1. Introduction

The demand for custom-shaped colloidal microparticles ranges across a breadth of applications [1] including advanced films and coatings [2], cell scaffolding [3], drug delivery [4], diagnostics [5], and optical devices [6]. Many properties of a bulk colloidal suspension including diffusivity and rheology are strongly influenced by the shape of the constituent particles [7,8]. Additionally, tuning the colloid chemistry can produce a variety of particle properties [1], including hydrogels that can be made to swell in response to external stimuli for sensing applications [9]. Once fabricated, particles can be assembled using approaches such as holographic optical tweezers [10] or self-assembly [11].

There are many established and high-throughput methods for fabricating spherical microparticles. Suspension polymerization [12], emulsion polymerization [13], or droplet-based polymerization techniques [14] each offer control of the sphere’s diameter but do not always produce truly monodispersed colloidal suspensions and cannot be readily adapted to produce general non-spherical shapes.

Many surface-patterning approaches, such as casting [15], lithography [16], and the PRINT method developed by DeSimone and colleagues [17,18], can produce upwards of billions of individual particles in a single patterning step, but their geometries are limited to 2D extrusions and must be patterned on a surface. To create freely-floating colloidal objects, Doyle’s group first proposed the stop-flow lithography (SFL) approach, in which UV-curable photopolymer is delivered to a projection lithography system via a polydimethylsiloxane (PDMS) microfluidic channel and 2D particles are fabricated while suspended [19]. The SFL approach was limited in throughput by the need to stop the flow of liquid pre-polymer in the channel prior to each curing step. Further improvement in throughput was demonstrated with the invention of continuous-flow lithography (CFL), where lithographic curing occurs while the liquid pre-polymer flows without stopping [20,21]. However, most flow-lithography approaches are limited in resolution, particle shape (i.e., they can only be extruded 2D shapes), and require an oxygen inhibition layer (limiting the use of pre-polymers that would swell PDMS).

In recent years, new approaches have been demonstrated to introduce a third dimension of geometric control. Inertial flow shaping [22] provides a means of creating 3D particles, but particle geometry is limited because axial features must be created by complex microfluidic flow profiles. Pisignano and associates applied a highly-localized two-photon lithography (TPL) process to a CFL system (i.e., two-photon continuous-flow lithography, or TP-CFL) [23], which generated 3D colloidal microsprings with very high 3D resolution. However, 3D positioning of the writing spot with respect to the photopolymer medium was performed entirely via piezo-based positioning of the microfluidic chip, limiting the fabrication rate. Furthermore, this approach did not employ a method for rapid on/off control of the laser beam and was limited to fabricating shapes generated by continuous laser-scan paths.

The purpose of this work is to increase the complexity of fabricated particle geometries and dramatically raise the throughput of the TP-CFL approach to produce large numbers of fully-arbitrary 3D shapes with submicron-sized features. To accomplish this, we introduce (1) a new system that allows for 3D scanning-TPL by synchronizing a galvanometer, piezoelectric stage, and an acousto-optic switch, and (2) a technique for modifying the laser’s scan path to compensate for the high-flow-rate liquid precursor. We demonstrate fabrication rates of over 30 particles per second or 10⁵ particles per hour. Although this fabrication rate is limited when compared to optimized CFL (i.e., nearly 1000 particles per second which is over 10⁶ particles per hour [24]) and a commercial PRINT approach (i.e., 1.2 grams per hour for sub-200nm particles or 30 grams per hour for 5µm particles [25]), the scanning TP-CFL approach is unmatched in its potential to create complex 3D microparticles with arbitrary shapes at submicron resolution. Also, as we discuss below, the throughput of the proposed approach also shows promise for producing comparable or superior throughputs given additional research.

2. Experimental setup

The introduction of a fast laser-scanning component (e.g., an acousto-optic modulator (AOM) or scanning mirror galvanometer) vastly increases the flexibility and speed of TP-CFL when compared to fixed-focal-point systems [23]. Galvanometers can rapidly scan arbitrary in-plane paths with linear writing rates surpassing 1mm/sec [26]. Scanning only in the microscope objective’s focal plane produces intricate 2D colloidal objects, but by changing the height of the microfluidic chip by piezoelectric actuation, the medium can be quickly repositioned to write on different layers and produce 3D parts [27]. Other means to rapidly change the writing plane include adaptive-optical components like spatial light modulators (SLMs) [28], however, a discussion of these methods is outside the scope of this paper.

The experimental setup shown in Fig. 1 includes a tunable femtosecond laser (Spectra-Physics MaiTai eHP DS) at 760nm and an AOM (IntraAction ATM-802DA2), which serves as a variable power attenuator and rapid beam shutter. The 2D mirror galvanometers (Thorlabs GVS012) scan the beam within the microchannel and can also temporarily divert it into a power sensor to measure writing power (Thorlabs S142C and PM100USB). The microscope objective is a 100x oil-immersion objective (Olympus Plan Apo Lambda, NA = 1.45). A ± 15µm-range Z-axis piezo actuator allows for the height of the microfluidic chip to be changed during laser scanning. The scanning mirror galvanometers, AOM, and piezo controller (Thorlabs TPZ001) are synchronously driven by a four-channel analog output module (National Instruments NI-9263). The imaging system consists of a camera (Thorlabs DCC1545M) and custom tube lens at 40x magnification.

 

Fig. 1 The experimental setup includes a femtosecond laser (fs) for two-photon lithography, imaging system, and microfluidic system. M: mirror; BB: beam block; BE: beam expander; Meter: power meter; 4-F: 4-F telescope relay; DM: dichroic mirror; TL: tube lens.

Download Full Size | PDF

The pre-polymer liquid medium consists of Sartomer SR9035 acrylate (62% w/w, and a dynamic viscosity of 168cP), DI water (37% w/w), and photoinitiator Li-TPO (1% w/w) [11]. Flow is driven by a standard syringe pump (Chemyx Fusion 100).

The microfluidic channel used was cast in polydimethylsiloxane (PDMS) using a standard lithographically-produced mold. The PDMS was bonded to a coverslip ceiling after both surfaces were treated with oxygen plasma. These microfluidic channels were 40mm in length, 75µm in width, denoted as w, and 50µm in depth, h. The experimental system presented here is compatible with a variety of other microfluidic devices.

3. Writing scheme

In this section, we discuss how the laser’s scan path must be modified to produce non-warped parts in a rapidly flowing microchannel. The laminar flow within the microchannel acts analogously to a conveyor belt in macroscale manufacturing for delivering uncured liquid pre-polymer to the writing volume (i.e., the scanning laser’s focal spot) and for carrying away fabricated colloidal particles. Increasing the flow rate of the channel can allow the laser galvanometer to scan at its maximum bandwidth and produce colloidal objects with higher throughput.

A laser-scan path designed to write in a static medium is shown in Fig. 2(a), but when this path used in a flowing medium, the part shifts downstream during its fabrication and additional polymer will be cured due to the relative velocity between the scanning-laser’s frame of reference and the flowing medium, as shown in Fig. 2(b). To compensate for the flow in the microchannel, we continuously shift the laser spot downstream to track the moving part during its fabrication.

 

Fig. 2 Laser-scan files designed for static media (a) produce warped particles in high-flowrate microchannels (b). An overcompensated (c) and correctly-compensated (d) particle are shown. Each particle had a minimum feature size of 800nm.

Download Full Size | PDF

Although in pressure-driven laminar flows the velocity profile is generally parabolic for a Newtonian fluid (Poiseuille flow), we have found that a uniform-velocity-profile assumption produces satisfactory results under the fabrication conditions we have used and for particles that occupy a small fraction of the flow cross-section. It is only important to compensate each particle for the average flow speed of its local streamlines. While the velocity profile gradient across each particle causes an additional rotation as the particles travel downstream, fabrication occurs much more rapidly than rotation is induced.

The details of the compensation algorithm are as follows: the lateral coordinate (along the flow direction of the microchannel) of each 3D point in the scan path (indexed as n) is shifted by an offset, δ_n, downstream as a function of the flow velocity, v_medium, and the time during the scan: δ_n = t_n · v_medium. The resulting corrected scan path is modified such that the laser travels faster when moving downstream and slower when moving upstream so the correct laser dose exposure is maintained. An example of an overcompensated particle is shown in Fig. 2(c). The corrected path produces a part with no warping effects as shown in Fig. 2(d).

We measured the fluid velocity for the compensation algorithm by using a video-tracking-based measurement. First, the algorithm recorded a background image of the channel with no objects present, and then two successive images of a flowing object at a known time difference. The flowing object can be a simple polymerized spot – no additional calibration particles are necessary. The fluid velocity was extracted by subtracting the background image from the latter two images and calculating the distance that the object travelled via centroid tracking.

4. Results

To take advantage of the full scanning bandwidth of the galvanometers and to utilize the maximum volumetric fraction of the photopolymer, fabrication sites were spatially multiplexed along a direction perpendicular to the flow (e.g., 3 colloidal objects were fabricated in parallel in Figs. 3 and 4, each of which was compensated for its own flow velocity). The planar objects shown in Fig. 3 and Visualization 1 were created at a rate of 31 particles per second at a linear flow rate of 110µm/sec at the center of the channel. The 3D objects shown in Fig. 4 and Visualization 1 were created at a rate of 15 particles per second at a linear flow rate of 100µm/sec at the center of the channel. Writing was performed in the middle of the channel’s depth where the gradient of the velocity field was minimal. The focal spot used 70mW of writing power and a linear scan rate of 1mm/sec. The minimum written feature size (i.e., the smallest feature that can be fabricated) was observed to be 800nm for patterns in Fig. 3 as observed by our optical microscope. Videos of the fabrication process are provided in Visualization 1.

 

Fig. 3 Three examples of 2D structures fabricated with scanning TP-CFL: 10µm stars (a), 5µm stars (b), and X-box patterns (c).

Download Full Size | PDF

 

Fig. 4 3D gears (a) and pyramids (b) are rotated by shear forces in the microfluidic channel’s boundary layer, allowing them to be imaged on their side.

Download Full Size | PDF

To image features along the writing beam’s axis, we fabricated 3D particles at approximately 10µm from the coverslip, where the gradient of the velocity profile in the channel’s boundary layer produces a torque that rotates the particles as shown in Fig. 4. Measurements via optical microscope revealed that the planar writing resolution was 800nm, and by viewing 3D pyramid-shaped particles that have rotated about an axis in the focal plane, we observed that the axial writing resolution was approximately 3µm. This value is considerably smaller than prismatic particles fabricated by most other flow lithography approaches that use a one-photon polymerization reaction. For such reactions the median particle thickness is equal to the thickness of the channel itself minus a small oxygen inhibition layer (e.g. 15µm-thick particles in 20µm channels [20]).

The fabrication rate is dependent on the maximum scan rate and the length of the scan path, so further improvements can be expected with a faster galvanometer or scan paths with fewer coordinates. Using a galvanometer with a 10-fold increase in scan rate and a modest increase in the laser writing power of approximately 2x [26], the flow rate can be increased by 10-fold, thereby correspondingly increasing the fabrication rate. Micromirror arrays hold the potential to reach higher scan rates than conventional mirror galvanometers due to their reduced inertia [29]. Diffractive optical elements may be used for further speed-up (e.g. 2-4x) by splitting a single beam for parallelized writing of multiple identical parts simultaneously [28].

5. Conclusion

This work demonstrates how scanning TP-CFL provides a means to fabricate complex 3D shapes with submicron printing resolution at rates surpassing 10⁵ parts per hour. Fabrication rates surpassing 10⁶ particles per hour are anticipated with faster scanning galvanometers and flow rates. To compensate for the continuously-flowing photopolymer, we present a method to adjust the laser’s scan path to prevent deformation as the part moves with the flow during fabrication. The scanning TP-CFL approach provides a promising path towards high-throughput fabrication of complex 3D colloidal microparticles for a variety of applications in materials and biological science.