Scaling up the throughput of microfluidic droplet-based materials synthesis: A review of recent progress and outlook

Abstract

The last two decades have witnessed tremendous progress in the development of microfluidic chips that generate micrometer- and nanometer-scale materials. These chips allow precise control over composition, structure, and particle uniformity not achievable using conventional methods. These microfluidic-generated materials have demonstrated enormous potential for applications in medicine, agriculture, food processing, acoustic, and optical meta-materials, and more. However, because the basis of these chips' performance is their precise control of fluid flows at the micrometer scale, their operation is limited to the inherently low throughputs dictated by the physics of multiphasic flows in micro-channels. This limitation on throughput results in material production rates that are too low for most practical applications. In recent years, however, significant progress has been made to tackle this challenge by designing microchip architectures that incorporate multiple microfluidic devices onto single chips. These devices can be operated in parallel to increase throughput while retaining the benefits of microfluidic particle generation. In this review, we will highlight recent work in this area and share our perspective on the key unsolved challenges and opportunities in this field.

I. INTRODUCTION

Microfluidic chips can precisely control fluid flows on the micrometer scale, and this capability has been leveraged with enormous success to generate highly uniform and precisely defined single- and higher-order emulsions. These precisely defined emulsions have been used as templates for the generation of functional micro- and nano-particles, with exquisite control over composition, shape, and surface functionalities. In particular, particles generated by microfluidic chips have been developed with unique capabilities for the encapsulation, delivery, and controlled release of cargo, such as drugs, enzymes, or cells, for applications in targeted cancer therapeutics and in food processing.^1–3 Microgels with precisely defined geometries have been produced using microfluidic droplets for applications in wound healing and injectable drug delivery systems that have superior properties, such as degradation and molecule release, over their conventional counterparts.⁴ Moreover, microfluidic-generated materials have also demonstrated enormous potential in meta-materials, chromatography, catalysis, and a multitude of other applications.^5–7

Despite many published manuscripts that have demonstrated the capabilities and unique opportunities of microfluidic-generated materials, the translation of the technology from academic labs to industry has been limited, and primarily constrained by a single fundamental issue. The throughput of microfluidic materials generation is constrained by the physics governing the flow of immiscible fluids confined within micrometer scale channels,⁸ and this throughput tends to be several orders of magnitude lower than what would be necessary for commercial and clinical applications. In particular, the droplet generation mechanisms in microfluidic geometries, such as cross-flow, co-flow, flow-focusing, or step emulsification, generate homogenous droplets only with sufficiently low flow rates of the continuous and dispersed phases.⁸ As such, microfluidic droplet generators have an upper limit on the throughput at which they can operate, which for micrometer-scale emulsions tends to be <10 ml/h.

An increasingly popular strategy to address this problem has been the incorporation of multiple generators onto a single chip; each device is operated in the low flow rate conditions where they perform best, and throughput can be increased proportionally to the number of devices operated in parallel. Work in this field began with pioneering research by Nisisako et al.^9,10 Over the last decades, there has been a steady march of progress where fabrication strategies and architectures to incorporate increasing numbers of increasingly complex devices have been developed to generate both single- and higher-order emulsions at high throughput.^9–16 This rapidly developing field is now entering an exciting stage of development, where for the first time, devices are reaching commercially relevant flow rates (∼10  l/h) for many applications, including pharmaceutical formulations.¹⁷ These technologies, which can now incorporate ∼10⁴ microfluidic generators on a single chip with a single set of inlets and outlets for fluids, can produce a wide pallet of microfluidic engineered materials at throughputs ∼10⁴ times greater than single microfluidic devices.

In this review, we highlight recent progress in the field of scaling up the production rate of microfluidic chips using parallelization, including works on increasing the throughput of the individual droplet generating devices, works on architectures to incorporate multiple microfluidic devices onto a single chip, and works on the incorporation of multiple unit operations into the parallelized architectures for the high throughput generation of engineered micro- or nanoscale materials (e.g., solvent extraction or UV curing). For each of these topics, in addition to reviewing progress in the field, we will also share our view on the fundamental mechanisms that dictate the challenges and progress in these areas and our perspective on promising future directions.

II. EMULSION GENERATION MECHANISMS CONSIDERATIONS

In microfluidic emulsion generation, a continuous phase and a dispersed phase, or multiple dispersed phases for higher-order emulsions, are brought together in a channel geometry that confines the fluid flows at the micro-scale. The physics of droplet generation in microfluidic devices is a complex phenomenon and there is a rich and thriving sub-field of fluid dynamics that has studied this topic extensively.⁸ In this paper, we will focus on the key features and insights that are relevant to high throughput particle generation; those who are interested in learning about the physics of droplet generations are referred to the excellent reviews found in Refs. 8 and 18. We will discuss four categories of droplet generators: flow-focusing [Fig. 1(a)], co-flow [Fig. 1(b)], cross-flow [Fig. 1(c)], and step emulsification [Fig. 1(d)].

FIG. 1.

Passive microfluidic droplet generator geometries. (a) Flow-focusing generator (FFG). (i) Schematic of a typical FFG device. Microscopic image of droplet formation under (ii) squeezing regime, (iii) dripping regime, and (iv) jetting regime. Reprinted with permission from Kovalchuk et al., Chem. Eng. Sci. 176, 139 (2018). Copyright 2018 Elsevier.¹⁴² (b) Co-flow. (i) Schematic of a typical co-flow device. Microscopic image of droplet formation under (ii) dripping regime, (iii) jetting regime with narrowing jet, and (iv) jetting regime with widening jet. Reprinted with permission from Utada et al., Phys. Rev. Lett. 99, 094502 (2007). Copyright 2007 American Physical Society.²¹ (c) Cross-flow. (i) Schematic of a typical cross-flow device. Microscopic image of droplet formation under (ii) squeezing regime, (iii) dripping regime, and (iv) jetting regime. Reprinted with permission from Zagnoni et al., Langmuir 26, 9416 (2010). Copyright 2010 American Chemical Society.⁴⁹ (d) Step emulsification. (i) Schematic of a typical step emulsification device. Microscopic image of droplet formation under (ii) dripping regime and (iii) jetting regime. Reprinted with permission from Eggersdorfer et al., Proc. Natl. Acad. Sci. U.S.A. 115, 9479 (2018). Copyright 2018 United States National Academy of Sciences.¹⁴³

Due to the complexity of the physics of droplet formation, there generally do not exist exact solutions of governing equations to describe droplet generation, but rather phenomenological models that relate droplet generation phenomenon to characteristic geometry dimensions and dimensionless parameters. The relevant forces involved in droplet generation are typically characterized using dimensionless numbers, such as Reynolds number $Re=\rho vl/\mu$, Capillary number $Ca=\mu v/\sigma$, Weber number $We=\rho v^{2}l/\sigma$, and the flow rate ratio between the dispersed and continuous phases $\varphi =Q_d/Q_c$, where $\rho$ is the fluid's density, $\mu$ its viscosity, and $\sigma$ the surface tension between the dispersed and continuous fluids.¹⁸ The characteristic dimension $l$ and the characteristic velocity $v$ are defined differently depending on the particular droplet generation geometry being studied.¹⁹ The Capillary and Weber numbers are particularly relevant dimensionless numbers for characterizing the throughput of droplet formation, as they relate the relative importance of the viscous and inertial forces to interfacial forces, respectively, which can dictate the flow rate regimes in which homogenous droplets can be generated.^8,20,21 The Reynolds number, which represents the ratio of inertia to viscous force, tends to be small ( $Re$ ≪ 1),²² due to the micrometer-scale channel dimensions. These dimensionless numbers and the characteristic dimensions of the device geometry have been phenomenologically related to droplet size, droplet generation frequency, and the flow rate regimes in which homogenous droplets can be generated for the various device geometries^21,23–26 that we consider in this review [Figs. 1(a)–1(d)].

In addition to the passive methods covered here, there are also active methods in which additional forces are used to control droplet formation, such as electric, magnetic, acoustic, or centrifugal forces.¹⁸ In this review, we focus only on the passive generation techniques rather than active microfluidic generators. Although active generators have shown promising results at the individual droplet generator scale and have potential for parallellization,²⁷ their utility for high throughput materials generation has not yet been extensively explored. In contrast, the generation of single emulsion with the major four geometries has been extensively studied, and characterized to have different operational regimes.^21,28–31 We will discuss the characteristic parameters of the devices and their relationship with the operational regimes and properties of generated emulsions.

A. Flow-focusing generator (FFG)

In FFGs, two immiscible phases meet shortly before flowing through a geometrically confined orifice, and at, or after, the orifice, the inner dispersed phase breaks up into droplets^32–34 [Fig. 1(a)]. The performance of FFGs is controlled by the height of the channel h, the width of the dispersed inlet $w_d$, the width of the continuous inlet $w_c$, the distance between inlet to nozzle $\Delta z$, the width of nozzle $w_n$, and the width of downstream channel $w_o$. These characteristic dimensions define a characteristic velocity $U=G\hspace{0.17em}*\hspace{0.17em}{a}_0$, dependent on an elongation rate $G=\frac{Q_c}{h\Delta z}\left(\frac{1}{w_n}-\frac{1}{2w_c}\right)$, and a characteristic length $a_0=\frac{w_d}{2}$. In this system, the Capillary number for the continuous phase can be expressed as ${Ca}_c=\frac{{\mu }_{c}U}{\sigma }=\frac{{\mu }_{c}G{a}_0}{\sigma }$ and the Weber number for the dispersed phase ${We}_d=\frac{{\rho }_d{U}^2{a}_0}{\sigma }$, where ${\rho }_d$ is the density of the dispersed phase and $\sigma$ is the surface tension between both phases.

At low Ca value (Ca < 0.01), the device operates in a squeezing regime, wherein the orifice becomes blocked by the dispersed phase, which induces a pressure increase in the continuous phase upstream of the orifice that causes the dispersed phase to break off to form a droplet.^35,36 In the squeezing regime, the volume of the droplet scales linearly with the flow ratio $\varphi$ with the relation $V_d=\alpha +\beta \varphi$, where α and β are adjustable parameters based on channel geometries.³⁵ As the Ca increases to an intermediate value ${\left[{Ca}_d\sim \mathcal{O}\left(0.1\right)\hspace{0.17em}\text{and}{Ca}_c\sim \mathcal{O}(0.1)\right],}_{}$³⁷ the droplets break up at the end of the orifice, due to a mixed effect of shear force and Rayleigh instability. This operational mode is called the dripping regime, which is characterized by the dispersed phase not physically blocking the orifice, resulting in the generation of droplets that are typically smaller than the channel size of the orifice.^38,39 In the dripping regime, monodisperse droplets form at a higher frequency than in the squeezing regime, by approximately two orders of magnitude [O(0.1–10 kHz)].^40,41 In this dripping regime, the droplet diameter scales as $d\propto {{Ca}_c^{-1/3}}_{}^{}$, when the droplet size is smaller than the channel height,⁴² and $d\propto {Ca}_c^{-1}$, when it is larger.³⁷ Further increasing the Ca value leads to a transition from the dripping mode to a jetting mode when ${Ca}_c$ > 0.1. In the jetting regime, the domination of the viscous force over surface tension elongates the dispersed phase into a thread, which breaks up at points downstream of the junction due to Rayleigh instability.⁴³ In the jetting regime, droplet size can be predicted as $\frac{d}{h}=3.1{(\frac{Q_d}{{2Q}_c})}^2$, with poor droplet homogeneity compared to the squeezing and the dripping mode.³⁷

To generate droplets for materials synthesis, FFG devices are typically operated in the squeezing or dripping regimes, to produce droplets in the range of $\mathcal{O}$(10–100 $\mu m$) with high uniformity (CV< 5%), and a dispersed phase flow rate of $Q_d$ < 1 ml/h [Fig. 2(a)]. The jetting regime is typically not used for materials generation due to its low homogeneity, including the formation of satellite droplets, leading to a polydispersity of droplets by roughly two- to threefold greater than those generated by squeezing or dripping.⁴⁴ The jetting mode, however, has found utility to synthesize fibrous materials.⁴⁵ Additionally, under specific conditions of surfactant concentration and flow rates, nanometer-sized droplets can be generated using an additional mode of operation for FFG devices coined tip-streaming.³⁸ The FFG geometry has been widely adopted for material synthesis applications, which particularly benefit from the capability to tune the droplet size by controlling the relative flow rates of the inputs.

FIG. 2.

Characteristic features of various step emulsification devices. (a) Microchannel type emulsifier (MC). (i) Schematic of droplet breakup in a single-channel MC. h and w indicate the height and width of the generator, respectively. (ii) Normalized droplet generation frequency dependence on w/h ratio, and corresponding operational regimes (blue circles). Reprinted with permission from Montessori et al., Physics of Fluids 31, 021703 (2019), Copyright 2019 AIP Publishing.²⁵ (b) Triangular-shaped nozzle emulsifier. (i) Top view of the design layout of the device. (ii) Close up image of dispersed phase expanding in the triangular terrace and breakup at the step. Reprinted with permission from Amstad et al., Lab Chip 16, 4163 (2016). Copyright 2016 Royal Society of Chemistry.¹² (c) EDGE step emulsification device. (i) Schematic of an EDGE device with wide plateau. (ii) Schematic of a modified EDGE device with the plateau compartmentalized by square-shaped partitions. Reprinted with permission from Sahin and Schroën, Lab Chip 15, 2486 (2015). Copyright 2015 Royal Society of Chemistry.¹⁴⁴ (d) Grooved-type step emulsification device. (i) Schematic of a grooved-step emulsifier with a hybrid geometry of MC and EDGE structure. (ii) h is the height of the partition, H is the height of the groove, L_g is the width of the partition, and w_g is the width of the device. Partition-to-groove ratio (PGR) is defined as h/H. Reprinted with permission from Opalski et al., Lab Chip 19, 1183 (2019). Copyright 2019 Royal Society of Chemistry.⁶⁷

B. Co-flow

Co-flow devices are most commonly constructed with glass capillaries nested within one another. In the co-flow geometry, the dispersed phase and an immiscible continuous phase are brought together in two coaxial channels [Fig. 1(b)]. The co-flow geometry is not discussed in detail here because the glass capillary fabrication method is not amenable to parallelization. The geometry can be characterized by the diameter of channels that carry the continuous phase, $w_c$ and dispersed phase, $w_d$ and the diameter of the downstream channel $w_o$, respectively. The characteristic velocity of the outer fluid can be represented as $U_c=\frac{4Q_C}{{\pi w_o}^2}$, and the Ca and We are defined as they are in the FFG geometry above. When the flow rates of continuous and dispersed phase are small, with ${Ca}_c$ and ${We}_d$ < O (1), the device operates in the dripping regime, where each droplet grows at the tip of the inner channel until breakup by interfacial force. The droplet size decreases with an increase in $Q_c$/ $Q_d$ ratio, as shear force increases, and scales linearly with flow rate ratio $\varphi =\frac{Q_d}{Q_c}$.²¹ The system transitions to the jetting regime as the Ca and We increase, forming a narrowing jet [Fig. 1(b)(ii)] when viscous shear stresses are balanced by surface tension forces, and a widening jet [Fig. 1(b)(iii)] when inertial forces are balanced by surface tension forces.^20,46 For a widening jet, the droplet size scaling follows $d\sim {(\frac{Q_d}{U_c})}^{0.5}$.

Co-flow devices are capable of generating droplets ranging roughly from 10 to 500 μm, with a throughput typically ranging from 10 μl/hr to 50 ml/h. The utility for applications is similar to FFGs but mostly limited to single-channel operation due to its fabrication involving the use of glass capillaries. Because its downstream channel can be made long, the device is well suited for making fibrous materials by operating the device in the jetting regime.⁴⁷

C. Cross-flow

For cross-flow droplet generators, the dispersed phase intersects a channel carrying an immiscible continuous phase [Fig. 1(c)]. The width of both inlets and that of the downstream channel $w_d{,w}_c,$ and $w_o$ define the operation of cross-flow devices, with the characteristic velocity $U_c=\frac{Q_C}{wh}$, where w and h are the width and depth of the downstream channel.^48,49 Similar to an FFG configuration, cross-flow devices operate in a squeezing regime when ${Ca}_c$ < 0.01,^50,51 in which the dispersed phase periodically occludes the junction and when the pressure becomes sufficiently large to overcome surface tension, a droplet breaks off. The droplet volume V can be expressed as $\frac{V}{h{w}^2}=\frac{V_{\mathit{fill}}}{h{w}^2}+\alpha \varphi$, where V_fill and α are two fitting parameters that are dependent on the channel geometry.⁵² As the Ca increases, shear stress dominates, and the device enters the dripping regime.^31,50 The droplet volume is dependent on the Capillary number following a power law relationship: $V\sim {Ca}_c^{-\frac{1}{3}}.$ The transition from dripping to jetting occurs when the flow rates of both phases are sufficiently large such that ${Ca}_c{\hspace{0.17em}\mathrm{and}\hspace{0.17em}Ca}_d$ > 1, where the dispersed phase stream flows along one side of the channel and breakup downstream into polydisperse droplets due to Rayleigh instability.⁵³

In droplet-based materials synthesis, cross-flow devices are typically used to produce larger droplets than those produced using the FFG geometry, from ∼50 to 500 μm, with a throughput less than 10 ml/h. They have been often used when multiple inlets of reagents are brought together to form droplets with varying compositions, for applications such as reagent screening of protein crystallization.⁵⁴

D. Step emulsification

Step emulsification has been developed with a variety of channel geometries, such as terrace,⁵⁵ trapezoid,⁵⁶ rectangular,⁵⁷ straight-through microchannel⁵⁸ (MC), edge-based droplet generation⁵⁹ (EDGE), and triangular.¹² In these geometries, the dispersed phase flows through a confined channel to a step in channel height where the dispersed phase meets the continuous phase, which rapidly changes the Laplace pressure of the droplet [Fig. 1(d)]. The device performance is characterized by the width of the end of the terrace $w_t$ and height of the step $h$.⁵⁶ Capillary and Weber numbers are used to characterize the device, which are calculated from the viscosity and velocity from the dispersed phase, ${Ca}_d=\frac{{\mu }_d{U}_d}{\sigma }$ and ${We}_d=\frac{{\rho }_d{U_d}^{2}h}{\sigma }$, where the characteristic velocity is the flow rate of the dispersed phase divided by width and length of the nozzle $U_d=\frac{Q_d}{w_{t}h}$ . The droplet size is primarily dependent on device geometry and wetting conditions and is insensitive to the change in flow rate because the breakup relies only on Rayleigh-Plateau instability.⁶⁰ Monodisperse droplets form in a dripping regime, when Ca is smaller than the critical Capillary number Ca*, which is a function of the device geometry ${Ca}^{*}=\frac{\pi }{6}\frac{{d_0}^3}{\mathit{hwa}}\frac{{\lambda }^{0.8}}{(1+\alpha \lambda )(1-\beta {\lambda }^{-0.2})}$. Here, λ is the viscosity ratio of dispersed and continuous phases, and α, β, and a are fitting parameters dependent on drop breakup timescale and channel dimension.⁶¹ Droplets form in the dripping regime when Ca_d is less than 10⁻² and We_d is less than 10⁻³.⁶¹ The volume of the droplet scales linearly with the step height h when operating in the dripping regime.⁶² As flow rate increases, the device transitions into jetting regime, producing larger droplets with lower uniformity.^30,63

There have been several seminal papers published on the scaling of the performance of step emulsifiers and their geometries. MC (microchannel) is the most widely adopted geometry for step emulsifiers, where the dispersed phase is injected through a long, narrow microchannel to meet continuous phase. A key feature that governs the throughput in the MC type step emulsifiers is the width-to-height (w/h) ratio of the microfluidic nozzle [Fig. 2(a)(i)]. Kobayashi et al. showed an oblong-shaped nozzle with an aspect ratio larger than 3 produces more uniform droplets than nozzles with low aspect ratio or a circular nozzle.^57,64 Montessori et al. have shown that the droplet breakup in a step emulsifier undergoes a topological breakup when w/h is larger than a critical value of 2.6, since the fluid experiences a series of fluid-dynamical rearrangements, while the nozzle is blocked by the isotopically growing liquid jet, when w/h is smaller than 2.6 [Fig. 2(a)(ii)].²⁵ In addition, the droplet generation frequency is linearly dependent on the width w, with a fixed h. In a triangular-shaped nozzle device (millipede device), it is found that the w/h value needs to be between 5.5 and 19 such that the droplet precursor can maintain a cylindrical shape, to produce droplets with narrow size distribution¹² [Fig. 2(b)]. Within this range, the maximum flow rate before dripping-to-jetting transition scales with h^3/2. Moreover, the authors have shown that increasing the step height can shift the dripping-to-jetting transition, thus increasing the maximum throughput of the device. Using this device, 20–160 μm droplets are formed with a CV around 3%.

EDGE devices adopt a design geometry where the dispersed phase flows over a wide plateau (>500 μm) to meet the continuous phase, and droplets form at multiple locations along the edge of the plateau [Fig. 2(b)]. An increase in the applied pressure on the dispersed phase increases the number of droplet formation locations until the plateau is filled, and further pressure increase results in polydisperse droplets.^24,65 The ratio between the height of the plateau and droplet size, called the scaling factor, is used to characterize the device performance. A scaling factor of 5.5–6.5 is found for a conventional EDGE device [Fig. 2(b)(i)], and 4.5 for a modified EDGE device, with square-shaped partitions added to the edge of the main plateau [Fig. 2(b)(ii)], while achieving a higher droplet generation frequency by two orders of magnitude.⁶⁶ In a grooved step emulsifier, which has a hybrid geometry of MC and EDGE architectures⁶⁷ [Fig. 2(c)(i)], a geometrical parameter, the partition-to-groove ratio (PGR) is defined; a PGR of 0 represents MC (no partition) and that of 1 represents EDGE (full partition) [Fig. 2(c)(ii)]. The droplet size increases with PGR and is relatively invariant with the flow when PGR is less than 0.5. A PGR value of 0.2–0.3 is optimal for generating monodisperse droplets, at a throughput > 3 times than an EDGE device with the same footprint.

For a single-channel step emulsification device, the flow rate for making droplets in the dripping regime is typically less than 10 ml/h for a droplet diameter on the order of 100 μm,⁶⁸ less than 100 μl/h for droplet diameter of the order of 10 μm,⁶³ and less than 1 μl/h for sub-ten micrometer droplets.⁶⁹ Step emulsification devices are particularly useful when there is fluctuation in the flow rates, for example, from fluctuations that may result from pumps, because the droplet size is insensitive of flow rates.

E. How to choose which type of droplet generator to use

While all four geometries discussed above—flow-focusing, co-flow, cross-flow, and step emulsification—can produce droplets with high uniformity (CV< 5%),¹⁸ we would like to share a framework for choosing which type of droplet generator to be used for specific applications. We have plotted the droplet size and the maximum throughput (i.e., operating at their maximum Ca and We numbers before the drip-to-jet transition), for several published works (Fig. 3).^{37,41,49,63,70} By visual inspection, there are several useful trends that emerge. When producing droplets with similar diameter, FFG geometry has superior throughput compared to other geometries because the dripping-to-jetting transition occurs at a higher frequency due to the enhanced shear effects from geometrical confinement of the nozzle. This advantage is amplified when the device is parallelized because the throughput increases in an individual device is multiplied by the number of devices.⁷¹ Compared to step emulsification, the FFG geometry is more suitable for producing higher-order emulsions since it is amenable to the incorporation of multiple phases, allowing precisely metered dimensions of higher-order emulsions. Cross-flow geometry can be advantageous when plugging and squeezing are desired to utilize the wall effect for mixing reagents or when multiple nozzles are used to create alternative drop pairs, to prevent premixing of the reagents.⁷² Step emulsification most differentiates itself from the other geometries by its flow-invariant drop generation,⁷³ making it particularly suitable for applications where the instrumentation required to provide precise flow rates to the input is impractical (e.g., expensive pumps, flow meters, etc.), such as point of use or portable applications. Furthermore, for processing droplets in a multi-step manner, a combination of droplet generators and droplet processing units, such as pico-injectors, can be adopted to process droplets in series.⁷⁴

FIG. 3.

Comparison of performance between FFG, cross-flow, and step emulsification devices. All data points in the graphs correspond to the device operating before the drip-to-jet transitions. (a) Plot of the generated droplet size with corresponding dispersed phase flow rate. (b) Plot of the droplet generation frequency with corresponding dispersed phase flow rate data points. Reprinted with permission from Ward et al., Electrophoresis 26, 3716 (2005). Copyright 2005 John Wiley and Sons, Inc.⁷⁰ Reprinted with permission from Cubaud et al., Physics of Fluids 20, 053302 (2008). Copyright 2008 AIP Publishing.³⁷ Reprinted with permission from Yobas et al., Lab Chip 6, 1073 (2006). Copyright 2006 Royal Society of Chemistry.⁴¹ Reprinted with permission from Zagnoni et al., Langmuir 26, 9416 (2010). Copyright 2010 American Chemical Society.⁴⁹

III. PRINCIPLES OF SCALING UP EMULSION PRODUCTION

There are a number of strategies to increase the throughput of a single-channel microfluidic droplet generator, by altering the geometry such that the dripping-to-jetting transition occurs at a higher flow rate. These strategies include performing parameter sensitivity analysis (e.g., ANOVA) to optimize channel geometries,²³ separating the processes of droplet formation and droplet stabilization using two sequential cross junction channels,⁷⁵ or including droplet splitters downstream of the droplet generators.⁵¹ However, the orders of magnitude increase in throughput necessary to translate microfluidic droplet–based material production for commercial applications cannot be achieved by altering the design of the individual droplet generator. Although operating multiple chips separately is fundamentally viable, it is impractical and not economical to build instrumentation to operate the N > 1000 individual chips necessary to achieve the required throughputs, as each chip will require its own set of pumps and tubing to drive fluid through the device and collect its output. Because of the need for log orders improvement of throughput and the impracticality of operating thousands of chips, the field has been highly motivated to develop chips with many microfluidic devices that can be used in parallel with a single set of inlets and outlets.

A. Design principles governing the throughput of parallelized flow-focusing and cross-flow devices

The first consideration for parallelizing a droplet generator is choosing the scheme by which fluids are distributed to and collected from the individual droplet generating devices. The two most commonly used architectures are the tree network and the ladder network^76,77 [Fig. 4(a)]. In the tree network, each phase enters from one inlet, the flow branches out into two separate channels, and each of those two channels breaks into two additional channels, allowing the flow to be distributed to 2^N channels for a tree with N layers of branching.⁷⁸ Because of the symmetry of the branching geometry, the fluidic resistance of the chip's input, or output, to any of the individual devices is the same, and as such, each device is driven with identical flow rates or pressures.

FIG. 4.

Schematic representations of two common layouts of parallelized microfluidic channel geometries. (a) Eight-channel parallel design with tree geometry. (b) Eight-channel parallel design with ladder geometry. Both layouts have a set of two inlets for both phases. Continuous phase is marked in yellow, dispersed phase is marked in blue, and emulsion phase is marked in green. Reprinted with permission from Tetradis-Meris et al., Ind. Eng. Chem. Res. 48, 8881 (2009). Copyright 2009 American Chemistry Society.⁷⁶

In the ladder geometry, the droplet generators are organized in a series of rows [Fig. 4(b)]. For each row, there are delivery channels that run the length of the row, in a different vertical layer of microfluidics than that of the droplet generators, which deliver fluid to each of the individual device inlets and collect each of their output. The individual droplet generators are connected to these delivery channels through vertical vias that connect between the microfluidic layers. Multiple rows can be connected to form a two-dimensional array of generators using arterial lines that supply or collect fluid to/from each of the delivery channels. In the ladder geometry, the flow resistance of the delivery channels is designed to be orders of magnitude lower than that of the individual droplet generators. As such, the fluidic resistance of the chip's input, or output, to any of the individual devices is not significantly different, and as such, each device is driven with negligible differences in flow rates or pressures.

In recent years, the ladder geometry has been more popular than the branching geometry due to two key advantages.^13,79,80 First, the ladder geometry requires a significantly smaller footprint (area of entire chip) than the tree geometry, and this difference becomes increasingly pronounced as the number of individual devices on a chip grows. For example, Tetradis-Meris et al. showed a 12-fold smaller footprint of the chip for the same number (N = 180) of individual droplet generating devices.⁷⁶ Second, the tree geometry is more sensitive to clogging and is more sensitive to small errors in fabrication. This sensitivity arises from the fact that the tree geometry relies on symmetry to distribute the flow evenly, and any variation in the branching network geometry arising from fabrication variability or from device clogging can cause the flow distribution amongst the devices to become non-uniform. In contrast, in the ladder geometry, because all of the devices are connected in parallel, a clog in one of the devices has a small and evenly distributed effect on the distribution in fluid flow to the unclogged devices.

To design chips using the ladder geometry, a key design goal is to ensure that each droplet generator produces uniform droplets by making sure that each device is driven with identical flow rates for each of the phases. Romanowsky et al. proposed a design rule to ensure that ladder geometry–based parallelized devices meet these criteria.¹⁵ This design rule ensures that the flow resistance from the inlet or outlet to any device in the row is not significantly different, i.e., that each individual droplet generator operates as if it were connected in parallel to the inlet or outlet.

$$2N\left(\frac{\mathrm{R}\mathrm{D}}{\mathrm{R\{dev}}\right)<0.01.$$ (1)

The fluid resistance in a rectangular channel can be calculated, assuming that the height h of the channel is less than that of the width w,²²

$$R=12\left(\frac{\mu L}{w{h}^3}\right){\left(1-0.63\left(\frac{h}{w}\right)\right)}^{-1}.$$ (2)

Microfluidic droplet generators in a ladder geometry can be modeled using a lump circuit model [Fig. 5(a)]. In this model, Q_i is the total flow rate flowing through each individual device, R_D is the resistance of the delivery channel between each generator, and R_dev is the resistance of the droplet generating device. Because the flow ratio between the first and second channel follows the relationship Q₁/Q₂ = 1 + 2(R_D/R_dev), the resistance ratio between the first and N^th device, scales as 2N(R_D/R_dev). Therefore, by satisfying the design rule (Eq. 1), the difference in the flow rates among all the devices is small, which ensures uniform distribution of the fluids among N drop generators.

FIG. 5.

Illustrations of a lump circuit model. (a) Hydrodynamic resistance in ladder geometry. R_D indicates the resistance of distribution channel between each two generators and R_dev indicates that of the drop generator channel. Q₁-Q_N represent the flow rates of each device. (b) Addition of flow resistor in ladder geometry. R_R indicates the resistance of flow resistor and R_μF indicates the resistance of the drop generator. R_dev indicates the sum of the resistance of an individual device. (c) An SEM image of a drop generator representation of circuit shown in (b). Reprinted with permission from Yadavali et al., Nat. Commun. 9, 1222 (2018). Copyright 2018 Springer Nature.⁷¹

One challenge of the ladder geometry is that there is a design trade-off between the design considerations to incorporate greater numbers of microfluidic droplet generators onto a single chip and the design considerations to make individual generators have higher throughput. The design considerations to incorporate more generators on a chip (Eq. 1) encourage devices with increased flow resistance, and thus smaller channels and nozzles. However, the design considerations to increase the throughput of individual devices encourage larger channels and nozzles such that higher throughput can be achieved in the dripping regime (i.e., within the constrains set by the maximum Capillary and Weber numbers). This trade-off limits the number of devices that can be incorporated onto a chip before there are diminishing returns. Yadavali et al. developed an approach to break this trade-off by incorporating fluid resistors upstream of the drop generators [Fig. 5(b)].¹⁴ The addition of the resistors decouples the device resistance R_dev to the channel geometry by adding a flow resistor R_R upstream that has much greater resistance than the droplet generating device. With the addition of these flow resistors, the droplet generators can be designed with larger cross sectional areas, allowing operation at higher volumetric flow rates before reaching the dripping-to-jetting transition point.^14,71 Moreover, the use of upstream fluidic resistors decouples the design of the individual fluidic generators from the considerations of parallelization, thus allowing a modular parallelization approach that can be readily adapted to any individual microfluidic droplet generator design without the need for redesign.

IV. DEVICE FABRICATION AND MATERIALS CONSIDERATIONS

Scaling up of the throughput of microfluidic droplet–based materials synthesis poses unique device microfabrication challenges. For example, variance in the dimensions of the microfabricated droplet generating devices affects droplet polydispersity. The device materials chosen (PDMS, silicon, glass, etc.) can constrain the choice of solvents and limit the operation temperature and pressure. Additionally, the ladder geometry architecture, which requires connections between multiple layers of microfluidic channels, demands more sophisticated multilayer fabrication than conventional single-layer microfluidics.⁸¹ The surface wetting properties of the device play an important role in stable generation of droplets, and in some cases, the hydro-philicity/phobicity must be patterned on the chips such that specific areas of the chip have different wetting properties. Depending on the application, each fabrication strategy has its strengths and weaknesses. We review the various fabrication strategies that have been applied to high throughput microfluidic droplet–based materials synthesis.

A. Elastomer-based microfabricated devices

Elastomer, in particular, poly(dimethylsiloxane) (PDMS), is the most commonly used material for making microfluidic chips. PDMS devices are typically fabricated using soft lithography, which has been widely adopted because of its high modularity and accessibility.⁸² Soft lithography can achieve minimum channel dimension of less than 1 μm and an aspect ratio as high as 1:10.^83,84 The high elasticity of cross-linked PDMS allows tunable deformation of the device, enabling the incorporation of microvalves that can be used to pinch the nozzle in droplet microfluidic devices, and thus controlling the diameter of generated droplets.⁸⁵

Devices with varying channel heights can be constructed in PDMS by performing multiple rounds of photolithography on the SU-8 molds, creating channels with different heights.⁸⁶ For example, the millipede chip is constructed through a two-layer soft lithography fabrication process to create the step height.¹² Furthermore, Hati et al. used a three-layer lithography process to implement a three-dimensional shunt design based on the millipede device [Fig. 6(a)(i)]. The addition of the shunt introduces backflow from the continuous phase, facilitating the droplet breakup process⁸⁷ [Fig. 6(a)(ii)]. Under the same condition, the flow rate threshold for dripping-to-jetting transition can be increased up to 75% with the shunt design, compared with conventional millipede device. Additionally, the shunt structure allows the device to process viscous fluids (155cP).

FIG. 6.

Elastomer-based parallelized high throughput droplet generation devices. (a) (i) Parallelized step emulsification chip with modified triangular nozzles with shunt design created by multilayer lithography. (ii) Schematic showing the backflow from continuous phase in the shunt channel. Reprinted with permission from Håti et al., Lab Chip 18, 648 (2018). Copyright 2018 Royal Society of Chemistry.⁸⁷ (b) A PDMS chip with 1000-channel parallelized FFGs. (i) “Double-sided imprinting” fabrication process flow for making device with 3D geometry. (ii) Close up images of the channel features of the chip. Reprinted with permission from Jeong et al., Lab Chip 15, 4387 (2015). Copyright 2015 Royal Society of Chemistry.¹³

Fabrication of PDMS devices with a ladder geometry poses a fabrication challenge beyond multilayer soft lithography because it requires the three-dimensional fabrication of multiple layers of fluid channels, which can cross one another. To overcome this challenge, Jeong et al. proposed a double-sided imprinting fabrication strategy and used it to integrate 1000 FFG units on a 30 cm² single device.¹³ In double side imprinting, uncured PDMS is sandwiched by a hard SU-8 master with multi-height features and a soft PDMS master to create vias, distribution channels, and droplet generators onto a single PDMS piece [Fig. 6(b)]. The alignment between both masters can be performed under a microscope with the uncured polymer acting as a lubrication layer. The authors demonstrated a generation of > 30 × 10⁹ droplets per hour (∼1500 ml/h) with a diameter as small as 36μm with a CV around 7%.

Despite the merits of PDMS soft lithography, there are limitations on using it for massively parallelized droplet generation. Parallelized chips often have to be operated at high pressures (>0.7MPa), in particular when devices use flow resistors to aid parallelization. These high pressures make the commonly practiced plasma bonding, which provides a maximum bonding strength of ∼0.5MPa, untenable.⁸⁸ The choices of solvents are also limited because of PDMS's incompatibility with many organic solvents.⁸⁹ Although strategies to make PDMS more solvent compatible have been developed, their applicability to parallelized devices has not been extensively tested.⁹⁰ In a recent study, perfluoropolyether (PFPE)-based elastomer, more solvent compatible than PDMS, was molded using soft lithography to fabricate microfluidic devices. The surface wetting properties of these PFPE devices can be changed by using PFPE–poly (ethylene glycol) (PEG) copolymer elastomer to enable parallelized generation of oil-in-water or water-in-oil emulsions.⁹¹

An additional limitation of soft lithography-based parallelization is variance in device dimensions that arise from fabrication error, such as non-uniformity in SU-8 thickness or non-uniform development/UV illumination during master fabrication. Previous work has studied how variations in channel dimensions in a parallelized chip can cause variations in hydrodynamic resistance and flow rates (Fig. 7), which in turn lead to non-uniformities in the generated emulsion diameters (CV  > 5%).⁹²

FIG. 7.

Parallelized PDMS device channel dimension variation. (a) Schematic of channel dimensions and their variations of a single droplet generator. (b) Heat map representation of the variations of channel resistance and generated emulsion diameter among 400 channels on a chip due to fabrication process. A total of three chips are shown from top row to bottom row. Left images indicate the resistance of the orifices, middle images indicate the resistance of the outlets, and right images indicate the droplet diameters. Reprinted with permission from Jeong et al., Lab Chip 17, 2667 (2017). Copyright 2017 Royal Society of Chemistry.⁹²

B. Hybrid-machined and microfabricated devices

Several fabrication techniques have been developed that offer alternatives to soft lithography, which have benefits for applications detailed below and offer an advantage when a clean room facility cannot be easily accessed. These methods include thermomoulding, 3D printing, and CNC milling with materials such as thermosets, thermoplastics, paper, hydrogel, or a hybrid of these materials.⁸³ Some of the methods and materials have already been demonstrated for fabricating high throughput parallelized devices, which are highlighted in this section.

Conchouso et al. used CNC milling to create microchannels on four separate layers of PMMA [poly(methylmethacrylate)] pieces and stacked them to form the final device.⁹³ In this device, 512 FFGs are parallelized using a tree geometry with a circular-array arrangement, achieving production of 160 μm diameter droplets at a maximum throughput of ∼1 l/h [Fig. 8(a)]. The use of precision CNC milling eliminates the need of a clean room facility, making prototyping more accessible. The high stiffness of PMMA provides rigidity to the device, allowing a quicker flow response and an ability to operate at higher pressures (up to 0.75MPa), compared to elastomer devices. However, CNC milling has limited resolution (hundreds of micrometers),⁹⁴ and creates variations in channel geometry that can cause a high polydispersity (CV ∼ 20%).

FIG. 8.

Hybrid-machined and microfabricated high throughput droplet generation devices. (a) PMMA-based droplet generation chip parallelized with 512 flow-focusing channels. (i) Schematic illustration of multilayer geometry and assembly of the device. (ii) Photograph of the chip showing inlets and outlets (top) and microscopic images of the flow-focusing nozzle (bottom). Reprinted with permission from Conchouso et al., Lab Chip 14, 3011 (2014). Copyright 2014 Royal Society of Chemistry.⁹³ (b) Parallelized flow-focusing device fabricated by stereolithography. (i) Schematic illustration and micrograph of device with 28 droplet generators, with a zoom-in image showing parallel droplet generation. (ii) Microscopic images of the channel and vias. Scale bar is 300μm. (iii) Microscopic images of the generated water droplets in oil (left) and microgels (right). Scale bars are 500μm. Reprinted with permission from Femmer et al., ACS Appl. Mater. Interfaces 7, 12635 (2015). Copyright 2015 American Society of Chemistry.⁹⁹ (c) Hybrid-machined step emulsifier using FEMTO laser. (i) Schematic illustration of device fabrication process flow. (ii) Rendering of parallelized droplet generators with 192 nozzles. (iii) Microscopic image of device during operation. (iv) Histogram of droplet diameters produced at different flow rates. Reprinted with permission from Tovar et al., Biomicrofluidics 12, 024115 (2018), Copyright 2018 AIP Publishing.¹⁰⁰

3D printing is an emerging fabrication method for microfluidic devices. The technology allows for low-cost fabrication of 3D structures with continuously improving reslolution.⁹⁵ Although not widely adopted, emulsion generators have been 3D-printed in recent years and have demonstrated the capability to produce single or double emulsions.^96–98 Femmer et al. demonstrated a parallelized device using stereolithography, where a bath of photoresin precursor is polymerized by focused UV light to construct the device in a layer-by-layer fashion⁹⁹ [Fig. 8(b)]. The device consists of 28 parallel channels with circular cross sections with a minimum channel size of ∼200μm that can generate droplets and microgels with a diameter of approximately 500μm, with a CV of ∼5%. Because the device is made using methacrylate, it has solvent compatibility with alkaline, dilute inorganic acids, and aliphatic hydrocarbons. Given the current rapid development and advances in 3D printing technologies, we expect this area will continue to evolve to produce smaller droplet diameters and achieve higher uniformity than currently possible.

Hybrid microfluidic devices combine multiple types of materials and fabrication methods to create multilayer device or non-conventional features, such as channels with variable heights. For example, Tovar et al. fabricated a multi-height PDMS device using femtosecond laser machining and wet etching¹⁰⁰ [Fig. 8(c)]. By changing the repetition rates and speed, the FEMTO (femtosecond) laser can pattern channels at a sub-ten micrometer scale with a channel tolerance of ±1μm and ±3 μm for width and height, respectively. In this process, the glass substrate is subsequently developed in acid and used as a replica mold for PDMS. This work demonstrated a single-step fabrication of a 192-nozzle step emulsifier with a step jump of 100 μm, producing ∼70μm droplets with 4% CV.

C. Silicon- and glass-based microfabricated devices

Silicon (Si) and glass offer unique advantages for fabricating microfluidic devices for droplet-based materials synthesis. By taking advantage of the established fabrication technologies developed by the semiconductor industry, Si and glass can be processed with standard photolithography and etching techniques.¹⁰¹ Moreover, both Si and glass have excellent solvent compatibility that allow the use of acids, bases, and solvents not compatible with polymeric devices, and can be operated under high temperature (>600 °C) and high pressure (> 7 MPa) conditions, owing to anodic and fusion bonding techniques that can provide high bond strength.¹⁰² Si and glass fabrication also provides high uniformity¹⁰³(<500 nm error) over device features, leading to uniform droplets (CV< 3%), and has the capability to be adapted to channels with features below 100 nm.⁸³

The early use of Si as parallelized microfluidic emulsifiers can be traced back to the early 1990s, when Si and glass step emulsifiers were reported.^55,104 In one example, a straight-through microchannel emulsifier was developed by etching from both sides of a Si-on-insulator (SOI) wafer using deep reactive ion etching (DRIE). The two sides are joined by etching through a 1 μm deep box layer, forming a large number of through Si vias (TSVs) [Fig. 9(a)]. This process enables fabrication of through-holes with various channel aspect ratios that are used to generate emulsions from sub-ten to hundreds of micrometers with throughput of up to liters per hour.^57,58,105 In another example, Kobayashi et al. have developed an asymmetric straight-through chip with 24 772 devices on a 40 × 40 mm² chip, demonstrating a production rate of up to 1.4 l/h of monodisperse droplets with diameter of 90 μm and a CV < 2%.¹⁰⁶ A more recent study used a similar structure with a total number of 173 173 devices and demonstrated production of droplets with 5 μm diameter. However, the throughput in this device is compromised by the small microchannel size, and the corresponding critical Capillary number below which uniform droplets are formed, and results in a total throughput of less than 5 ml/h. In another type of step emulsifier implemented in Si and glass, a terrace structure is fabricated that is connected by microchannels. Using such an approach, a chip with 364 triangular nozzles has been developed by Ofner et al.¹⁰⁷ [Fig. 9(b)]. Two pieces of glass wafers are lithographically patterned and wet-etched by hydrofluoric (HF) to form shallow and deep channels, in the two glass substrates respectively, followed by thermal bonding of the two wafers. This chip is capable of generating 80μm diameter droplets with a CV= 2.8%, at a maximum throughput of 25ml/h before transitioning to the jetting regime. A follow-up work by the same group showed that double emulsions can be formed by re-injecting emulsions from one chip to another.¹⁰⁸

FIG. 9.

Silicon- and glass-based parallelized high throughput droplet generation devices. (a) Straight-through microchannel (MC) chip containing 92 575 channels. (i) Schematic illustration of the MC chip from side view and top view with dimensions. (ii) Step-by-step fabrication process flow of the silicon MC chip with oblong channels. Reprinted with permission from Kobayashi et al., Microfluid Nanofluid 4, 167 (2008). Copyright 2008 Springer.⁵⁸ (b) Glass-based parallelized step emulsification chip. (i) Schematic illustration of the chip with 364 nozzles. (ii) Chip fabrication process via wet etching. (iii) Chip maximum throughput before drip-to-jet transition and corresponding droplet size. Reprinted with permission from Ofner et al., Macromol. Chem. Phys. 218, 1600472 (2017). Copyright 2017 John Wiley and Sons, Inc.¹⁰⁷ (c) Parallelized silicon-and-glass chip with 10 260 flow-focusing generators. (i) Photograph of the chip with two sets of inlets and outlets. (ii) SEM image showing the 3D ladder geometry in the chip. (iii) Step-by-step fabrication process flow based on dry etching. Reprinted with permission from Yadavali et al., Sci. Rep. 9, 12213 (2019). Copyright 2019 Springer Nature.¹⁴

Both parallelized FFG and cross-flow microfluidic chips have also been implemented in Si and glass. Yadavali et al. developed a high throughput device based on Si and glass, incorporating more than 10 260 droplet generators on a 4-inch diameter wafer,⁷¹ and in a more recent version, 20 160 generators.¹⁴ In this work, by adding flow resistors upstream of each FFG or cross-flow device, as described in Chapter 2, uniform flow distribution can be maintained throughout the chip, and dripping-to-jetting transition regime is shifted to high flow rates, enabling an overall throughput above 6 L/h for droplets with diameter as small as 20μm. This three-dimensional device is fabricated using multilayer lithography and dry etching using DRIE. To enable this fabrication, a novel TSVs approach was developed to achieve a small footprint to allow a high density of features to be packed into a single 4-inch wafer and minimize debris generation and wafer warping for reliable bonding of the Si chip to glass¹⁴ [Fig. 9(c)]. The Si chip is encapsulated by glass wafers through anodic bonding, allowing the device to operate at high pressure P_max > 100psi.

D. How to choose which parallelization approach to use

Several factors determine the number of parallelized drop makers, the choice of material, and the fabrication technique used for parallelization. PDMS and hybrid chips are often most suitable when ∼10²–10³ droplet generators and a throughput below ∼10⁹ droplets per hour are required. PDMS chips are capable of producing droplets with good uniformity (3% < CV< 10%) and can be rapidly prototyped using widely accessible soft lithography. Si and glass devices are more suitable when a higher throughput (10¹³ droplets per hour), excellent emulsion uniformity (CV< 3%), or solvent compatibility are required. Additionally, Si/glass devices are ideal when high temperature or pressure are required for the processes. However, Si/glass devices require access to the microfabrication facilities which may not be readily available. Alternatively, microfabrication methods, such as laser cutting and 3D printing of materials like PMMA or PFPE, can be chosen for rapid prototyping, with some compromise on the device precision and emulsion uniformity.

V. APPLICATIONS OF MATERIALS GENERATED FROM PARALLELIZED MICROFLUDIC DEVICES

Great progress has been made using parallelized microfluidics to generate an ever-expanding menu of precisely defined materials at high throughputs (Table I). From an industrial perspective, for these approaches to be adapted, microfluidic generated particles need to have unique, valuable properties that cannot be obtained through conventional, often less expensive, methods.¹⁰⁹ In the past decades, researchers have successfully demonstrated the synthesis of a wide range of materials with superior functionality for a wide range of applications using droplet microfluidic platforms. The produced particles include porous polymeric microparticles, metal nanoparticles, oxides, quantum dots, and nanocrystals.^27,110–116 We highlight opportunities and challenges in device development related to increasing the throughput of the generation of materials with high potential for commercial use.

TABLE I.

Comparison of parallelized droplet generation devices in recent years.

Flow-focusing/cross-flow
Reference	Measured dispersed phase throughput (l/h)	Dispersed phase per area (l/m2 *h)	Emulsion Diameter (μm)	CV (%)	Number of parallelized channels
10	7.3	200	96.4	1.3	256
113	6	89	90.7	2.2	144
79	3	9	86.1	9	512
93	1.5	32	100	6–20	128–512
99	0.008	429	500	4.8	28
13	1	500	45	6.6	1000
71	0.32	930	22.5	3	10 260
14	0.18	764	21	4	20 160
Step emulsifier
114	10	10	26	4	23 489
106	1.4	2431	87	2	23 348
12	0.0600	600	75	3	550
69	0.03	3	5	5	176 176
68	0.15	8	30	5	400
	0.025	2667	1000	5	120
107	0.0058	28	80	2.8	364
115	0.0015	0.0714	25	5	1850
116	0.00005	N. A	243	4	300

One promising area in the application of parallelized microfluidics to materials generation is in the area of polymeric particles, with many applications where microfluidics adds great value, such as long-lasting pharmaceutical injectables, cell encapsulation, tissue engineering scaffolds, and solid surfactants. One of the earliest examples of this work is the synthesis of bi-colored Janus particle demonstrated by Nisisako and coworkers⁹ [Fig. 10(a)]. The Janus-type droplet can be achieved by co-flowing two dyed monomer streams using a Y-junction upstream of FFG droplet generators. Benefiting from the low Reynolds number flow in microfluidic devices, the mixing of the two monomers in the droplet is only diffusive, which provides enough time to thermally polymerize the particles, locking in the Janus structure. Sixteen of these Janus particle–generating devices were integrated onto a single chip, and a maximum throughput of 20 g/h was achieved. More recently, Lan et al.¹⁴⁵ demonstrated a different approach to prepare snowman-like Janus particle with green materials derived from plants, which can be used to stabilize multiphasic mixture as solid surfactants [Fig. 10(b)]. After being generated in a parallelized FFG chip, the pre-polymer droplets are exposed to UV off-chip, which leads to polymerization and phase separation, forming the asymmetrical particles. The synthesis is scaled-up using an Si-glass microfluidic device with over 10 000 parallel droplet generators, achieving liter-scale production.

FIG. 10.

Microfluidic generated functional particles using parallelized devices. (a) Bi-colored Janus particles synthesis. (i) Microscopic image of 16 parallel Y-junction droplet generators. (ii) Microscopic image of Janus particles with two colors. Reprinted with permission from Nisisako et al., Lab Chip 12, 3426 (2012). Copyright 2012 Royal Society of Chemistry.¹¹³ (b) Large-scale synthesis of green Janus particles. (i) 10 260 parallel flow-focusing parallel channel generation of pre-polymer droplets. (ii) Microscopic image of Janus particles after off-chip UV polymerization. Reprinted with permission from Lan et al., ACS Sustain. Chem. Eng. (2020).¹⁴⁵ Copyright 2020 American Society of Chemistry.¹⁴⁴ (c) Cell-encapsulated microgel synthesis. (i) Schematic of 8-channel flow-focusing droplet generation device. (ii) Image of generated PEG-4MAL microgels with cells encapsulated. Green and red color represent live and dead cells, respectively. Reprinted with permission from Headen et al., Microsyst. Nanoeng. 4, 17076 (2018). Copyright 2018 Springer Nature.¹¹⁹ (d) pH-induced cross-linking of microgels using a 200-channel step emulsification device for building microgel scaffold. Reprinted with permission from de Rutte et al., Adv. Funct. Mater. 29, 1900071 (2019). Copyright 2019 John Wiley and Sons, Inc.¹²⁰

Hydrogel microparticles have shown great potential for numerous biomedical applications from building tissue scaffolds, to cell therapies, to 3D bioprinting.¹¹⁷ Producing microparticles at greater throughputs than is possible with single microfluidic devices is crucial to meet the needs of constructing scaffolds for tissue engineering (typically > cm),¹¹⁸ or to increase cell viability by decreasing the encapsulation time. To this end, an eight-channel chip was developed by Headen et al. for cell encapsulation in microgels¹¹⁹ [Fig. 10(c)]. A suspension of cells in four-arm PEG (PEG-4MAL) was emulsified in eight parallel FFGs. Downstream of each FFG, the droplets were brought into contact with another stream of continuous phase that contain emulsified cross-linkers to initiate the polymerization. The authors have shown high cell encapsulation concentration (10⁷ cells per ml) in microgels smaller than 100 μm, which enables immobilization in small blood vessels. De Rutte et al. used a millipede device with 200 channels to fabricate microgel particles as building blocks for a tissue construct¹²⁰ [Fig. 10(d)]. In this device, PEG-based microgels are cross-linked by adding base in the oil phase to induce a pH change. By changing the monomer concentration, the stiffness of the microparticles can be controlled. Moreover, the flow-invariant nature of the millipede device provides uniform microparticle production for over 12h at a maximum production rate of ∼25 ml/h.

Although particle synthesis has been parallelized as highlighted in the last paragraph, most of these examples rely on off-chip solidification of droplets which may cause emulsions to lose their uniformity, lead to unwanted non-uniformities within particles,¹²¹ or limit the types of particles that can be produced. In addition to ensuring particle uniformity, on-chip solidification can enable precise control over the kinetics of solidification such that the properties and morphology of microparticles can be tailored in ways that are not possible with off-chip bulk solidification methods. For example, on-chip microfluidic synthesis of macroporous materials, i.e., materials with pores with diameters greater than 50 nm, has attracted growing interest as it provides an opportunity to fabricate microparticles with precisely defined macroporous properties.¹¹² Complex structures, such as radially hierarchical porous^122,123 and inverse opal,^124,125 have been engineered via different microfluidic processes, which provide control over material properties not attainable with conventional methods. Here, we describe some of the on-chip solidification processes that have been demonstrated, mostly on single-channel devices.

Three main categories of on-chip solidification of droplets for particle synthesis have been reported. The first type relies on the extraction of the solvent or reactant in or out of the droplet. Typically, this method involves the dissolving of a compound in a volatile solvent followed by removal of the solvent to form solid particles. For example, PLGA [poly (lactic-co-glycolic acid)] microspheres [Fig. 11(a)] are formed through evaporation of the solvent from droplets, allowing PLGA to polymerize into particles. By controlling the solute/solvent [PLGA/DMC (dimethyl carbonate)] concentration, particles with precisely defined diameters (5–30μm) can be prepared from the emulsion templates. To synthesize nanometer-sized (>70 nm) PLGA particles, solvent extraction can be used by merging PLGA–DMSO and water droplets, wherein DMSO (dimethyl sulfoxide) is extracted into the water and PLGA particles precipitate due to their low water solubility.¹²⁶ In a different study, mesoporous silica microspheres with controlled surface morphologies have been generated through diffusion-induced self-assembly [Fig. 11(b)]. Droplets containing ethanol, TEOS (tetraethyl orthosilicate), P104 (Pluronic^® P104), and dilute HCl are emulsified in hexadecane, a continuous phase that has high solubility in ethanol. The authors speculated that the diffusion of precursor solution to oil creates an interfacial subphase, and the fast diffusion from the subphase to oil are the key kinetic processes that drive the formation of corrugated surface morphology. Switching to an oil (mineral oil) with low ethanol diffusion resulted in smooth surface morphology. The produced particles have a diameter around 17μm with a CV < 5%.¹²⁷

FIG. 11.

Droplet-based microfluidic templated particles synthesis with different processes. (a) Micrometer/nanometer-sized PLGA particles prepared through extraction or evaporation methods. Reprinted with permission from Hung et al., Lab Chip 10, 1820 (2010). Copyright 2010 Royal Society of Chemistry.¹²⁶ (b) Morphology-controlled polymeric microcarriers formation through solvent diffusion. Reprinted with permission from Lee et al., Adv. Funct. Mater. 18, 4014 (2008). Copyright 2008 John Wiley and Sons, Inc.¹²⁷ (c) Alginate microgel solidified through acid diffusion inside the pre-polymer. Reprinted with permission from Utech et al., Adv. Healthc. Mater. 4, 1628 (2015). Copyright 2015 John Wiley and Sons, Inc.¹²⁸ (d) Thermal-induced gelation of agarose microgel particles. Reprinted with permission from Velasco et al., Small 8, 1633 (2012). Copyright 2012 John Wiley and Sons, Inc.¹⁴⁶ (e) Temperature-controlled perovskite nanocrystal synthesis. Reprinted with permission from Lignos et al., Nano Lett. 16, 1869 (2016). Copyright 2016 American Chemistry Society.¹³⁰ (f) High-temperature synthesis of bi-metallic nanomaterials with controlled morphology. Reprinted with permission from Sebastián and Jensen, Nanoscale 8, 15288 (2016). Copyright 2016 Royal Society of Chemistry.¹³¹ (g) Janus particle synthesis through on-chip photopolymerization. Reproduced with permission from Nie et al., J. Am. Chem. Soc. 128, 9408 (2006). Copyright 2006 American Chemistry Society.¹³⁵ (h) “Stop-flow lithography” for on-chip photosynthesis of microparticles with high complexity. Reprinted with permission from Dendukuri et al., Nat. Mater. 5, 365 (2006). Copyright 2006 Springer Nature.¹³⁶

Mass transfer of reactants can also work in the reverse fashion of solvent extraction to transport reagents from the continuous phase into the dispersed phase to induce solidification. Utech et al. demonstrated that ionic cross-linking of alginate particles can be achieved by diffusion of the acid from the continuous phase into the dispersed phase.¹²⁸ The pre-polymer solution is mixed with chelated calcium complex, which releases calcium ions upon pH change [Fig. 11(c)]. The diffusion of the acid into the droplet triggers cross-linking; the rate of cross-linking reaction is limited by the diffusion rate of H⁺ into the droplet, thus avoiding channel clogging due to rapid solidification.¹²⁸

Temperature control provides another method for solidifying particles on-chip. Thermally induced gelation, for example, takes advantage of sol–gel phase transition that occurs above or below a critical temperature to produce solid particles. For example, an agarose solution can be transformed to microgel particles by dropping the temperature of the downstream microfluidic channel from 37 °C to 2 °C¹²⁹ [Fig. 11(d)].

Temperature-initiated processes are particularly useful for synthesizing inorganic nanoparticles. For example, cesium lead halide perovskite nanocrystals can be synthesized using a drop-based microfluidic system by changing the channel temperature from 90 °C to 230 °C.¹³⁰ It has been shown that the particle diameter of these particles, and therefore photoluminescence characteristics, can be controlled by setting the reaction temperature [Fig. 11(e)]. In a separate study, a library of bi-metallic nanomaterials is constructed using a Si-and-glass microfluidic device.¹³¹ Bi-metallic nanoparticles, such as Pt-Pd, Ag-Pd, Pt-Ru, Pt-Ni, and Pt-Co, were used to form complex nanostructures, such as nano dendrites, core-shell, nano dumbbells, or nanosheets, based on the residence time and temperature, which controls the nucleation and growth kinetics [Fig. 11(f)]. Using a drop-based reactor gave uniform thermal history to all droplets, resulting in highly monodisperse nanoparticles with fine control of size and shape.¹³²

Photoinitiated solidification/polymerization is another method that has been used to enable the on-chip conversion of liquid droplets to solid particles. Photoreaction has several advantages, including fast reaction kinetics and spatiotemporal control over material formation.¹³³ In photoinitiated droplet solidification, droplets flow through an area of a microfluidic chip that is exposed to a light source, initiating the photoreaction; droplets subsequently undergo solidification as they move downstream to the outlet. One of the main advantages of this method is the ability to control the droplet geometry.¹³⁴ The formed particles can retain the shape of the channel where the UV curing occurs, or non-equilibrium morphologies can be achieved because the particle can be solidified immediately after droplet generation. Biphasic Janus particles or particles with ternary structures can be synthesized using this method¹³⁵ [Fig. 11(g)]. Additionally, continuous-flow lithography techniques have been developed, wherein a pre-polymer solution flows through a microfluidic channel and is exposed to UV light that passes through a photomask to control photoinitiation spatially, allowing the generation of microparticles with a wide variety of shapes¹³⁶ [Fig. 11(h)].

Parallelizing these on-chip solidification strategies poses a number of challenges. For extraction-based synthesis, the timescale required for diffusion-based mass transfer can lead to a large footprint for the microfluidic devices, compromising the number of devices that can be placed on a single chip. For temperate-based processes, it is necessary to maintain the uniformity of temperature over the many parallel devices and in some cases, a specific thermal gradient across each device on the chip. The high thermal conductivity of most fabrication materials (e.g., glass) make establishing thermal gradients challenging. For photoinitiated processes, light scattering in microfluidic devices needs to be minimized or eliminated as it can trigger undesired solidification of reactants, possibly leading to clogging of the device.

VI. OUTLOOK

Scaling up the throughput of microfluidic droplet–based materials synthesis has already shown enormous promise in a number of important, seminal papers highlighted in this review. As the field continues its evolution toward higher throughput and the capability to generate increasingly complex emulsions, we foresee multiple emerging opportunities and challenges.

There are several opportunities that we envision to further develop parallelized microchips. One outstanding technical challenge in the parallelization of microfluidic droplet generators is the reliable, spatial patterning of the hydro-philicity/-phobicity. The patterning of hydro-philicity/-phobicity is necessary for the generation of many higher-order emulsions. For example, a double emulsion w/o/w generator, that consists of two flow-focusing droplet generators in a series, would require a hydrophobic coating after the first generator and a hydrophilic coating after the second generator. While many methods have been developed to pattern hydro-philicity/-phobicity in single device chips, these strategies often rely on selectively flowing coating reagents to select regions of the chip and the required stringent control of flow has proven challenging to apply to parallelized devices.^137–139 An additional challenge is the incorporation of multiple on-chip processing steps, building on the work reviewed in Sec. V, to avoid off-chip bulk processing of the emulsion templates that can compromise particle uniformity and functionality.

There are several opportunities that we see to further develop design rules and modeling to account for a wider range of materials than are currently being generated. One outstanding challenge is the generalization of the design rules and strategies outlined in this paper to fluids where Newtonian fluid behavior cannot be assumed. There are many exciting applications that require emulsification of non-Newtonian fluids, such as those with highly concentrated polymers. Design rules and strategies that can account for the non-Newtonian nature of highly viscous fluids for high throughput generation have yet to be developed. Additionally, there are many exciting applications to emulsify viscous fluids, which can be challenging with existing designs due to the high required pressures and the low flow rate transitions into jetting modes. A recent work demonstrated an exciting strategy to this problem by using a phase inversion method, wherein the viscous fluid that is intended to be emulsified is used as the continuous phase in a droplet generator and the emulsion is then inverted downstream by reversing the surface wettability in the channel or by abruptly changing the device dimenions.^140,141 Deeper theoretical understanding and simulation tools are expected to guide the design of microfluidic geometries for systems that have complex reagents to reduce the trial-and-error iteration of device design to fabrication to testing.

Another important outstanding challenge is the incorporation of quality control in parallelized droplet generators to produce emulsions over long timescales. Specifically, how can chips be designed to minimize clogging or fouling of surfaces over long-term use, which can lead to dispersion in particle characteristics? Future efforts are needed to incorporate feedback systems that can track and monitor chip performance so that either operation can be halted upon device failure or flow conditions are adjusted to maintain constant emulsion size and uniformity. These feedback systems will be necessary for application of these chips in industry settings where particle quality must be stringently controlled and guaranteed, e.g., pharmaceuticals.

DATA AVAILABILITY

The data that support the findings of this study are available within the article.