Time-of-Flight Flow Imaging of Two-Component Flow inside a Microfluidic Chip
Elad Harel, Christian Hilty, Katherine Koen, Erin E. McDonnell, and Alex Pines*
Materials Sciences Division, Lawrence Berkeley National Laboratory, and Department of Chemistry, University of California,
Berkeley, California 94720, USA
(Received 6 August 2006; published 5 January 2007)
Here we report on using NMR imaging and spectroscopy in conjunction with time-of-flight tracking to
noninvasively tag and monitor nuclear spins as they flow through the channels of a microfluidic chip. Any
species with resolvable chemical-shift signatures can be separately monitored in a single experiment,
irrespective of the optical properties of the fluids, thereby eliminating the need for foreign tracers. This is
demonstrated on a chip with a mixing geometry in which two fluids converge from separate channels, and
is generally applicable to any microfluidic device through which fluid flows within the nuclear spin-lattice
DOI: 10.1103/PhysRevLett.98.017601PACS numbers: 76.60.Pc, 47.61.Ne, 82.56.?b
Microfluidic technology holds great potential for ad-
vancing research in a wide range of areas, for example,
screening of ligand binding for drug development , as
well as for performing fundamental studies of chemical,
physical, and biological processes . Naturally, fluid flow
inside microfluidic chips plays a key role in their function-
ing and must be well understood to enable better device
design for specific applications . In this respect, finding
methods to characterize the properties of the microfluidic
device noninvasively and optimize the vast number of
possible configurations and designs that govern efficiency
and usability is critical.
Here we report on using NMR imaging and spectros-
copy to investigate multiphasic fluid flow inside a micro-
fluidic chip with a simple mixing geometry. To date, the
most common method for monitoring mixing has been
based on optical detection, which typically requires the
use of a fluorescent dye or other foreign tracer .
However, optical detection suffers from a few serious
drawbacks. For example, it is completely precluded in
samples or sample environments that are opaque, confining
it to chips fabricated from optically transparent materials.
And while optical detection can provide chemical specif-
icity, it can do so only in cases where the substance con-
tains a chromophore, or where a molecular sensor has been
designed to detect a specific substance . Also, in some
cases, addition of fluorescent dyes in a microfluidic system
can alter the flow properties of the system through non-
specific interactions with the walls of the microchannels
. On the other hand, sensitivity is an inherent problem
for absorption optical detection resulting from the short
optical path length through microchannels .
Magnetic resonance (MR) is capable of bypassing some
of those limitations. As an analytical technique NMR has
achieved a significant impact across nearly all disciplines
of science from structural biology  and drug discovery
 to materials chemistry , to name just a few.
Together with magnetic resonance imaging (MRI), it is a
chemically sensitive method, suitable for microfluidic chip
analysis provided that spin active nuclei such as hydrogen
are present. Unfortunately, sensitivity has been a serious
limitation to the more widespread use of magnetic reso-
nance methods because the filling factor of the analyte in a
microfluidic channel is very small due to the necessarily
large coil that surrounds the chip. For purely spectroscopic
applications, several groups have combated this limitation
by fabricating surface coils directly onto the chip itself to
increase sensitivity . However, in addition to requiring
a major effort in the way of chip fabrication which is not
compatible with many of the more sophisticated chip
designs already in place, such an approach enables acquis-
ition of an NMR spectrum from only a single point on the
device. Remote detection NMR provides a solution to both
of these problems by physically separating the encoding
and detection steps of conventional MR experiments so as
to individually optimize each . It is especially well
suited for studies of flow as it naturally correlates spectro-
scopic or imaging information about the fluid or fluid
environment with the time of flight (TOF) of the encoded
nuclear spins from any location inside the sample to the
detection region .
Previously, we have demonstrated the ability of the
remote methodology to examine the flow of hyperpolar-
ized129Xe gas inside a porous Bentheimer rock  and a
single-channel microfluidic device . More recently, we
used the large chemical-shift sensitivity of129Xe to its
local environment to characterize its flow and dispersion
through a sample of silica aerogel by encoding for the
chemical shift of xenon in different pore environments
For molecules, in particular, liquids, each chemical spe-
cies present shows a unique chemical-shift signature as it
passes through the detector. This offers several advantages:
first, it allows the flow from each species, regardless of the
number, to be differentiated and thus imaged in a single
remote experiment as long as resonances from different
fluids are sufficiently resolvable and provided that each
fluid reaches the detection coil on the order of T1after
PRL 98, 017601 (2007)
5 JANUARY 2007
© 2007 The American Physical Society
encoding. For conventional, direct NMR detection, any
magnetic field inhomogeneity across the entire microflui-
dic chip will lead to poor discrimination between the
various nuclei, or the fluid channels may be too close in
space for effective spatial selection. The advantage of the
present method lies in the fact that because the fluids retain
their chemical composition throughout travel to the detec-
tion region their NMR signature is preserved. Therefore,
each species’ fluid flow can be monitored separately by
selecting out the appropriate peak or peaks in the spectra
obtained in the detection coil. This concept is illustrated in
Fig. 1 on a T-mixer chip with two inlets. The basic remote
detection pulse scheme consisting of an encoding se-
quence, storage pulse, and stroboscopic detection is illus-
trated in Figs. 1(a)–1(c). The width of each travel curve
[Fig. 1(d)] is a measure of the axial dispersion of the fluid
flow, while the signal minimum is the most abundant TOF
for travel from the encoding to detection region. More
precise information is gained if a magnetic resonance
imaging technique is applied during encoding, such as a
gradient for phase encoding. Similarly, it is possible to
encode chemical shifts, in which case a reconstructed
We demonstrate the potential of this approach by study-
ing mixing in a simple microfluidic chip. The chip contains
two channels, each with a width of 200 ?m and a depth of
100 ?m, which converge at a T junction into one larger
400 ?m wide channel. Pure ethyl alcohol flows through
one of the channels and distilled water through the other.
The encoding coil is a commercial imaging coil (Varian,
Palo Alto,California) large enoughto encompass the entire
microfluidic chip. For detection, a small solenoid coil with
a diameter of 500 ?m was wound to allow a thin capillary
(150 ?m inner diameter) to pass through. Based on pre-
vious measurements, this coil is about 2 orders of magni-
tude more sensitive than the encoding coil for comparable
volumes . All experiments were performed on a Unity
Inova NMR spectrometer (Varian Inc., Palo Alto, CA) at a
magnetic field of 7.05 T, corresponding to a1H frequency
of 300 MHz. Figure 2 shows a 2D representation of a
remote spectral vs TOF dimension acquired by applying
an incremented evolution time between the excitation and
storage pulse and Fourier transforming the resultant indi-
rect interferogram. The flow through each channel was
adjusted so as to demonstrate that a priori knowledge of
the origin of each fluid combined with such 2D spectral
TOF data can give useful information about dispersion of
each fluid’s flow through the device even when full imag-
ing is not possible due to experimental or measurement
time constraints. In the resulting spectrum, the resonances
stemming from the channel carrying ethyl alcohol are
present at early times (ttravel< 500 ms), while the single
peak from the channel carrying water arrives at a later time
(ttravel< 750 ms). Notably, any natural magnetic field gra-
dient across the chip will manifest itself in a change in the
apparent chemical shift. In this regard, separating the
chemical shift in the detection coil offers an advantage
since such undesirable gradients in the encoding region do
not affect the detected spectra.
Figure 2(b) shows the results of an experiment where
spins were inverted in 2 mm slices parallel to the direction
of flow. Signal at early times arises from spins encoded
near the outlet of the chip, while later times correspond to
the inlet region. Such remote inversion recovery experi-
ments are a convenient and rapid method (<1min) to
measure dispersion along the channels. The dispersion as
measured by remote detection is in fact a convolution of
contributions from the flow inside the sample itself and
from travel of the fluid from the sample to the detection
coil. Considerable dispersion may arise during the travel
time of the fluid as it travels from the outlet to the detection
coil. However, it is still possible to compare the dispersion
from two separate regions inside the chip because each
must traverse the same path from the outlet to the detection
Imaging was performed in a point-by-point fashion by
phase encoding in reciprocal (k) space. Such indirect phase
modulation is necessary in remote detection since the
readout dimension, here replaced by a storage pulse, is
not present as in conventional imaging or spectroscopy.
The longer experiment time that arises due to this scheme
is, however, compensated for by the additional time-of-
flight information that is obtained. Figure 3 shows the
results of a remote 2D time-resolved, flow imaging experi-
ment in which each fluid is selected according to its spectra
in the detection spectra [Fig. 3(b)]. This three-dimensional
α β γ
FIG. 1 (color).
experiment applied to a microfluidic chip. (a) Fluids with differ-
ent chemical-shift signatures (red and blue arrows) flow through
each inlet channel before converging into the outlet channel.
(b) Basic remote detection pulse sequence. See Refs. [12–16] for
a more detailed description of remote detection NMR pulse
Schematic of a remote detection TOF imaging
PRL 98, 017601 (2007)
5 JANUARY 2007
experiment contains dispersion data for any voxel inside
the chip, allowing the two fluids’ times-of-flight to be
easily compared [Fig. 3(c)]. Because the acquisition for
each detected spectrum is limited by the flow rates of
encoded spins, a high time resolution necessarily implies
poor spectral resolution in the detection coil as the resi-
dence time of the spinsis rather short,onthe order of10ms
in this set ofexperiments. This time resolution, on the other
hand, is sufficient to capture the relevant dynamics as
shown in the higher-resolution image in the region follow-
ing the T junction, which clearly shows that the two fluids
do not mix while traversing the single-channel part of the
microfluidic chip (Fig. 4). At the low Reynolds numbers of
ethyl alcohol and water at the flow rates used here, this is
not surprising for the current mixer geometry . This
experiment, however, highlights the potential of the pre-
sented remote detection methodology to evaluate the per-
formance characteristics of any mixing geometry.
Using remotely detected time-of-flight NMR with
chemical-shift selection, we have demonstrated a powerful
218 ms263 ms308 ms353 ms
218 ms263 ms308 ms 353 ms
0.4 0.8 1.2 1.6
0.4 0.8 1.2 1.6
FIG. 3 (color).
encoded remotely reconstructed partial images of water (red, top panels) and ethyl alcohol (blue, bottom panels) as they flow through
the microfluidic T mixer. (b) Spectrum from the detection coil. Peaks that were chosen for the reconstruction of the images in (a) are
marked with dashed lines of the corresponding color. Broadening of the peaks is due to a short residence time inside the detection coil.
(c) Position in x versus time of flight obtained by projecting the three-dimensional data set along z.
Time-resolved remotely detected images of spectrally resolvable fluid components. (a) Two-dimensional, phase-
FIG. 2 (color).
versus TOF dimension. (a) Remotely re-
constructed spectrum of a 2.5 mm slice
encoded across the chip, acquired by
incrementing ?delay and Fourier trans-
The flow rate of each fluid was adjusted
to separate the resonances along the TOF
dimension. (b) 2 mm slices along x were
chosen by a combination of x gradient
and rf pulse to give a ? rotation of the
magnetization. Blue (ethyl alcohol) and
orange (water) curves in the lower panel
represent cross sections taken at the po-
sition indicated by the dashed lines.
Frequency and spatial
PRL 98, 017601 (2007)
5 JANUARY 2007
method for detecting fluid flow of multiple species inside a
microfluidic chip, noninvasively. This offers significant
advantages over optics-based methods as it does not re-
quire tracers or markers to be introduced into the fluid
stream, and it allows the use of opaque materials for chip
fabrication. As a means of detection, coupling magnetic
resonance with microfluidics makes available the multi-
tude of methods routinely used in high-resolution NMR
spectroscopy and imaging. Future work will focus on
imaging flow in chips with geometries designed to enhance
mixing, for tracking chemical reactions, and monitoring
We would like to thank T. Logan for help with micro-
fabrication and J. Granwehr for helpful discussions. The
microfluidic chip was fabricated in the Berkeley Microlab.
E.H. is supported by the U.S. Department of Homeland
Security under DOE Contract No. DE-AC05-00OR22750.
This work is supported by the Director, Office of Science,
Office of Basic Energy Sciences, and Materials Sciences
Divisionsof the U.S.Department ofEnergyunder Contract
Electronic address: firstname.lastname@example.org
 T. Chovan and A. Guttman, Trends Biotechnol. 20, 116
 J.W. Hong, V. Studer, G. Hang, W.F. Anderson, and S.R.
Quake, Nat. Biotechnol. 22, 435 (2004).
 H.A. Stone, A.D. Stroock, and A. Ajdari, Annu. Rev.
Fluid Mech. 36, 381 (2004).
 A.D. Stroock, S.K. Dertinger, A. Ajdari, I. Mezic,
H.A. Stone, and G.M. Whitesides, Science 295, 647
 A.K. Singh, E.B. Cummings, and D.J. Throckmorton,
Anal. Chem. 73, 1057 (2001).
 D. Ross and L.E. Locascio, Anal. Chem. 75, 1218 (2003).
 L. Zhu, C.S. Lee, and D.L. DeVoe, Lab Chip 6, 115
 K. Wuthrich, NMR of Proteins and Nucleic Acids (John
Wiley & Sons, New York, 1986).
 P.J. Hajduk, R.P. Meadows, and S.W. Fesik, Q. Rev.
Biophys. 32, 211 (1999).
 P.T. Callaghan,
 H. Wenink, F. Benito-Lopez, D.C. Hermes, W. Verboom,
H.J.G.E. Gardeniers, D.N. Reinhoudt, and A. van den
Berg, Lab Chip 5, 280 (2005).
 A.J. Moule, M.M. Spence, S.I. Hans, J.A. Seeley, K.L.
Pierce, S. Saxena, and A. Pines, Proc. Natl. Acad. Sci.
U.S.A. 100, 9122 (2003).
 J. Granwehr, E. Harel, S. Han, S. Garcia, A. Pines,
P.N. Sen, and Y.Q. Song, Phys. Rev. Lett. 95, 075503
 C. Hilty, E.E. McDonnell, J. Granwehr, K.L. Pierce, S.I.
Han, and A. Pines, Proc. Natl. Acad. Sci. U.S.A. 102,
 E. Harel, J. Granwehr, J.A. Seeley, and A. Pines, Nat.
Mater. 5, 321 (2006).
 E.E. McDonnell, S.I. Han, C. Hilty, K. Pierce, and
A. Pines, Anal. Chem. 77, 8109 (2005).
 An estimate for the Reynolds number is given by R ?
?Uh=?, where ? is the density, U is the mean velocity, h
is the channel depth, and ? is the shear viscosity. For the
data in Figs. 3 and 4 the flow rate was approximately
13 cm=s, giving a Reynolds number of about 15 for water
and ethyl alcohol. At these low Reynolds numbers, flow is
expected to be laminar, and the primary mixing mecha-
nism is due to diffusion across streamlines. The Peclet
number, which is defined as the ratio of convective to
diffusive transport, is given by Pe ? uzh=D, where D is
the molecular diffusivity and uzis the component of the
velocity along the direction of flow. Under these experi-
mental conditions it has a value greater than 102, indicat-
ing that convective transport is much faster than diffusive
transport. The axial distance needed for complete mixing
can be estimated by ?zm? Pe ? h, which gives a value of
about 10 cm, significantly longer than the 1 cm region
over which mixing is allowed to occur from the T junction
to the outlet of the microfluidic chip.
218 ms 263 ms308 ms353 ms
FIG. 4 (color).
water inside the outlet channel of the microfluidic chip. Resolution along x is 75 and 2.5 mm along z. The flow conditions were
identical to those obtained for Fig. 3.
High-resolution time-resolved images of mixing inside the microfluidic chip. Contour plots of ethyl alcohol and
PRL 98, 017601 (2007)
5 JANUARY 2007