Channel integrated optoelectronic tweezer chip for microfluidic particle manipulation

Light patterned electrical fields have been widely used for the manipulation of microparticles, from cells to microscopic electronic components. In this work, we explore a novel electromechanical phenomenon for particle focusing and sorting where the electrical field patterns are shaped by a combination of the light patterned photoconductor and the channel geometry. This effect results from the combination of particle polarisation described by the Clausius–Mossotti relation and the engineering of large electric gradients produced by choosing the channels height to suit the size of the particles being manipulated. The matched geometry increases the distortion of the field created by a combination of the illuminated photoconductor and the particles themselves and hence the non-uniformity of the field they experience. We demonstrate a new channel integration strategy which allows the creation of precisely defined channel structures in the OET device. By defining channels in photoresist sandwiched between upper and lower ITO coated glass substrates we produce robust channels of well controlled height tailored to the particle. Uniquely, the top substrate is attached before photolithographically defining the channels. We demonstrate versatile control using this effect with dynamically reconfigurable light patterns allowing the retention against flow, focusing and sorting of micro particles within the channels. Contrary to traditional designs, this channel integrated device allows patterned micro channels to be used in conjunction with conductive top and bottom electrodes producing optimal conditions for the dielectrophoretic manipulation as demonstrated by the rapid flow (up to 5 mm s−1) in which the particles can be focused.

S Supplementary material for this article is available online (Some figures may appear in colour only in the online journal) Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.

Introduction
Light patterned electrical fields are a rapidly developing noncontact alternative for the manipulation of microscopic objects to traditional approaches such as probes, micro needles or pipettes. Applications include the patterning, sorting and sequestration of cells into predestined pens in research and industrial settings [1,2]. The use of moving light patterns in optoelectronic tweezer (OET) devices [3,4] or virtual sieves [5] have been used for the manipulation and separation of different cell types, DNA [6] the purification of circulating tumour cells [7] as well as the electroporation [8], transfection [9] and lysis of cells [10][11][12]. OET has also proven to be useful for cell analysis such as determining a cell's physical properties [13] or discrimination between healthy and non-viable cells [5] and development stages of complex systems such as mouse embryos [14], identifying viable sperm cells [15] and healthy oocytes for fertilisation experiments [16]. As well as these varied life science applications OET has also been developed for use in a second area, namely the assembly of sub-millimetre electronic and electrical devices. From original work assembling nanoparticles [17] and nanowires (separating metal and semiconductor nanowires, lateral OET with PL nanowires) this has recently been extended to the manipulation of standard electrical components with dimensions in the hundreds of microns. Surface Mount Technology components such as capacitors [18], integrated circuit (IC) chips [19] and optoelectronic components from micro disk lasers [20] to standard InP laser die [21] have all been assembled with OET. The motivation for manipulating these components is to develop a printer like assembly system that could, instead of printing with ink, print electrical components onto a circuit [19]. Recently micro robots controlled by OET have been demonstrated that can manipulate mammalian cells and other micro objects [22]. For this to be achieved, or for the potential biological applications to be realised, there is a need to integrate OET into a microfluidic channel. This is hindered by the need for the top of the channel to be conductive to give the top electrode of the device and for the bottom surface of the channel to be photoconductive to give the pattern-able electrode. This cannot be achieved with standard PDMS soft lithographic fabrication techniques and has limited many studies to batch processing without the benefit of a flow channel.
The inherent need for microfluidics integrated with the OET device has prompted several groups to develop varied approaches including patterning SU-8 then epoxy gluing the top substrate to the channel walls [23][24][25][26], using PDMS modified with gold/titanium mesh or carbon nanotubes as the top electrode and channel structure [27] and defining the channel in a hydrogel placed on a spin coated organic photoconductive layer [24]. These approaches suffer from incomplete bonding [23], weak forces [27] or low durability and incompatibility with solvents [24].
In this work, we demonstrate the integration of precisely controlled channels into a robust microfluidic chip that allowed us to tune the electric properties of the OET chip. The process we developed uses negative photoresist (SU8 3000 series) to create microfluidic channels. As shown in previous studies, SU8 has good adhesion to glass, mechanical and chemical stability making it an excellent material for many microelectromechanical system applications. Moreover, lithography-defined SU8 structures and layers have shown to be biocompatible making it a suitable material for mammalian cell handling applications.

OET principle
The conventional optoelectronic tweezing device structure (figure 1) consists of transparent top and bottom ITO coated glass electrodes while one of the electrodes (bottom) is covered with a photoconductor such as amorphous silicon (aSi). A bias across the plates and selective illumination of the aSi creates a patterned electrical field in a liquid placed between the plates whose gradients place dielectrophoresis forces onto any polarizable particle present as [28]; Here, a is the radius of the particle. ε r ε 0 is the absolute permittivity of the medium. ∇E 2 is the gradient of the electric field squared. Re [K(ω)] is the real part of the Clausius-Mossoti factor (CM) which is defined through frequency dependent properties of the materials (particle/medium). The CM factor determines whether the DEP force is directed towards regions of low (negative DEP) or strong (positive DEP) electric fields and is given by [28]: Here, ε * p and ε * m are the complex permittivity of the particle and the medium, σ the conductivity of the materials and ω the angular frequency of the applied AC signal.
To estimate the DEP force acting on the particles retention experiments were carried out. Particles were trapped against the laminar flow (parabolic flow profile at low Reynold's number of 10 −2 ) by balancing the negative DEP force against the drag force F D : where η is the viscosity of the fluid medium (10 −4 Pa s), a the particle radius and v the velocity of the particle (10 −4 m s −1 ). We scaled the force (3×) according to Faxen's corrections assuming particles being close to the surface [29].

Device fabrication
High resolution polymer-emulsion film masks ( The OET chip was constructed from an ITO and aSi coated glass slide and an ITO coated cover slip. First, the ITO cover slip had inlet and outlet holes drilled according to the microchannel mask design. A Dremel drill (MultiPro, Robert Bosch GmbH, Germany) mounted to a work station was used in conjunction with Tungsten drill bits of 0.5 mm diameter (Diama, UK). The substrates were then cleaned in acetone, methanol and deionised water. This was followed by a dehydration bake at 90 °C for 30 min. The OET chip was assembled by sandwiching SU8 between the modified cover slip and glass slides. SU8 has good tensile (73 MPa) and dielectric strength (115 V µm −1 ). SU8 3050 and SU8 3025 have different viscosities. SU8 3050 has a viscosity of 12000 cSt, while SU8 3025 has a viscosity of 4400 cSt. With that, each SU8 resist yields a different thickness on top of the photoconductor when adjusting the spin coating speed. SU8 3050 was spin coated at 1000 rpm and 4000 rpm for 30 s with an initial coating step at 500 rpm for 10 s. SU8 3025 was spin coated at 500 rpm for 10s followed by 4500 rpm for 30 s. After spin coating the substrate was placed on the hotplate at 95 °C. The cover slip containing predrilled holes for inlets and outlets was bonded immediately onto the freshly prepared SU8 film by carefully placing it on top of the bottom substrate with a small overhang of several millimetre to provide wire connection points. This was followed by a soft bake at 95 °C for 25 minutes, 45 min and 20 min for SU8 3050 coated at 4000 rpm, 1500 rpm and SU8 3025 coated at 4500 rpm, respectively. Following the soft bake, the SU8 was exposed (i-line, 365 nm, 200 mJ cm −3 , Mask aligner MA 6, SÜSS MicroTec AG, Germany) through the mask aligned to the predrilled holes in the cover slip for 60 s and 90 s, for SU8 3050 (4000 rpm), SU8 3025 (4500 rpm) and SU8 3050 coated at 1500 rpm, respectively. The post exposure bakes for SU8 3050 (4000 rpm), SU8 3025 (4500 rpm) and SU8 3050 (1500 rpm) coated substrates were set to 5 min and 10 min at 95 °C with an initial bake at 65 °C for 2 min. The developing of the unexposed SU8 was performed through drilled inlet and outlet holes using Microposit EC solvent (Shipley, USA). The whole chip was placed in a beaker containing the developer. The developing process was monitored by observing the phase change (from solid to liquid) of the unexposed SU8 in the microchannel. A clear interface between the solid and the liquid phase was visible during this process. Once the interface vanished, the unexposed SU8 was fully dissolved. To remove any residues left in the channel structures, the chip was placed in an ultrasonic bath with acetone for 5 min to 10 min. The chip was rinsed with acetone, followed by blow drying until the acetone inside the chip was fully removed. The inlet and outlets were connected to PTFE tubing (#30, Cole and Parmer, UK) using shortened 10 µl pipette tips glued to the substrate.
The wire connection points were cleaned by stripping the unexposed SU8 using acetone. On the bottom aSi substrate, the aSi layer was scratched off using a sharp blade and opening a connection point to the ITO layer underneath. Wires were connected with conducting silver paint prior to strong fixation by applying epoxy glue (BondLoc, UK).
Dektat profilometer, SEM and standard brightfield microscope were used to characterise lateral and vertical dimensions of the fabricated OET chips.

Experimental setup
A TG5011 (TTi, UK) signal generator was used to apply AC signals to the OET chip. A microscope (BX51, Olympus, Japan) equipped with a dual port (U-DP, Olympus, Japan) attachment was used for observation of the micromanipulation and for camera recordings as well as image projection onto the photoconductor (setup shown in figure S1 (stacks. iop.org/JMM/30/045004/mmedia)). The dual port allowed for simultaneous operation of camera and data projector (Dell 1510X). An Orca Flash4.0 CMOS-camera (Hamamatsu, Japan) was used for recordings. The projector was connected to a PC and images for light induced electric fields were created using the drawing toolset in Microsoft PowerPoint. The microscope was equipped with a 4 × (NA: 0.1), 10 × (NA: 0.4), 20 × (NA: 0.4) and 40 × (NA: 0.65) objectives. An increase in the numerical aperture increased the sharpness of the light pattern as well as the light intensity. Measurements of the optical power using an optical power meter (Wilcom FM1318, USA) resulted in a power range shown in table S1. These values were converted into optical intensities and then corresponding aSi conductivities (figure S2).
Spherical polystyrene microparticle (Bangs Laboratories) of different sizes (3 µm, 6 µm, 10 µm) were suspended in deionised water containing 0.1% tween 20. The conductivity was adjusted by adding potassium chloride to the deionised water and tween 20 suspension prior to the addition of the particles. The suspension was injected into the chip using a syringe pump (NE-1000, New Era Pump Systems, USA). Recordings of particle retentions experiments were analysed using ImageJ (v1.47). The velocities of moving particles under various conditions were extracted to quantify retention and focussing capabilities and DEP forces.

Device simulation
In order to understand the device characteristics of a micro scaled OET device a range of simulations were undertaken using the finite-element modelling software COMSOL Mutliphysics (v3.5, AC/DC module, in-plane electric currents). A simple 2D cross-section model which depicts the liquid layer and the photoconductor layer in the OET device and their actual thicknesses was used to introduce field distributions, magnitudes and components in the microdevice before the influence of parameters such as the conductivity of the liquid medium and the photoconductor, vertical channel dimensions, electrode pattern sizes as well as potential drops across the layers were specifically explored by experiments. Figure S3 outlines the boundary conditions as well as subdomains used for the model. It describes a typical configuration for a conventional OET device. The model parameters for the different materials are shown in table S2. The photoconductive effect was modelled as a change in the electrical conductivity of the aSi layer. The microscope in this work was equipped with a set of different objectives leading to different optical powers and hence different light intensities (figure S2). The influence of light intensity on the conductivity of the aSi layer was considered by defining values for the dark-and light-state for each objective. To simplify the model the ITO layer was removed as the potential change across it is negligible. The sides of the cross-section were assumed to be electric insulators owing to the fact that a polymer (SU8) was used as microchannel wall in this work. The upper side of the geometry was given an electric potential, while the lower side of the photoconductor was set as ground ( = 0). The enclosed sides (interfaces to layers) were subjected to the continuity boundary condition. The layers of the OET device were defined as subdomains and specified by defining geometrical.
GNU Octave (v3.6.4) was used to model the dielectrophoretic response of different sized particles under varying conditions.

OET Device fabrication
Integration of the conventional OET structure was achieved by sandwiching the photoresist SU8 between the OET electrodes before photopatterning. The resist acted as bonding agent and the microchannel defining layer with adjustability in terms of layer thickness and channel layout. The former was controlled by tuning the spin-coating speed while the latter was defined by the mask layout. Channel thicknesses of 15 µm to 70 µm were achieved using SU8 3000 series (SU8 3025, SU8 3050) and spin coating speeds of 4500 rpm to 1500 rpm (figure 2(a)). Figures 2(b)-(f) shows SEM images of cross-sections of fabricated OET devices with integrated microchannels and an example OET device. It has to be noted that the standard SU8 protocol had to be altered to cater for the unconventional fabrication process. This included pre-bake and exposures times to ensure strong bonding of the OET electrodes and welldefined channel layouts. For instances pre-bake times and exposure times have been increased to reduce solvent content of the SU8 layer after ITO cover slip bonding and compensate UV light absorption losses in the ITO layer. The resulting microchannels were characterised by a round channel-wall profile, likely due to the sandwich structure, adhesion properties of the resist and the action of developing along the channel rather than from the top. Lateral channel dimensions were not true to the original mask layout but slightly increased as result of mask to resist distance of 160 µm-190 µm (the thickness of the top ITO coated cover slip). Overall, microchannels were residue free once developed (figure 2(f)), the bonding area of the OET electrodes with the SU8 was 100% and the OET chips were durable (leakage-free), and reusable during experiments when conducting appropriate wash and rinsing steps using aqueous and organic solvents.

OET Device characterisation
3.2.1. AC amplitude dependency. The dielectrophoretic performance of the OET chips were characterised by particle retention experiments where the particles experienced negative DEP imposed by a virtual electrode projected along the width of the microchannel (figure 3(a)). The DEP force was calculated from bead velocity measurements once the drag force was balanced against the negative DEP force under continuous fluid flow. DEP force scaling was investigated by increasing voltage amplitude and flow rates while keeping the frequency at 50 kHz, using an optical power of 2 W cm −2 , a virtual electrode width of 45 µm and medium conductivity of 5 mS m −1 . The DEP force scales with the square of the gradient of the electric field (equation (1)). Hence, the field should be proportional to the square of the applied voltage. However, our results demonstrated a rather linear relationship in the developed OET device ( figure 3(b)). A voltage range from 2.5 V to 20 V enabled retention of particle with velocities of 15 µm s −1 to 522 µm s −1 , corresponding to DEP forces of 4 pN to 150 pN. The electric field gradient under these conditions is shown in the simulations in figure 3(c) for a voltage amplitude of 20 V. The highest field magnitudes at distances of 2.5 µm to 5 µm to the aSi surface varied between 2.2 · 10 17 V 2 m −3 to 8.0 · 10 15 V 2 m −3 and can be found at the edges of the virtual electrode. The direction of the DEP force for insulating particles is indicated by black arrows in the surface plot. A negative DEP force with lateral and vertical components was acting on the polystyrene beads resulting in repulsion from high field regions. The exact reason for the linear relationship between DEP force and applied voltage was unclear and requires more investigations. However, one may attribute this effect to the particles being unable to experience the full negative DEP force at higher speeds because of an upward motion. The vertical and the lateral field components scale both with the applied voltage while having a similar magnitude. Under the influence of the negative DEP force a lifting of particles attributed to the vertical field component may reduce the retention effect induced by the lateral field component. This would result in a movement of particles towards lower field regions near the top electrode which then allowed particles to pass the virtual electrode where the drag force overcomes the DEP force. The described behaviour was supported by simulations of the components of ∇E 2 . In figure 3(e), ∇E 2 y and ∇E 2 x were determined for different voltages and distances from the virtual electrode edge in the photoconductor layer. At distances relative close to the surface, where the DEP force had its maximum, ∇E 2 y was dominating over ∇E 2 x . This promoted upward motion of particles over the potential barrier and decreased the retention effect. Recently, this phenomenon has also been described by Zhang et al who used a conventional OET setup without continuous fluid flow but dynamic light pattern [30].

3.2.2.
Influence of medium conductivity. The medium conductivity is a crucial parameter when using conventional OET devices. The developed OET chip can be understood as a simple lumped equivalent circuit containing impedance elements which represent the photoconductor and the liquid medium ( figure 4(a)). The applied voltage drops across the photoconductor in the absence of light. The impedance of the photoconductor is higher compared to the liquid medium. When the photoconductor is illuminated the voltage drops across the liquid creating an electric field. The latter case requires the impedance of the liquid and hence the conductivity to be sufficiently low. The voltage drop across the photoconductor was modelled for different medium conductivities ( figure  4(b)). The model depicts a voltage drop across the centre of the illuminated area on the photoconductor layer for medium conductivities of 5 mS m −1 to 100 mS m −1 using an AC signal of 20 V at 50 kHz and a light intensity of 2 W cm −2 . At low conductivities (5 mS m −1 ) the voltage dropped mainly across the liquid layer generating a strong non-uniform electric field. At high medium conductivities (100 mS m −1 ) the voltage dropped mainly over the photoconductor producing a low electric field magnitude in the liquid layer.
The performance of the OET device was tested at varying liquid conductivities using retention experiments as described above. It has to be noted that an appropriate particle size (10 −5 m) should be used for this experiment. From equation (2), it is known that the CM factor is dependent on the frequency and the complex permittivity of particle and medium. When the liquid conductivity increases the CM factor decreases. From equation (1) it is clear, the CM factor partially contributes to the DEP force. Furthermore, for small particles the surface conductance effect in the electric double layer influences the particles conductivity and hence the DEP response [31]. The width of the electric double layer is altered with changing ion concentration in the solution which again influences the CM factor. Therefore, when relating measured particle velocities to the device performance it should be verified that the DEP response of the particle was not altered by increasing conductivities. A large particle (⩾10 µm) has a relative constant CM factor (only ~10% change on CM) for increasing conductivities while smaller particles (~1 µm) can undergo significant changes ( figure 4(c)). Also, the CM factor considers ideal dielectric spherical particles and medium. More complex particles (e.g. cells, protein, DNA) and below size scales of 10 −6 µm the application of the CM factor has limitations [32]. Moreover, at higher conductivities (>50 mS m −1 ) electrothermal effects from temperature gradients due to Joule heating can drive microflows which can compete with electrokinetic effects like DEP [33]. Temperature increases by illumination of the photoconductor can be excluded in the OET chip due to low optical intensities (2 W cm −2 ) [34]. The high electric fields (10 5 V m −1 ) generated in the OET chip suggest the ability to produce Joule heating and promote electrothermal microflows. However, at higher conductivities the applied voltage drops mainly across the photoconductor producing low field strength in the liquid which combined with the low channel height [35] suppresses electrothermal microflows.
The retention performance of the OET device for various conductivities was investigated ( figure 4(d)). Velocities of 522 µm s −1 to 17 µm s −1 were measured for conductivities of 5 mS m −1 to 100 mS m −1 . This corresponded to DEP forces of 150 pN to 5 pN. A sharp drop in the DEP force occurred within These media are not applicable for a conventional OET device. The electric field in the liquid layer would become 3 to 4 magnitudes lower (∇ 2 ~ 10 13 -10 14 V 2 m −3 ) resulting in DEP forces not sufficient for continuous particle manipulation. However, a Ph-OET device developed by Hsu et al was capable of overcoming this restriction and enabled cell and particle handling in cell culture medium [36].

Influence of gap between the electrodes (channel height).
The performance of the OET chip can be improved by considering how the electric field is dependent on the gap between photoconductor and the ITO electrode. The voltage drop across the liquid layer reduces while the electric field magnitude increases when decreasing the gap between the photoconductor and the ITO electrode. Simulations of the electric field gradient ∇ 2 were carried out to obtain insight into the significance of channel heights in the OET device. In figures 5(a) and (b), ∇ 2 is shown for varying electrode gaps. In particular, figure 5(a) shows the decay of ∇ 2 above the edge of a virtual electrode for electrode gaps of 15 µm to 110 µm. The field distribution is shown for the first 15 µm above the illuminated aSi layer. The increase in the channel height decreases the field magnitude and promotes a steeper decay above the photoconductor. Furthermore, larger particles experience a greater volume of the channel and so we need to consider how the field varies higher up in the channel where the change in the magnitude becomes more pronounced (figure 5(b)). Therefore, the OET device performance is significantly influenced by the device dimensions combined with the particle size. The device fabrication method introduced above enables the construction of variable and well controlled SU8 thickness. With that the performance change was tested by creating channels with heights of 70 µm and 15 µm. A 20 µm virtual electrode (2 W cm −2 ) and an AC signal of various voltages at 100 kHz were used for retention experiments of 6 µm particles suspended in a medium of 5 mS m −1 ( figure 5(c)). Using a channel height of 15 µm the obtained DEP forces for voltages of 6.5 V-20 V were 16 pN to 132 pN while at a gap of 70 µm the obtained DEP forces were only 3 pN to 22 pN. As the channel's height was reduced the impedance of the liquid layer decreased, decreasing the voltage dropped across it so that the field did not increase linearly with decreased channel height. Our results show that despite this, the forces obtainable increase at a greater rate than the channel height decreases which can be understood by considering that making the channel smaller has the same geometrical effects as considering larger particles [37].
These results demonstrated that relatively fast manipulation of small particles is feasible but requires good dimensional control over the conventional OET device structure. However, reducing the electrode gap is also accompanied with an increased shear stress imposed onto particles. While this is less important for artificial particles, mammalian cells might be affected, although some cells (e.g. red blood cells) are able to withstand higher shear rates. Furthermore, high sample concentration (10 7 -10 8 beads ml −1 with particle diameters >5 µm) should be avoided as channel blocking by particle aggregates inhibits efficient manipulations.

Light pattern size and intensity.
The device performance was dependent on several factors as shown above and each of these need to be adjusted to obtain forces of appropriate magnitude for particle manipulation under continuous flow. Moreover, the virtual electrode projected on to the photoconductor is an additional independent parameter that requires adjustment. The magnitude of the light intensity which is projected onto the photoconductor through an objective influences the magnitude of the electric field in the sample. The light intensity increases with increasing magnification and hence increases the conductivity of the aSi layer. Also, the dimension of the electrode (e.g. width) as well as the pixel intensity of the virtual electrode defined by the imaging software influence the electric field created in the liquid medium. We show that the virtual electrode width and transparency (pixel intensity) can be used to tune the electric field magnitude and hence the DEP force experienced by the particle. Experiments were carried out with increasing virtual electrode width (7 µm to 120 µm) and increasing electrode transparency (0% to 60%). Again, retention of 6 µm particles under continuous flow were investigated using an OET chip with 15 µm gap height, a voltage signal of 10 V and 15 V at 100 kHz in a medium of 5 mS m −1 . The virtual electrode width was varied between 7 µm and 120 µm ( figure 6(a)). The measured velocities for that width range were between 54 µm s −1 and 300 µm s −1 ( figure 6(b)) which corresponded to DEP forces of 9 pN to 51 pN.
A non-linear behaviour characterised by rapid increase in the DEP force was observed when increasing the virtual electrode width from 7 µm to 20 µm followed by saturation of the force for width of more than 20 µm. This device behaviour may be associated with the change of the gradient of the electric field across the channel height at constant voltage. Simulation of the electric field gradient for several electrode widths showed similar magnitudes of the field close the photoconductor surface (<5 µm) (figure S5). However, the field strength above the surface decayed much steeper for the 7 µm electrode. An increasing electrode width resulted in a convergence of ∇ 2 which agreed well with the asymptotic behaviour observed in the experiments. In addition, reflection at the interfaces in the OET device may alter the light intensity which reaches the photoconductor. This influence may be pronounced when using small virtual electrode patterns and could affect the photoconductivity of the aSi layer.
The light intensity of a virtual electrode can be easily changed by defining the pixel intensity in the imaging software. This gives spatial fine control over the DEP force. Altering the pixel intensity or transparency from 0% to 60% (0% being full intensity of 2 W cm −2 and 60% being reduced intensity, figure 6(c)) resulted in retention velocities of 190 µm s −1 to 37 µm s −1 with corresponding DEP forces of 11 pN to 2 pN (figure 6(d)). Thus, controlling the light pattern enables the generation of tuneable electric field gradients with corresponding DEP forces at arbitrary positions in the OET device, while other parameters (e.g. voltage, frequency, objective) can be kept constant. For instance, this can be used to selectively manipulate particles of different sizes.

Continuous particle manipulation
Conventional dielectrophoresis with fixed metal electrodes has been commonly applied for different microfluidic operations under pressure driven flows. Examples included focussing of particles into a single stream [38,39] or separation of a particle mixture [40] based on physical properties (e.g. volume). Focussing particles into tight streams can be considered as an essential step in microfluidic flow cytometer [41,42] where the properties (e.g. fluorescence) of single particles can be investigated prior to subsequent processes (e.g. separation, detection). It can also be used to concentrate particles into specific regions within the microchannel for counting, detecting, enrichment purposes [43] or to separate particles from a carrier liquid [44][45][46] (e.g. washing, functionalisation).
Here we demonstrate that that the developed OET device can create arbitrary virtual electrode patterns in real time and that these can be used for particle concentration, focussing and sorting. Figure 7(a) shows the focussing of particles into the centre of the channel. In this case the DEP and Stokes drag force experienced by the particles combined to a net force that guided particles along the virtual electrode. This resulted in particles being lined up once past the electrode pattern. Combined with the hydrodynamic force under laminar flow conditions a stable and tight single stream of particles was maintained. Depending of the angle (15°-75°) of the virtual electrodes and the applied voltage signal (10 V, 20 V) particles with velocities of 0.4 mm s −1 to 5 mm s −1 were tightly focussed into single particle stream ( figure 7(b)). Compared to previous studies on OET devices, these results demonstrated improved particle manipulation speed with velocities in the mm/s range using oblique virtual electrodes combined with a shallow microchannel geometry. This will be of use for application such as a microfluidic flow cytometer where flexible electrode arrangements suited for different kinds of particles would give the device a greater flexibility through real time reconfiguration although it would not be able to compete on speed with state of the art flow cytometers using velocities of 1 m s −1 to 10 m s −1 for high throughput analysis (500 000 particles s −1 ) [47].
Channel integrated OET devices can also be utilized for particle sorting, which may reduce the complexity of a cytometer setup. A simple example is shown in figure 7(c) where a particle mixture was guided towards a separation zone consisting of an angled virtual electrode with width of 22 µm and two virtual channels (Movie S1). While flow rate and voltage signal were kept constant, the electrode width was adjusted to achieve a separation between the particle sizes. While the negative DEP force was sufficient for deflecting 6 µm particles, the drag force on 3 µm particles dominated causing these particles to pass the virtual electrode with only minor deflection. Furthermore, oblique electrodes can also be used to deflect particles laterally in a microfluidic channel (figure S6). The channel integration then has a second advantage in that different patterns with multiple inlets or outlets can be integrated into the design in the same layer that defines the channel height. This could be used for automated biological assays integrated into continuous microfluidic assays. Such a concept has been shown using solid rails which guide particle across several adjacent laminar streams [46]. Each stream can carry a certain reagent to wash and modify (e.g. protein coupling, surface functionalisation) particles prior to a reaction in another laminar stream which carries a sample liquid (e.g. blood) with an analyte of interest. The size based response of the DEP force may also be used for the separation of cell populations (e.g. white bloods cell from red blood cell or malaria infected cells) or in droplet microfluidics to maintain monodispersity.

Limitations
The device characterisation experiments showed unwanted particle-particle and particle-wall interactions. This was attributed to a voltage drop into the liquid medium in the absence of illuminations of the photoconductor but when applying a voltage signal. It resulted in a leaked and mainly uniform electric field within the channel that interacted with suspended particles and causing a weak non-uniform field distribution along the channel walls. The former caused particle-particle interactions due to mutual dielectrophoresis. This pearl-chaining effect is the result of the uniform field being spatially altered creating low and high field regions around the particle surface. The latter was caused by the insulating properties of SU8. The electric field strength being low close to the SU8 wall forced particles to move towards the channel boundaries occasionally.
The leakage fields can be explained by the fact that the used photoconductor was not an ideal insulator in the absence of illumination. Then a combination of voltage amplitudes and low medium conductivities created these leakage fields.
To address this problem without increasing medium conductivities and sacrificing strong DEP forces the thickness of the photoconductor can be increased. While this seems trivial because deposition of thicker photoconductor films can be easily performed, our trials with increased film thickness showed residual stresses in these films leading to delamination or blistering. The deposition process has to be optimised to create stable films. Moreover, an increase in the thickness of the photoconductor may require increased optical power because of a relative high absorption coefficient within 1 µm of aSi which otherwise would not fully increase the photoconductivity throughout the layer.

Conclusion
A method for integrating a conventional OET arrangement into a microfluidic chip was introduced in this work. We have demonstrated the importance of channel geometry in the OET device and developed a method to finely control the channel dimensions. To achieve this negative photoresist SU8 3000 series was used and a novel step of attaching the top slide before photopatterning the resist was produced. The advantages of this method were low bonding temperature (65 °C to 95 °C), homogenous bonding area, stability to organic solvents for cleaning purposes and flexibility in creating microchannels of various heights and patterns. The device characterisation showed that the device performance can be tuned by controlling a range of parameters including the voltage signal, channel heights, the virtual electrode size and transparency. The flexibility of quickly creating virtual electrodes of arbitrary shape and intensity allows useful microfluidic functions including particle focussing which can be combined with continuous subsequent particle separation steps.