ARTICLES

PUBLISHED ONLINE: 21 MARCH 2010 | DOI: 10.1038/NNANO.2010.34

Direct seawater desalination by ion concentration polarization

Sung Jae Kim1, Sung Hee Ko2, Kwan Hyoung Kang2 and Jongyoon Han1,3 *
A shortage of fresh water is one of the acute challenges facing the world today. An energy-efcient approach to converting sea water into fresh water could be of substantial benet, but current desalination methods require high power consumption and operating costs or large-scale infrastructures, which make them difcult to implement in resource-limited settings or in disaster scenarios. Here, we report a process for converting sea water (salinity 500 mM or 30,000 mg l21) to fresh water (salinity <10 mM or <600 mg l21) in which a continuous stream of sea water is divided into desalted and concentrated streams by ion concentration polarization, a phenomenon that occurs when an ion current is passed through ion-selective membranes. During operation, both salts and larger particles (cells, viruses and microorganisms) are pushed away from the membrane (a nanochannel or nanoporous membrane), which signicantly reduces the possibility of membrane fouling and salt accumulation, thus avoiding two problems that plague other membrane ltration methods. To implement this approach, a simple microuidic device was fabricated and shown to be capable of continuous desalination of sea water (99% salt rejection at 50% recovery rate) at a power consumption of less than 3.5 Wh l21, which is comparable to current state-of-the-art systems. Rather than competing with larger desalination plants, the method could be used to make small- or medium-scale systems, with the possibility of batterypowered operation.

resh water is a vital resource for human life. However, population growth and enhanced living standards, together with the expansion of industrial and agricultural activities, are creating unprecedented demands on clean water supplies all over the world. The Organization for Economic Co-operation and Development (OECD) and the United Nations (UN) have reported that 0.35 billion people in 25 different countries, particularly in the Middle East and Africa, are currently suffering from water shortage, and this will grow to 3.9 billion people (two-thirds of the world population) in 52 countries by 2025 (refs 1,2). Converting sea water into fresh water could provide the solution to the worldwide water-shortage problem, because about 97% of the total water resource on Earth is sea water, and only 0.5% of the total comprises potable, fresh water. Historically, distillation has been the method of choice for desalination of sea water, in spite of its high capital and energy costs, but this is suitable only for countries where the fuel required for distillation is relatively inexpensive. The other standard approaches to seawater desalination are reverse osmosis (RO) and electrodialysis (ED), both of which have relatively good energy efciencies ($5 Wh l21 for RO; $1025 Wh l21 for ED)3. The RO process requires the generation of high pressure to overcome the osmotic pressure of sea water (up to $27 times atmospheric pressure) across semi-permeable membranes. The ED process uses electric currents to move ions selectively through perm-selective membranes, leaving pure water behind. Current seawater-desalination techniques can produce freshwater at a cost of $$0.50.8 per 1,000 l, if operated at the scale of a large plant. However, areas affected by acute water shortage are often in the poorest, most underdeveloped countries, which lack the necessary power and water-delivery infrastructures. This presents a signicant global challenge, because the lack of clean water also creates considerable health, energy and economic challenges to the populations of

these countries. For this reason, small-scale or portable seawaterdesalination systems with low power consumptions and high throughput would be very useful in fullling important government, civilian and military needs, including humanitarian operations in disaster-stricken areas or in resource-limited settings. Another signicant challenge in seawater desalination is detecting and removing micro- or macroparticles, bacteria and other pathogens contained in the source water. These particles and microorganisms cause membrane fouling, which is a major issue for both RO4 and ED systems5,6. It is for this reason that the forward osmosis process (in which sea water is converted into even saltier liquid, followed by reverse osmosis) is considered a viable (ltration) technology for seawater desalination, even at the cost of additional energy consumption7.

Ion concentration polarization phenomena

We report a desalination process that uses ion concentration polarization (ICP) for membraneless direct desalination of sea water. ICP is a fundamental electrochemical transport phenomenon that occurs when an ion current is passed through ion-selective membranes810. Often called ion depletion or enrichment, this phenomenon arises owing to the mismatch of charge carriers at the membrane interface. The membrane (either nanochannel or nanoporous membrane) preferentially conducts only cations (cation exchange membrane) that do not match the ion conductivities in the bulk electrolyte. As a result, ion concentration gradients are generated on both sides of the membrane. Once ICP is triggered near the cation exchange membrane, the concentrations of both cations and anions decrease on the anodic side of the junction (ion depletion) and increase on the cathodic side (ion enrichment)11,12. Furthermore, any charged particles, cells and other small colloids will be similarly depleted or enriched. In combination with an

1 Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, Massachusetts 02139, USA, 2 Department of Mechanical Engineering, Pohang University of Science and Technology, San 31, Hyojadong, Gyeongbuk, 790-784, Korea, 3 Department of Biological Engineering, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, Massachusetts 02139, USA. * e-mail: jyhan@mit.edu

NATURE NANOTECHNOLOGY | VOL 5 | APRIL 2010 | www.nature.com/naturenanotechnology 2010 Macmillan Publishers Limited. All rights reserved.

297

ARTICLES

NATURE NANOTECHNOLOGY
a

DOI: 10.1038/NNANO.2010.34

external pressure-driven ow, it is possible to achieve a well-dened steady-state depletion zone using a device as shown in Fig. 1a,b. The inlet microchannel in this gure has a width of 500 mm and depth of 100 mm. The Naon nanojunction is connected to two microchannels13. Using bifurcated channels (each with a width of 250 mm and depth of 100 mm) as shown in Fig. 1, one can separate the desalted and salted streams, achieving continuous and steady-state desalination. Strong ICP zones are established, even under a condition of high ionic strength, because the pore size of the Naon nanojunction is small ($5 nm). They therefore remain cation-selective even at an ionic strength of 500 mM (30,000 mg l21; sea water). One important characteristic of this arrangement is that salt ions (and other charged debris) are also driven away from (not towards) the membrane, fundamentally eliminating the potential for membrane (nanojunction) fouling.

External pressure
V
Nano junct ion

Salte

V V oMicr ode r elect

Desalted

GND SG desalter GND

b
ANY charged species V+ pressure Sea water i Ion-depletion boundary

Bri

ne

Demonstration of ICP desalination operation

We experimentally tested the direct desalination of sea water at the scale of a single unit device. Figure 2a shows a desalination experiment carried out with natural sea water (obtained from Crane Beach, Ipswich, Massachusetts, pH % 8.48.5). NaOH was added at a nal concentration of 1 mM to the natural sea water to precipitate Ca2 ions, this being known to cause signicant precipitation in brine solution14. This process did not signicantly decrease the salinity of the sea water. The seawater sample was then physically preltered to remove any precipitation and large debris (larger than the channel dimension) such as dirt, sand and seaweed. The seawater sample was intentionally mixed with FITC (uorescent dye as a molecular marker), polystyrene nanoparticles, r-phycoerythrin (r-PE) and human whole blood that included uorescently stained white blood cells (WBCs; Hoechst). The mixture (pH % 9.19.2) was then loaded (using an external pressure pump) into the reservoir of the device at a dened ow rate ($0.120 ml min21, depending on the dimensions of channel used). Once the ICP process was initiated, a depletion zone formed within 1 s and diverted charged ions (represented by dye molecules) into the salted stream, as shown in Fig. 2a (see Supplementary Video). The ICP layer was also shown to act as a virtual barrier for any charged particles found in sea water (both negative and positive), including most solid particles, microorganisms and biomolecules (including proteins, bacteria, viruses, red blood cells and WBCs) (the device, shown in Fig. 2b, was selected to have inlet microchannel dimensions of 100 mm (width) 15 mm (depth) to allow the movements of WBCs to be clearly visualized; see Supplementary Video). This is because most water-borne microorganisms and microparticles have a non-zero (usually slightly negative) zeta potential. This means that both small salt ions and large microorganisms can be removed simultaneously from the output desalted stream, making this process highly attractive for direct desalination of sea water from natural sources. Because most ions are removed from the desalted stream, the pH of the output desalted stream was measured to be $7.07.5, which falls within the World Health Organizations recommended range of acidity for drinking water15. We conrmed that the desalination operation could be maintained over 1 h without any membrane clogging. After 1 h of operation, the reservoir at the end of the brine channel became concentrated with particles and dyes, as shown in Fig. 2c, but the reservoir of the desalted channel remained free from debris (Fig. 2d).

Desalted water

Nanojunction

GND

GND

Figure 1 | ICP desalination scheme. a,b, Schematic of micro/nanouidic desalination system with embedded microelectrode for the measuring potential drop (a) and electrokinetic desalination operations associated with the external pressure eld (b).

Quantitative analysis of ICP desalination

The ionic concentration of sea water in the desalted stream was signicantly lower than the original concentration owing to the repulsion of salts from the ion-depletion zone. To quantify the concentration in the desalted stream, in situ conductivity measurements of the desalted stream were carried out using an embedded microelectrode (Fig. 1a).
298

When an above-threshold voltage was applied and the ion depletion zone established as shown in Fig. 3a, the conductivity of the output desalted stream dropped to $0.5 mS cm21 ($3 mM; $180 mg l21) from $45 mS cm21 ($500 mM; $30,000 mg l21), the conductivity of the original sea water. In another experiment using 100 mM phosphate buffer solution ($15 mS cm21, a model for brackish water), the conductivity of the output desalted stream was also reduced, to $0.3 mS cm21 ($2 mM; $120 mg l21), as shown in Fig. 3b. Note that the salinity of potable water should be lower than 10 mM (600 mg l21). The ow rate at the desalted stream realized in this initial proof-of-concept device was $10 ml min21 (the inlet ow rate was 20 ml min21, equally split into two 10 ml min21 streams), with jEj % 75 V cm21. The total steady-state current required in our device was found to be between $1 mA (seawater desalination output at 0.25 ml min21 in a device with a microchannel cross-section of 100 mm 15 mm) and $30 mA (10 ml min21 in a device with a cross-section of 500 mm 100 mm; see Supplementary Information). Thus, the power consumption was $752,250 mW per unit device. The energy efciency of this desalting mechanism is therefore between $5 Wh l21 (75 mW/0.25 ml min21) and $3.75 Wh l21 (2,250 mW/10 ml min21). The steady-current level of ICP is much lower than that of ED for the following reasons. First, the ED current must be equivalent to the actual number of ions being displaced, but ICP simply deects the highly concentrated neutral solution into one microchannel and the lowconcentration neutral solution into the other microchannel. Because both channels are net neutral, electric current is not required for the separation of all the ions in solution. Second, ICP forms a large, low-conductivity diffusion layer (ion-depletion zone) near the membrane, which allows the current level to be low. In addition, the energy required to push the uid through the microchannel with a

NATURE NANOTECHNOLOGY | VOL 5 | APRIL 2010 | www.nature.com/naturenanotechnology 2010 Macmillan Publishers Limited. All rights reserved.

NATURE NANOTECHNOLOGY
a
500 m Seawater sample at Q = 20 lmin1

DOI: 10.1038/NNANO.2010.34

ARTICLES
Brin e

Sal

ted

Ion-depletion boundary Salted Desalted water

Nanojunction

Voff (t < 0.0 s)

Von (t = 0.5 s)

b
100 m
Sal t+ o wh le b loo d

Seawater sample at Q = 250 nl min1

Stained WBC Desalted water

Stained WBC

Nanojunction Voff (t < 0.0 s) Von (t = 0.5 s)

c
500 m

Reservoir connected to the salted stream

Reservoir connected to the desalted stream

Figure 2 | ICP desalination. Experimental demonstration of the desalination system. a, Fluorescent image tracking of desalination processes with an external ow rate (Q) of 20 ml min21 and applied electric eld of 75 V cm21. The inlet microchannel has a width of 500 mm and depth of 100 mm. The seawater sample was injected and split into salted and desalted streams. b, Fluorescent dyes (representing salts) and WBCs (representing micrometre-sized particle) passed only through the salted stream when ICP was triggered. To obtain a clear visualization of micrometre-sized particles, a small microchannel (100 mm width 15 mm depth) was used. c,d, Microscopic image of each reservoir (salted (c) and desalted (d)) after desalination for 1 h, demonstrating the cleanness of the desalted stream.

syringe pump is $0.0411.55 mWh l21 (see Supplementary Information for a detailed calculation). Therefore, the power needed for uid delivery is negligible, mainly because the uidic channel in this system does not pose such signicant uidic resistance as does the RO membrane. As a result, a simple gravity-fed uid delivery is sufcient for operation, as demonstrated in this study (see Supplementary Information). In contrast, the energy consumption of the RO process alone is $2.5 Wh l21, but the true power requirement is signicantly higher, at 5 Wh l21, because of the additional ow intake, pre-treatment, recirculation and distribution processes that need to be taken into account16. Furthermore, the state-of-the-art power efciency of the RO process ($5 Wh l21) is only achievable in large plant-scale RO facilities. Small-scale RO systems, such as shipboard desalination systems, tend to have a much lower power efciency. Therefore, the total energy consumption of the ICP desalination system would be at least comparable to the state-of-the art RO facility, and much lower than existing small-scale RO systems. Similar to RO systems, the power consumption can be further lowered by using source water with a lower salinity than sea water (such as brackish water). The theoretical lower bounds of the energy required for desalination are $0.81, 0.97 and 1.29 Wh l21 for desalted water recoveries of 25, 50 and 75%, respectively3.

Concept of small- and medium-scale systems

The unit microuidic device can be further optimized by implementing a number of improvements. First, the microelectrodes can be integrated to apply voltages near the nanojunction to drastically decrease the voltage and enhance power efciency for compatibility with existing battery technology or small-scale solar power arrangements. Second, the length of the Naon nanojunction can be controlled, signicantly reducing power loss arising from the initiation and maintenance of the ion-depletion zone. Finally, one can optimize the design of the main/brine channel to minimize the overall size of the unit chip so that maximum parallelization can be achieved within a given system size. Figure 4a shows a possible improved design of the unit device. Instead of using a liquid buffer channel, the metal microelectrode is itself used as the buffer channel, and the Naon nanoporous polymer is coated on top of the microelectrode to maintain the perm-selective ion transport. One signicant benet of this design is that the strength of the depletion force is constant along the entire microelectrode, which allows additional design exibility, critical for scaling up the system. The width of the microchannel can be increased freely, by more than a millimetre, and the device only requires two electrical connections, one inlet channel and two outlets
299

NATURE NANOTECHNOLOGY | VOL 5 | APRIL 2010 | www.nature.com/naturenanotechnology 2010 Macmillan Publishers Limited. All rights reserved.

ARTICLES
a 50
40 Conductivity (mS cm1) Low microchannel was completely desalted

NATURE NANOTECHNOLOGY
b
~500 mM Conductivity (mS cm1) 15

DOI: 10.1038/NNANO.2010.34

~100 mM 12

30

9 Completely desalted 6 Partially desalted 3 ~2 mM

20 Sea water evenly passed through Partially desalted divided microchannel sample passed along low microchannel 50 60 70 cm1) 80

10

~4 mM

0 50 60 70 80 Applied electric eld (V cm1)

Applied electric eld (V

Figure 3 | Quantitative analysis of ICP desalination. a,b, Conductivity of desalted stream of experiments with a seawater sample (a) and 100 mM phosphate buffer solution (b) as a function of applied electric eld. In both cases, the conductivity of the desalted stream dropped to the level of a few mM once the electric eld value reached a threshold. This also coincides with the establishment of ICP zones, observed by the uorescent trackers.

Bri

ne

Sea water V+ pressure Desalted water

Nanojunction GND

b
250 m
are Cs WB wing o

nd ou Gr trode n n c o ele Na nctio ju no na

Ion-depletion zone

n n o Na nctio oju an nd ou Gr trode c ele

No WBC ow

Figure 4 | ICP desalination using a metal buffer channel device. a, Schematic of a metal buffer channel desalination system. b, Demonstration of the system with uorescent stained WBCs with a ow rate of 10 ml min21 of 100 mM phosphate buffer solution at an electric eld of 60 V cm21. Dimensions of the microchannel: 1 mm (width) 15 mm (depth).

(desalted and salted). Figure 4b demonstrates a successful desalting operation, with stained WBCs, at an applied voltage of 30 V (60 V cm21; see Supplementary Video). Because the critical salt removal step occurs within a relatively short distance into the microchannel, the area required for a unit microuidic device is estimated to be $4 mm 5 mm (see Supplementary Information.) Massive parallelization of the unit device over a $68-inch wafer ($18,30032,400 mm2, allowing multiplication by a factor of $1,0001,600) would allow a throughput of $180288 ml min21 in a small-scale system, which is ideally suited to portable seawater desalination applications. In comparison, gravity-fed common household water purication systems have a throughput of $200 ml min21, and one of the commercially available desktop (but not portable) seawater desalination systems using RO
300

technology has a ow rate of between $262 ml min21 and 11.8 l min21 with a low energy efciency of $3595 Wh l21. Although massive parallelization of the unit device does require signicant engineering efforts, such levels of parallelization of small unit devices over a large area are not unprecedented, one example being found in the eld of photovoltaic power generation.

Conclusions
We have demonstrated a new, efcient and fouling-free desalination process based on the ICP phenomenon, for direct desalination of sea water. This process has several unique and attractive features for such applications. It has a power efciency that more or less matches that of current state-of-the-art RO plants. In a single-step operation, $99% of the salt contained in the sea water is removed,

NATURE NANOTECHNOLOGY | VOL 5 | APRIL 2010 | www.nature.com/naturenanotechnology 2010 Macmillan Publishers Limited. All rights reserved.

NATURE NANOTECHNOLOGY

DOI: 10.1038/NNANO.2010.34

ARTICLES

with 50% of the incoming sea water being recouped as desalted water (with a salinity less than 180 mg l21). In addition, it can eliminate any charged species, ranging in size from small salt ions to large particles or cells, without membrane fouling and clogging. This can signicantly reduce the complexity and cost of direct seawater desalination. Unlike ED or RO technologies, the particles and salts are driven away from the critical nanojunction and continuously re-routed into a different stream, thereby inherently preventing clogging and fouling of the nanojunction. This will allow robust long-term operation without the need for cleaning or exchanging the membranes. However, the present mechanism cannot be applied to the removal of neutral organic compounds. So, to obtain truly drinkable water, conventional methods for removing neutral compounds, such as charcoal absorption, should be used in combination with this method. The technology presented would not compete directly with large-scale RO or ED plants for general seawater desalination (for agricultural and industrial purposes), at least in the immediate future. However, the desalination of sea water or brackish water is straightforward using this system because it does not require high-pressure pumping/recirculation systems, and the technique can be realized in a small portable unit with low power consumption, with the possibility of batterypowered operation in the future. It is therefore suitable for seawater desalination in disaster- and poverty-stricken areas where the necessary infrastructures for operating large-scale desalination systems are unavailable. Given that the basic water requirements for human survival (through drinking and ingesting food) amount to only 2 l per person per day (ref. 17), the ow rate achieved in the projected portable system could address the water challenge in such areas. Ultimately, we anticipate that this technology will become a high-efciency alternative to ED or RO, if large-scale operation and efciency of scale can be realized.

voltage divider. As shown in Fig. 1a, platinum wires (Sigma Aldrich) were placed into each reservoir to enable proper external electrical connections. The potential drop (DV) between the microelectrodes was measured with Keithley 6514 electrometers, and current (i) through the desalted stream was determined using Keithley 6487 picoammeters. The conductivity (s) could be calculated simply using the relation s i/jEjA, where i is the current passing through the microelectrodes and A the cross-sectional area of the microchannel. E can be estimated from the potential drop divided by the gap between the microelectrodes. For calibration, we measured the conductivity of sea water of different dilutions using a benchtop conductivity meter (VWR sympHony), giving values of 45 mS cm21, 9.91 mS cm21, 1,292 mS cm21, 127.8 mS cm21, 12.4 mS cm21 and 1.8 mS cm21 for 1, 5, 50, 500, 5,000 and 50,000 dilutions, respectively. This number was used to convert the measured conductivity to the salinity of the desalted stream.

Received 8 December 2009; accepted 5 February 2010; published online 21 March 2010

References
1. United Nation Population Fund Population and Sustainable DevelopmentFive Years After Rio 136 (UNFPA, 1997). 2. Robert Engelman, L. P. Sustaining Water: Population and the Future of Renewable Water Supplies 747 (Population Action International, 1993). 3. Shannon, M. A. et al. Science and technology for water purication in the coming decades. Nature 452, 301310 (2008). 4. Greenlee, L. F., Lawler, D. F., Freeman, B. D., Marrot, B. & Moulin, P. Reverse osmosis desalination: water sources, technology and todays challenges. Water Res. 43, 23172348 (2009). 5. Davis, T. A. in Handbook of Industrial Membrane Technology (ed. Porter, M. C.) 482510 (Noyes Publications, 1970). 6. Lindstrand, V., Sundstrom, G. & Jonsson, A.-S. Fouling of electrodialysis membranes by organic substances. Desalination 128, 91102 (2000). 7. Kessler, J. O. & Moody, C. D. Drinking water from sea water by forward osmosis. Desalination 18, 297306 (1976). 8. Probstein, R. F. Physicochemical Hydrodynamics: An Introduction (Wiley-Interscience, 1994). 9. Rubinstein, I. & Shtilman, L. Voltage against current curves of cation exchange membranes. J. Chem. Soc. Faraday Trans. II 75, 231246 (1979). 10. Holtzel, A. & Tallarek, U. Ionic conductance of nanopores in microscale analysis systems: where microuidics meets nanouidics. J. Separation Sci. 30, 13981419 (2007). 11. Pu, Q., Yun, J., Temkin, H. & Liu, S. Ion-enrichment and ion-depletion effect of nanochannel structures. Nano Lett. 4, 10991103 (2004). 12. Kim, S. J., Wang, Y.-C., Lee, J. H., Jang, H. & Han, J. Concentration polarization and nonlinear electrokinetic ow near nanouidic channel. Phys. Rev. Lett. 99, 044501 (2007). 13. Kim, S. J. & Han, J. Self-sealed vertical polymeric nanoporous junctions for highthroughput nanouidic applications. Anal. Chem. 80, 35073511 (2008). 14. Kapp, E. M. The precipitation of calcium and magnesium from sea water by sodium hydroxide. Biol. Bull. 55, 453458 (1928). 15. World Health Organization. Guidelines for Drinking-Water Quality (WHO, 2008). 16. Veerapaneni, S., Long, B., Freeman, S. & Bond, R. Reducing energy consumption for seawater desalination. J. Am. Water Work Assoc. 99, 95106 (2007). 17. World Health Organization International Standards for Drinking Water (WHO, 1971). 18. Lee, J. H., Song, Y.-A. & Han, J. Multiplexed proteomic sample preconcentration device using surface-patterned ion-selective membrane. Lab Chip 8, 596601 (2008).

Methods
The systems under study are composed of two parallel microchannels connected by nanochannels (or a nanoporous membrane) as shown in Fig. 1a. Compared with classical membrane geometry, in which the membrane blocks a straight channel, uid ow in our design is not blocked by the membrane but rather ows in directions tangential to it. We fabricated polydimethylsiloxane (PDMS) microuidic chips with perm-selective nanojunctions using previously published methods13. The polymeric nanojunction was created by inltrating Naon polymer solution (Sigma Aldrich; 5 wt%) between the gaps created by mechanical cutting across the microchannels on a PDMS. The PDMS can seal itself to the heterogeneous polymeric nanoporous material in the gap. The PDMS substrate was then bonded with a glass plate by means of plasma treatment. The inlet microchannel and buffer microchannel had dimensions of 500 mm (width) 100 mm (depth); the bifurcated microchannels had a width of 250 mm and depth of 100 mm. For visualization of micrometre-sized particles and WBCs, devices with dimensions 100 mm (width) 15 mm (depth) were also fabricated. Gold microelectrodes, with titanium as an adhesion layer (100 mm wide, 110 nm high, electrode spacing of 100 mm), were deposited at the inlet, desalted and salted microchannels for electric potential measurement, using a standard evaporation/lift-off process (titanium, 10 nm; gold, 100 nm). For the metal buffer channel system, the Naon polymer was coated on top of the microelectrode using a microchannel ow method18. Natural sea water was obtained from Crane Beach, Massachusetts (pH % 8.48.5), and sodium hydroxide at a concentration of 100 mM was mixed (sea water:NaOH, 100:1) to pre-remove Ca(OH)2 precipitation. Although it is possible to directly carry out ICP with natural sea water, there is the possibility that Ca2 will precipitate, causing instability. The mixture was then ltered to remove the precipitation and any large particles such as seaweeds and sands. FITC (1 mg ml21; Invitrogen) and r-PE (1 ng ml21; Invitrogen) were intentionally added to enable uorescent tracking together with 2% v/v of 250-nm polystyrene particles (1,000 dilution; BangsLab) and human whole blood with uorescently stained WBCs (Hoechst; 2,000 dilution, Innovative Research) to allow micrometre-size particles to be traced. The pH of the sample increased slightly to $9.19.2 owing to the addition of NaOH. The external ow rate was generated with a syringe pump (Harvard apparatus, PHD 2200). All ow patterns and particle motions were imaged with an inverted uorescence microscope (Olympus, IX-51) and a CCD (charge-coupled device) camera (SensiCam, Cooke Corp.). Sequences of images were analysed by Image Pro Plus 5.0 (Media Cybernetics). A d.c. power supply (Stanford Research System) was used to apply an electric potential to each reservoir by means of a custom-made

Acknowledgements
This work was mainly supported by the National Science Foundation (CBET-0854026) and Innovation grant from SMART Innovation Centre. S.H.K. and K.H.K. were supported by the Korea Research Foundation (nos. R0A-2007-000-20098-0 and KRF-2006-331D00058). The MIT Microsystems Technology Laboratories are acknowledged for support in fabrication.

Author contributions
S.J.K. and J.H. conceived the idea and designed the experiment. S.J.K. carried out the desalting experiments and analysed the data. S.H.K. fabricated the devices and built the conductivity measurement system. K.H.K. and J.H. supervised the study. The manuscript was written by S.J.K. and J.H.

Additional information
The authors declare no competing nancial interests. Supplementary information accompanies this paper at www.nature.com/naturenanotechnology. Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/. Correspondence and requests for materials should be addressed to J.H.

NATURE NANOTECHNOLOGY | VOL 5 | APRIL 2010 | www.nature.com/naturenanotechnology 2010 Macmillan Publishers Limited. All rights reserved.

301