Microchannels with Self-Pumping Walls.

When asymmetric Janus micromotors are immobilized on a surface, they act as chemically powered micropumps, turning chemical energy from the fluid into a bulk flow. However, such pumps have previously produced only localized recirculating flows, which cannot be used to pump fluid in one direction. Here, we demonstrate that an array of three-dimensional, photochemically active Au/TiO2 Janus pillars can pump water. Upon UV illumination, a water-splitting reaction rapidly creates a directional bulk flow above the active surface. By lining a 2D microchannel with such active surfaces, various flow profiles are created within the channels. Analytical and numerical models of a channel with active surfaces predict flow profiles that agree very well with the experimental results. The light-driven active surfaces provide a way to wirelessly pump fluids at small scales and could be used for real-time, localized flow control in complex microfluidic networks.

One of the current barriers to developing completely standalone micro- and nanofluidic devices is that such devices still require external pressure sources to drive fluid through the small channels. Although in certain cases it is possible to exploit simple physical principles to achieve a functional device, such as capillary imbibition used in modern glucose meters, volumetric pumps remain essential for the development of any continuous-flow device.[1,2] The most widespread pumping methods are either mechanical[3] (which require fluid lines, pressure regulators, and syringes) or electro-/magneto-kinetic[4−8] (which require wires, batteries, and field generators). Because of the pumping hardware, such devices are often constrained to benchtop use, and a relatively small number of fluid paths can be controlled in a flow network at one time. In more complex flow networks, a global driving pressure thus affects the entire network and cannot easily be used to achieve precise temporal or spatial control of the flow in specific sections.

Self-powered active surfaces have recently grown in popularity as a technique for localized, microscale fluid pumping. The principles governing many active surfaces have been borrowed from self-phoretic chemical micromotors, which harvest chemical energy to propel themselves through a fluid.[9−14] By fixing such motors onto a wall, the same chemical activity is expected to lead to hydrodynamic pumping.[15] A variety of recent studies have shown that a Janus (2-patch) architecture can chemically drive two- and three-dimensional flows radially converging (or diverging) toward (from) the patch. Examples of these pumping systems include gold patches on silver[16] or platinum[17] substrates, photocatalytic platinum on silicon,[18] enzyme patches on glass,[19−21] and ion-exchange resins on glass.[22] However, to date, the pumping action in such systems remains limited to recirculating or radial flows toward (or away from) the active patch. A possible solution to produce unidirectional pumping is to introduce an asymmetry to the Janus patches, similar to what has been reported for electro-osmotic micropumps.[7,8] Although it has been theoretically argued that it should be possible to drive unidirectional flow in an active channel by properly selecting and arranging the active patches on the wall,[23−25] an experimental demonstration of such self-pumping active channels has remained elusive.

In this article, we show that wall-driven unidirectional pumping can be achieved in a microchannel by decorating the walls with photocatalytically active Janus (TiO2–Au) micropillars. The pillars promote water splitting when irradiated with UV light, leading to local osmotic flows around the pillars. The ensemble of these localized flows generate an effective wall-slip that drives a bulk flow in the channel. The process is depicted schematically in Figure . We demonstrate that pumping is possible with a three-dimensional (3D) pillar geometry, but not with a two-dimensional (2D) disk one, and that the pumping performance depends significantly on the pillar spacing. The bulk fluid responds instantaneously when illuminated, demonstrating our pump’s ability to rapidly control fluid motion within the channel. We show that a simple fluid dynamic model of an active channel, in which the active surface provides a wall-slip velocity, can accurately describe the experimentally measured flow profiles within the channel. By combining active top and bottom walls with different orientations of the pillar arrays, adjustable linear and parabolic flow profiles can be realized within the channel. The results suggest that this architecture would scale down well to smaller scales, where it could enable finely tuned control over complex micro- and nanofluidic networks.

Figure 1

Left: A Janus pillar illuminated with UV light catalyzes a water splitting reaction, which gives rise to osmotic flow (blue arrows) around the pillar. Center: Within an ensemble of oriented pillars, a cooperative effect of these local osmotic flows leads to an unidirectional macroscopic flow (red arrow) along the alignment (TiO2 → Au direction) axis of the pillars. Right: Within a channel, the active surface provides an active traction and drives bulk flow.

Results and Discussion

Fabrication and Characterization of Photochemically Active Patterned Surfaces

The self-pumping surfaces are fabricated on glass substrates using photolithography and physical vapor deposition. The full fabrication procedure is described in detail in the Methods and Experimental Section below. In short, photoresist microholes are patterned on the glass substrate with a tunable diameter and spacing, as shown in Figure . The micropillars are then successively deposited with Au, TiO2, and SiO2 using the physical shadow growth method GLAD.[26] By controlling the substrate orientation during the deposition, the coating materials can be selectively deposited onto different parts of the pillars, creating the Janus structure. Finally, after removing the photoresist and annealing the sample, a regular array of TiO2–Au Janus pillars are left on the substrate.

Figure 2

Fabrication of active surfaces by shadow deposition onto photolithographically patterned substrates. (a) Glancing angle deposition of Au, followed by deposition under normal incidence of (b) TiO2 and (c) SiO2. (d) Lift-off of the photoresist results in the TiO2–Au micropillars as shown in the SEM image in e.

When immersed in water and exposed to UV light, the micropillars catalyze a water splitting reaction.[27] Electron–hole pairs are created in the semiconductor TiO2, which react with the ambient water to produce O2 and H+ dissolved in H2O. Electrons are transferred to the metal (Au) side, where they recombine with H+ to produce H2 dissolved in H2O, reflecting an electrokinetic flux of H+ through the solution from the TiO2 side to the Au side. Simultaneously, gradients in the chemical composition of the solution (schematically depicted in Figure ) develop in the vicinity of each pillar because of the release of O2 and H2 at opposite sides of the Janus structure. These gradients induce diffusio-osmotic flows around the pillars. The superposition of the two effects causes a net ”osmotic slip” along the lateral surface of the pillars, as schematically indicated by the blue arrows in Figure . The direction of the osmotic slip is empirically determined to be from the TiO2 toward the Au side of the pillars (see experimental results below).

The pillars used in this study have a height of 1.5 μm and a diameter of 2 μm (see Figure e). To test the effect of the pillar spacing on pumping performance, we fabricated grids of pillars with spacings between 2 and 18 μm. To test the role of the pillar geometry, we also measure the pumping performance of a grid of 2D (100 nm tall) Janus microdisks and a grid of 3D Janus bars. The fabrication of these additional geometries is described in the Methods and Experimental Section below. To measure the pumping performance within microchannels, we constructed channels with heights of 110 and 170 μm by placing the active or inactive active surface over another active surface with appropriately sized spacers between them.

In all of the experiments below, the pumping velocity is measured by seeding the fluid with 1 μm polystyrene tracer particles, and tracking the motion of the individual particles through an optical microscope. Using image processing, we track the 3D position of particles above the active surfaces and in the channels (see Methods and Experimental Section and Supporting Information Note 2 for more details). The average flow velocity along the channel is then calculated from the x-position of the particles as a function of time. For tracer particles very close to the wall (within 5 μm), the particle motion wobbles because of hydrodynamic interactions with the pillars (e.g., see Figure ). Therefore, particle velocities measured near the wall underestimate the true pumping velocity.

Figure 3

Pumping speed as a function of the light intensity and the spacing between micropillars. The surface with micropillars is covered by a 300 μm thick water film containing tracer particles. The tracer particles are imaged in a plane 1.5 μm above the micropillars. (a) Tracer particles undergo Brownian motion when the UV light is off, as opposed to (b) when the illumination is on and the tracer particles reveal directional flow along the channel. The small oscillations in the particle trajectories are caused by hydrodynamic interactions with the pillars and lead to a slight underestimate of the true slip velocity at the wall. The scale bar indicates 5 μm. The white dashed arrow at the left bottom indicates the flow direction. (c) The pumping speed increases linearly with the UV light intensity (photocatalytic activity). The dotted line is a linear fit to the data. (d) Pumping speed is also seen to depend on the spacing s between the micropillars. The maximum flow speed is observed for a spacing of approximately 2 μm.

Active Wall Pumping Performance and Phenomenological Model

The pumping ability of our device is first demonstrated for a single active surface with pillars spaced by s = 2 μm. The surface is covered by a 300 μm thick water film containing tracer particles. Without illumination, there is no flow within the channel and the tracer particles undergo Brownian motion (Figure a). When the UV light (wavelength 365 nm) is turned on, the fluid above the surface responds immediately, flowing in a single direction that corresponds to an osmotic slip from the TiO2 to the Au side of the pillars (Figure b).

Measurements of tracer particles at a height of h = 1.5 μm above the pillars reveal the osmotic slip velocity at the active surface, as shown in Figure c. The slip velocity increases linearly with the optical irradiance I, up to 3 μm/s for I = 320 mW/cm2.[28−31] Writing the flow speed at the wall as uw = αI, the proportionality constant in our experiments is α = 0.009 μm s–1/(W cm–2). This linear dependence of u on I agrees well with the expectation that the flow is driven by photocatalytically established osmotic gradients. In a first approximation, the osmotic flow velocity should be directionally proportional to the pillar’s reaction rate, which itself is linear in the light intensity for photocatalytic reactions (below saturation).

The pumping model based on the assumption of a cooperative effect between the osmotic flows around the pillars suggests a strong dependence of the pumping performance on the pillar spacing. On the one hand, if two pillars are too close to each other, the Au face from one pillar is more exposed to the inhomogeneities produced by the reaction at the TiO2 face on the neighboring pillar rather than to the ones from the other side of its own pillar. This cross-talk between the reactions at neighboring pillars leads to reduced inhomogeneties in the chemical composition around each pillar and thus to reduced, or completely stopped, pumping within the channel. On the other hand, if pillars are spaced too far apart, the energy density of the pumping surface will decrease, resulting in low pumping velocities. Therefore, one expects that an optimal pillar spacing exists, at which the slip velocity is maximal. Intuitively, the optimal pillar spacing should be on the order of the pillar diameter, so that the source-sink pairs on a single pillar are closer to each other than they are with their neighbors.

To test this expectation, we measured the wall-slip velocity for active surfaces with pillar spacings from 2 to 18 μm. As shown in Figure d, indeed an optimal spacing seems to exist: the wall-slip velocity is maximal for a pillar separation s ≈ 2 μm, which is comparable to the pillar diameter. The velocity drops rapidly for a spacing above 2 μm.

In addition to the pillar spacing, the 3D cylindrical geometry is a critical component of the pumping performance. As described in Supporting Information Note 3, we repeated the pumping experiments with flat Janus disks and horizontal Janus rods. In the case of the disks, no slip velocity or pumping behavior was observed. In the case of the rods, the active surface could pump fluid in the channel, but with a much smaller velocity than the cylinders. These results are summarized in Figure and Supporting Information Figure S3. It remains unclear what specific mechanisms causes this strong geometric dependence on the pumping performance, but one possibility is that the flow around the 3D structures leads to more stable pumping, whereas the chemical gradients over 2D patches or stripes can more easily counteract each other and lead to lower bulk flow rates.[27] Although such effects are still the subject of ongoing research, our measurements are consistent with reports of Janus microswimmers, whose swimming velocity has been observed to depend strongly on particle shape and the positioning of the active elements.[32] In the case of fixed Janus structures, we observe that vertical Janus cylinders can drive a bulk flow more effectively than flat disks or horizontal cylinders.

Figure 4

Pumping is seen with 3D pillars but not with 2D disks. (a) Schematic of the two-dimensional Janus microdisks (top) and corresponding SEM image at bottom (scale bar = 1.5 μm). (b) Schematic (top) and SEM image (bottom, scale bar = 1.5 μm) of the three-dimensional Janus micropillar array. (c) Flow rate measurements above the two different surfaces reveal no pumping and pumping for the geometries shown, respectively, in panels a and b. Pumping speeds were measured using tracer particles at a height of 1.5 μm above the disks and pillars, respectively.

Controlling the Flow Profile in an Active Channel

When the active surfaces are embedded within a microchannel, not only is unidirectional pumping possible, but the flow velocity profile can be tuned by adjusting the slip velocity on the channel walls. Here we focus on three profiles that can be realized with a 2D wall-driven flow: a symmetric parabolic profile, a linear profile, and a skewed parabolic profile. As shown in Figure , the parabolic profile is realized by two active surfaces pumping in the same direction (symmetric), the linear profile by two active surfaces pumping in opposite directions (antisymmetric), and the skewed parabolic profile by one active surface and an inert, no-slip wall (skew).

Figure 5

Flow profile in a 2D microfluidic channel can be engineered with self-pumping walls. Different flow profiles are achieved with (a) a single active surface, (b) two symmetric active surfaces, and (c) two antisymmetric active surfaces. We note that because of experimental variability between identically fabricated samples, the channels with two active surfaces do not always have identical slip velocities (see, e.g., panel c). Blue arrows in the upper schematics indicate the pumping direction of the top and bottom surfaces, respectively. In all three cases, experimental measurements agree with results of analytical and numerical modeling.

We fabricated these three channel designs by fixing an active surface at the bottom of the channel and placing the second required surface, properly aligned, on top of a spacer to enclose the channel with a defined height. The observed flow profiles as a a function of height within the channel are plotted in Figure for each type of channel. The three different channel designs lead to qualitatively different flow profiles in the channel, making it clear that the flow profile can be engineered by simply adjusting the relative slip velocities of the two walls.

The flow in the channel is accurately described by a planar Stokes flow. In the experiments, each channel is bordered by an inactive inlet and outlet region, which affects the flow within the channel. Therefore, we model the flow in the channel as a flow through three sequential regions as shown in Supporting Information Figure S4. The complete channel consists of an inert inlet of length L1, an active central region of length L0, and an inert outlet of length L1. The channel is submerged in a reservoir of water so that the end-boundary conditions are zero-pressure at the inlet and outlet. By solving the Stokes equations in each section, and then relating the solutions by enforcing mass conservation between the sections, we can describe the fully developed flow profile in each section as a function of only the wall slip velocity uw, the channel height h and the lengths of the channel sections (see Supporting Information Note 4 for full derivation):Substituting experimentally measured slip velocities uw, along with the channel dimensions L0, L1, and h, we find that the analytical model agrees well with the measured flow profiles presented in Figure (see also the details provided in Supporting Information Note 6). Note that because measurements right near the walls are expected to underestimate the true wall velocities (see discussion above), we used the measurement at the next observable height as estimates for the corresponding uw.

To further validate the model, we numerically calculated the expected flow profile in the channels using COMSOL Multiphysics (see Supporting Information Notes 5 and 6 for details). The flow profile from the center of the active channel section is plotted in Figure . It agrees well with the analytical model and the data, but one notes that both the analytical model and the numerical results slightly underestimate the flow velocity in the skew and symmetric channel configurations, particularly where the flow velocity approaches u = 0. The overall shape of the flow are nevertheless predicted well by the model, which provides an analytic description that can be used to tailor the flow in custom channels.

A noteworthy implication of the model is that wall-driven flows can contain both positive and negative flow velocities, with planes of constant z containing zero flow velocity u = 0. This unusual feature arises because the active walls generate a positive pressure difference ΔP, which works against the forward flow within the active channels, leading to a back-flow along the channel centerline. Note that, despite the region of negative velocity, the overall volume flow rate is still positive. This behavior is only possible when there are inert channels on either side of the active section, corresponding to a flow resistance that the pump is working against. A symmetric channel with very long inactive sections (L1 ≫ L0, corresponding to a high flow resistance), will experience a strong back-flow and will pump vanishing amounts of fluid forward as L1/L0 → ∞. In this case, localized vortices will arise within the active section, and the centerline velocity will approach a minimum value u = −uw/2. By contrast, a completely active channel (L1/L0 → 0) would simply carry the fluid forward in a plug flow. Such a variety of flow profiles are not possible with traditional pressure-driven pumps, which further highlights some of the complementary capabilities enabled by active-wall micropumps.

Scalability

The analytical model also lets us quantify the active wall pumping performance and evaluate how well the technique would scale to smaller channels. For example, the flow model directly describes the pressure boost provided in the channel by the active-wall pumping in terms of the channel geometry and wall velocity (see Supporting Information Notes 4). In the symmetric configuration, which would generate the highest pressure boostFor the experiments presented in Figure b, this corresponds to a pumping head of 99 μPa. Equivalently, the active channel section generates a pressure gradient of 1 × 10–6 Pa/μm and a flow rate of 4 pl/s.

The model also allows us to analyze how well the flow performance would scale down to smaller channels. The flow rate produced by a slip velocity of uw is Q̇ = uwhL0/L, where L = L0 + 2L1 is the total channel length. Rearranging, we can write that such a flow rate requires a slip velocity of . Given that the slip velocity is linearly proportional to the UV irradiance (see Figure c), uw = αI, the total power (per unit depth of the channel, keeping with our 2D modeling) needed to maintain a flow rate Q̇ with active walls can be expressed asTherefore, for a fixed light intensity, the flow rate produced by an active wall scales as Q̇ ∼ h, and the power required increases for smaller channels as Wa ∼ Q̇h–1.

Creating smaller active channels with this architecture will likely be limited by two key factors: resolution limitations in the fabrication process, and chemical activity limitations that suppress the flow creation for very small Janus structures. The fabrication of Janus pillar-lined walls should be possible down to 30 nm, given that 30 nm Janus microswimmers have been made with the same fabrication technique.[33] However, at this scale, the active particles do not display coherent swimming, but rather enhanced diffusion. Therefore, the dominant limitation to scale the pumping surfaces down will likely be the size limits for the active structures to generate a coherent flow. Janus particles can generally self-propel down to sizes on the order of 100 nm, and we expect this to also be the lower limit for coherent pumping with Janus pillars. Finally, we expect that two active surfaces should be separated by at least one pillar height in order to provide a channel for bulk flow without chemical cross-talk between the surfaces. Thus, the active channels we describe should be scalable to channel heights of about 300 nm.

Conclusions

In this study, we have demonstrated that surfaces decorated with photocatalytically active TiO2–Au micropillar arrays can actively pump water along a microchannel. The Janus geometry of the micropillars defines the direction of the flow, which is induced by local self-osmotic flows via photocatalytic water splitting at the pillars. Stable, directional flows are obtained with flow speeds up to 4 μm/s, which is comparable to the fastest existing chemical pumps involving water.[34] The shape, orientation, and spacing of the pillars determine the flow speed and direction, and the 3D structure is critical to achieving a bulk flow. By lining a channel with our active walls, fluid can be pumped without physical pressure connections or wiring. Using a 240 μm long active channel, we create a pressure gradient of 1 μPa/μm that drives a volume flow rate of 4 pl/s through the 4.24 mm long microchannel. Unlike pressure-driven flows in static channels, chemically active self-phoretic walls offer the possibility to shape the flow profile in the microchannel. Such local flow shaping can have important implications for flow control, unjamming of suspension in microchannels, and analyte transport in chromatography. Finally, we demonstrated that the pumping performance of our active surfaces should scale down to channel sizes of a few hundred nanometers with pumping power scaling as W ∼ Q̇h–1. By using such optically active channels in micro- and nanofluidic systems, we envision that complex flow networks can be precisely controlled in space and time, allowing for noncontact, real-time rerouting of flows, and tuning of the flow profiles.

Methods and Experimental Section

Fabrication of the Micropillar Array

The Janus micropillar array was fabricated on a standard cover glass using photolithography and physical vapor (glancing angle) deposition.[26] First an array of 2 μm diameter holes was patterned in 2.6 μm thick maP1215 photoresist (micro resist technology GmbH). After transfer to a vacuum chamber, 80 nm of Au were deposited at an angle of 18° by e-beam evaporation. This was followed by 100 nm TiO2 and then 1400 nm SiO2 at normal incidence (0°). After lift-off in acetone and air plasma cleaning, the sample was annealed for 2 h at 450 °C in air. SEM and EDX images of the substrate were acquired using a Zeiss Ultra 55, and are shown in Figure S1.

Fabrication of the Microdisk Array

The flat microdisk arrays were similarly fabricated. However, the evaporation sequence differed: a 100 nm TiO2 layer was first evaporated at 0°, which was followed by 20 nm Au deposited at 32°.

Channel Fabrication and Fluid Flow Measurement

A 10 mm × 10 mm × 0.3 mm chamber was formed to hold the microchannels. Two thin parallel pieces of tape with thicknesses, respectively, of 110, 170, and 170 μm were positioned in the center of the bottom substrate to construct each of the three channels. The sample was covered with water containing 1 μm polystyrene particles that serve as flow tracers. A second surface either with or without a pillar array was placed on top of the channel spacer, and the entire system was covered in Milli-Q water. Finally, the sample was mounted on a Zeiss Axio inverted microscope and the flow was observed using a 63× objective. Videos were recorded at a rate of 20 frames per second with an Andor Zyla 5.5 camera. Particle positions were tracked using ImageJ (version 1.52i, NIH). The x–y positions of the particles were extracted from the particle centers and the z-position (height within the channel) was determined using the calibrated defocusing method described in Supporting Information Note 2. Each data point in the plots above represents the average x-velocity (along the flow) of at least 10 particles, and the error bars represent the standard deviation across these measurements.