the theory of Arkilic et al. (1997b) was vali-
dated at these low flow rates. Although this series of experiments indicated excellent
agreement with theory, the extraction of the TMAC requires an even more sensitive
and more extensive series of measurements. It is this that forms the focus of the
present paper.
Finally, it is worth noting that numerical approaches to rarefied flows in mi-
crochannels have also been explored using both slip-flow Navier–Stokes equations
with higher-order boundary conditions (Beskok & Karniadakis 1999) as well as di-
rect simulation Monte Carlo techniques (Piekos & Breuer 1995, 1996; Beskok &
Karniadakis 1993, 1999 and most recently Cai et al. 1999). Such approaches have
confirmed experimental and analytical results, although they still need the TMAC as
an empirical input.

1.3. Present work

The present paper extends significantly the experimental work presented previously
by our group (Arkilic et al. 1997b) in a number of ways. First and foremost, the
sensitivity and accuracy of the pressure and mass flow measurements have been
greatly enhanced using new hardware and improved experimental protocols. This
additional precision allows not only verification that slip flow occurs at the micron-
scale (not surprisingly), but has further allowed determination of the TMAC from
those mass flow measurements. The determination of the TMAC is accomplished
using a regression analysis, presented in § 2.1. Lastly, our previous work was conducted
solely for helium flows, in which rarefaction is apparent in relatively large channels
and at high pressures since the mean free path is so large. The present paper reports
on experiments (and corresponding TMAC) for nitrogen, carbon dioxide and argon –
all gases of larger molecular weight in which the accurate measurement of rarefaction
is harder to achieve.

2. Background theory
A theoretical model for slip flow in long channels was reported by Arkilic et al.
(1997b). The key assumptions and results are reported here for completeness and
convenience. In this model, the gas is assumed to be isothermal and represented by
the ideal gas equation of state. The compressible Navier–Stokes equations are solved
32 E. B. Arkilic, K. S. Breuer and M. A. Schmidt
with a Maxwellian slip boundary condition which has the non-dimensional form

2 − σ dũ 
ũ|w = K , (2.1)
σ dỹ  w

where ũ and ỹ are the non-dimensionalized streamwise velocity and wall-normal

coordinate (normalized by the mean exit velocity and channel height, H, respectively)
and K is the Knudsen number, defined as the ratio of the mean free path, λ, to
the channel height (K = λ/H). The subscript ‘w’ denotes evaluation at the wall,
y/H = ±1/2. This boundary condition is strictly only valid for small Knudsen
numbers (K < 0.3).
Upon expansion of the governing equations in the small parameter defined by the
channel height-to-length ratio, Arkilic et al. (1997b) derived a system of equations,
which upon solving at the lowest order describe a locally two-dimensional Poiseuille
flow. However, due to the large pressure drop, the gas density changes along the
length of the channel, resulting in a nonlinear pressure distribution, p̃, given by
2−σ
p̃(x̃) = p(x/L)/Po = −6 Ko
σ
s 2    
2−σ 2−σ 2−σ
+ 6 Ko + 1 + 12 Ko x̃ + P2 + 12 Ko P (1 − x̃),
σ σ σ

(2.2)
where P is the inlet-to-outlet pressure ratio, Pi /Po , and the subscripts i and o refer to
inlet (x = 0; x̃ ≡ x/L = 0) and outlet (x = L; x̃ ≡ x/L = 1) conditions respectively.
The mass flow is also found from the zeroth-order solution:
 
H 3 W Po2 2 2−σ
ṁ = P − 1 + 12 Ko (P − 1) , (2.3)
24µLRT σ
where µ is the fluid viscosity, T the temperature and R the gas constant. L and W
are the channel length and width respectively. Note that, for a given inlet and outlet
pressure condition, the rarefaction acts to increase the observed mass flow and that as
the pressure ratio decreases, the effect of slip makes a more significant contribution
to the total mass flow. Also, as the outlet pressure increases for a given inlet pressure,
or as the characteristic dimension increases, the rarefaction contribution diminishes.
With the exception of the TMAC, all variables in the expression for the mass flow
(equation (2.3)) are known or are measurable and we can thus use this expression to
determine the TMAC from mass flow–pressure characteristics of the microchannel.
2.1. Determination of TMAC
We can use and extend this analytic result to determine the TMAC for a given fluid–
surface interaction. Making an analogy to Ohm’s law where the differential pressure
applied across the channel (∆P ) is analogous to the potential and the mass flow is
analogous to the current, the flow conductance, C, can be defined by
ṁ H 3W H 3W 2 − σ
C= = P̄ + Ko Po , (2.4)
∆P 12µLRT 2µLRT σ
where ∆P = Pi − Po is the differential pressure across the channel. The conductance
can be written as
C = AP̄ + B, (2.5)
 Tangential momentum accommodation in microchannels 33
where
H 3W
A= , (2.6)
12µLRT
H 3W 2 − σ
B= Ko Po (2.7)
2µLRT σ
are constants, and P̄ = (Pi + Po )/2 is the average of the inlet and outlet pressures
(from here on referred to as the mean pressure).† Note that A is independent of the
TMAC.
One could extrapolate this line to P̄ = 0 to determine the intercept, B, and thus
derive the TMAC. However, this is problematic, since it relies heavily on data in the
free-molecular flow regime, where the theory becomes more and more tenuous. A
preferable alternative is to normalize the measured mass flow by (Pi2 − Po2 ):
ṁ H 3W H 3W 2 − σ 1
= + Ko Po , (2.8)
Pi2 − Po
2 24µLRT 4µLRT σ P̄
or
ṁ A B1
= + , (2.9)
Pi2
− Po2 2 2 P̄
where A and B are defined above. If the measured mass flow (weighted by the inverse
of the difference between the squares of the inlet and outlet pressures, ṁ/(Pi2 − Po2 ))
is plotted versus inverse mean pressure 2/(Pi + Po ), then the intercept of this line
(which corresponds to the continuum limit) is independent of the TMAC while the
slope of the line can be used to extract the TMAC. This weighting of the data
provides a more attractive method of extracting the TMAC and is similar to the
temperature jump (TJ) method which is used to derive experimental values for the
energy accommodation coefficient (Goodman & Wachman 1876).

3. Experimental apparatus and procedure

The microchannels were fabricated in the MIT Microsystems Technology Labo-
ratory using a fabrication process similar to our previous experiments (Arkilic et
al. 1997b) and reviewed here for convenience. The fabrication sequence is shown
schematically in figure 1, and a scanning electron micrograph of the resulting channel
is shown in figure 2. To begin, a thick (1.33 µm) thermal oxide layer was grown on a
double-polished silicon wafer. This defines the channel thickness. The thermal oxide
was chosen as it forms a very uniform film over the wafer, thus ensuring good control
of the channel thickness. The channel geometry, entrance and exit plena were then
etched into the silicon oxide using buffered hydroflouric acid. A silicon nitride layer
was deposited to protect the channel. The entrance and exit ports were then patterned
and etched using a 20% potassium hydroxide solution. The channel was sealed by
fusion bonding a second ‘capping’ wafer onto the top of the channel wafer. Bonding
was achieved by contacting the wafers and then annealing at 1000 ◦ C for 70 minutes.
The bonded wafer was finally diced into smaller chips for handling, thus completing
the fabrication sequence.
The fabrication sequence has a number of features that are worth highlighting.
† Note that, due to the nonlinear pressure distribution within the channel, P̄ is not the same as
the mean pressure within the channel.
34 E. B. Arkilic, K. S. Breuer and M. A. Schmidt

Wafer oxidation determines

Silicon wafer channel height

Oxide is patterned to define

channel planform

Ports are etched from back to form

inlet and outlet ports

Silicon wafer
Wafer bond is used to complete
the structure

Gas flow

Figure 1. Schematic depiction of channel fabrication sequence.

Adapted from Arkilic et al. (1997b).

53.5 microns

51.0 microns

Figure 2. Scanning electron micrograph of microfabricated channel, shown end-on.

(a) The channels are long enough (7500 microns) such that entrance effects are
negligible and can be safely ignored (L/H > 5000). In addition, the theory, which
relies on a small channel height-to-length ratio is clearly valid.
(b) The channels are wide, with an aspect ratio (W /H) of approximately 40.
Since three-dimensional effects are proportional to the inverse of the aspect ratio
squared (White 1991), we can safely assume two-dimensional flow (also assumed in
the analysis).
(c) The oxide layer that defines the channel height is very uniform over the channel
length. This is a characteristic of thermal oxidation of silicon and one which can
be easily confirmed during the channel fabrication process, prior to wafer bonding.
This property is one of the reasons that this processing route was adopted for these
experiments.
 Tangential momentum accommodation in microchannels 35

Parameter Nominal value (µm) Uncertainty µm)

Length (L) 7490 ±1
Width (w) 52.3 ±0.3
Height (H) 1.33 ±0.01
Maximum surface roughness 6 0.65 × 10−3 —
Table 1. Geometry of microfabricated channels. The measurements were made using Dektak and
Wyco surface profilometers.

(d) The top and bottom surface of the channel are identical and are composed of
polished silicon crystal with a thin layer of ‘native’ oxide (oxide that grows naturally
on a silicon surface exposed to ambient air). This surface is very smooth, uniform
and, as with the channel height, can be well characterized before the wafer-bonding
step. This feature is in contrast to channels used by both Harley et al. (1995) and
Pong et al. (1994). Harley et al. (1995) used a plasma-etched silicon channel with a
glass capping layer. This resulted in a somewhat rough lower surface and dissimilar
materials on the lower and upper surface. Pong et al. (1994) used a dissolved sacrificial
layer of PSG to define their channel. Although they did not report surface roughness,
it is likely to be higher than that of polished silicon. In their case, the lower surface
was silicon while the upper surface was formed by a silicon nitride membrane.
(e) The surfaces that form the channel are very thick (approximately 500 microns)
and will not bend under the large pressures that are required to drive the flow
through the channel. This ensures that the channel geometry is invariant with flow
pressure. Although this has similar structural integrity to the channels used by Harley
et al. (1995), the microchannels used by Pong et al. (1994) used a thin (1.2 micron)
silicon nitride membrane to define the upper surface. This has the drawback that it
is susceptible to deformation and rupture under the high pressures required to drive
a flow through the long channel.
Table 1 summarizes the channel geometry as measured prior to the bonding step.

3.1. Mass flow measurement

As mentioned earlier, a highly accurate mass flow measurement was required for
these experiments. Earlier experiments, both by our group (Arkilic 1994) and others
(e.g. Pong et al. 1994) have used accumulation techniques in which the gas flowing
through the channel accumulates in a reservoir, resulting in either a change in pressure
(constant-volume technique) or a change in volume (constant-pressure technique). The
difficulty with these methods is that the small mass flows are very difficult to measure
accurately using a single tank accumulation. In particular very small changes in
temperature can overwhelm the mass flow measurement due to thermal expansion of
the gas. The present results use a very sensitive dual-tank accumulation technique,
described and analysed in detail by Arkilic, Schmidt & Breuer (1998). The system
is schematically shown in figure 3. Gas flows through the test channel and into
the accumulation tank. The mass flow through the system can be inferred from
a measurement of the differential pressure between the accumulation tank and a
reference (or ‘system’) tank. This arrangement is very insensitive to temperature
fluctuations and allows highly sensitive measurements of mass flows, with a sensitivity
demonstrated to be as low as 7 × 10−15 kg s−1 .
36 E. B. Arkilic, K. S. Breuer and M. A. Schmidt
Thermal
MKS baratron isolation
Microchannel chamber

Valve A Valve B Valve C

Flow

Reference tank
Thermistor
Flow tank

Copper block

Isolation
valve

Two-way
selector 1 Two-way
selector 2

Pressure
transducer Pressure
regulator

Vacuum High-
Vacuum pump ballast pressure
gas

Figure 3. Schematic of mass flow measurement system. Adapted from Arkilic et al. (1998).

3.2. Testing procedure

The testing procedure was as follows. The reference tank and flow tank were first
exposed to the desired outlet pressure by opening valves A, B, C and the isolation
valve and setting the outlet pressure. Valve A was then closed and the system allowed
to settle. The isolation valve was then closed. The valves, upon closing, inject mass
into the piping due to the compression of the bellows. It was found that this unsteady
mass injection led to thermal transients in the system. These were eliminated by
closing the isolation valve very slowly, over 30 minutes (this was accomplished using
a voltage-controlled pressure regulator to drive the pneumatically controlled isolation
valve). After the isolation valve was closed, the system was again allowed to settle.
The flow test sequence consisted of three 10-minute intervals during which the
differential pressure between the flow and reference tanks was measured. In the
pre-flow measurement interval (interval I), the differential pressure was recorded for
10 min before the microchannel was pressurized. The flow was then initiated through
the resistance by increasing the inlet pressure of the microchannel to the desired
 Tangential momentum accommodation in microchannels 37
–22
(× 10 )

Normalized mass flow (kg s–1 Pa–2) 8

0 0.4 0.8 1.2 1.6

Inverse mean pressure (1/Pa) (× 10–5)

Figure 4. Data and least-squares fit (which is used to derive a value for the TMAC) for a
representative set of argon flow experiments exhausting to an outlet pressure of 13 kPa. Two data
points at each pressure represent measurements acquired on ramping the inlet pressure up and then
down.

value by means of the voltage-controlled pressure regulator. It was found that by

slowly ramping up the pressure over 10 s we were able to maintain quasi-steady
(isothermal) conditions. Once at the desired inlet pressure, ∆p was recorded for an
additional 10 min (interval II). Finally, the inlet pressure was ramped back down over
a time period of 10 s and ∆p was recorded for another 10 min (the post-flow interval
or interval III). Any ambient thermal variations during these three intervals was
easily recognizable from measurements taken during the pre- and post-flow sequences
(intervals I and III). These effects, if present, were removed using a multi-segment
least-squares algorithm (Arkilic et al. 1998).

4. Results and discussion

The test conditions for the flow experiments are shown in table 2. At this range of
working conditions, both the Reynolds and Mach numbers are approximately 10−2 ,
clearly in the region of validity for the isothermal, variable-density theory of Arkilic et
al. (1997b). For each working gas and at each outlet Knudsen number, a suite of mass
flow versus inlet pressure measurements were taken. The data were then normalized
as described earlier and plotted as the normalized mass flow, ṁ/(Pi2 − Po2 ), versus
the inverse mean pressure, 2/(Pi + Po ). A representative example of the experimental
data normalized in this manner is shown in figure 4. The uncertainty bars shown
in the figure are calculated using the definition of the normalized mass flow and
the uncertainty in the mass flow, inlet and outlet pressures previously defined. As
predicted by the theory (equation (2.8)) the data are well-described by a straight line
fit. For the case shown in figure 4, the coefficient of determination for the linear
least-squares fit is better than 0.99 and for all of the results presented in this paper
the coefficient of determination was greater than 0.96. Note that the time taken
to acquire each of these data sets (approximately 20 data points at a given outlet
pressure) is roughly 12 hours due to the long times needed to ensure system stability
and experimental accuracy.
38 E. B. Arkilic, K. S. Breuer and M. A. Schmidt

Gas Channel Pi (kPa) Po (kPa) Ko

Ar 1A, 2A 130–430 13–102 0.05–0.41
N2 1B 132–430 16–101 0.05–0.34
CO2 1B 135–430 15–100 0.03–0.44
Table 2. Range of test conditions.

4.1. Experimentally determined viscosity

As discussed earlier, the intercept of the normalized mass flow data as shown in
figure 4, is determined solely by geometry and known gas quantities. This was used
as a check of the quality of the experimental data. Using the theoretically derived
definition of the intercept, A (2.6), we can derive an empirically determined coefficient
of viscosity, µe :
H 3W
µe = , (4.1)
12ALRT
which can then be compared with the tabulated viscosity for the experimental flow
conditions.
The empirical viscosities are shown as a function of outlet Knudsen number in
figures 5(a), 5(b) and 5(c) for experiments conducted in argon, nitrogen and carbon
dioxide respectively. Two sets of data are presented for argon. These represent
measurements taken in two different channels. Each data point is plotted with its
corresponding uncertainty. The dotted line indicates the tabulated value of µ for the
working gas.
For the majority of the experiments the empirical viscosity is equal to the accepted
value for the viscosity to within the uncertainty of the measurements, indicating that
the experimental data are of high quality. The exception to this are three points at
the low outlet Knudsen numbers for the carbon dioxide series, shown in figure 5(c).
It is believed that the most likely reason for this anomalous result is a gas impurity
introduced during the exchange of high-pressure gas tanks. (During flow experiments
a substantial amount of high-pressure gas is ‘consumed’ by the pressure regulators,
i.e. bled to the atmosphere.) However, since records of tank changes were not kept,
this belief cannot be verified and the results cannot be outright rejected. However,
measurements of TMAC based on these three data points should be regarded with
caution. The flow measurements at this condition (K = 0.032) were repeated five
times in a second series of experiments designed to check this anomalous result and
to provide some data on repeatability. The experimentally determined viscosity for
these points are also shown in figure 5(c). This series shows excellent repeatability
and lies within the range of experimental uncertainty. These results increase our
confidence in the data and reinforce the belief that the anomalous data present in
figure 5(c) was, as suspected, due to accidentally introduced gas impurities. It also
emphasizes the utility of the empirical viscosity (or any other means for comparing
the measured value of A to its predicted value) as a valuable warning on the integrity
of the data in these difficult experiments.
Although the theory developed earlier is strictly valid only at small values of K,
the obtained values of µe suggest that it might remain valid at much higher values
of K. This is, perhaps, not surprising. Asymptotic expansions are commonly found
to be accurate well beyond their strict range of validity. In this case, however, one
 Tangential momentum accommodation in microchannels 39
2.5
Viscosity (Pa s)
(×10–5)

2.0

(a)
1.5
0 0.1 0.2 0.3 0.4 0.5
2.5
(b)
Viscosity (Pa s)

(×10–5)

1.5
0 0.1 0.2 0.3 0.4 0.5
1.8
(× 10–5) (c)
Viscosity (Pa s)

1.6 Second series of (5) repeated measurements

1.4

1.2
0 0.1 0.2 0.3 0.4 0.5
Ko
Figure 5. Empirical viscosity vs. outlet Knudsen number (a) Working gas: argon (data from two
different channels). (b) Working gas: nitrogen. (c) Working gas: carbon dioxide. A second series of
data was taken at one Knudsen number (K = 0.032) to assess repeatability.

possible reason for this extended validity might be due to the fact that the Knudsen
number along most of the channel remains quite low (due to the high pressure along
the length of the channel) and thus, even at relatively high exit Knudsen number
conditions, most of the fluid in the channel sees conditions that are well within the
strict definition of slip flow. In addition, the channel geometry is so simple that many
of the correction terms that should come in at higher order (for example, from the
Burnett equations) are not present due to the geometric simplicity of the long channel
geometry. Thus, although we should place a limit of K < 0.1, the model might well
hold for values of the Knudsen number far higher.

5. Normalized flow and TMAC measurements

Having established confidence in the accuracy of the data, we now proceed to
determine the TMAC, σ, from the slope of the normalized mass flow. The results of
this are shown in figures 6(a), 6(b) and 6(c) for argon, nitrogen and carbon dioxide
respectively. As with the empirical viscosity measurements, two sets of argon data
are reported, representing two different flow channels. Each of the data points in
each figure corresponds to a separate set of flow experiments similar to those shown
40 E. B. Arkilic, K. S. Breuer and M. A. Schmidt
1.2 1.2
(a) (b)

0.8 0.8
TMAC

TMAC
0.4 0.4

0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
Ko Ko

1.2 Suspect measurements (c)

(based on anomalous values of µe)

0.8
TMAC

Second series of (5)

repeated measurements

0.4

0 0.1 0.2 0.3 0.4 0.5

Ko

Figure 6. TMAC values as a function of outlet Knudsen number. (a) Argon: data from two distinct
microchannels are presented (Series I and II). (b) Nitrogen. (c) Carbon dioxide: second a series of
five measurements was made at Ko = 0.032 after observing the anomalous values of µe in order to
check data consistency and repeatability.

in figure 4 and the figures are derived from the same data used to extract the
experimentally determined viscosity, shown in figure 5.
There are two striking features of these results. First, for the majority of the tests,
there is a significant deviation from unity of the TMAC value, with most of the values
ranging between 0.75 and 0.85. Secondly, the value of the TMAC derived from the
data is more or less constant with outlet Knudsen number. For the carbon dioxide
data at low Knudsen number, the values of the TMAC shown are greater than one.
However, these data should be discounted based on the anomalous values of µe
corresponding to these conditions. The TMAC based on the second series of carbon-
dioxide experiments (five experiments at the same outlet condition of K = 0.032)
indicate sub-unity values of TMAC and are consistent with, though lower than, the
values of TMAC derived from the first series of experiments.
Another interesting observation from figure 6(b) is an apparent increase in the
TMAC as the outlet Knudsen number approaches zero. At this stage, we are unable
to determine whether there is any physical reason for this change (such as channel
‘aging’) or whether this is simply a statistical quirk or an artifact of the experimental
procedure. It is worth noting that (in all cases) the data were collected starting at
high Knudsen number, and progressing to lower values of K. In addition, for this
case, the channel was ‘pristine’ at the start of the experimental series. This might
suggest that the TMAC increases slightly as the channel ages. However, at this stage
the relevance of this observation is still not clear and will hopefully be addressed in
future experiments.
 Tangential momentum accommodation in microchannels 41
1.2

(× 1010)

1.0

0.8
Mass flow (kg s–1)

0.6

0.4

0.2

0
1 2 3 4 5
Pressure ratio
Figure 7. Mass flow versus pressure for argon with outlet Knudsen number of 0.053: ——, the
continuum solution (K = 0); · · · · · ·, the analytic solution for rarefied flow with complete accommo-
dation (K = 0.053, σ = 1); - - - - -, the analytic solution for rarefied flow, incomplete accommodation,
using the value of TMAC derived from the experimental data (k = 0.053, σ = 0.8); x, experimental
measurements and their size represents the uncertainty in the experimental measurements using the
current configuration.

As a final step, it is useful to take the value of the TMAC derived using the
above procedures and to plot the dimensional mass flow against the applied pressure,
compared with the corresponding theoretical prediction. This is shown in figure 7 for
the case of argon, flowing at Ko = 0.053 and using σ = 0.8. The close agreement is
evident, confirming both the accuracy of the experiments and the good applicability
of the low Knudsen number theory to this flow regime. The relevance of the sub-
unity accommodation can be seen by comparing the dotted and dashed lines which
correspond to complete and incomplete accommodation respectively.

6. Concluding remarks
The results presented here clearly show a sub-unity accommodation coefficient
for prime silicon substrates. The TMAC is (as expected) independent of pressure,
reinforcing the credibility of these results. What is particularly intriguing about
these results is that they are achieved at relatively standard operating conditions
for MEMS and that the surfaces and gases used are not by any means unusual (in
comparison, for example, with some of the previous TMAC measurements in which
the surfaces were painstakingly prepared and isolated from contamination). Thus,
one would expect that these phenomena are observed in standard MEMS devices and
that such sub-unity accommodation should be accounted for in modelling of viscous
interactions.
A word should be added on the difficulty of such measurements. The mass flows
in MEMS are so small and so sensitive to small variations in geometry, temperature,
etc. that very careful measurements are required to differentiate between slip, TMAC
42 E. B. Arkilic, K. S. Breuer and M. A. Schmidt
and other extraneous effects. In the present work, this necessitated very carefully
controlled device fabrication, and the development of a mass flow measurement
system that was previously not available. With this system now in place, we anticipate
extending such studies to include different gas–surface combinations and the effects
of surface roughness and adsorbants such as moisture or chemical contaminants.

We are most grateful for the financial support of NSF, DARPA, AFOSR and
Physical Science Incorporated who, at various times and at various levels, make
the work possible. We should also like to acknowledge Harold Wachman and Isaac
Greber who both contributed valuable suggestions and insight. Some of the data
reported in this paper was presented at the Rarefied Gas Dynamics Symposium,
Beijing, China (Arkilic, Schmidt & Breuer 1997a).

REFERENCES
Arkilic, E. 1994 Gaseous flow in micron-sized channels. Master’s thesis, Massachusetts Institute
of Technology, Cambridge, MA.
Arkilic, E., Schmidt, M. & Breuer, K. 1994 Gaseous flow in microchannels. In Application
of Microfabrication to Fluid Mechanics, ASME Winter Annual meeting, Chicago, IL (ed. P. R.
Bandyopadhyay), pp. 57–65.
Arkilic, E., Schmidt, M. & Breuer, K. 1997a TMAC measurement in silicon micromachined
channels. In Proc. 20th Symp. on Rarefied Gas Dynamics, Bejiing, China (ed. C. Shen). Peking
University Press.
Arkilic, E., Schmidt, M. & Breuer, K. 1998 Sub-nanomol flow measurement near atmospheric
pressure. Exps. Fluids 25, 37–41.
Arkilic, E. B. & Breuer, K. S. 1993 Gaseous flow in microchannels. AIAA Paper 93–3270.
Arkilic, E. B., Schmidt, M. A. & Breuer, K. S. 1997b Gaseous slip flow in long microchannels.
J. Microelectromech. Systems 6, 167–178.
Beskok, A. & Karniadakis, G. 1993 Simulation of heat and momentum transfer in micro-
geometries. AIAA Paper 93–3269.
Beskok, A. & Karniadakis, G. 1999 A model for flows in channels, pipes and ducts at micro- and
nano-scales. J. Microscale Thermophys. Engng 3, 43–77.
Cai, C.-P., Fan, J., Boyd, I. & Candler, G. 1999 Direct simulation methods for particle simulation
of micro-channel flows. AIAA Paper 99–3801.
Goodman, F. & Wachman, H. 1976 Dynamics of Gas-Surface Scattering. Academic.
Harley, J., Huang, Y., Bau, H. & Zemel, J. 1995 Gas flow in micro-channels. J. Fluid Mech. 284,
257–274.
Maxwell, J. C. 1879 On stress in rarefied gases arising from inequalities of temperature. Phil.
Trans. R. Soc. Lond. Reprinted in The Scientific Papers of James Clerk Maxwell. Dover.
Nakarjakov, A., Borisov, S., Sharipov, F. & Suetin, P. 1995 The effect of the surface chemical
composition on a free molecular gas flow in a cylindrical channel. In Rarefied Gas Dynamics
(ed. J. Harvey & G. Lord). Oxford University Press.
Pfahler, J., Harley, J., Bau, H. & Zemel, J. 1991 Gas and liquid flow in small channels. In
Micromechanical Sensors, Actuators, and Systems, ASME Winter Annual meeting, Atlanta, GA,
pp. 49–59.
Piekos, E. & Breuer, K. S. 1995 Computational modelling of micromachined devices. AIAA Paper
95–2089.
Piekos, E. & Breuer, K. S. 1996 DSMC modeling of micromechanical devices. Trans. ASME:
J. Fluids Engng 118, 464–469.
Pong, K., Ho, C.-M., Liu, J. & Tai, Y.-C. 1994 Non-linear pressure distribution in uniform
microchannels. In Application of Microfabrication to Fluid Mechanics, ASME Winter Annual
meeting, Chicago, IL, pp. 51–56.
Seidl, M. & Steinheil, E. 1974 Measurement of momentum accommodation coefficients on
surfaces characterized by Auger spectroscopy, Sims and LEED. In Rarefied Gas Dynamics (ed.
M. Becker & M. Fieberg). DFVLR Press.
 Tangential momentum accommodation in microchannels 43
Shih, J., Ho, C., Liu, J. & Tai, Y. 1996 Monatomic and polyatomic gas flow through uniform
microchannels. In Application of Microfabrication to Fluid Mechanics, ASME Winter Annual
meeting, Atlanta, GA, pp. 197–203.
Thomas, L. & Lord, R. 1974 Comparative measurements of tangential momentum and thermal
accommodations on polished and on roughened steel spheres. In Rarefied Gas Dynamics (ed.
K. Karamcheti). Academic.
White, F. M. 1991 Viscous Fluid Flow. McGraw-Hill.

View publication stats