3D-Printed Micro-Junction Array

• 3D-printed micro-junction arrays are defined by programmable 3D geometries that enable efficient routing, mixing, switching, and coupling of physical signals and fluids.
• They employ diverse fabrication techniques such as resin LCD, DLP, and two-photon lithography, offering tailored solutions across microfluidics, optics, and quantum ion transport.
• Optimized junction geometry enhances performance by reducing pressure drops in droplet generation and improving trap depths and confinement in ion transport systems.

A 3D-printed micro-junction array is a microscale architecture in which repeated junction elements are fabricated by additive or laser-based 3D structuring to route, mix, switch, confine, or couple physical signals. In the recent literature, the junction element may be a droplet generator in a microfluidic chip, a closeable segment in a vertically stacked valve array, an X-junction in a trapped-ion transport network, a facet-attached optical coupler, a micro-lenslet registered to a multicore fiber, or a freeform lightguide that redistributes sampled image pixels. Reported implementations span resin LCD-printed droplet generators for rapid in-droplet mixing (Barzoki et al., 2024), auxetic metamaterial valve arrays for multiplexed fluidic control (Supakar et al., 2023), 3D-printed ion-trap junction networks for QCCD architectures (Taniguchi et al., 21 Sep 2025, Xu et al., 2023), facet-attached micro-optics and micro-lens arrays for optical interconnects and spectroscopy (Maier et al., 2022, Anagnos et al., 2021), and 3D-printed glass lightguide arrays for snapshot hyperspectral imaging (Hong et al., 2022).

1. Definition and architectural range

The common structural feature is the deliberate use of three-dimensional geometry at the junction itself rather than only in the surrounding package. In microfluidics, this means T-junctions, cross-junctions, and asymmetric flow-focusing junctions with inlet angles of $30^\circ$, $45^\circ$, or $60^\circ$ (Barzoki et al., 2024). In fluidic control, it means vertically stacked arrays of “closeable” and “always open” segments embedded in auxetic metamaterials (Supakar et al., 2023). In ion trapping, it means X-junction units that join loading, gate, and readout branches while preserving confinement during transport (Taniguchi et al., 21 Sep 2025). In optics, it includes freeform facet-attached microlenses, merged microlens arrays on multicore fibers, and curved lightguide arrays that remap spatial samples into spectrally separable outputs (Maier et al., 2022, Anagnos et al., 2021, Hong et al., 2022).

Domain	Junction element	Representative reported feature
Droplet microfluidics	T-, cross-, asymmetric flow-focusing junctions	Asymmetric $30^\circ$ design yields MI close to T-junction with droplet diameters $\sim 100\,\mu\mathrm{m}$ smaller (Barzoki et al., 2024)
Fluidic valving	Auxetic closeable/always-open segments	9 control regions address 96 flow channels (Supakar et al., 2023)
Ion trapping	QCCD X-junction units with 3D RF electrodes	$>200$ densely packed microtraps on one chip (Taniguchi et al., 21 Sep 2025)
Optical coupling and imaging	FaMLs, MLAs, glass lightguides	Coupling losses down to $0.35\,\mathrm{dB}$; 21 wavebands from 450–650 nm (Maier et al., 2022, Hong et al., 2022)

A plausible unifying interpretation is that the “array” designation refers not merely to repetition of identical junctions, but to programmable spatial organization of many junction-mediated transfer sites within one monolithic printed object.

2. Fabrication modalities and material systems

The fabrication literature is heterogeneous. One microfluidic route uses resin LCD 3D printing on an Anycubic Photon mono X 6k with $35\,\mu\mathrm{m}$ XY resolution and $50\,\mu\mathrm{m}$ layers, using transparent plant-based UV resin. The workflow begins in SolidWorks, exports STL, applies 25 s UV exposure for the first 5 layers and 1.1 s for subsequent layers with 1 s off at the end of each layer, then proceeds through isopropyl alcohol washing, 2 min post-curing, a final ethanol+water wash, PMMA bonding, and pressing at $80^\circ\mathrm{C}$ for 1 day to minimize trapped bubbles. The resulting chips have three inlets, one outlet with 2.5 mm holes, pneumatic fittings, and $45^\circ$0 channel width (Barzoki et al., 2024).

A second route uses commercial DLP printers in a mechanically active rather than purely fluid-conducting role. The vertically stacked valve array employs an Elegoo Mars-series printer and a flexible resin mixture of SuperFlex resin with Plant-Based UV Resin in a 10:1 ratio. Geometry generation is parametric in OpenSCAD, exported as STL, and followed by IPA washing, sonication, and UV curing (Supakar et al., 2023). Here the printed body is itself the actuation substrate; the metamaterial is not ancillary support but the logic-bearing structure.

High-resolution two-photon and multi-photon lithography support smaller and more topologically intricate junction arrays. In trapped-ion devices, commercially available two-photon lithography such as Nanoscribe writes suspended 3D polymer structures directly on a surface-electrode chip, after which Au or Al is deposited by electron-beam evaporation to form conductive electrodes (Taniguchi et al., 21 Sep 2025). In micro-optical systems, two-photon polymerization writes IP-Dip microlens arrays directly on polished multicore-fiber facets using a 780 nm femtosecond laser and a Zeiss $45^\circ$1, NA = 1.4 objective, with 100 nm hatching and slicing for shape fidelity (Anagnos et al., 2021). Facet-attached microlenses for VCSELs, PDs, fiber arrays, and multicore fibers are similarly fabricated in situ by high-resolution multi-photon lithography using negative-tone photoresists with refractive index $45^\circ$2 (Maier et al., 2022).

Inorganic printed junction arrays have also been demonstrated. The glass lightguide array is produced from a pre-condensed liquid silica resin, then pyrolyzed at $45^\circ$3 into fully inorganic transparent silica glass with peak-to-valley deviation $45^\circ$4 and surface roughness $45^\circ$5 (Hong et al., 2022). A distinct glass-network route uses ultrashort pulse laser direct writing inside fused silica, selective KOH etching, and subsequent $45^\circ$6-laser sealing of extra access ports to realize embedded freeform microfluidic networks (Lin et al., 2019).

Rapid prototyping at the micron scale has additionally been enabled by a resonant-scanning two-photon microscope used as a direct laser writing system. That platform reports maximum single-print size of approximately $45^\circ$7, mean beam scanning speed of $45^\circ$8–$45^\circ$9, and fabrication time of about 20–25 s for a fully dense $60^\circ$0 block, with practical feature sizes of roughly $60^\circ$1 in $60^\circ$2 at standard zoom (Pearre et al., 2018). At the multi-material end of the spectrum, directed capillary assembly combined with two-photon direct laser writing links PS, silica, and PNIPAM colloids into junction-bearing 1D and 2D structures; assembly yields $60^\circ$3 are reported for large $60^\circ$4 particle arrays, and typical link-printing rates are $60^\circ$5 (Kesteren et al., 2022).

3. Droplet-generation junction arrays in microfluidics

The clearest microfluidic instantiation of a 3D-printed micro-junction array is the comparative study of T-junction, cross-junction, and asymmetric flow-focusing generators fabricated on printed chips (Barzoki et al., 2024). The system uses olive oil as continuous phase and DI water plus dyed water as dispersed phases, driven by insulin syringes at dispersed-phase flow rates of 5–20 $60^\circ$6 and a continuous-phase flow rate of 50 $60^\circ$7. Mixing is evaluated from frame-by-frame microscope videos using MATLAB on five droplets per condition.

The numerical model treats the device as quasi-2D and solves incompressible laminar Navier–Stokes flow coupled to a phase-field Cahn–Hilliard description and a convective–diffusive transport equation. The reported mixing metric is the mixing index

$60^\circ$8

where $60^\circ$9 is the number of pixels or cells in the droplet, $30^\circ$0 is the dye concentration at pixel or cell $30^\circ$1, and $30^\circ$2 is the average dye concentration inside the droplet (Barzoki et al., 2024).

The geometric dependence is strongly vortex-mediated. The T-junction forms a single recirculation vortex during the filling stage; only one side of the dispersed phase is sheared by the continuous phase, and the resulting higher-velocity vortex yields the highest MI. The cross-junction forms two independent symmetric vortices, each mixing primarily within its own hemisphere, which produces the lowest MI. The asymmetric junctions also form a single vortex, but its strength depends on the inlet angle: decreasing $30^\circ$3 from $30^\circ$4 to $30^\circ$5 strengthens the dominant continuous-phase stream and increases MI. Experimentally and numerically, the asymmetric generators outperform the cross-junction but fall slightly short of the T-junction in mixing efficiency; the asymmetric $30^\circ$6 design gives MI close to the T-junction while producing droplet diameters about $30^\circ$7 smaller (Barzoki et al., 2024).

Droplet size and eccentricity follow consistent geometric trends. The T-junction yields the largest droplets and highest eccentricity, the cross-junction the smallest droplets and lowest eccentricity, and within the asymmetric family both droplet size and eccentricity increase as $30^\circ$8 decreases, so that $30^\circ$9 in size. Increasing dispersed-phase flow rate increases droplet size and eccentricity in all geometries while lowering MI, because less time is available for vortex-driven mixing during droplet formation (Barzoki et al., 2024). The practical implication stated in the study is that sufficiently optimized junction geometry can remove the need for downstream micromixers, reducing complexity and pressure drop.

4. Addressable flow-control and embedded network architectures

Junction arrays in fluidics are not limited to passive droplet breakup. The vertically stacked valve array based on auxetic metamaterials uses re-entrant “bowtie” unit cells with re-entrant angle $\sim 100\,\mu\mathrm{m}$0 and tilt angle $\sim 100\,\mu\mathrm{m}$1 to create bands of programmable compliance (Supakar et al., 2023). Under applied load, wireframe tilted bands collapse before more rigid un-tilted bands, localizing closure to selected regions. The fluidic logic is built from two segment types: “closeable” segments, which collapse and seal under compression, and “always open” segments, which remain patent even when compressed.

The addressability is combinatorial. For $\sim 100\,\mu\mathrm{m}$2 control regions, the number of controlled flow channels is given as

$\sim 100\,\mu\mathrm{m}$3

The device algorithm assigns each flow channel a unique pattern of closeable and always-open segments so that every combination of $\sim 100\,\mu\mathrm{m}$4 pressed control regions selects a single channel (Supakar et al., 2023). Experimental demonstrations include a 6-control-region device with 18 individually addressable flow channels and a 9-control-region device with 96 flow channels in a compact vertically stacked format. FEA and flow measurements show that banded layered structures focus stress into the compliant band, and Coriolis-sensor tests confirm that only the intended closeable segments block flow (Supakar et al., 2023).

A different large-footprint realization of junction arrays is the freeform microfluidic network encapsulated in printed glass objects (Lin et al., 2019). Here, ultrashort pulse laser writing at 1030 nm, with typical pulse energy around $\sim 100\,\mu\mathrm{m}$5, 50 mm/s writing speed, and NA = 0.3, defines the buried channel network in fused silica. Selective chemical etching in 10 mol/L KOH at about $\sim 100\,\mu\mathrm{m}$6 with ultrasound creates hollow channels. The key network-level design element is the insertion of a string of extra access ports along long channels and near junctions to prevent over-etching at open ends and under-etching in deeper regions. After etching, these ports are sealed by localized $\sim 100\,\mu\mathrm{m}$7-laser melting, typically at 30–36 W for about 60 s with 10–12.5 cm defocus distance (Lin et al., 2019).

The resulting glass networks support arbitrary branches, crossings, helices, and intertwined channels with uniform diameters in the 150–350 $\sim 100\,\mu\mathrm{m}$8 range and lengths up to at least 7 cm. A fused-silica hand of approximately $\sim 100\,\mu\mathrm{m}$9 encapsulating a vascular-style network demonstrates the scale of the approach (Lin et al., 2019). Compared with the auxetic valve array, this architecture provides passive routing and encapsulation rather than actively reconfigurable gating, but both cases show that the array logic is encoded geometrically at the junction network level.

5. Ion-transport junctions for quantum processors

In trapped-ion QIP, the micro-junction array becomes a transport network for QCCD operation. One recent architecture integrates three-dimensionally structured micro RF electrodes above a conventional surface-electrode trap to control electric fields in both linear and junction regions (Taniguchi et al., 21 Sep 2025). Each X-junction unit spans a $>200$0 square and connects loading, gate, and readout branches. The overhanging RF electrode sits $>200$1 above the surface, has a diamond-like cross-section of $>200$2, and is supported by diamond-shaped beams. Minimum RF-DC spacing and RF width are both $>200$3. The authors state that more than 200 densely packed microtraps can be printed on a single chip (Taniguchi et al., 21 Sep 2025).

Away from the junction, the confinement improvement over planar traps is explicit. For a $>200$4 ion at 190 V and 44.3 MHz, the 3D-printed trap gives ion height $>200$5, trap frequency $>200$6, and trap depth $>200$7 along $>200$8. A planar trap at the same conditions gives $>200$9 and only $0.35\,\mathrm{dB}$0 depth; even when the planar trap RF power is increased to match frequency, depth reaches only $0.35\,\mathrm{dB}$1. The 3D design therefore yields a 17% higher trap frequency and more than 30 times deeper trap depth at equal RF power, while requiring 37% less RF power to achieve a given frequency and height. Third- and fourth-order anharmonic coefficients $0.35\,\mathrm{dB}$2 are smaller by factors of 2–10, and residual pseudopotential in linear regions is $0.35\,\mathrm{dB}$3 across a $0.35\,\mathrm{dB}$4 span, corresponding to excess micromotion amplitude of about 38 nm (Taniguchi et al., 21 Sep 2025).

The junction optimization is centered on bringing the constant total confinement path and the pseudopotential minimum path into near coincidence. The reported geometry reduces the pseudopotential barrier along the CTC path by about one order of magnitude relative to the best prior planar-surface designs. At the junction center, the trap height drops only to about $0.35\,\mathrm{dB}$5, the trap frequency falls only to half its linear-region value rather than to $0.35\,\mathrm{dB}$6, the RF-noise-induced heating rate is $0.35\,\mathrm{dB}$7 quanta/s at the worst spot, and the total motional excitation for a full round-trip shuttle can be as low as $0.35\,\mathrm{dB}$8 quanta per round-trip (Taniguchi et al., 21 Sep 2025). The study further notes that overhanging electrodes partially obstruct fluorescence collection, although meshed electrodes are proposed as an adaptation for integrated photonics.

A complementary experimental platform demonstrates the broader feasibility of printed 3D ion-trap arrays (Xu et al., 2023). In that system, two-photon polymerization prints the trap directly on sapphire, followed by deposition of a $0.35\,\mathrm{dB}$9-thick gold layer. The electrode set consists of four vertical RF pillars $35\,\mu\mathrm{m}$0 tall with $35\,\mu\mathrm{m}$1 spacing between opposing electrodes, plus a $35\,\mu\mathrm{m}$2 grid of planar DC electrodes. The paper estimates array densities $35\,\mu\mathrm{m}$3 traps/cm$35\,\mu\mathrm{m}$4, reports room-temperature confinement of single $35\,\mu\mathrm{m}$5 ions at 51.6 MHz with radial trap frequencies from 2.09 MHz to 24.15 MHz, and achieves $35\,\mu\mathrm{m}$6, near the theoretical limit of 0.911. At 21.29 MHz, Doppler cooling alone yields average motional occupation $35\,\mu\mathrm{m}$7, and coherent optical-qubit control shows $35\,\mu\mathrm{m}$8 contrast for many carrier $35\,\mu\mathrm{m}$9-pulses with error rates $50\,\mu\mathrm{m}$0 per $50\,\mu\mathrm{m}$1-rotation (Xu et al., 2023). Taken together, these results position the junction array as both a transport topology and a confinement-optimization problem.

6. Optical coupling, spectroscopy, and light-redistribution arrays

In photonics, the micro-junction array is often a coupling or remapping interface rather than a transport intersection in the fluidic sense. Facet-attached microlenses are printed directly on VCSEL, PD, fiber-array, and multicore-fiber facets by high-resolution multi-photon lithography (Maier et al., 2022). The printed elements may include freeform beam shapers, TIR mirrors, collars, and alignment markers; fabrication uses negative-tone photoresists with $50\,\mu\mathrm{m}$2, and sub-100 nm alignment accuracy is reported during in-situ printing. The system demonstrates coupling losses down to $50\,\mu\mathrm{m}$3 and lateral 1 dB alignment tolerances in excess of $50\,\mu\mathrm{m}$4, enabling passive machine-vision assembly. Importantly, the method directly connects individual cores of a multicore fiber to standard linear arrays of VCSELs or PDs without additional fiber-based or waveguide-based fan-out structures. A 3 $50\,\mu\mathrm{m}$5 25 Gbit/s transceiver assembly in a small form-factor pluggable module fulfills many IEEE 802.3 metrics, including a line rate of 25.78125 Gbit/s per channel and TDEC of 3.5 dB against a requirement of $50\,\mu\mathrm{m}$6 dB (Maier et al., 2022).

A related but astronomically oriented implementation is the 3D-printed MLA integrated with a custom single-mode multicore fiber in the 3D-M3 instrument (Anagnos et al., 2021). The fiber contains 19 cores in a hexagonal arrangement with 40 $50\,\mu\mathrm{m}$7 pitch and $50\,\mu\mathrm{m}$8 mode-field diameter at 980 nm, and the printed aspheric lenslets are generated directly from the measured core positions on the polished facet. The plate scale is 0.45 arcseconds/mm, so each core samples 18 mas on sky, and 7 cores span an instantaneous 54 mas field. Simulations give total coupling efficiency of 48.8%; laboratory measurements show average per-core throughput of $50\,\mu\mathrm{m}$9 and maximum throughput of $80^\circ\mathrm{C}$0, with surface roughness of 37 nm RMS. Throughput is empirically related to Strehl ratio by $80^\circ\mathrm{C}$1 (Anagnos et al., 2021). The junction-array function here is spatial segmentation and modal matching at diffraction-limited scale.

The glass lightguide array for snapshot hyperspectral imaging implements an image-to-sparse transformation in a monolithic printed optic (Hong et al., 2022). The guides are printed in a pre-condensed liquid silica resin, converted to transparent silica glass at $80^\circ\mathrm{C}$2, and can be as small as $80^\circ\mathrm{C}$3 in diameter. The prototype contains $80^\circ\mathrm{C}$4 lightguides, while scalability to resolutions $80^\circ\mathrm{C}$5 is claimed. Measured average transmission for a $80^\circ\mathrm{C}$6-diameter, $80^\circ\mathrm{C}$7-long guide exceeds 80% across the visible. Spectral dispersion is supported by sparse output reformatting, and the demonstrated system resolves 21 wavebands from 450 nm to 650 nm in 10 nm steps. The coordinate transformation is formalized as

$80^\circ\mathrm{C}$8

Biological inspiration is explicit: curved input and output faces mimic retinal and compound-eye arrangements, while cladding-free dense packing maximizes fill factor (Hong et al., 2022). In contrast to the FaML and MLA systems, which optimize coupling between pre-existing channels, the lightguide array creates the channels themselves.

7. Cross-cutting design logic, performance criteria, and limitations

Across domains, performance is controlled less by the mere presence of 3D printing than by junction-level geometry. In droplet generators, decreasing asymmetric inlet angle from $80^\circ\mathrm{C}$9 to $45^\circ$00 strengthens the dominant internal vortex and raises MI (Barzoki et al., 2024). In auxetic valves, compliance localization arises from the contrast in spring constant $45^\circ$01 between tilted and untilted bands, with $45^\circ$02 increasing with $45^\circ$03 and decreasing with $45^\circ$04 (Supakar et al., 2023). In ion traps, the decisive geometric criterion is the near-overlap of CTC and pseudopotential minimum paths at X-junctions (Taniguchi et al., 21 Sep 2025). In freeform optical couplers, off-axis cores require explicit beam steering, such as the 14.6° beam tilt used for outer channels in multicore-fiber coupling (Maier et al., 2022). This suggests that the junction is best understood as a local field-transforming element rather than simply a connection point.

A recurrent misconception is that additive fabrication automatically removes planar constraints or obviates reduced-order modeling. The fluidic literature explicitly states that many devices remain quasi-two-dimensional and that 2D simulations are valuable mainly for qualitative and semi-quantitative analysis in such layouts (Barzoki et al., 2024). The valve-array work begins from the observation that many fluidic devices are still often restricted to quasi-planar structures and frames the vertically stacked metamaterial specifically as a way to overcome that restriction rather than as evidence that it has already been solved (Supakar et al., 2023). In ion trapping, 3D structure improves confinement but introduces new packaging tradeoffs such as partial obstruction of fluorescence collection by overhanging electrodes (Taniguchi et al., 21 Sep 2025).

Manufacturing constraints also remain architecture-specific. Resonant-scanning two-photon printing is fast but confined to roughly $45^\circ$05 per field, and feature anisotropy is strongest along the resonant $45^\circ$06-axis, where the measured minimum feature is 2.5–5.6 $45^\circ$07 at $45^\circ$08 zoom (Pearre et al., 2018). In glass microfluidic networks, uniform etching of long or highly branched channels requires extra access ports, and practical sealing reliability declines for port diameters above about $45^\circ$09 (Lin et al., 2019). In optical systems, performance may depend on post-print alignment or detector mapping even when the printed optic itself is monolithic (Maier et al., 2022, Hong et al., 2022).

An emerging extension is the use of printed micro-junctions as material connectors rather than flow or field transformers. “Printing-on-particles” combines capillary-assisted assembly with two-photon direct laser writing to connect colloids into 1D and 2D architectures, including checkerboards, lattices, curved links, and out-of-plane bridges, while preserving the possibility of harvesting and redispersion in water (Kesteren et al., 2022). A plausible implication is that future micro-junction arrays may be simultaneously structural, mechanical, and functional, with the junction serving as both interconnect and material interface.