Membraneless microseparation by asymmetry in curvilinear ...

Journal of Chromatography A, 1162 (2007) 126–131

Membraneless microseparation by asymmetry
in curvilinear laminar flows

Jeonggi Seo ∗, Meng H. Lean, Ashutosh Kole

Palo Alto Research Center, Palo Alto, CA 94304, USA
Available online 21 June 2007

Abstract

Membraneless microseparation by asymmetric inertial migration is studied in curvilinear laminar flows and evidence of the microseparation is
presented. Along a curvilinear laminar flow, transverse particle migration involves competition between three shear-flow effects; the tubular pinch
effect, centrifugal force, and Dean’s vortex. Equilibrating control of migration allows for particle separation to different outlets. No filter-media
or external force is necessary for the microseparation utilizing only shear-flow characteristics. A double-spiral design effectively controls the
migration to optimize microseparation. The concentration ratio of 10 ␮m beads from the two different outlets was 660 times at 92 mm/s of flow
velocity. This new technology has great potential for high-throughput and low cost in bio-agent and particulate separation at both macro and micro
scales.
© 2007 Elsevier B.V. All rights reserved.

Keywords: Microfluidics; Particle separation; Microfiltration; Asymmetric tubular pinch effect; Dean’s vortex; Spiral channel; Membraneless separation; Curvilinear
laminar flow

1. Introduction cles by their electromagnetic properties. Laminar flow profiles
at microfluidic junctions cause particles to follow different paths
Recently, microseparation is becoming an important tech- [11,12]. Even traditional sedimentation and centrifugation incor-
nology to safeguard human life from harmful bio-agents. porated microfluidic techniques for more refined performance
Contaminated air or water by natural outbreaks or intentional [13,14].
attacks imposes dangerous threats to humans. The outbreak of
Escherichia coli O157:H7 in 2006 [1] and the anthrax attack However, present demands in microseparation require unin-
in 2002 [2] are typical examples. Separating, concentrating and terrupted, low cost, and high-throughput techniques. Current
detecting low levels of harmful bio-agents – including toxins microseparation methods employ batch processes requiring
such as ricin, bacteria such as anthrax (Bacillus anthracis), pro- frequent down time for system preparation or necessitate
tozoans such as Giardia, and numerous viruses – are extremely comparatively large investments in supporting equipment.
important for scientific study, health, and safety. Microsepara- Traditional centrifugation works only for batch processes. Sed-
tion is one of the essential biological security tools of the modern imentation of microparticles is not effective, requiring long
biotechnology revolution. processing times and large surface areas. Membrane separation
degrades in performance over time due to clogging and chemical
Recent advances in microfluidics have brought about a num- degradation. Dielectrophoresis and immunomagnetic separation
ber of new techniques for microseparation. A combination of have been applied for low-throughput uses. Microfluidic FFF
microfluidics and field flow fractionation (FFF) resulted in relies on applied external forces complicating experimental set-
microfluidic FFF [3–5]. Periodic microstructures in microflu- ups. Scaling issues in microfluidic FFF still need to be addressed
idic channels induce deterministic particle paths on the basis of [15]. Other microfluidic separation devices in laminar flows use
particle size, allowing separation [6]. Dielectrophoresis (DEP) comparatively slow flow rates, losing their efficiency at higher
[7–9] and immunomagnetic techniques [10] manipulate parti- flow rates [16,17].

∗ Corresponding author. Tel.: +1 650 812 4887; fax: +1 650 812 4251. In this study, we propose a new separation technology, based
E-mail address: [email protected] (J. Seo). solely on the intrinsic flow characteristics in curvilinear chan-
nels. Dispersed particles in a curvilinear channel at low Reynolds
0021-9673/$ – see front matter © 2007 Elsevier B.V. All rights reserved. numbers make transverse migration and become ordered for
doi:10.1016/j.chroma.2007.05.110

J. Seo et al. / J. Chromatogr. A 1162 (2007) 126–131 127

Fig. 1. Asymmetry in a curvilinear channel. The centrifugal force on fluid elements generates the secondary flow or ‘Dean’s vortex’. This flow perturbs the tubular
pinch effect to induce the asymmetry in the curvilinear channel. Particles are retained at inside equilibrium where the three forces such as centrifugal force, lift force,
and lateral force are balanced.

microseparation. Transverse particle migration involves compe- dual recirculation paths along the top and bottom walls to sat-
tition between three effects from shear flows; the tubular pinch isfy fluidic continuum; creating transverse rotational secondary
effect, centrifugal force, and Dean’s vortex. The ‘tubular pinch’ flows.
effect (TP effect) balances three lateral forces to move particles
to equilibrium positions in a straight channel [18–29] depend- A curvilinear channel at low Reynolds numbers adds the cen-
ing on density. The first two are simple-shear-induced [19] and trifugal force to the TP effect, resulting in asymmetric lateral
stresslet-velocity-field-induced [25], while the third is due to inertial migration (Fig. 1). The centrifugal force on particles
wall lift [28,29]. The Dean’s vortex is the transverse secondary competes with three forces from the TP effect and the centrifu-
flow in a curvilinear channel [30–33]. The no slip boundary con- gal force on a fluid induces the transverse secondary flow or
dition generates velocity gradient in the primary flow, creating Dean’s vortex. This perturbation establishes new equilibrium
a fast moving core and slower flows near boundaries. Along the positions depending on the size and mass of particles. Parti-
curvilinear channel, the fast flowing core in the channel is moved cles at separate equilibrium positions run into different outlets
outward by the centrifugal force. The displaced fluid creates and achieve microseparation. The spiral flow separator demon-
strates microseparation in a curvilinear channel. It consists of

Fig. 2. A double-spiral design to maximize separation efficiency. Two spiral channels are joined to form a double-spiral channel. The sample suspension is introduced
to the inlet and flow into the center junction in a counterclockwise direction. After the center junction, the sample suspension flows out to the bifurcated outlet in a
clockwise direction. Dispersed particles in the suspension were ordered into bands through three-stage separation.

128 J. Seo et al. / J. Chromatogr. A 1162 (2007) 126–131

microfluidic channels in a flat spiral. Its uninterrupted separa-
tion process relies solely on internal shear flow, eliminating the
need for media-filters or external force fields. Simplicity in the
microseparation allows this device to provide inline integration
with other downstream processes as well as to serve as stand-
alone applications. Its dynamic separation capacity makes it
suited for both high-throughput macroscale and fine microscale
lab-on-chip applications.

2. Experimental

2.1. Materials

Si (100) wafers were purchased from Silicon Quest Inter-
national (Santa Clara, CA, USA). SU-8 2100 was obtained
from MicroChem (Newton, MA, USA). Sylgard 184 (poly-
dimethylsiloxane, PDMS) was acquired from Dow Corning
(Midland, MI, USA). Suspended polystyrene beads of differ-
ent sizes were obtained from Polysciences (Warrington, PA,
USA). Self-adhesive silicone sheet was purchased from Fralock
(Valencia CA, USA).

2.2. Microdevice design and fabrication

A microdevice for separation has two different spiral chan- Fig. 3. Fabrication processes and experimental set-ups. (a and b) 100 ␮m thick
nels in the same plane as shown in Fig. 2. The first channel SU-8 on a Si wafer is patterned by microfabrication techniques to serve as a
starts from an inlet and proceeds in a counterclockwise direc- mold. (c) A pre-mixture of PDMS is poured on the mold. (d and e) The cross-
tion to the center. The second channel forms a junction with linked PDMS is detached and holes are punched. (f) Plastic tubes are inserted
the first one at the center but changes its flow direction clock- after a thin PDMS sheet was bonded. (g) A completed device is shown (the
wise, going out to a bifurcated outlet. Each spiral makes 12 turns channel is filled with a blue dye solution for visual clarity). (h) The completed
from the edge to the junction. The channel dimension is 300 ␮m device is placed between a pump and collection tubes for separation.
wide and 100 ␮m high. Two adjacent channels are 300 ␮m
apart. bubbles. DI water was also added to the original bead solu-
tions to dilute their concentration. A syringe pump was used to
The prototype fabrication sequence for building the mem- introduce a mixed sample solution into the device. Separated
braneless spiral micro-separator employed soft-lithography with samples flowed out through a bifurcated outlet channel. Out-
PDMS [34,35]. First, a 100 mm silicon wafer (Fig. 3a) had a let valves were adjusted for flow uniformity between outlets,
100 ␮m thick SU-8 2100 layer. Mask exposure with 365 nm if necessary. The collected samples were diluted 50 times for
wave light and chemical development patterned the SU-8 layer particle counting (Z2 Coulter Counter, Beckman Coulter, CA,
to have reverse patterns of the spiral microstructures (Fig. 3b). USA).
The silicon wafer with SU-8 structures served as a mold during
the following PDMS casting process. A PDMS pre-mixture con- 3. Results and discussion
tained a base and curing agents of PDMS with a 10:1 ratio. After
a degassing step in a vacuum chamber, the mixture evenly coated 3.1. Particle transverse migration
the mold, cross-linked in an 80 ◦C oven for 1 h (Fig. 3c). The
hardened PDMS mechanically detached from the mold. A razor Shear flow in a flat double-spiral channel (Fig. 2) transformed
cut step made ∼25 mm × 25 mm microfluidic chips (Fig. 3d). A particle distribution from a random and homogeneous pattern
blunt metal capillary tube was aligned to the inlets and outlets into ordered streams. We divided this transformation into three
and used to punch 0.5 mm through holes to allow fluidic access consecutive stages; the first and third stages with the main TP
(Fig. 3e). A 25 mm × 25 mm piece of self-adhesive 500 ␮m thick effect along small curvature paths and the second stage with
silicone sheet covered the top of the channels. Plastic capillaries the primary centrifugal force at large curvature paths. In order
tubes were fitted to the holes and the other ends of tubes are for clean visualization, a sample suspension of only 10 ␮m
routed into a pump and collection tubes (Fig. 3f). A glass slide
provides mechanical support during microseparation (Fig. 3g).
It should be noted that a set-up for spiral microseparation does
not include any filter-media or electrical device except a pump
(Fig. 3h). Before separation, deionized (DI) water was used to
prime the micro device. Special care was taken to prevent air

J. Seo et al. / J. Chromatogr. A 1162 (2007) 126–131 129

polystyrene beads was flowed into a spiral device at an average fluid was also pushed outward by the centrifugal force, inducing
velocity of 92 mm/s. an additional effect since the fluid is commonly an incompress-
ible continuum medium. The secondary flow is simulated and
The first stage had the TP effect as the major contribution. shown in Fig. 4b. Arrows represent the flow vector. Arrows near
After two complete turns, the shear flow formed a partially con- the middle plane pointed more outward than those near the top
centrated particle stream as well as a clean fluid stream along and bottom. Particle tracing might be used to characterize the
an inside boundary (Fig. 4a). The distance between the inside particle migration by the Dean’s vortex. First, a particle flow
boundary and particle stream was ∼0.4D, where D is a half of image was captured. Serial marks on particle traces showed real
the channel width. As previously discussed, the forces from the paths of slow moving particles (Fig. 4c). Image analyses com-
TP effect and centrifugal force competed to find equilibrium pared the real particle paths with the simulated fluid streamlines
positions. Since the lift force from the wall is most effective in in the middle plane (50 ␮m from the top – plane A) and the plane
the regular TP effect [18], the TP effect as well as the centrifugal adjacent to the top (12.5 ␮m from the top – plane B). The parti-
force developed an initial concentration along the inside bound- cle path and the streamlines of plane B matched well (Fig. 4e).
ary. However, the centrifugal force suppressed the lift force along The inward circulation-induced migration by the Dean’s vortex
the outside boundary by pushing the particles outward. This per- was confirmed. On the third stage, again, the TP effect became
turbation in the symmetry differentiates the curvilinear TP effect dominant.
from the regular TP effect. As the flow came into the junction of
the spiral channel, the particle distribution made the transition 3.2. Particle separation
to the second stage.
Particles are separated in shear flows by different efficien-
On the second stage, the centrifugal force played two major cies in transverse migration. The efficiency results from the
roles on the particles and the fluid. Particles saw more outward competition of the three involved effects. Inertial migration by
centrifugal force due to the larger curvature of the channel. The
width of the inside clear fluid stream is seen to increase. The

Fig. 4. Particle transverse movement in a curvilinear channel. (a) Along small curvature regions of the channel, the tubular pinch effect plays an important role in
moving particles transversely. The lift force and centrifugal force push particles outward and develop an inside clear gap between the bead stream and the inside
wall. However, the centrifugal force also perturbs the symmetry of the tubular pinch effect. (b) Fluidic simulation shows the Dean’s vortex generation near the center
junction. The curvature of the channel becomes larger to make the centrifugal force play a more distinctive role in moving particles. (c) An image is analyzed to
show the traces of particles. (d) Simulated stream lines at the middle plane are compared with the real particle trace. (e) Simulated stream lines at 12.5 ␮m from the
top of the channel are compared with the traces of particles.

130 J. Seo et al. / J. Chromatogr. A 1162 (2007) 126–131

Fig. 5. Effects of average flow velocity on separation; A mixture of 6 and 10 ␮m polystyrene beads flow through the flat double-spiral device at different average
flow velocities. (a) At 14 mm/s, the TP effect is not distinctive. (b) Two times faster velocity makes the TP effect more pronounced. (c and d) The faster velocities
move particles more inward.

the TP effect is proven to be a function of the Reynolds num- 52 mm/s. The 10 ␮m particle band made a large degree of over-
bers and particle size [19]. The Dean’s vortex is a function of lap with a wide band of 6 ␮m particles. Faster speed resulted in
the fluidic parameters and can be characterized by the geome- more distinctive TP and Dean effects on the particles to form
try ratio and Reynolds numbers [30]. The centrifugal force is narrower bead band streams.
a function of particle mass, curvature of channels, and flow
velocity. With a given device and particle mixture, the size To confirm the effect of particle sizes on separation, two dif-
of particles as well as flow velocity determines the mode of ferent bead mixtures were prepared: 3 and 10 ␮m and 6 and
operation. Large particles migrate more efficiently inward than 10 ␮m. They flowed through the double-spiral devices at three
small ones. At a bifurcated outlet, large particles above a cer- different flow velocities: 28, 55 and 92 mm/s. The collected sam-
tain cut-off size flows out through the inside outlet, resulting
in the depletion of the same size particle in the outside out- Fig. 6. Particle counting results from separations of two mixtures; 3 and 10 ␮m
let. Particles below a threshold scattered across the channel. and 6 and 10 ␮m polystyrene beads. In both cases, 10 ␮m beads show effective
Fig. 5 shows the effect of flow velocity on separation and Fig. 6 separation between inside and outside collections. The operating velocity cannot
summarizes the effects flow velocity as well as the size of discriminate 3 ␮m beads effectively. The separation factor of 6 ␮m beads is better
particles. than that of the 3 ␮m beads.

A mixture with 10 and 6 ␮m polystyrene beads flowed though
the spiral channel (Fig. 2) with different flow velocities. Images
were captured at the bifurcated outlet. At 14 mm/s, a clear fluid
layer of the asymmetric TP effect was not observed along the
inside wall (Fig. 5a). Particles were partially concentrated near
the inside wall and dispersed across the channel. It is believed
that 10 ␮m particles were stagnated near the inside wall during
weak Dean flows. The dimensionless Dean number is the ratio
of the product of inertial and centrifugal forces to viscous forces,
which is κ = (d/R)0.5 Re, where d is the channel hydraulic diame-
ter, R is the flow path radius of curvature, and Re is the Reynolds
number [33]. In the double-spiral channel (Fig. 2), the maximum
Dean number at 14 mm/s is ∼2. The weak Dean’s vortex was
previously demonstrated at the small Dean number range of 2–5
[33]. When the flow sped up, the TP effect was more clearly
developed. At 28 mm/s, particles migrated from the inside wall
and 10 ␮m particles formed a wide streaming band. 92 mm/s of
flow velocity achieved a narrower band and cleaner gap than

J. Seo et al. / J. Chromatogr. A 1162 (2007) 126–131 131

ples from each outlet were compared by counting the number [3] N. Narayanan, A. Saldanha, B.K. Gale, Lab. Chip 6 (2006) 105.
of particles in the different size ranges. The counting results are [4] H.J. Sant, B.K. Gale, J. Chromatogr. A 1104 (2006) 282.
shown in Fig. 6. The concentrations of samples were normalized [5] K.-H. Nam, W.-J. Chang, H. Hong, S.-M. Lim, D.-I. Kim, Y.-M. Koo,
as the original mixture concentration was 100. Ten micromo-
lar beads were effectively separated to the inside outlet in both Biomed. Microdev. 7 (2005) 189.
separation experiments. With the 6 and 10 ␮m mixture, the con- [6] L.R. Huang, E.C. Cox, R.H. Austin, J.C. Sturm, Science 304 (2004) 987.
centration difference of 10 ␮m beads between the inside and [7] N.G. Green, H. Morgan, J.J. Milner, Biochem. Biophys. Methods 35 (1997)
outside outlet, was 660 times at 92 mm/s of flow velocity. The
concentration differences for 6 and 3 ␮m beads were 9.3 and 1.4 89.
times, respectively. The counted data confirmed that the larger [8] Y. Huang, K.L. Ewalt, M. Tirado, T.R. Haigis, A. Forster, D. Ackley, M.J.
particles were effectively separated from the outside outlet. In
summary, faster flow velocity increases the sensitivity to the cut- Heller, J.P. O’Connell, M. Krihak, Anal. Chem. 73 (2001) 1549.
off size with particles above the cut-off being separated from the [9] H.B. Li, R. Bashir, Sens. Actuators B 86 (2002) 215.
outside outlet and collected into the inside outlet. [10] D.W. Inglis, R. Riehn, R.H. Austin, J.C. Sturm, Appl. Phys. Lett. 85 (2004)

4. Conclusion 5093.
[11] X. Zhang, J.M. Cooper, P.B. Monaghan, S.J. Haswell, Lab. Chip 6 (2006)
A continuous spiral flow separation method and device using
fluidic shear flow is presented. The intrinsic properties from 561.
shear flow in a curvilinear channel effectively separates parti- [12] Y. Sai, M. Yamada, M. Yasuda, M. Seki, J. Chromatogr. A 1127 (2006)
cles with no physical filter interfaces or external fields involved.
The TP effect and Dean’s vortex including the centrifugal force 214.
causes transverse particle migration across flow streamlines. The [13] T. Chiane´a, N.E. Assidjo, P.J.P. Cardot, Talanta 51 (2000) 835.
efficiency of this process is a function of a particle size with the [14] T.-S. Leu, P.-Y. Chang, Sens. Actuators A 115 (2004) 508.
given fluidic parameters, resulting in microseparation. Separa- [15] J.C. Giddings, J. Microcolumn Sep. 5 (1993) 497.
tion efficiency can be optimized by changing fluidic parameters [16] S.S. Shevkoplyas, T. Yoshida, L.L. Munn, M.W. Bitensky, Anal. Chem. 77
such as fluid velocity, channel geometry or intrinsic fluidic prop-
erties. Multistage separation can also be envisioned to increase (2005) 933.
efficiency. This spiral microseparator can be applicable to sep- [17] J.A. Davis, D.W. Inglis, K.J. Morton, D.A. Lawrence, L.R. Huang, S.Y.
arate bio-agents such as cells, bacteria, and biohazard particles.
Simplicity in the spiral microseparation concepts allows great Chou, J.C. Sturm, R.H. Austin, Proc. Natl. Acad. Sci. USA 103 (2006)
potential for inline integration with other applications or stand- 14779.
alone applications. Future improvement includes the design with [18] G. Segre, A. Silberberg, Nature 189 (1961) 209.
more outlets for finer resolution and rigid fabrication prototype [19] P.G. Saffman, J. Fluid Mech. 22 (1965) 385.
to withstand faster flow rates. [20] B.P. Ho, L.G. Leal, J. Fluid Mech. 65 (1974) 365.
[21] P. Vasseur, R.G. Cox, J. Fluid Mech. 78 (1976) 385.
References [22] J. Feng, H.H. Hu, D.D. Joseph, J. Fluid Mech. 277 (1994) 271.
[23] E.S. Asmolov, J. Fluid Mech. 381 (1999) 63.
[1] H. Pearson, Nature 445 (2006) 8. [24] E.S. Asmolov, Phys. Fluids 14 (2002) 15.
[2] C.F. Chyba, Foreign Aff. 81 (2002) 122. [25] J.-P. Matas, J.F. Morris, E´ . Guazzelli, J. Fluid Mech. 515 (2004) 171.
[26] B.H. Yang, J. Wang, D.D. Joseph, H.H. Hu, T.-W. Pan, R. Glowinski, J.
Fluid Mech. 540 (2005) 109.
[27] E.E. Michaelides, J. Fluids Eng. 125 (2003) 209.
[28] P. Cherukat, J.B. McLaughlin, Int. J. Multiphase Flow 16 (1990) 899.
[29] P. Cherukat, J.B. McLaughlin, A.L. Graham, Int. J. Multiphase Flow 20
(1994) 339.
[30] S.A. Berger, L. Talbot, L.-S. Yao, Ann. Rev. Fluid Mech. 15 (1983)
461.
[31] Yu.P. Gupalo, Yu.V. Martynov, Yu.S. Ryazantsev, Fluid Dyn. 12 (1977)
109.
[32] W.R. Dean, Proc. R. Soc. London Ser. A 121 (1928) 402.
[33] A.P. Susarsan, V.M. Ugaz, Proc. Natl. Acad. Sci. USA 103 (2006)
7228.
[34] Y. Xia, G.M. Whitesides, Annu. Rev. Mater. Sci. 28 (1998) 153.
[35] J. Seo, C. Ionescu-Zanetti, J. Diamond, R. Lal, L.P. Lee, Appl. Phys. Lett.
84 (2004) 1973.