Fabrication of Embedded Microfluidic Chips with Single Micron Resolution Using Two‐Photon Lithography

Two‐photon lithography (TPL) is an advanced high‐resolution additive manufacturing technique for objects with feature sizes between 100 nanometers to tens of micrometers and an overall footprint of up to hundreds of micrometers. With recent advances in the TPL technique, writing speeds are becoming faster, rendering the method feasible to print high‐resolution microfluidic chips with a footprint in the centimeter range within a reasonable time frame. In this work, a process flow to fabricate embedded microfluidic chips with channel diameters down to 30 µm is developed. To address the particular difficulty of washing the embedded channels free of uncured material, introduces a developing scheme based on a 3D printed chip‐to‐world‐interface to connect the chips to a pressure‐driven pump. This setup is leakage‐free up to a pressure of 6.9 bar for faster and safer development of embedded microfluidic devices. It manufactures meander chips with channel lengths up to 20 cm, droplet generator chips, and cell sorting chips based on deterministic lateral displacement with pillar diameters of 30 µm and pillar spacing of 4 µm. TPL of microfluidic chips will enable rapid manufacturing of novel designs, significantly reducing concept‐to‐chip times with high resolution in a reasonable amount of time.


Introduction
Microfluidic chips have become an important tool for analytical chemistry and molecular biology because they offer many useful advantages such as the ability to employ very small amounts of DOI: 10.1002/admt.202300667samples and reagents, less waste generation, easy process integration, and low cost. [1,2]Additionally, separation and detection can be carried out in short times with a high resolution as well as sensitivity. [3]6][7][8][9][10] However, these techniques only allow for the manufacturing of 2.5D structures and require a subsequent bonding process to enclose the channels and the manufacturing of the master structures, which makes the overall process cost-intensive and time consuming. [11,12]Due to these disadvantages, additive manufacturing (AM) became very popular in the microfluidic community.The advantages of AM over replicative processes are higher flexibility of the chip design, such as truly 3D channel structures, as well as short concept-tochip intervals, as the tedious master mold manufacturing step can be omitted. [13]In recent years, various AM techniques such as fused deposition modeling (FDM), [14,15] inkjet printing (IJP), [16,17] and stereolithography (SL), [13,18] have been used to manufacture polymeric microfluidic chips with embedded microfluidic channels.Compared with printing open channels, manufacturing embedded channels has the advantage of not requiring sealing or bonding protocols that can lead to leakage or critical failure of the microfluidic chip.In FDM, a solid filament is fed through a heated nozzle and deposited on a substrate.The limiting factors of FDM are low resolution and low transparency of the build materials. [2]IJP is based on jetting a liquid photocurable resin, which is subsequently polymerized by UV light. [2]The main drawback of this method is the necessity of a sacrificial supporting material for any embedded void spaces, e.g., channels, in the microfluidic chip. [17,19]For channel dimensions <200 μm and designs that include bends and meanders, the removal of the support material is very challenging. [19]SL is based on photocuring a liquid photoresin using a laser or a digital light projector (DLP) in a layer-by-layer fashion.The most common approach is the bat configuration, or constrained surface technology, where the build plate is immersed in the vat containing the photoresin and the object is printed hanging down from the build plate. [19,20]Of the aforementioned methods, SL is the one allowing for the manufacturing of the smallest channels so far with feature sizes down to a few tens of micrometers.To achieve this, there are two key challenges when printing objects with void spaces such as embedded channels.The first is the potential curing of the embedded channels, which is addressed by carefully designing a photoresin with an appropriate absorber that mitigates the light and avoids polymerization in already printed layers. [21][23] The second key challenge is the developing step after the printing.This step is necessary, in order to void internal volumes such as channels from uncured photoresin.However, the smaller the channel dimensions, the more challenging it is to remove this uncured material.It was recently reported that the limiting factor in determining the minimum achievable channel dimension was not the printing capability or the optical properties of the photoresin, but the ability to flush out the uncured photoresin. [24]This has, so far, limited the usable materials range that precursors of sufficiently low viscosity, most prominently poly(ethylene glycol) diacrylate 250 (PEGDA 250). [24,25]However, there is a high demand for various functional materials such as soft -like PDMS materials or combined hydrophilic and hydrophobic materials for spatial wettability, that require higher viscous precursors. [26,27]wo-photon lithography (TPL) is an emerging tool for highresolution 3D printing, where a photoinitiator, e.g., for a radical polymerization, is activated by two-photon absorption (TPA).TPL is arguably among the AM techniques with the highest possible resolution with feature sizes below 100 nm. [28,29]Such a high resolution can be achieved due to the non-linear absorption profile of TPA. [30]Considering photopolymerization induced by single photon absorption (SPA), the photoinitiator will be activated along the whole pathway of the beam, which makes it more difficult to structure hollow objects such as microfluidic channels, because the material above and below the focal plane is continuously exposed to light. [31]In contrast to SPA, the TPA approach reduces the volume activated by the photoinitiator (so-called voxel) to a very confined space within the focal point. [32]The downside of this technology is that a very high resolution generally comes with a slow fabrication speed and, in the case of TPL, a very limited printing volume. [33][36][37] This method has the advantage that submicron features can be implemented into microfluidic devices.However, this requires a precise alignment to ensure that the internal features are placed correctly and anchored sufficiently inside the microfluidic device. [38]Furthermore, the material of the internal feature has to be compatible with the existing chip.In a different approach, the so-called definition-reinforcementsolidification (DRS) strategy aims to enhance the processing efficiency by defining (D) the surface profile of the 3D structure using near-threshold polymerization, followed by a second scanning of the inner above surface for reinforcement (R) and a flood illumination to solidify (S) the inner uncured resin. [39]Due to recent advances in TPL, much higher throughput can nowadays be achieved, making TPL an attractive option for the manufacturing of entire microfluidic chips with a footprint in the centimeter range while still maintaining high resolution. [40]n this work, we printed, for the first time, microfluidic chips using a commercial high-resolution TPL printer with channel resolution down to 30 μm and feature sizes down to 4 μm, respectively, on lateral chip footprints of 2.5 cm (see Figure 1).For printing the chip, we used the commercial TPL system NanoOne (UpNano) in the so-called vat configuration mode (see Figure 1a).The device is equipped with a 1 W laser, which allows very high writing speeds of up to 1000 mm s −1 resulting in a throughput of up to 200 mm 3 h −1 on a footprint of 4 cm.This work demonstrates the manufacturing of usable microfluidic devices with feature resolutions in 3D topographies, such as a low surface roughness below 50 nm, that can only be achieved by TPL so far.By using an objective with a low magnification of 10x, the system reaches a throughput of 22 mm 3 h −1 , enabling comparatively short printing times at a resolution that surpasses DLP, thus filling the resolution gap between in situ TPL within pre-fabricated microchannels and DLP.Using TPL to print entire holistic chips has the advantage over in situ processes that it is a one-step technique, which does not require time-consuming alignment steps, objective exchanges, or the need for compatible materials.To enable the washing of higher viscosity material out of the printed embedded channels we introduce a developing scheme, which is based on a chip-to-world interface (CWI) we previously described (Figure 1 b). [41]The CWI consists of two components: a top part containing holes for the tubes as well as a compression element, and a bottom part including a cavity for the chip, aligning its inlets/outlets to the tubes from the top part.The flanged tubes are pressed onto the inlets using four screws on each side of the CWI.A pressure-driven pump is utilized to force the developing solution through the channels, which dramatically speeds the developing step up.This setup has the advantage that the connectors can be packed at a higher density, which reduces the overall chip size, thus reducing the printing time. [41]e demonstrate the manufacturing and development of embedded microfluidic channels with diameters down to 30 μm and a total length of 2.5 mm as well as meander chips with a channel diameter of 100 μm and a total length of 20 cm.The versatility of this approach is validated by creating a droplet generator and a cell sorting application based on a deterministic lateral displacement (DLD) chip with a separation efficiency of 89.4 %.

Chip Fabrication and Development Using the CWI
Depending on the smallest feature size, the chips were printed either horizontally or vertically with respect to the substrate and channel plane (see Figure S1, Supporting Information for the printing direction).Using a 10x objective, the smallest printable feature size is 0.73 μm in the XY-plane and 9.2 μm in the Z-plane due to the elongated shape of the voxel.If necessary, the printing direction was changed according to the voxel dimension and smallest feature size.For example, a chip with a channel crosssection below 85 μm 2 needed to be printed vertically, whereas a chip with a high lateral resolution (e.g., a DLD chip) needed to be printed horizontally.After the prints were finished, the components were immersed in the developer solution for 10 min for pre-developing and then connected to the CWI for the full developing step.Figure S2 (Supporting Information) shows the disassembled CWI, with the compression elements for the flanged tubes and the cavity for the alignment, as well as the assembled CWI with a mounted DLD chip and the connection to the pressure driven pump using standard HPLC (high-performance liq-uid chromatography) connectors.The flanged tube ends had a diameter of ≈1.5 mm, therefore the inlets of the microfluidic chips were spaced with a minimal pitch of 3.4 mm to avoid overlapping of the flanged tube ends, which would lead to leakage.Slight unevenness or surface roughness of the microfluidic chips are compensated by pressing the flexible flanged tube ends onto the channel inlets.The compression elements had a diameter of 3.4 mm and a height of 1 mm.The developer was flushed through the channels using a pressure of 6.9 bar which the CWI sustained without any leakage, even if the screws are only tightened by hand.The maximum pressure of 6.9 bar was limited by the capabilities of the pressure-driven pump.Using standard HPLC connectors allows for a fast and easy assembly as well as disassembly and renders this system compatible with follow-up experiments.

Microfluidic Channel Structuring
To showcase the geometrical freedom that TPL allows, various channel geometries such as rectangular, circular, triangular, and trapezoidal with features below 100 μm and a channel length of 500 μm were printed using the commercial photoresin UpFlow (UpNano, Austria) (see Figure 2).The scanning electron microscope (SEM) images show two sets of channel geometries, which were printed in a horizontal and vertical fashion.Since the resolution depends on the orientation, printing the channels horizontally resulted in a better representation of the designed geometries.Printing the channels vertically led to slight deformations; nonetheless, the shapes can be readily recognized.
Microfluidic chips with embedded channels and with channel geometries <100 μm were printed using UpFlow.The CWI was used to assess if the limiting factor of manufacturing embedded microfluidic chips is either the printing or the developing step.However, the viscosity of the resin plays an important role in the feasibility of the developing step, whereby a lower viscosity facilitates flushing the uncured material out of the printed channels.Recent publications in the field show that PEGDA 250, which has a low viscosity of 13.4 mPa s, was mainly utilized because of that reason thus limiting the choice of material significantly. [21,24,25,42]he most commonly used development protocol consists of immersing the chips in 2-propanol using diffusion to flush out uncured material.However, if microfluidic chips with different material properties than those of PEGDA 250 are required, the viscosity will be the limiting factor in material selection as higher molecular weight precursors will be difficult to remove.The same holds true for many commercial materials such as, e.g., the photoresin UpFlow (with a viscosity of 555.4 mPa s) used as a model material in this work.Table 1 summarizes the channel dimensions, the resulting pressure drops per flow rate, and the developability by applying a pressure of 6.9 bar to flush out the uncured material.If there was no observable flow after 3 h of developing, the chip was determined as not developable.We found developability strongly depended on the channel length.While channels with a diameter of 100 μm were developable up to a length of Table 1.Evaluation of the developability of microfluidic channels using Up-Flow with respect to their diameter and length.20 cm, a diameter of 50 μm was only developable up to 1 cm, and a diameter of 30 μm up to 0.25 cm.The pressure drops per flow rate of the channels were calculated as a guideline, to determine whether or not a channel design could be developed using this method.The calculated pressure drops assume a straight channel and the viscosity of pure UpFlow.It was observed that the developing part is indeed the limiting factor when it comes to manufacturing embedded microfluidic channels.However, by applying the developing scheme using the CWI, it was possible to free microfluidic chips in significantly smaller dimensions.Future work with specially designed low-viscosity photoresins for TPL and our developing scheme could enable even smaller channel networks.
Using UpFlow we manufactured a meander chip and a droplet generator (see Figure 3).The meander chip had a channel width of 100 μm and three channels were stacked on top of each other with a vertical distance of 400 μm between them.Printing the channels on different levels has the advantage that a channel network can be compacted and printed on a much smaller footprint, thus saving printing time.Figure 3b shows the printed meander chip with the filled channels using dyed water.The longest channel (in red) has a length of 20 cm.
To showcase an application of a printed embedded microfluidic chip, a droplet generator with a nozzle size of 50 μm was manufactured.The design of the channel is shown in Figure S4 (Supporting Information).Figure 3c shows the nozzle of the droplet generator and highlights the channels of the continuous phase (CP) and the disperse phase (DP).For the CP, dodecyl acetate was injected into the chip, and for the DP dyed water.The pressure-driven pump was set to 300 mbar for the DP and 330 mbar for the CP. Figure 3d shows the resulting droplets with a diameter of 45 μm.
The surface roughness R a of the printed channels was measured using white light interferometry (WLI).As an example, a cross-section of the nozzle of the droplet generator was printed, because it shares channels that are oriented horizontally and vertically (see Figure S5, Supporting Information, for the nozzle cross-section).The R a for the horizontally and vertically oriented channel was 8.6 and 47.8 nm, respectively.This difference in R a reflects the above-mentioned difference in resolution due to the voxel shape.However, the measured values of R a are much lower compared to microfluidic devices manufactured by SL.5]

Deterministic Lateral Displacement
We demonstrated the capability of this manufacturing method by fabricating a DLD device.The DLD chips were printed using the commercial photoresin UpPhoto (UpNano, Austria), which has a slightly higher viscosity of 785.7 mPa s compared to UpFlow.UpPhoto was mainly utilized because of its fluorescent behavior, which made it possible to assess if all of the uncured material in the chips was flushed out entirely to ensure that said material would not interfere and possibly damage the cells within the solution.A UV LED flashlight was faced to the tubes of the outlets during the developing step.After there was no more fluorescence visible, the chips were flushed for an additional 5 min prior to use.DLD is a useful tool for the continuous separation of micrometer-sized particles such as bacteria, parasites, and tumor cells in the blood. [46,47]In general, this process utilizes laminar flow through a periodic array of obstacles (e.g., pillars) in a microfluidic device (see Figure 4), where the total fluid flux between each gap of the obstacles is divided into multiple streamlines. [46]Each streamline contains an equal fluid flux.If a particle is smaller than the critical diameter D c , it stays within its streamline and travels straight through the microfluidic channel in the so-called "zig-zag mode" (see red path in Figure 4a).However, if a particle is larger than D c , it is "bumped" to the next streamline after each obstacle and travels through the channel in the so-called "bumper displacement mode", thus being shifted to the channel wall (see purple path in Figure 4a). [48]The design parameters for DLD chips include the center-to-center distance , the lateral shift between adjacent rows Δ, the gap between each obstacle G, and the row shift fraction  and angle  (see Equations (S1)-(S4), Supporting Information). [46,49]ecently, our groups reported the separation of cells with rodlike and spherical morphologies with high efficiency of 75.5% by inserting a mixed solution containing both cell types into a DLD chip and using a flow rate of 50 μL min −1 . [47]Within the present work, we manufactured DLD chips with a row shift angle of 1°a nd a D c of 800 nm and 1.4 μm, respectively.Using Equation (S4) (Supporting Information) we calculated a pillar spacing G of 4 μm for a D c of 800 nm and of 7 μm for a D c of 1.4 μm.A schematic of the working principle of the DLD chip is shown in Figure 4b and the entire chip is shown in Figures S6 and S7 (Supporting Information).The channels with the incorporated pillar array were 1.2 mm wide and 100 μm tall.The pillars had a diameter of 30 μm and a height of 100 μm.
The cell solution containing elongated rod-shaped cells (the so-called parental cells), as well as small roughly spherical cells (the so-called minicells) was injected into inlet I 2 and the remaining inlets I 1 and I 3 were blocked.Samples from each outlet O 1-3 were collected to analyze the ratio of parental cells versus minicells.The two chips (D c of 800 nm and 1.4 μm) were compared in terms of throughput as well as separation efficiency.The flow rates at various applied pressures of both chips are summarized in Table 2.At the maximum pressure of 6.9 bar, the chip with a D c of 800 nm showed a flow rate of 100 μl min −1 and the chip with a D c of 1.4 μm a flow rate of 180 μl min −1 .The separation efficiency was calculated using the following equation: The ratio of parental cells to minicells was calculated prior to injection and compared to the resulting ratio after separation using the output of the center outlet O 2 .Cells whose long axis is shorter than or equal to 1 μm were identified as minicells and cells whose long axis is longer than 1 μm were identified as parental cells.

Conclusion
In this work, we showed that recent advances in TPL technology made it possible to print microfluidic devices with a footprint in the centimeter range in a reasonable timeframe without exceeding printing times of 16 h.By using TPL, we manufactured embedded microfluidic meander channels with a channel length of 20 cm and a channel diameter of 100 μm as well as channels with a diameter of 30 μm and a length of 2.5 mm.These highresolution channels are further enabled by a developing scheme using a chip to world interface, which allows higher viscosity noncured resins to be washed out of the channel in a controlled manner after the printing process.The capability of this method was demonstrated by the manufacturing of a droplet generator chip with a nozzle size of 50 μm and a channel height of 35 μm to generate microdroplets with a diameter of 45 μm.Additionally, DLD devices at a high resolution with pillar spacings down to 4 μm (D c 800 nm) were manufactured.The DLD devices were tested by separating spherical minicells with diameters that are smaller than or equal to 1 μm from parental rodlike cells up to 10 μm at an efficiency of 89.4%.
Surface Functionalization: The borosilicate substrates were treated for 30 min in acidic methanol (methanol:HCl, 1:1 (v:v)).Subsequently, the substrates were washed with IPA and deionized (DI) water and dried with nitrogen.The substrates were then immersed in a 100 mM solution of MACS in dry toluene for 60 min.The substrates were again washed with IPA and DI water and subsequently dried using nitrogen.
Two-Photon Lithography: TPL was performed using a NanoOne highresolution printing system (UpNano GmbH, Austria) equipped with a 10x air immersion objective (NA 0.4, UPLXAPO10X, Olympus, Austria) in vat mode.The laser (80 MHz repetition rate, 90 fs pulse length, 780 nm center wavelength) was focused through a cover glass into a material vat containing the resin, and the focal point was maintained at a constant height above the glass.The intensities at focus for the printing conditions were calculated in Equation (S6), Supporting Information) and are displayed in Table S1 (Supporting Information).For layer-wise 3D structuring, the laser was scanned along the XY-plane by a galvanometer scanner, and the objective together with the vat was lowered along the Z-axis using a piezo stage.The printing parameters for the different materials and objectives are summarized in Table S2 (Supporting Information).The printing volumes, printing times, and throughput of the presented microfluidic devices were displayed in Table S3 (Supporting Information).The DLD chips were printed using UpPhoto and were pre-developed in IPA for 10 min.The Sub100 μm chips were printed with UpFlow and were pre-developed in PGMEA for 10 min and subsequently in IPA for 5 min.
Chip-To-World Interface: We used a CWI which we have previously described and modified it to be used for 3D printed microfluidic chips. [41]he principle is shown in the Supporting materials in Figure S2 (Supporting Information).In brief, it consists of a top part containing openings for the tubes and compression elements, while the bottom part contains a cavity to align the inlets of the chip to the tubes from the top part.To form a tight seal between the tubes and the chip, flanged tubes are pressed by the compression elements onto the inlets using four screws on each side, respectively.The CWI parts were 3D printed using an SLA printer of type Asiga MAX X27 (Asiga, Australia) and PlasCLEAR 2.0 (Litholabs, Germany).After printing, the CWI parts were developed in IPA.The PTFE tubes were flanged using a thermoelectric flanging tool from BOLA (Bohlender, Germany).
Development of The Microfluidic Chips: After the microfluidic chips were pre-developed, the DLD chips as well as the sub-100 μm chips still contain uncured material within the channels.To flush out the remaining material, the chips were connected to a pressure driven pump of type LineUp Flow EZ (Fluigent, France) using the CWI and the channels were flushed using IPA for UpPhoto and PGMEA and subsequently IPA for Up-Flow.The pressure was set to 6.9 bar until the channels were washed free from the uncured material.
Droplet Generation: To generate microdroplets using the printed chips, colored water using red food colorant was utilized as the disperse phase and DA as the continuous phase (see Figure S4, Supporting Information) for the CAD design of the droplet generator chip).Both solvents were introduced into the chip using the pressure driven pump.The droplet formation was observed using a digital microscope of type VHX-5000 (Keyence, Germany).
Cell Sorting By Deterministic Lateral Displacement: The cell solution was prepared as described elsewhere. [50,51]In brief, E.coli MG1655 ΔminB cells were transformed with pUC19-pLacO-sfGFP plasmid.An overnight culture was started from the glycerol stock of cells into 25 mL of Luria-Bertani (LB) medium at 37 °C and 200 rpm.On the following day, the culture was washed two times and subsequently, the medium of the culture was replaced with the same volume of phosphate buffered saline (PBS).The culture was directly used in the chip.The sample solution was injected into the DLD chips using a pressure driven pump at a pressure of 6.9 bar.Samples were collected from each outlet and further analyzed using microscopy.
The cell solutions were imaged using a Zeiss Axio Observer Z1/7 (Carl Zeiss, Germany) inverted wide-field microscope, equipped with a Colibri seven LED light source, an Alpha Plan-Apochromat × 100/ 1.46 oil DIC (UV) M27 objective, filter set 38 HE (ex.450-490, dichroic beam splitter 495, em.500-550; sfGFP) and an Axiocam506 Mono camera.The samples were embedded into 1 % agarose pads on metal microscopy slides prior to imaging.To determine the efficiency, cell numbers were counted manually from the images.
Material Characterization: Scanning electron micrographs were taken using an ultrahigh resolution focused ion beam SEM of type Scios 2 Dual-Beam HiVac (Thermo Fisher Scientific, Germany).Surface roughness was measured using a WLI of type NewView 9000 (Zygo, USA).The viscosity of the photoresins was measured using a rheometer of type MCR 302 (Anton Paar, Germany).

Figure 1 .
Figure 1.Schematics of the TPL principle and the CWI.a) Schematic of TPL in vat mode.The vat containing the resin is placed on top of the objective.The laser is focused through a high precision cover glass window on the vat bottom.The focal point is maintained at a constant height above the glass window and the polymerization only occurs within the focal point (voxel).A substrate holder including the substrate are dipped into the material vat.For a layer wise structuring, the laser is scanned along the xy-plane and the vat together with the objective are lowered on the z-axis.b) Schematic section of the CWI utilized in this work.The top part of the CWI contains openings for the tubes as well as compression elements and the bottom part has a cavity to align the inlets/outlets of the chip to the tubes from the top part.The flanged parts of the tubes are pressed by the compression element onto the chip inlets using four screws.Since the flanged ends of the tubes are flexible, slight unevenness of the chip is compensated and a tight homogenous sealing across all tubes is ensured.Using a pressure driven pump, the developer is flushed through the channel to the waste disposal.c) Micrograph of a printed microfluidic chip with embedded channels utilized in a deterministic lateral displacement (DLD) application to separate parental cells from minicells (scale bar 200 μm).d) Magnified section of the pillar array showing a spacing of 4 μm (scale bar 100 μm).

Figure 2 .
Figure 2. SEM images of channel cross-sections with various geometries and feature sizes below 100 μm with a channel length of 500 μm.a) Channels in horizontal printing direction.The channel geometries are well represented.b) Channels in vertical printing direction.The geometries are slightly warped because of the lower resolution in the Z-plane (scale bars 25 μm).
drop per flow rate was calculated using the viscosity of UpFlow.

Figure 3 .
Figure 3. Meander and droplet generator chips.a) Schematic of the meander chip in two different angles.The three channels are on three different levels, the channel diameter is 100 μm, the green channel is 7 cm, the blue channel 13 cm, and the red channel is 20 cm long.b) Printed meander chip with filled channels using dyed water.The colors are as in (a) (scale bar 1 mm).c) Micrograph of the nozzle of the droplet generator.The continuous and dispersed phases and their respective direction of flow are highlighted (scale bar 100 μm).d) Inset of the droplet generator from (c).The droplets have a diameter of 45 μm (scale bar 100 μm).

Figure 4 .
Figure 4. Cell sorting application based on DLD using printed microfluidic chips with embedded channels.a) Schematic of the separation process using DLD. is the center-to-center distance between each pillar, G is the gap between each pillar and D P is the pillar diameter.Δ is the lateral shift between adjacent rows and  is the migration angle.The pale red, yellow, and blue areas represent the individual streamlines.A particle smaller than the critical diameter D c follows the fluidic flow (zig-zag mode) maintaining its initial streamline.A particle larger than D c is bumped to the next streamline and pushed to the wall of the microfluidic channel (bumper displacement mode) after each obstacle (pillar).b) Schematic of the working principle of the DLD chips.The sample solution containing parental cells as well as minicells is introduced to the chip through inlet 2 (I 2 ) while I 1 and I 3 are blocked.After passing the pillar array, the minicells are concentrated through outlet 2 (O 2 ) and the parental cells through O 1 and O 3 .c-e) Microscope images of the cell sorting experiments using the chip with a D c of 800 nm, where c) shows the results of O 1 , d) of O 2 , and e) of O 3 , respectively.The yellow arrows point to representative parental cells and the red arrows to representative minicells (scale bars 5 μm).
Figure4c,d,eshows microscope images of the output of outlets O 1-3 .The output of the center outlet O 2 contains predominantly spherical minicells, since they travel in zig-zag mode through the channel, whereas outlets O 1 and O 3 contain mainly the rodshaped parental cells, since they travel in bumper displacement mode and are shifted to the channel walls.The separation efficiency of the chip with D c = 800 nm is 89.4 %, while that of the chip with D c = 1.4 μm is 40.3 %.The difference in separation efficiency shows that a successful DLD application strongly depends on a carefully adjusted D c .An AM technique such as TPL proves to be a powerful tool for the rapid prototyping of various iterations of such devices, where its undisputed high resolution surpasses other AM methods such as FDM, IJP, or SL.

Table 2 .
Summary of the resulting flow rates of the DLD chips with a D c of 800 nm and 1.4 μm, respectively, at various pressures.