SUPERCELLS: a novel microfluidic reactor architecture for ultra-fast sequential delivery of chemical reagents†

Naghmeh Fatemi ^a, Ahmed Taher ^a, Jelle Fondu ^a, Lei Zhang ^a, Tinne De Moor ^ab, Kherim Willems ^a, Olivier Henry ^ac, Peter Peumans ^a and Tim Stakenborg *^a
aImec, Life Sciences Technologies, Kapeldreef 75, 3001 Leuven, Belgium. E-mail: tim.stakenborg@imec.be
^bDepartment of Electrical Engineering (ESAT), KU Leuven, Kasteelpark Arenberg 10, 3001 Leuven, Belgium
^cDepartment of Soft Matter and Biophysics, KU Leuven, Celestijnenlaan 200d, 3001 Leuven, Belgium

First published on 8th August 2024

---

---

Introduction

Compared to traditional liquid handling, microfluidic devices typically operate in a laminar flow regime, as it allows for more predictable and controllable fluidic actions. Other benefits include low sample and reagent consumption, short time-to-result, and enhanced temperature control.^1,2 Despite their many benefits, limited standardization and remaining integration challenges limit the widespread use of microfluidic systems.³ Additionally, the need for many external components affects their reliability and results in traces of impurities. Even for very simple microfluidic channels with a single inlet and outlet (see Fig. 1a), the need for external valves and pumps results in relatively large dead volumes, while stopping the flow leads to reagent cross-contamination by diffusion. Several washing steps, often for the entire flow cell and system, are typically required to avoid reagent carry-over.⁴ Besides reagent purity concerns, external pumps and valves also limit the speed of reagent exchange. This is especially relevant for precise time-resolved measurements where a fast exchange of reagents is crucial to minimize diffusion.

To avoid the need for extensive washing steps and to enable rapid reagent exchange, different microfluidic solutions have been proposed. One solution for fast buffer exchange is the implementation of multi-stream microfluidic chips in which multiple inlets provide a parallel fluidic stream separated by laminar boundaries. At these boundaries, only limited local mixing takes place by diffusion and the position of the diffusion boundaries can be controlled by controlling the fluid pressure.⁵ Typically only simple designs – with two or few inlets converging into a single channel – have been used, especially focussing on single molecule applications.⁶ Another solution proposed is the use of multi-reagent cartridges pre-filled with a sequence of reagents.⁷ This naturally limits flexibility during operation and complicates assembly. Also, electrowetting on dielectric (EWOD) or digital fluidic chips to actuate droplets of liquid in a controllable order has been proposed to avoid the need for external valves,⁸ but requires an active array of electrodes. Alternatively, using on-chip microvalves or pumps (see Fig. 1b) has been described to reduce the need for many large, external components.⁹ Complex networks with even thousands of miniaturized valves and hundreds of individually addressable picolitre-scale reaction cavities have been fabricated in silicone devices.¹⁰

Here, we propose a novel microfluidic reaction cavity structure, named supercells, to repeatedly load and unload different reagents with a fast-switching time and without the need for valves integrated on the fluidic chip (see the concept in Fig. 1c and 2). We also show that the concept can be extended to an array of supercells, enabling large-scale parallelization. Using an array of small reaction cavities is beneficial, compared to one big reaction cavity, as smaller individual volumes will result in faster reagent switching per supercell and less overall reagent consumption during switching. This concept is particularly advantageous for applications that require repeated reagent exchange while avoiding cross-contamination, such as next-generation DNA sequencing or DNA synthesis.

Materials and methods

Design concept

As a proof-of-concept, a simple supercell geometry was designed enabling rapid switching between two reagents with the notion that the system can be scaled to a larger number of both reagents and supercells needed for most practical implementations.

For the concept to work and enable the rapid replacement of the reagents with minimal cross-contamination, as schematically illustrated in Fig. 2, two features were incorporated into the supercell design. Firstly, to avoid dead-zones within the main cavity, the traditional single outlet was replaced with a multitude of channels. As depicted in Fig. 2a, in case of only one outlet channel connected to the main cavity, the main flow path of the introduced reagent (shown in red) will result in dead-zones (shown in blue). This could be resolved by adding more outlet channels to the main cavity (see Fig. 2b). Not only the number but also the flow resistances of the different outlet channels can be tuned to ensure reagents to fill up the entire cavity with convection rather than diffusion. However, even with this added feature, if the flow rate of reagent A is zero (Q_A = 0), reagent B will diffuse back to the inlet channel of reagent A while if the flow rate of reagent A is not zero (Q_A > 0), the main cavity cannot be filled purely with reagent B. Therefore, as the second feature, side drains connected to each reagent inlet are implemented as diffusion barriers to prevent reagent mixing at their inlets or within the main cavity. In other words, the side drains near the inlets enable the diversion of the (temporarily) unwanted reagents (reagent A) to the outlet or waste, while at the same time ensuring reagent B can remain of the highest purity within the main cavity. As shown in Fig. 2c, this condition is realized by controlling the flow rates between the different reagent flows. More specifically, the flow rate of reagent A (Q_A) is very small compared to the flow rate of reagent B (Q_B). This ensures a low (non-zero) flow rate of reagent A (Q_A) being maintained to prevent back diffusion of reagent B from the main cavity contaminating the other reagent fluid stream without compromising the purity of B in the main cavity. Hence, the side drains act as diffusion barriers to prevent cross contamination from one reagent into the other inlet, with the condition that the mass flux of reagent B due to diffusion is equal to or less than the opposite mass flux due to advection.

The concept as such allows for a rapid switching between reagents by adapting their pressures and thus associated flow rates. More specifically, to introduce each reagent into its respective inlet: Q_A ≫ Q_B when introducing reagent A into the main cavity replacing reagent B, and Q_B ≫ Q_A for vice versa. The working principle can be extended to an array of supercells in a hierarchical branching structure by using supply channels to provide reagents to the inlets of all the supercells. The outlet of each supercell is connected to a common outlet which removes the reagent from the array. To create an array configuration, the hierarchical branching is realised using a 3D fluidic network (see Fig. 3).

Specific design implementation and numerical simulation

A supercell with 2 reagent inlets (reagent A – Q_A and reagent B – Q_B) was simulated and designed to demonstrate the working principle (see Fig. 3). Although different designs are naturally possible, we have chosen a square cavity (W × L × H: 150 μm × 150 μm × 5 μm; see Fig. 3c). The main cavity of the supercell was surrounded by an outlet busbar (width 20 μm) connecting multiple outlet channels (width ranging from 3 to 8 μm) distributed around the perimeter of the main cavity. The hydraulic resistances of the outlet channels were selected to equalize the advection times for all the corners of the main cavity, while the busbar ensures a much lower hydraulic resistance (larger size). As such, the design allows fluids to be effectively replaced in the main cavity without observable dead-zones where species transport occurs primarily through diffusion as explained above. For testing and to demonstrate scalability, a 10 × 10 array of supercells was designed as shown in Fig. 3a. To connect all supercells, a hierarchical multilevel fluidic network was constructed. All fluidic inlets of the supercells in the array are connected to common supply channels (width 100 μm, height 35 μm) and further connected to common inlets (width 200 μm, height 65 μm) giving fluidic access to the reagents. Similarly, the outlet of each supercell in the 10 × 10 array is connected to one main common outlet allowing for only one outlet for the whole array that needs to be connected externally via external tubing to an outlet drain (see Fig. S3a†). The use of larger dimensions for the supply channels and the common inlet/outlet (hence, lower hydraulic resistance) compared to the busbar and outlet channels connected to the main cavity minimizes the pressure drop over the entire system and ensures each reagent to arrive at the inlet of each supercell with almost equal pressure.

For this specific supercell design, transient 3D numerical simulations were conducted using Ansys Fluent 18.0. The studied geometry domain is one supercell with a structured, hexahedral mesh and a uniform, finite volume cell size of 0.5 × 0.5 × 0.5 μm³ (x, y, z). The advection–diffusion/scalar transport equation was used to model the transport of the reagent species within the flow domain. Spatially-uniform, time-varying velocity boundary conditions were applied to inlets, uniform, time-invariant normalized reagent species concentrations of 0 and 1 were applied to the two inlets, respectively, and a uniform time step of 0.001 s was applied in the simulation. The full description on the simulation method is provided in the Fig. S1,† together with specific boundary conditions used for the results presented in this work.

Device fabrication

To prove the principle, an array of 10 × 10 array supercells was designed and fabricated (see Fig. 3). Considering the large range of dimensions and aspect ratio requirements, three silicon wafers were sequentially patterned, etched and bonded with intermediate grinding and polishing steps to obtain the desired multilevel structure. As shown in Fig. 3d and S2,† wafer 1 (final thickness of 50 μm) contained the smallest fluidic vias (width 10 μm, height 50 μm) to connect the fluidic inlets/outlets of the supercells. The supercells were patterned and etched in a 5 μm thick SiO₂ layer. Sequentially bonded to wafer 1, wafer 2 (100 μm thick) and wafer 3 (150 μm thick) each contained one supply channel layer and one via layer (width 90 μm and 800 μm, respectively), both realized by a deep reactive ion etch (DRIE) Bosch process. The obtained stack was finally bonded to a quartz wafer to cap the supercell and allow optical access. All the wafers were bonded by oxide–oxide fusion bonding using infrared alignment with 2 μm overlay accuracy. Fusion bonding was also used for the bonding of the Si stack to the quartz wafer. The bonding of multiple wafers was made possible by exerting tight control over the bow of the stack by managing the internal stress levels of the intermediate SiO₂ layers and applying a special compensating adjustment during bonding. Further details of the fabricated structures are shown in the Fig. S2.†

Experimental set-up

To test the function of the fabricated supercell array, a bespoke fluidic set-up with optical inspection was built. Fluid actuation was performed using a pressure driven pump (4-channel MFCS-EZ, Fluigent) connected to fluid reservoirs (Fluiwell-4C units, Fluigent) to adjust the flow rates. The flow rates were monitored using a flow controller and flow sensor (MFCS Flow board, Fluigent). Each reagent reservoir was connected to a 3-port/2-way microfluidic switch valve (2-switch, Fluigent) using PTFE tubing (OD 1/16′′, ID 0.5 mm). The outlet of the full array was also connected to the pressure pump via PTFE tubing (OD 1/16′′, ID 1 mm). Switching between the reagents was performed by alternating the connections from the on-chip inlets to a low resistance line (for reagent B) and a high resistance line (for reagent A) using a switch board (ESS Switchboard, Fluigent) and a signal generator device. More specifically, tubing with an ID of 20 μm and a length of 30 cm was used for ensuring a higher resistance (R_H = 6.80e + 16 Pa s m⁻³), while tubing with an ID of 0.5 mm and a length of 30 cm was used for the lower resistance (R_L= 1.74e + 11 Pa s m⁻³) for each of the reagents. For an individual single supercell, this resulted in a ratio between the high and low upstream resistances of 3.91e + 05. As the hydraulic resistance of a single supercell was estimated to be 1.3e + 15 Pa s m⁻³, the estimated hydraulic resistance for the 10 × 10 array (100 supercells connected in parallel) was estimated at 1.3e + 13 Pa s m⁻³. For optical inspection, an inverted microscope (IX71, Olympus) connected to a high-speed camera (Os4, IDT Vision) was used. The schematic of the experimental set-up is depicted in Fig. S4.†

Measurement procedure

The 10 × 10 supercell array was initially cleaned from the outlet using acetone to remove any potential leftover residues from the fabrication process. To demonstrate the fluidic function and to enable optical visualization, deionized (DI) water and a solution of 25 μM fluorescein (pH = 9) were used. Before starting, special care was given to remove any air bubbles within the tubing and the chip itself by rinsing steps at a high flow rate (by applying pressures up to 5 bars). Switching between reagents was enabled with external valves and adjusted based on the flow rates using a signal generator. Images were captured at 20 fps using a fluorescence microscope. For measuring the switching times, an in-house generated Python code was used to determine the fluorescence intensity within the main cavity over time. The fluorescence intensity was recorded and normalized to the highest intensity. Hence, a recorded intensity near zero meant that the supercell was dark and the cavity was filled with DI water, while an intensity equal to 1 meant that the cavity was filled with the fluorescent solution. The switching time was defined as the time duration when the fluorescent solution (or the DI water) enters the main cavity till it was replaced by the pre-filled DI water (or the fluorescent solution) in the cavity. The switching times were recorded for individual single supercells of the array and determined for at least 4 cycles (i.e., switching from fluorescent solution to DI water and back). Switching between the reagents was examined for different flow rates ranging from 0.1 to 10 nL s⁻¹ per supercell. For all these cases, the flow rate of the reagent to be replaced (i.e. the reagent connected to the high resistance line) was kept close to zero. For data analysis, the area of a supercell cavity was divided into 16 even regions, each with a size of approximately 37 × 37 μm, and the average fluorescence intensity for each region was measured.

Results

Optimization of the supercell design

To prove the supercell concept, a square shaped flow cell for switching between two reagents was conceptualized. To optimize the geometry of such a single supercell for fast switching while using a minimum amount of replacement volume, the liquid flow through the supercell was simulated numerically using 3D ANSYS Fluent software. Different design iterations and optimization steps were taken.

Initially, a fixed width of all the connecting channels, the outlet busbar, and the side drain channels was used. Such a design, however, led to an uneven flow through the connecting channels of the main cavity. This resulted in the newly introduced reagent to take a shortcut from the inlet to the outlet (see Fig. S1a†). To further improve the design, the widths of the connecting channels were tuned to achieve the same flow rate at each connecting channel. However, the closer the connecting channel to the inlet was, the faster the replacement of the reagent. In other words, the furthest connecting channel took a much longer time than the closest one to the inlet and increased the total switching time. Therefore, the target flow rate at each connecting channel was set to be proportional to its distance from the inlet. This was achieved by modulating the hydraulic resistance of each channel (see Fig. S1b†). Using this method, switching times were decreased to minimum using the same inlet flow rate (i.e. minimum replacement volume). The simulation results for the initial filling of the supercell from inlet 2 are shown in Fig. 4a. The as-obtained dimensions, shown in Fig. 3, were used for the final design and for the proof-of-concept experiments.

Simulations of experimental results

To match and be able to better understand the observed experimental results, the velocity boundary conditions of the 3D simulations were adjusted (see Fig. S1†). These simulation results together with the experimental results are depicted in Fig. 4b–e for different time points and a flow rate of 10 nL s⁻¹ of reagent B. The simulations show that the adjusted hydraulic resistances of the outlet channels result in reagent B filling up the main cavity in all the directions with almost equal speed, not taking any shortcuts. This also verifies that using a hydraulic resistance of the outlet busbar ∼5 times lower compared to that of the outlet channels is sufficient to reduce the resistance differences among different fluidic paths between outlet channels and the common outlet, hence improving the equal filling of the main cavity. The working principle of the “diffusion barrier” resulting from the side drains is shown as well. The inset in Fig. 4a shows the velocity vector field in the vicinity of the 2 inlets (inlet 1 and inlet 2) to the main cavity and the side drains. The arrows superimposed on the inset show the direction of the bulk-fluid flow through the channel network. If Q_B (inlet 2) is sufficiently higher than Q_A (inlet 1), then reagent B will be introduced into the main cavity (see arrows B1 and B3) while generating a backflow (see arrow A3) that flushes reagent A and later, a portion of reagent B out of the main cavity. Considering the non-zero flow of reagent A introduced from inlet 1 (see arrow A1), the replaced liquid from the main cavity will be flushed into a side drain (see arrow A2). Hence, the main cavity can be filled by reagent B with high purity, while reagent A at inlet 1 is not being contaminated.

Experimental results for fast switching

All experimental tests were performed using 10 × 10 supercell arrays. To demonstrate fast fluidic switching, most experiments were performed using a flow rate of 10 nL s⁻¹ for reagent B and a flow rate close to zero for reagent A (∼0.1 nL s⁻¹). The fluorescence intensity in individual supercells was recorded during liquid switching. Fig. 4b–e show the images for a single supercell at 4 time points in one experiment as well as the matching simulation data. The fluorescent solution from inlet 2 started to replace the water from the nearest corner of the main cavity (the bottom right corner in the image) and then filling upwards to the top corner on the right, the centre of the main cavity and lastly, the farthest corner from inlet 2. At the same time, the fluorescent solution also filled the outlet channels surrounding the main cavity as well as the outlet busbar, reaching the common outlet (waste) channel simultaneously from both sides. This is expected as the supercell was designed to achieve equal flows in all directions within the main cavity avoiding dead-zones (shown in Fig. 3a). The entire reagent exchange happened within 0.15 s which is the best experimental result obtained aligned with the simulation data. To gain more insight into the liquid replacing phenomena, the cross-sectional views of the reagent mass fractions at different locations in a supercell at different time points were obtained in the same numerical simulation (see Fig. S4 and S5†).

To characterize the switching time reproducibility within a 10 × 10 array, different supercells located at different positions were closely examined. Exact switching times were determined by measuring the average fluorescence intensity of 16 different regions of single supercells alternating from bright to dark or vice versa. Imaging 16 regions per supercell helped to analyse how fast each corner of the main cavity was flushed with the desired reagent. Fig. 5a depicts the average normalized fluorescence intensity for seven different supercells within a single array (for 5 repeat cycles) at a flow rate of 10 nL s⁻¹, where the average intensity was taken from the 16 regions per supercell. A switching time of 0.23 ± 0.02 s was measured at a flow rate of 10 nL s⁻¹. Experiments also showed that adding pressure to the outlet line (∼100 mbar) helped the better synchronization between the observed cells and more stable flow rates, likely due to avoiding any influence of small air pockets or bubbles. Besides the overall switching time, the synchronization was examined in more detail over different cycles. Fig. 5b and c show the propagation of the fluorescent solution in the 10 × 10 array of supercells at 10 nL s⁻¹ per supercell. Fig. 5b shows the initial filling of 48 supercells with the fluorescent solution and Fig. 5c shows these supercells fully filled after introducing the fluorescent solution entering from the inlets. These results confirm the good synchronization for all supercells within the array. Due to the limitation of the field of view in the experimental setup, only 48 cells could be observed at the same time; hence, the synchronization of 100 cells was checked at different time points. A video of the well synchronized array of 48 supercells is provided as an example (see video in the ESI†).

Besides synchronization of supercells within a single array, additional experiments were conducted on various arrays and at different flow rates (see Fig. 5d). The reported data were taken from at least 3 supercells per array from at least 3 different arrays at flow rates ranging from 0.1 to 10 nL s⁻¹ per supercell. The switching times ranged from 3.80 ± 0.24 s at a flow rate of 0.1 nL s⁻¹ to 0.23 ± 0.09 s at a flow rate of 10 nL s⁻¹. The switching times per array at a flow rate of 10 nL s⁻¹ are shown in the Fig. S5.† Notably, the normalized intensity upon switching went from 0 to 1 (or vice versa) for all imaged regions, indicating the high purity achieved for each reagent. The intensity also remained stable between switching, indicating that cross-contamination could be avoided as expected.

Discussion

A novel, passive microfluidic concept, called supercells, was proposed to enable a rapid and repeated fluidic reagent exchange. Using a hierarchical branching structure with supply lines, the liquid of the different fluidic reagents is brought to the entrance of every individual cavity or supercell. The supercell concept was here worked out for a specific design for the switching of 2 reagents. Naturally, designs can be made for many more reagents and inlet lines using an equal amount of Si layers. We also provided a design with the inlets on one side and outlets on the other side, but a more symmetric, e.g. radial, design could be considered. Furthermore, a design where the side channels are in a different layer than the main cavity would maximize the coverage of the supercell. Independent of the design choice, the close proximity of the reagents to the supercell ensures that every liquid only needs to travel a short distance to enter the cavity which brings several benefits. First, the concept allows – as demonstrated – fast switching times. For the current design, a fast switching in between two reagents was achieved, namely around 3.80 s at a flow rate of 0.1 nL s⁻¹ and 0.23 s at a flow rate of 10 nL s⁻¹ per supercell. As expected, increasing the reagent flow rate resulted in a shorter switching time as the reagents are replaced faster. Achieving these fast switching times to exchange reagents is nearly impossible with external valves and would typically take a few seconds or more due to the need to additionally replace the relatively large volume of liquid between the valves and the cavity. Second, the concept minimizes reagent use and ensures a minimal dead volume as no cleaning or rinsing of long supply lines is required. The design of the supercell eliminates dead volume by allowing the fluid to flow in all directions out of the central cavities, independent of the flow rate used. It must be noted, however, that increasing the flow rate to achieve faster switching times does result in more reagent consumption (switching volume) per switching cycle. For example, for the current design, the switching volume at a flow rate of 1 nL s⁻¹ was 0.38 nL whereas it increased to 2.35 nL at a flow rate of 10 nL s⁻¹. An alternative method to further reduce the volume is to further reduce the supercell height at the expense of a higher pressure drop and risk of clogging. Overall, the design and switching times must be tuned to the application needs and operational conditions. Third, the concept ensures high reagent purity. This is achieved by keeping a different, but non-zero flow for all reagents. As such, the fastest flowing liquid (achieved by applying a higher differential pressure) pushes out the former liquid from the cavity, while the other liquid keeps flowing at a low, but non-zero flow speed. This ensures that the unused reagents are diverted to a side drain (i.e. diffusion barrier). As mixing only happens in these side drain channels, the main supply lines are kept 100% pure. This also means that – as there are no integrated valves – stopping the flow would lead to diffusion and would result in reagent mixing. Depending on the application and if high switching speeds are not necessary, the flow rate can be maximally decreased to the diffusion limit; still, the inability to stop the flow (to ensure high reagent purity) may pose a limitation to the concept for certain applications. Fourth, the pressure drop can be limited by design due to the hierarchical branching structure with relatively large supply lines enabling almost unlimited scaling options. As shown, the wider supply lines also allow for very well synchronized propagation of the fluorescent solution in all the supercells for the tested 10 × 10 array. Other benefits include the ability to use a passive structure without the need for integrated active elements, the limited need for external instruments (in principle, only one pressure actuator per reagent independent of the number of supercells in an array) and the CMOS compatibility of the fabrication method ensuring a rather straightforward route for manufacturing.

While different applications could benefit from this concept, we envision its value particularly in applications such as DNA synthesis and sequencing. Traditional flow cells on the market for DNA sequencing and synthesis are slow to switch between different reagents, which reduces throughput, and require significant volumes of reagents, which increases the operational cost significantly. The proposed supercell can be beneficial due to improvements to both reagent cycling speed and the volume of the consumed reagents.

Nomenclature

Abbreviations

ID	Inside diameter
fps	Frames per second
OD	Outside diameter

Roman symbols

d	Diameter
D AB	Diffusion coefficient [m2 s−1]
L	Length [m]
m A	Mass fraction
R	Resistance
R H	High resistance [Pa s m−3]
R L	Low resistance [Pa s m−3]
t	Time [s]
v	Velocity [m s−1]

Greek symbols

ρ	Density [kg m−3]
μ	Viscosity [Pa s]

Data availability

The data supporting this article have been included within the article as well as part of the ESI.†

Author contributions

NF: methodology; investigation, formal analysis, visualization, and writing – original draft; AT: conceptualization, investigation, software, and writing – review & editing; JF: investigation and supervision; LZ: writing – review & editing; TDM: software; KW: supervision; OH: supervision – review & editing; PP: conceptualization; TS: conceptualization, resources, and writing – review & editing.

Conflicts of interest

The authors declare that there are no conflicts of interest.

Acknowledgements

The authors want to thank Young Jae Choe for her help with the images.

References

1. D. Mark, S. Haeberle, G. Roth, F. Von Stetten and R. Zengerle, Chem. Soc. Rev., 2010, 39, 1153–1182,  10.1039/b820557b .
2. E. K. Sackmann, A. L. Fulton and D. J. Beebe, Nature, 2014, 507, 181–189,  DOI:10.1038/nature13118 .
3. L. R. Volpatti and A. K. Yetisen, Trends Biotechnol., 2014, 32, 347–350,  DOI:10.1016/j.tibtech.2014.04.010 .
4. S. K. Pramanik and H. Suzuki, Microfluid. Nanofluid., 2019, 23, 1–8,  DOI:10.1007/s10404-019-2188-z .
5. J. Madariaga-Marcos, R. Corti, S. Hormeño and F. Moreno-Herrero, Sci. Rep., 2020, 10, 1–12,  DOI:10.1038/s41598-020-74523-w .
6. X. Tan, M. Mizuuchi and K. Mizuuchi, Proc. Natl. Acad. Sci. U. S. A., 2007, 104, 13925–13929,  DOI:10.1073/pnas.0706564104 .
7. C. D. Chin, T. Laksanasopin, Y. K. Cheung, D. Steinmiller, V. Linder, H. Parsa, J. Wang, H. Moore, R. Rouse, G. Umviligihozo, E. Karita, L. Mwambarangwe, S. L. Braunstein, J. Van De Wijgert, R. Sahabo, J. E. Justman, W. El-Sadr and S. K. Sia, Nat. Med., 2011, 17, 1015–1019,  DOI:10.1038/nm.2408 .
8. M. G. Pollack, R. B. Fair and A. D. Shenderov, Appl. Phys. Lett., 2000, 77, 1725–1726,  DOI:10.1063/1.1308534 .
9. K. W. Oh and C. H. Ahn, J. Micromech. Microeng., 2006, 16, R13,  DOI:10.1088/0960-1317/16/5/R01 .
10. T. Thorsen, S. J. Maerkl and S. R. Quake, Science, 2002, 298, 580–584,  DOI:10.1126/science.1076996 .

---

Footnote

† Electronic supplementary information (ESI) available: Includes more details about the simulations, fabrication, set-up and results as well as a movie of an array of supercells. See DOI: https://doi.org/10.1039/d4lc00534a

This journal is © The Royal Society of Chemistry 2024