In a typical T-junction configuration, as depicted in Figure 1, the two phases flow through orthogonal channels and form droplets where they meet. This type of geometry was first demonstrated in 2001 by Thorsen et al. [28], who produced monodisperse droplets with pressure controlled laminar flow in microchannels. Since then many studies were performed using T-junction geometries, to achieve a better understanding of the droplet formation mechanism and the role of several physical parameters therein [25,29–35], as well as to develop various applications [36–42]. The size of the droplets depends on the flow rates of the two liquids [28], the dimensions of the channels [29], the relative viscosity between the two phases [43], and surfactants and their concentrations [44].

Three main regimes can be distinguished for drop formation as the parameters are varied: dripping, squeezing and parallel flowing stream. In the dripping regime, droplet breakup occurs when the viscous shear stress overcomes the interfacial tension, analogous to the breakup of spherical droplets. If the capillary number is chosen large enough, the droplets are emitted before they can block the channel. Alternatively if the capillary number is low, the formed droplets will obstruct the channel and hence restrict the continuous phase. This causes a dramatic increase of the hydrodynamic pressure in the upstream part, which in turn induces the pinch-off of droplets. This is the so-called squeezing regime, which has been described by Garstecki et al. [29]. One theoretical study about the transition from squeezing to dripping based on the influence of Ca and viscosity ratio was reported by Menech et al. [45]. Also Lattice Boltzmann simulations have been performed to increase the understanding of drop formation at T-junctions. Van der Graaf et al. [24] obtained a scaling rule for the drop size. Gupta et al. [25] found that the transition from droplet formation to parallel flows is strongly dependent on the Ca of the continuous phase.

A slightly different geometry having similar features as the above explained T-junction geometry is the so-called head-on device (see Figure 2a). Shui et al. demonstrated droplet formation in such a device, where two liquids come from opposite directions of two straight channels and form droplets upon meeting [36,46,47]. Also a Y-shaped junction has also been studied, for example by Steegmans et al. [34,37]. As illustrated in Figure 2b, droplets can be formed in the dripping regime in such a Y junction geometry. The mentioned authors studied the mechanism of droplet formation and derived a general model predicting the droplet size. They also demonstrated that such a flat Y-junction can be used as a microfluidic tensiometer, i.e., a device that can measure dynamic interfacial tensions.

For certain applications, a single T-junction is clearly not enough. To perform chemical reactions or to produce droplets with alternating compositions, more sophisticated designs have been realized: for example double T-junctions to produce droplet pairs [48–51]. One example is shown in Figure 3 [51]. The authors of this paper demonstrated a perfect “one-to-one” droplet pair formation (self synchro-nization) with the use of additional connections in the upstream and downstream channels.

For the mass production of emulsion droplets using microfluidic devices, large scale integration of droplet generators is a necessity. For the case of T-junctions, this has been explored for up to 256 junctions in parallel [41,42]. The highest throughput was reported as 320 mL·h⁻¹ in a 4 cm × 4 cm chip with 256 droplet formation units. Further developments along this line would be needed to achieve production at industrial scales, but the perspectives are already there. One of the challenges that may have to be faced is to minimize detrimental cross-talk between the different droplet injectors. This could occur for example if the transient pressure variation associated with the creation of a droplet is transmitted to other droplet injectors and interferes with the droplet formation there.

The flow focusing (FF) geometry was first proposed by Anna et al. [52] and Dreyfus et al. [53]. As demonstrated in Figure 4, it consists of three inlet channels converging into a main channel via a narrow orifice. The dispersed phase, contained in the middle channel, is squeezed by continuous phase flows from two opposing side channels. Both phases pass through the small orifice that is located downstream of the three channels. Finally, the stream of the dispersed phase becomes narrow and breaks into droplets. The droplet size is determined by the flow rates of the two phases and by the flow rate ratio [54,55], in addition to the channel geometries [56] and the viscosities of the two phases [57,58]. This multitude of influential parameters in principle offers a lot of control over drop formation, but it is also true that in the absence of adequate (i.e., quantitatively predicting) theoretical models, each new combination of geometry, speeds and viscosities may need to be explored and tuned, in order to let the chip meet the demands (i.e., criteria for droplet size and formation rate). Many variations of the basic flow focusing device (FFD) geometry have recently been developed to improve the control over the size and size distribution of the droplets [26,59–61]. Since recently, these developments can be assisted by numerical (Lattice Boltzmann) simulations, which have seen their first application to flow-focusing [26] and cross-flow geometries [27].

Also so-called axisymmetric flow focusing designs have been presented (Figure 5). They allow the formation of monodisperse droplets with reduced size as compared to planar FFDs [59]. In these geometries, the dispersed phase is confined in the central axis of the microchannel, and pinches off by a combination of shear stresses and wetting upon contact with the inner surfaces of the channel.

Four different droplet breakup regimes have been identified in planar FFDs: squeezing, dripping, jetting and thread formation (tip-streaming), shown in Figure 6. As mentioned, there are no general scaling laws that can predict the transitions between these regimes, and the same applies for the size and generation frequency of droplets. This is due to the large number of variable parameters. Recently, Funfschilling et al. concluded from velocity field measurements that the squeezing regime is governed by the build-up of a pressure difference, as a response to the partial and temporal blocking of the orifice by the advancing finger [62]. Lee et al. stated that the squeezing and dripping regimes depend solely on the upstream geometry and the related flow field, while the thread formation mode depends only on the downstream channel and its associated flow field [56]. It is clear that unraveling the mechanisms of droplet break-up in FFDs still needs further investigation.

Some aspects, namely the transition between dripping and jetting was studied in detail for the somewhat simpler geometry of co-flowing liquid jets [17,19,20]. Building on earlier work for liquid jets injected into an unconfined bath of another immiscible liquid [63,64], Guillot et al. performed a linear stability analysis of a liquid jet of the to-be-dispersed inner phase in a co-flowing stream of the outer fluid confined to a cylindrical capillary. They showed that the difference between dripping and jetting is primarily controlled by the Ca number of the outer phase fluid and be the radius of the inner jet (measured in units of the radius of the capillary). Small Ca and small radii correspond to absolutely unstable jets, i.e., any perturbation of the radius propagate towards the nozzle and induce drop generation there. This corresponds to the dripping regime. For large Ca and large radii, perturbations of the radius are advected by the flow downstream along the jet where they may eventually lead to the formation of drops. In this case, the jets are said to be convectively unstable. For jet radii close to the radius of the capillary, the confinement of the inner jet plays a key role in the suppression of the break-up. Utada et al. [17] demonstrated yet another mechanism of dripping to jetting transition: for moderate Ca (<O(1)) of the outer phase jetting can be induced by increasing the flow rate of the inner phase. In this case, the We number of the inner phase was shown to be the control parameter governing the transition.

Returning to applied aspects of drop generation, it was demonstrated that multiple FFDs could be combined in parallel in either linear [65,66] or circular circuits [41] in order to increase the droplet production rate. Li et al. demonstrated a quadruple droplet generator with a weak parametric coupling between the different parallel FFDs. By choosing different geometries for the individual FFDs, these authors were able to simultaneously produce several populations of droplets with distinct sizes, where each of the populations had a narrow size distribution (Figure 7). Also Hashimoto et al. [66] studied the dynamic mechanism of droplet formation in parallel FFDs. They found a weak hydrodynamic coupling as well.

To increase the flexibility of FFDs, additional (active) elements have also been incorporated into devices. Several groups have applied electrical means to obtain more control over droplet formation in FFDs [67–70]. For example, Gu et al. integrated the functionality of electrowetting (EW) [68,69] by embedding insulator-covered electrodes underneath the channel. In this case the wetting angle between the oil/water interface with the channel wall can be tuned via the voltage applied over the insulator layer (see Figure 8). This is dictated by the so-called electrowetting number η = ɛɛ₀U²/2dσ with ɛɛ₀ the permittivity and d the thickness of the insulator layer, U the voltage and σ the oil/water interfacial tension [71]. Three different droplet formation regimes could be achieved using the combination of hydrostatic and electrical driving: dripping, tip-streaming and conical spray (see Figure 9). It is clear from these results that the additional electric control can provide an extension of the size range in which droplets can be produced: from tens of micrometers down to a few microns in the presence of an applied voltage. Also the rate of droplet generation could be raised to very high values using EW.

The conical spray was found at high relative flow rates of the dispersed (i.e., water) phase and large electrowetting numbers (η > 1, corresponding to U ≈ 50 V). In this specific regime of electro-wetting, the droplets appear to repel each other due to finite electrostatic charges accumulated at their surfaces. Similar droplet spray patterns were observed by Kim et al. [67] and He et al. [70] who integrated an electrospray functionality into FFDs. In such devices, the droplet size can also be diminished by increasing the voltage. Yet for the formation of very fine droplets, one needs to be in the Taylor cone regime, which requires voltages above ≈1500 V.

Also membrane valves have been introduced into FFDs to vary the width of the orifice [72–75]. Abate et al. demonstrated that the droplet size and formation frequency in the dripping regime can be controlled by such an adaptable orifice [75]. The approach based on (local) adjustment of the temperature has been reported: here use is made of the temperature dependence of the viscosity and interfacial tension [76–78]. This method allowed independently controlling the droplet size and generation frequency.

Additional means to actively control drop formation can be obtained by adding particles that respond to external fields into the to-be-dispersed phase. Examples of this approach, which we will not address in more detail here, include magnetic control of ferrofluid droplets [79], electric control of electrorheological droplets [80] and temperature control of certain types of nanofluids [81].

In a large majority of the existing continuous flow devices, droplets are produced incessantly; the flow can be switched on and off, and the conditions of droplet generation can be modified, but the droplets will always appear in trains. In cases where this scenario is undesirable, and droplets need to become available one-by-one upon request, digital microfluidics applications [12,13] may come to mind first. However, continuous flow systems can also be adapted to deliver droplet-on-demand (DOD). Surprisingly, this possibility has hardly been explored, in spite of its strong potential for high throughput screening in microtiter technology or in the programmed coalescence of droplets (after a synchronized formation of droplet pairs). Especially the combination of on demand formation of droplets and a subsequent processing at high speed would make it interesting.

One of the possibilities for on-demand droplet formation is the use of integrated microvalves [82,84,85]. For instance, Zeng et al. incorporated a pneumatic valve made of polydimethylosiloxane (PDMS) into microfluidic devices (Figure 10a). By intermittently switching the valve on/off, individual droplets can be produced on demand [82]. Also piezoelectric actuators have been used in DOD applications [86,87]. In such systems the droplet size and frequency can be set with high accuracy through a conversion of the piezo voltage into a mechanical displacement. Churski et al. reported a DOD system that used external electromagnetic valves interconnected with the chip for the scanning of reaction conditions [88].

Alternatively, electric fields can also be used to produce droplets on demand. Malloggi et al. [83,89] used electrowetting as an active control mechanism to increase the wettability of the channel wall at the location of droplet formation. Combining pressure control over the two phases and electrical control over wetting, the size and/or generation rate of their droplets could be tuned within a certain range (Figure 10b).

In passive droplet merging, the design of the channel geometry is a key to achieve proper merging, since droplet synchronization is required and active means to compensate for any synchronization errors are missing. In principle merging can occur simply at a channel junction, if the generation and transport of each pair of droplets is such that both drops arrive there at the same time. However in practice this can be difficult to achieve, and therefore special designs of geometries are often used.

One widely used geometry for droplet merging consists of a widening channel follow by a narrower channel (Figure 11) [90–93]. In this geometry the droplet velocity decreases in the widening channel because of drainage of the continuous phase, after which it increases again upon entry in the narrow channel. Due to this changing flow field, two subsequent droplets are allowed to come close together and let the liquid that separates them drain away. Bremond et al. observed that the merging does not occur during the first interdroplet encounter in the extended channel, but rather during the separation stage of two droplets when the first droplet begins to enter the narrow channel (Figure 11c) [93]. The separation induces the formation of two facing protrusions (Figure 11d) which then bring the two interfaces close enough until they merge. Later Lai et al. reported a theoretical study based on this observation [111]. The created protrusions lead to a rapid increase of the surface area locally, and thus to destabilize the interface at certain locations. The conditions under which droplet merging occurs, can be predicted on the basis of their model. Alternatively in other channel geometries, droplets are merged by slowing down or stopping the leading droplet at a constriction [97,98], or in a channel with an array of pillar elements [94,95].

It should be noted that typically no surfactant is used in these passive merging experiments. However, the absence of surfactant has its drawbacks: unintended merging events can occur, and also the possibilities for further manipulation of the droplets after the merging can be limited. By exception, a case of passive merging of surfactant-covered droplets has also been reported. Mazutis et. al. demonstrated a channel design for merging droplets with significant asymmetry in size, both formed in the presence of surfactant [100]. However, undesired coalescence still occurred often. It is therefore often preferred to use surfactant stabilized drops and achieve merging with the help of external forces.

To achieve active and selective droplet merging, the most widely utilized method is to ensure the presence of an electric field at the location where two droplets meet. Link et al. showed that droplets can be merged by applying voltages with opposite sign across the two droplets during their formation. This is supposed to result in oppositely charged surfaces, which will attract each other strongly as soon as the droplets reach close proximity [72]. Alternatively, Chabert et al. achieved merging of individual droplet pairs via electrocoalescence (EC) [112]. This appears to be a promising method, although the mechanistic aspects of EC still remain to be understood [105,109,113,114].

The general picture of EC is sketched in Figure 12. Due to the electric field, the two droplets experience an electrical (Maxwell) stress σ_E that tends to deform their shape from spherical to prolate spheroid. This stress is then balanced by the interface tension and the viscous stresses due to the deformation rate [115]. For Newtonian fluids at low Reynolds and Bond numbers, this is described by: (2) μ ∇ 2 U   −   ∇ P   =   ∇ Twith μ the viscosity, U the velocity, P the pressure and T the stress field in each phase of the two-phase fluid. Since the velocity is continuous across the interface, the total stress difference (electric plus viscous) between inside and outside the droplet is balanced by the interfacial tension: (3) n T   =   n T N   +   n T E   =   σ n ∇ S nwhere n is the unit normal vector at the interface, σ is the interfacial tension, ∇_Sn is the mean curvature of the interface, T^E is the Maxwell stress tensor (proportional to the square of the applied electric field) and T^N is the tensor of viscous forces [115]. Hence the field, the viscosities and the interfacial tension all play a role.

Priest et al. argued that EC involves an electric-field-induced dynamic instability of the oil/water interface, which subsequently leads to the formation of a liquid bridge and coalescence (Figure 13a) [105]. Thiam et al. analyzed the merging of droplets as a function of their separation distance (Figure 13b) [109], and also explained their observations in terms of a competition between electrical stress and restoring capillary pressure. Qualitatively speaking, it is clear that the electric field near the droplet surfaces can be amplified by dipole-dipole interactions between the droplets, and hence become stronger as the droplets get closer. It is conceivable that this will lead to destabilization of the surfaces [116]. Furthermore, also the surfactant molecules can be involved. In the case of surfactants with dipolar head-groups, a redistribution or re-alignment along the electric field lines can take place. Also this can destabilize the interface and lead to coalescence [117].

One of the first applications of EC in two-phase flow microfluidics was presented by Tan et al. [118]. Two droplets containing biological molecules were brought into an expanded channel and merged there, due to an electric field generated by an embedded electrode. Later, several variations based on this geometry were adopted to implement EC in microfluidic chips [72,109,112]. For each of these EC-based systems, droplet synchronization and precise electrode alignment are required.

To overcome these limitations, Gu et al. used EW-induced on demand formation to obtain synchronization of two streams of produced droplets [119]. These two streams then meet at a T-junction where interdigitated electrodes are embedded (Figure 14a and b). Merging on demand can be achieved there based on EC. As illustrated in Figure 14c, Niu et al. depicted an alternative method, by combining a passive merging approach (a pillar array in the channel) with an active merging approach (built-in electrodes) [110]. In this scheme, the pillar array slows down and traps the droplet during the drainage of the oil phase. EC then occurs when droplets have reached close proximity. Also a double T-junction geometry with embedded electrodes has been reported in the context of active merging. In the system of Wang et. al., two series of droplets can be produced and merged at the same time [108].

Yet another method for the active merging of droplets is dielectrophoresis (DEP). A drawback of this method is that it requires rather high voltages, up to several kV [120–122]. Finally, thermo-capillary effects can also be cited as a mechanism to perform active merging of droplets [76,104,123,124]. Heating two adjacent droplets with a focused laser beam was reported to cause convective motions in the droplets, as well as depletion of surfactant molecules from the interface. Also this turned out to be effective for droplet merging.