Main

Three-dimensional (3D) microfabrication/nanofabrication technologies have attracted tremendous attention as they can create various functional microdevices/nanodevices, such as microrobots1,2, microactuators3, micrometre-scale metamaterials4,5,8 and microphotonic/nanophotonic devices6. Over the past two decades, two-photon polymerization (2PP) has emerged as the state-of-the-art 3D microfabrication/nanofabrication strategy owing to its high resolution (up to 100 nm), facile fabrication process and capability of printing intricate free-form 3D microstructures/nanostructures9,10,11,12. However, the material compatibility of 2PP remains highly limited to cross-linkable polymers. Recent efforts have aimed to expand printable materials beyond polymers, primarily through developing advanced photoresists with tailored chemistries, such as grafting photochemically bonding ligands onto inorganic colloidal nanocrystals13,14 or incorporating metal-coordination complexes into cross-linkable monomers15,16,17,18,19,20,21. Nonetheless, these approaches remain restricted to specific materials and continue to face challenges in achieving broad compatibility with diverse material systems.

As a promising approach to overcome these material limitations, direct assembly of material building blocks has proved effective for 3D microstructures/nanostructures22. Among various assembly techniques, optical assembly23,24,25,26,27,28,29,30 has been an appealing strategy for the construction of complex material assemblies. It uses the non-specific light–matter interactions (for example, optical gradient forces31 and light-controlled electric or temperature fields32) to trap microparticles/nanoparticles suspended in a solution. The trapped particles can then be transported and positioned one by one at designated locations, enabling high-precision single-particle assembly. Techniques such as optical tweezers33, opto-thermophoretic assembly34, opto-thermoelectric assembly35 and optothermal-flow-based assembly24,27,36,37 have been used to assemble colloidal particles with diverse sizes, shapes and surface properties. However, these approaches remain largely limited to 2D structural configurations and exhibit low assembly efficiencies, typically on the order of 101–103 particles min−1. More importantly, establishing a general optical assembly platform with broader material applicability is still challenging, as most techniques impose strict requirements on parameters, such as particle size/surface properties, solvent composition and other environmental conditions. A detailed comparison of various assembly strategies is provided in Supplementary Table 1. Therefore, innovations in general assembly mechanisms are urgently needed to achieve controllable, non-specific guidance and precise assembly of material building blocks at the microscale/nanoscale, enabling the creation of high-quality, truly volumetric free-form 3D architectures with broad material compatibility.

Here we propose a universal 3D microfabrication/nanofabrication strategy compatible with a broad range of materials. This technique uses the optofluidic interaction—light-driven flow—to efficiently assemble diverse micromaterials/nanomaterials into predefined 2PP-printed microtemplates, creating high-quality, truly volumetric free-form 3D microarchitectures/nanoarchitectures from a variety of materials, including diamond, metal, metal oxide, quantum dots and others. By integrating the optofluidic assembly process with 2PP, our optofluidic 3D microfabrication/nanofabrication strategy proposes a generalizable model that overcomes the material limitations of 2PP, opening new avenues for nanotechnology (Supplementary Fig. 1 and Supplementary Table 2).

Concept of optofluidic 3D micro-/nanofabrication

The working process of optofluidic 3D microfabrication/nanofabrication is illustrated in Fig. 1a and consists of the following key steps. First, a 3D hollow polymeric microstructure with an open hole (for example, a cube) is printed on a glass substrate by the 2PP process, serving as the 3D microtemplate. Next, the printed template is immersed in a solution containing uniformly dispersed nanoparticles (or few-micrometre particles). A femtosecond (fs) laser with a beam diameter of 2 µm is then applied near the open hole of the template, generating a sharp temperature gradient that induces a strong convective flow (up to several mm s−1), propelling the dispersed particles towards the open hole. As a result, these particles are transported into the hollow microtemplate and accumulate over time, ultimately assembling into the prescribed 3D shape dictated by the template. Following assembly, the polymer template is selectively removed through rational post-treatments (see details in Methods), yielding a free-standing 3D volumetric microarchitecture composed entirely of the targeted materials. For example, our approach enables the fabrication of a series of solid microcubes randomly assembled by SiO2 nanoparticles with an assembly efficiency of approximately 105 particles min−1, as shown in the scanning electron microscopy (SEM) image in Fig. 1b, its enlarged image in Fig. 1c and Supplementary Fig. 2. The resulting 3D microarchitecture exhibits high structural integrity, as these constituent nanoparticles are strongly bonded and stabilized by van der Waals forces. Even without further chemical bonding or annealing process for interfacial improvement, these 3D structures can be self-supporting and mechanically stable, owing to the strong van der Waals interactions in colloidal nanoparticle assemblies12. This is further evidenced by the successful fabrication of a dangling croissant-shaped superstructure with intricate 3D curved surfaces (Fig. 1d,e), demonstrating the robustness of this technique.

Fig. 1: Concept of optofluidic 3D microfabrication/nanofabrication.

a, Schematic illustration of the optofluidic 3D microfabrication/nanofabrication process, in which a localized temperature gradient induced by femtosecond (fs) laser heating generates a strong convective flow, guiding the 3D assembly of microparticles/nanoparticles within a confined hollow 3D microtemplate, printed by 2PP. b,c, SEM images of a SiO2 colloidal-particle-assembled microcube (b) and its zoomed-in view (c). d,e, SEM images of a dangling croissant-shaped microstructure with 3D curved surface assembled from SiO2 particles (d) and its enlarged view (e). The inset in d is the 3D model of the croissant structure. f, Simulation result showing the temperature distribution and fluid flow field around a hollow microcube following fs laser heating. g, Schematic illustrations and time-lapse optical images showing the assembly process of SiO2 nanoparticles within a hollow microcube. These images are extracted from Supplementary Video 1. Scale bars, 10 μm (b,d); 4 μm (c,e); 20 μm (g).

To gain deeper insights into the assembly process, we conducted simulations and recorded the experimental process to analyse the underlying dynamics. As shown in Fig. 1f, Supplementary Fig. 3 and Supplementary Video 1, when laser heating is applied at the open hole of a hollow microcube, a sharp thermal gradient near the hotspot is induced. This thermal gradient causes variations in fluid density, pressure and surface tension around the hole26,38, generating a strong, directed fluid flow towards the open hole, transporting dispersed SiO2 particles into the microcube. Furthermore, we note that bubble formation with a diameter up to 100 µm (Supplementary Fig. 4) occurs easily owing to solvent evaporation induced by femtosecond laser heating, leading to an extra flow from the Marangoni effect39, and this disturbance further promotes and accelerates particle assembly within the template (Supplementary Video 1). As shown in Fig. 1g and Supplementary Video 1, an assembly speed of around 700 µm3 s−1 is reached when assembling SiO2 colloidal particles with a diameter of 1 µm, which is about twice as fast as the typical 2PP process (Supplementary Fig. 5). By precisely manipulating the light-driven fluid flow through localized laser irradiation, any nanoparticles carried in the flow can be efficiently assembled into 3D microarchitectures. This non-selective mechanism makes the strategy inherently adaptable to a broad range of materials.

Assembly mechanism

The assembly process is primarily governed by two competing physical forces: inter-particle interactions and particle–fluid interactions. We start with a model system in which SiO2 colloidal particles (150 nm in diameter) assemble in aqueous solutions with varying ionic strengths. In this system, the inter-particle interactions consist of van der Waals attraction and electric double layer (EDL) repulsion, commonly described as the net DLVO interaction force40,41. The inter-particle DLVO force is highly sensitive to particle distance and particle properties and plays a crucial role in colloidal cluster formation. Meanwhile, the particle–fluid interactions are represented by hydrodynamic forces, particularly the Stokes drag force, which reflects how particles are carried by the surrounding flow. In our local assembly region, it tends to disperse the particles and counteracts aggregation (Fig. 2a). To determine whether colloidal particles will form clusters, we analyse the balance between inter-particle potential energy and hydrodynamic effects. The total interaction energy governing the system can be expressed as: Utotal = UDLVO + Wfluid, in which UDLVO represents the interaction potential between particles and Wfluid denotes the work done by hydrodynamic forces. The DLVO potential energy is derived from the integration of the inter-particle DLVO force over the separation distance, whereas the work done by the fluid force depends on the external flow field and particle motion. The tendency of the system to form clusters can then be evaluated by examining changes in Utotal. If ΔUtotal < 0, the inter-particle attractive interactions dominate, leading to energetically favourable conditions for colloidal clustering around the laser spot. If ΔUtotal > 0, particle–fluid interactions prevail, causing particles to move with the fluid flow and remain dispersed (Fig. 2a).

Fig. 2: Assembly mechanism.

a, Schematic illustration showing the competition between DLVO interactions and Stokes drag force and their influence on colloidal cluster formation following femtosecond laser heating. b, Cluster area (purple spheres) of SiO2 particles dispersed in aqueous solutions with varying NaCl concentrations after 60 s of laser heating at a power of 50 mW and a scan speed of 5 μm s−1, along with the corresponding zeta potential value (black diamonds) of these SiO2 particles. The insets are optical images of SiO2 clusters formed in 0.6 M and 1 M NaCl solutions. Data are presented as mean ± s.d. with at least four independent samples/measurements (n ≥ 4). c,d, Schematic illustrations and optical images depicting the assembly process of SiO2 particles within a hollow microcube in 1 M (c) and 100 mM (d) NaCl solution. The microcube measures 32 μm in length and width and 10 μm in height, with an opening of 26 μm in length and 10 μm in height on the side. e, Theoretical phase diagram (grey background) and experimental results (open circles and solid diamonds) showing the influence of various concentrations of NaCl and flow speed on SiO2 particle clustering. The light-grey region and open circles represent the domain in which SiO2 particles remain dispersed and no cluster is formed, whereas the dark-grey region and solid diamonds represent the domain in which SiO2 cluster formation occurs. Scale bars, 30 μm (b); 10 μm (c,d).

Source data

The template we used plays a crucial role in enabling deterministic 3D fabrication: it defines the overall geometry to ensure well-defined edges and symmetry, channels the optofluidic flow to reproducibly fill complex volumes and provides design versatility by allowing diverse 3D architectures. To effectively assemble dispersed SiO2 colloidal particles within a confined 3D space, the system must satisfy the condition ΔUtotal < 0 to promote colloidal cluster formation. This can be achieved through two approaches: (1) increasing the inter-particle attractive interaction or (2) decreasing the speed of fluid flow. We first investigate the effect of varying inter-particle interactions on SiO2 cluster formation. In aqueous solution, SiO2 particles are highly surface charged, resulting in strong electrostatic repulsion that stabilizes the dispersion42,43,44. By tuning the ionic strength of the solution, the zeta potential of SiO2 colloidal particles and the Debye length—a characteristic length scale over which electric fields are naturally screened—can be systematically adjusted, thereby influencing inter-particle electrostatic interactions. As illustrated in Fig. 2b, increasing the NaCl concentration progressively decreases the zeta potential of SiO2 particles and compresses the Debye length, thus reducing the electrostatic repulsion within the EDL. Consequently, the attractive component of the DLVO interaction is enhanced, leading to stronger particle aggregation. At higher NaCl concentrations, this effect facilitates the formation of larger SiO2 clusters, whereas no notable clustering is observed when the NaCl concentration is less than 200 mM. Correspondingly, as depicted in Fig. 2c, SiO2 particles effectively assemble and accumulate within a hollow template in a solution with 1 M NaCl concentration (Supplementary Video 2, part 1), whereas no assembly is observed at a lower NaCl concentration (100 mM; Fig. 2d and Supplementary Video 2, part 2).

Flow speed is another vital factor affecting the cluster formation of the SiO2 colloidal particles. Higher flow speeds generate stronger Stokes forces acting on colloidal particles, hindering their aggregation. By comparing the relative values between UDLVO and Wfluid, a critical flow speed of approximately 300 µm s−1 is theoretically determined and summarized in a phase diagram (Fig. 2e; see Methods for details). Above this critical speed, the Stokes force dominates and completely overcomes the DLVO attraction, preventing cluster formation regardless of NaCl concentration. Below this threshold, the DLVO force becomes comparable to the Stokes force, allowing cluster formation at high NaCl concentrations (for example, >0.5 M). By adjusting the laser scan speed, different flow speeds can be generated experimentally. Overall, the assembly process was systematically conducted and analysed across various NaCl concentrations and flow speeds. The experimental results (represented as discrete symbols in Fig. 2e) closely align with theoretical predictions. For example, SiO2 particles in 1 M NaCl solution aggregate into clusters at flow speeds less than 320 µm s−1, whereas they remain dispersed at higher flow speeds of about 900 µm s−1 (Supplementary Fig. 6 and Supplementary Video 3). Besides, an extremely weak flow field cannot provide sufficient driving force to continuously transport the particles towards the inside of a template, failing to fill up the template (Supplementary Video 4).

Assembly in various solvent systems

To broaden the compatibility of the optofluidic 3D microfabrication/nanofabrication technique, the 3D assembly of microparticles/nanoparticles can be extended to various solvent systems, as different colloidal particles in different solutions exhibit varying balances between inter-particle and particle–fluid interactions. For instance, replacing a good solvent with a poor solvent can enhance particle–particle attraction, leading to the clustering of the colloidal particles45,46 (Fig. 3a). When hydrophilic SiO2 particles are redispersed into hydrophobic solvents, such as immersion oil, mineral oil or oleic acid, the strong hydrophobic interaction facilitates continuous cluster growth (Fig. 3b), resulting in a faster growth of SiO2 cluster under laser illumination (Fig. 3c). Consequently, SiO2 particles can be efficiently assembled and fully filled within a hollow micro-template in these solvent systems (Supplementary Video 5). More importantly, the strong hydrophobic interaction enables efficient 3D assembly at a flow speed of several mm s−1, even under substantial Stokes force, greatly improving assembly efficiency. Furthermore, the assembly speed can be modulated by adjusting the laser dosage (the laser power and scan speed; see Supplementary Fig. 7). Notably, SiO2 particles dispersed in silicone oil exhibit poor clustering ability and fail to undergo 3D assembly within the hollow template (Supplementary Video 6). This might be because of weak attractive interaction among SiO2 particles, as their strong affinity to silicone oil (attributing to their mutual Si–O bonds) prevents clustering.

Fig. 3: Assembly optimization using various solvent systems.

a, Schematic illustration showing the clustering behaviours of colloidal particles in solutions with different particle–solvent affinities. b, Cluster area of SiO2 particles over time in different solvents. Data points are extracted from Supplementary Video 16 at 10-s intervals. c, Cluster area of SiO2 particles in various solvents after 60 s of laser heating at a power of 50 mW and a scan speed of 500 μm s−1. Insets are optical images of SiO2 clusters in oleic acid and silicone oil. d, Schematic illustration depicting the effect of surfactant on colloidal particle assembly in the aqueous solutions. The addition of surfactant reduces the surface tension gradient, attenuating the bubble growth and weakening Marangoni flow, thereby facilitating the particle clustering. e, Cluster area of SiO2 particles (purple spheres) and flow speed (black diamonds) in aqueous solutions with different concentrations of CTAB. Insets are optical images of SiO2 clusters in pure water and 1 mM CTAB solution. f, Cluster area of SiO2 particles in solutions containing different surfactants (1 wt% PF108, 1 wt% PEG and 8 mM SDS). Insets are optical images of SiO2 clusters in solutions of 1 wt% PF108 and 8 mM SDS. The laser power and scan speed for e and f are both 50 mW and 5 μm s−1, respectively, with a duration of 10 s. Data points are shown as mean ± s.d. with at least three independent samples/measurements (n ≥ 3). Scale bars, 30 μm (c,e,f).

Source data

Beyond solvent selection, surfactants can also enhance efficiency in 3D assembly in aqueous solutions. Laser illumination often induces bubble formation owing to high-temperature-driven evaporation, generating Marangoni flow caused by the surface tension gradients at the bubble–solution interface. Moderate flows promote SiO2 particle clustering, whereas a strong flow inhibits cluster formation (Fig. 3d,e and Supplementary Video 7), as the bubble formation frequently results in intense fluid flow exceeding 4 mm s−1. The addition of a surfactant, such as hexadecyltrimethylammonium bromide (CTAB), effectively reduces the surface tension, limiting bubble growth and weakening the laser-induced Marangoni flow39. Moreover, the cationic ions (CTA+) dissociated from CTAB molecules absorb onto the surface of negatively charged SiO2 particles, neutralizing their surface charge (Supplementary Fig. 8). This reduction in electrostatic repulsion enhances the DLVO attractions among the SiO2 particles, thereby promoting their aggregation. As a result, the SiO2 clustering is progressively enhanced by increasing CTAB concentration (Fig. 3e), enabling robust 3D assembly in 1 mM CTAB solution (Supplementary Video 8). Beyond CTAB (cationic surfactant), other surfactants, including anionic surfactant sodium dodecyl sulfate (SDS) and non-ionic surfactants such as polyethylene glycol (PEG) and Pluronic F-108 (PF108), exhibit similar effects on SiO2 particle clustering (Fig. 3f). Notably, SiO2 particles exhibit strong electrostatic repulsion in the solution of SDS (Supplementary Fig. 8), which may hinder the cluster formation. However, other inter-particle attractions, such as opto-thermophoretic force23,34,47 and depletion force48, may also arise in these surfactant systems, facilitating particle aggregation.

Broad compatibility with versatile materials

Our strategy is superior at constructing complex 3D microstructures using a diverse range of micromaterials/nanomaterials, regardless of their shape, size and surface chemistry. To demonstrate the universality of this approach, we first assembled SiO2 colloidal particles of varying sizes and surface chemistries into versatile 3D microstructures. These include a micro-gourd composed of 1-µm-diameter SiO2 (Fig. 4a–c), a micro-hexagon made of 600-nm-diameter SiO2 particles (Fig. 4d–f) and microcubes constructed from 1-µm-diameter green fluorescent SiO2 particles (Supplementary Fig. 9) and 3-µm-diameter and 10-µm-diameter SiO2 particles (Supplementary Video 9). Furthermore, particles of different sizes can be co-assembled to form heterogeneous 3D microarchitectures. For instance, a microsphere consisting of both 1-µm and 600-nm SiO2 particles is successfully co-assembled (Fig. 4g–i), demonstrating the capability of integrating different components into 3D microstructures.

Fig. 4: Wide compatibility with versatile micromaterials/nanomaterials.

a–c, 3D model (a) and SEM images (b,c) of a micro-gourd superstructure assembled with 1-μm SiO2 particles. d–f, 3D model (d) and SEM images (e,f) of a hexagon-shaped micro-superstructure assembled with 600-nm SiO2 particles. g–i, 3D model (g) and SEM images (h,i) of a microsphere co-assembled with 1-μm and 600-nm SiO2 particles. j–l, Model (j) and SEM images (k,l) of superstructures of the letters ‘P’ (assembled with 1-μm SiO2 particles) and ‘I’ (assembled with 600-nm SiO2 particles). m,n, 3D model (m) and SEM image (n) of a TiO2 nanoparticle (NP)-assembled screw-like microstructure. o, TEM image of TiO2 NPs. p,q, 3D model (p) and SEM images (q (i), (ii)) of the letter ‘E’ composed of Fe3O4 NPs. r, TEM image of Fe3O4 NPs. s–z, Model (s), SEM image (t) and EDS mapping (u–z) of a microcube assembled with various materials, including SiO2 (t), TiO2 NWs (u), WO3 NWs (v), diamond NPs (w), Al2O3 NWs (x), Fe3O4 NPs (y) and Ag NPs (z). The high-resolution surface morphology of the microstructures assembled with different nanomaterials, along with the corresponding component nanomaterials, can be found in Extended Data Figs. 1 and 2. Scale bars, 10 μm (b,e,h,k,t); 4 μm (c,f,i,l); 5 μm (u–z); 2 μm (n,q (i)); 800 nm (q (ii)); 200 nm (o); 80 nm (r).

Furthermore, architectures at distinct locations can be site-selectively assembled with particles of different sizes or components by locally addressing the fs laser, with no cross-interference among these assembled architectures. For example, the alphabet letters ‘P’ and ‘I’ are sequentially assembled using 1-µm and 600-nm SiO2 within a close mutual distance of about 10 µm on the substrate (Fig. 4j–l). Once assembled within a template, strong inter-particle interactions resist disturbances from the violent flow during sequential assembly process, preventing disassembly. This precise control enables the localized integration of several components for a targeted layout, paving the way for the fabrication of microdevices with spatially varying compositions and on-demand multifunctionalities (see details in Fig. 5). Beyond these micrometre-sized SiO2 particles, our method is compatible with a broad range of nanomaterials, enabling the fabrication of diverse 3D structures with nanoscale features. As illustrated in Fig. 4m–r, a screw-like microstructure with helical threads of around 320 nm in width (Fig. 4m,n) and an alphabet letter ‘E’ with a height of around 855 nm (Fig. 4p–r) are created using TiO2 nanoparticles with 90.2 ± 15.8 nm diameter and Fe3O4 nanoparticles with 16.7 ± 2.3 nm diameter, respectively. Notably, smaller nanoparticles with a uniform size distribution yield microstructures/nanostructures with smoother surfaces (Fig. 4q and Extended Data Fig. 1). Furthermore, as illustrated in Fig. 4s–z and Extended Data Fig. 2, we successfully assemble microcubes from various nanomaterials, including TiO2 nanowires (NWs), diamond nanoparticles (Supplementary Fig. 10 and Supplementary Video 10), Fe3O4 nanoparticles, WO3 NWs, Al2O3 NWs, Ag nanoparticles and CdTe quantum dots (Supplementary Video 11), respectively. Notably, although these architectures exhibit intrinsic mechanical robustness (as demonstrated by the dangling croissant-shaped superstructure in Fig. 1d,e) owing to strong intrinsic inter-particle interactions, post-treatments such as annealing can further promote intra-particle interfacial welding, substantially enhancing the mechanical properties of the 3D microarchitectures.

Fig. 5: On-demand construction of multifunctional microdevices.

a–e, Microfluidic sieving devices. a, Schematic illustration showing the capillary-driven separation process of a microfluidic sieving device. b, Time-lapse optical images showing the capillary-driven fluid flow through a microfluidic chip embedded with a colloidal microvalve (20 μm width, 40 μm length) composed of 1-μm SiO2 particles. Images are extracted from Supplementary Video 17. c, SEM image of the SiO2-assembled microvalve. d, Fluorescent image of 500-nm polystyrene (PS) nanospheres rejected by the microvalve. e, Fluorescent image showing 100-nm PLGA nanoparticles rejected by the microvalve. Inset is the variation of gray value along the microchannel. f–s, Multifield-driven microrobots. f,g, Schematic illustration (f) and trajectory (g) of the magnetic tumbling of a Fe3O4 nanoparticle (NP)-assembled cylinder microrobot. h–o, Schematic illustrations (h,j,l,n) and motion trajectories (i,k,m,o) of TiO2–Au microrobots with different shapes and material distributions. p–s, Schematic illustration (p) of an L-shaped microrobot integrated with four different materials (TiO2, Au, Pt and Fe3O4) and its three motion modes: magnetic pulling (q), light-driven anticlockwise rotation (r) and clockwise rotation in about5 wt% H2O2 (s). Scale bars, 50 μm (b,d,e); 60 μm (g,i,k,m,o,r,s); 30 μm (q); 10 μm (c).

Source data

On-demand creation of multifunctional microdevices

With the aid of precise spatial control and broad material applicability, microstructures spatially encoded with diverse functional materials can be created, highlighting the potential of our technique for developing microdevices with on-demand functionalities. As proof-of-concept demonstrations, microfluidic chips with tailored separation capabilities for tiny objects are first demonstrated (Fig. 5a–e). This microfluidic chip consists of a polymeric microchannel printed by the 2PP process with a colloidal-particle-assembled microvalve embedded inside (Fig. 5b,c). The microvalve, measuring 40 µm in length and 20 µm in width, is entirely assembled from 1-µm SiO2 particles, forming complex 3D porous channels (Fig. 5c). When a droplet of nanoparticle suspension is introduced at one end of the microfluidic chip, capillary forces instantaneously drive the solution into the microchannel (Fig. 5a). The colloidal microvalve, with feature porosity of several hundred nanometres, permits the rapid permeation of solvent flow in seconds (Fig. 5b), while efficiently rejecting and retaining nanoparticles (Fig. 5d). As solvent evaporation continues at the opposite end of the chip, nanoparticles progressively accumulate at the inlet side of the microvalves, forming an enrichment zone (Fig. 5a and Supplementary Fig. 11), which holds potential for trace substance detection. By modulating the size of the microvalves, we can selectively separate particles of different sizes. For instance, 100-nm poly(lactic-co-glycolic) acid (PLGA) nanoparticles are successfully sorted out using a SiO2 valve of 40 µm in both width and length (Fig. 5e). Furthermore, we also construct a microfluidic chip embedded with several concatenated microvalves composed of different materials, each featuring distinct porosities and cut-off capabilities, achieving the size-selective sorting of a mixture of different particles (Supplementary Fig. 12).

Subsequently, we demonstrate the application of our technique in fabricating multifield-driven microrobots with multimodal locomotion (Fig. 5f–s). As illustrated in Fig. 5f,g, a Fe3O4 nanoparticle-assembled cylinder microrobot exhibits magnetically controlled tumbling on the substrate (Supplementary Video 12). The magnetic response can be flexibly tuned by regulating the Fe3O4 nanoparticle content (Supplementary Fig. 13). Beyond magnetic actuation, Fig. 5h–o illustrates light-driven microrobots with controlled motion enabled by tailoring both the shape and spatial distribution of functional materials. For example, a cylinder heterojunction TiO2–Au microrobot demonstrates linear propulsion through self-electrophoresis49 under ultraviolet (UV) illumination in water (Fig. 5h,i, Supplementary Fig. 14 and Supplementary Video 13, part 1). Shape asymmetry further enables rotational motion: a T-shaped TiO2–Au microrobot rotates in tight circles (Fig. 5j,k, Supplementary Video 13, part 2), whereas a V-shaped design produces larger circular trajectories (Fig. 5l,m and Supplementary Video 13, part 3). Moreover, alternative spatial encoding of functional materials allows the same V-shaped microrobot to switch from rotation to linear motion (Fig. 5n,o and Supplementary Video 13, part 4). Notably, our technique also enables the integration of several functional materials into a single microrobot, achieving multistimulus responsiveness. For instance, we fabricate an L-shaped microrobot incorporating Au, TiO2, Pt and Fe3O4 (Fig. 5p–s), which exhibits three distinct motion modes: magnetic pulling, anticlockwise rotation under UV light and clockwise rotation in H2O2 (Supplementary Video 14).

In conclusion, our technique overcomes the material limitation in conventional 3D microfabrication/nanofabrication, enabling the creation of complex, volumetric 3D microstructures/nanostructures from a diverse range of materials. We believe that this step change in fabrication capability—bridging optical physics, colloidal science, fluid mechanics and device engineering—not only provides new insights into fundamental colloidal assembly but also opens new frontiers in various fields, such as reconfigurable photonics, multifunctional microdevices/microrobots and biologically integrated systems.

Methods

Materials

SiO2 colloidal particles with various sizes (140 nm to 5 μm, 5 wt%), green-fluorescent SiO2 (1 μm, excitation/emission: 497/530 nm, 2.5 wt%) and red-fluorescent polystyrene nanospheres (500 nm and 2.5 μm, excitation/emission: 530/607 nm, 2.5 wt%) were purchased from microParticles GmbH. Mineral Oil Rotational Viscosity Standard (494.0 mPa s), oleic acid (technical grade, 90%), silicone oil AR20 (viscosity: approximately 20 mPa s), CTAB (98%), SDS (98.5%), PEG (Mn: 400), PF108 (Mn: 14,600), TiO2 NWs (powder, 100 nm × 10 μm), WO3 NWs (powder, 50 nm × 10 μm), Al2O3 NWs (powder, 2–6 nm × 200–400 nm), CdTe quantum dots (powder, COOH functionalized, fluorescent emission: 710 nm), iron oxide (Fe3O4) powder (50–100 nm, 97%), Fe3O4 (20 nm, 5 mg ml−1, dispersed in H2O), TiO2 nanoparticles (150 nm, 900 nm, 5 wt%, dispersed in H2O), Au urchin nanoparticles (50 nm, in 0.1 mM PBS), Pt nanoparticles (powder, 50 nm, 99.9%), silver (Ag) powder (<100 nm, PVP as dispersant, 99.5%), orange fluorescent PLGA nanospheres (excitation/emission: 530/582 nm, 100 nm), diamond nanopowder (<10 nm, 97%), trichloro(1H,1H,2H,2H-perfluorooctyl) silane (97%), sodium chloride (NaCl, 99%), H2O2 (30% w/w in H2O) and propylene glycol monomethyl ether acetate (PGMEA, 99.5%) were purchased from Sigma-Aldrich. Isopropyl alcohol (IPA, 99.9%) was purchased from Carl Roth GmbH. IPS photoresist was purchased from Na