Ice-phase optothermal tweezers Authors: Jianxing Zhou, Yuhang Peng, Peng Du, Zhen Chen, Jinming Zhang, Zhengtian Jin, Xiaoqi Dai, Yinyue Ji, Meiting Wang, Yuye Wang, Yuebing Zheng, Jiajie Chen, Junle Qu, Yonghong Shao Published: 2026 ABSTRACT The deterministic and reconfigurable assembly of heterogeneous micro- and nanostructures remains challenging, where Brownian motion, nonspecific interfacial adsorption, and method-specific material constraints undermine precision and cross-material compatibility. Here, we present ice-phase optothermal tweezers (IOT), a platform that utilizes the ice–water interface for fast (10–100 μm/s), programmable manipulation of diverse targets, including dielectric/metal nanoparticles, proteins, DNA, and even microbubbles within ice, which have long been regarded as passive byproducts of freezing. In this scheme, the optothermally induced mobile melting zone within the ice matrix enables directed transport and nanostructure assembly with nanometer-scale resolution. We further fabricate multi-material heterostructures with tailored anisotropic and chiral optical responses, enabling applications in chiral photonics and polarization-controlled nano-optical devices under cryogenic environments. Accordingly, by unifying precise control of micro- and nano-entities across fluidic and solid-state regimes, IOT provides a general cryogenic platform for meta-structure fabrication, in-situ optical spectroscopy, and nanoparticle-interaction studies. Introduction The programmable assembly of heterogeneous nanoscale components into complex, functional architectures or patterns would unlock transformative technologies, from optical metamaterials to sophisticated bio-interfaces. Yet this goal remains fundamentally constrained by a persistent paradox: Liquid-phase manipulation enables nanoscale mobility but guarantees instability, while solid-state fixation ensures stability at the cost of dynamic programmability. Current high-precision techniques (e.g., optical/acoustic tweezers) inevitably suffer from disruptive Brownian motion and irreversible surface adsorption, precluding robust reconfiguration of nanoscale assemblies1,2. Conversely, directed self-assembly strategies, despite achieving thermodynamic stability, lack the arbitrary programmability or broad material compatibility required for constructing truly heterogeneous architectures with multi-component complexity3,4. This impasse has stifled progress toward applications demanding both structural integrity and biocompatibility, such as optical metamaterials, bio-interfaces engineering, and biosensing. Currently, various advanced optical manipulation techniques have been developed, including plasmonic tweezers5–9, hybrid electro-optical tweezers10–15, optothermal tweezers16–23, and low-temperature optothermal tweezers24–26, which enable low-power trapping of diverse colloidal particles across expanded size and material ranges. However, these delicate approaches remain fundamentally constrained by two persistent challenges. First, their operation necessitates stringent environmental conditions, such as specific particle surface charges, solvent conductivity, or the introduction of auxiliary solutes to generate thermophoretic or electrokinetic forces. Second, akin to conventional optical tweezers27, their manipulation fidelity is inevitably compromised by Brownian motion28,29, limiting nanoscale positional stability. Crucially, reconfigurable manipulation of particles immobilized within solid media is inherently difficult, as van der Waals, electric double-layer, and friction forces render adhesion effectively irreversible under gentle actuation. Current state-of-the-art strategies, including surfactant-modified substrates30–32 and pulsed optothermal modulation33, impose material-specific requirements (e.g., light-absorbing particles), provide only a limited manipulation size range, and compromise biocompatibility, thereby restricting their broader applicability. Here, we overcome this fundamental limitation by introducing a paradigm that harnesses the physics of ice-water phase change. We developed ice-phase optothermal tweezers (IOT), a technique that uses a focused laser to create and precisely steer a localized melting zone within a solid ice matrix. This mobile melting zone acts as a fluidic workspace where particles can be manipulated, while the surrounding ice matrix provides a cryogenic reference frame. The manipulation itself is driven by the forces exerted by the advancing solidification front, a process governed by a delicate interplay of intermolecular, thermal, and viscous forces. Because the manipulated particles are propelled by the slippery ice–water interface with a relatively large curvature on the micrometer scale, we achieve transport at exceptionally high speeds (10–100 µm/s), far surpassing previous surface-manipulation strategies based on optical forces or sol–gel phase transitions, which typically remain below 10 µm/s3,30. This approach enables the user-defined assembly of a broad spectrum of materials, including dielectric and metallic nanoparticles, proteins, DNA, and microbubbles. We demonstrate reversible merging, splitting, deformation, and programmable sculpting of microbubbles within ice, whereas previous studies achieved only passive, one-directional deformation under freezing dynamics34. This capability transforms bubbles from passive markers of freezing into programmable elements for probing cryogenic interfacial mechanics and engineering reconfigurable ice-templated materials35–37. Notably, we also achieve on-demand fabrication of gold–SiO₂ heterostructures that exhibit engineered anisotropic and chiral optical responses, establishing a route toward polarization-programmable nano-optical devices and chiral photonics38,39. Thus, IOT extends optical manipulation into frozen environments. By integrating dynamic, reconfigurable control with stable fixation, it bridges the fundamental gap between the flexibility of liquid-phase manipulation and the stability of solid-state fabrication. This platform enables functional microdevice fabrication and nanoparticle-interaction studies, while offering a biocompatible route for cryogenic biomolecular research. Results Working principle As illustrated in Fig. 1a, b, target particles were initially dispersed in a microfluidic cell (~30 μm height; Supplementary Note 1 and Supplementary Fig. 1) filled with water and mounted on a cooling stage. Lowering the temperature below the freezing point induced ice formation, immobilizing particles near the substrate through a downward-propagating ice–water interface. After complete freezing, a focused laser beam was directed onto the gold-coated substrate, where optothermal heating established a stable temperature gradient. Once the local temperature exceeded 0 °C, a hemispherical, dome-shaped melting zone formed, releasing particles into a confined liquid micro-environment. This melting zone exhibits a tilted arc interface in the ZX plane and a circular footprint in the XY plane. The melting zone confines a femtoliter-scale liquid volume at the microscale (Fig. 1c). Benefiting from the high thermal conductivity of the gold film, melting and re-solidification occur within ~10 ms, enabling a rapid steady state as confirmed by simulations (Fig. 1d). This fast response allows real-time particle manipulation. Figure 1e and f show that the melting zone radius depends on laser power and ambient temperature; tuning these parameters provides precise, high-throughput particle control with adjustable zone sizes. The close agreement between simulation and experiment further validates the model.Fig. 1Working principle of IOT.a Schematic of the suspension mode, where a particle is driven by the flow field in a vortex-like motion within the melting zone. The melting-zone height and radius are denoted as H and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{m}$$\end{document}Rm. b Schematic of the surface mode, where a particle is pinned at the ice–water–substrate interface and pushed by the advancing ice–waterfront. c Simulated temperature field and thermo-osmotic velocity distribution at −20 °C and 27 mW laser power in the confined melting zone. d Temporal evolution of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{m}$$\end{document}Rm under 27 mW laser power at −20 °C. e Comparison of measured and simulated \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{m}$$\end{document}Rm at different laser powers (−20°C). f Measured and simulated \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{m}$$\end{document}Rm at different ambient temperatures under a fixed laser power of 27 mW. g Thermo-osmotic flow field. The dotted pink rectangle highlights the trapping site of particles. Red and blue lines indicate isoclines where the velocity components \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${v}_{x}=0$$\end{document}vx=0, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${v}_{z}=0$$\end{document}vz=0, respectively. Particles are trapped within a certain range near the intersection of these isoclines (18 mW, −5°C). h Vortex radius (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{v}$$\end{document}Rv) extracted from trajectories of trapped 1-µm PSs at varying laser powers and trajectory of a single 1-µm polystyrene sphere (inset, 18 mW, −5°C). All data were obtained using a 20× objective (NA = 0.45). Working principle of IOT. a Schematic of the suspension mode, where a particle is driven by the flow field in a vortex-like motion within the melting zone. The melting-zone height and radius are denoted as H and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{m}$$\end{document}Rm. b Schematic of the surface mode, where a particle is pinned at the ice–water–substrate interface and pushed by the advancing ice–waterfront. c Simulated temperature field and thermo-osmotic velocity distribution at −20 °C and 27 mW laser power in the confined melting zone. d Temporal evolution of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{m}$$\end{document}Rm under 27 mW laser power at −20 °C. e Comparison of measured and simulated \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{m}$$\end{document}Rm at different laser powers (−20°C). f Measured and simulated \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{m}$$\end{document}Rm at different ambient temperatures under a fixed laser power of 27 mW. g Thermo-osmotic flow field. The dotted pink rectangle highlights the trapping site of particles. Red and blue lines indicate isoclines where the velocity components \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${v}_{x}=0$$\end{document}vx=0, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${v}_{z}=0$$\end{document}vz=0, respectively. Particles are trapped within a certain range near the intersection of these isoclines (18 mW, −5°C). h Vortex radius (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{v}$$\end{document}Rv) extracted from trajectories of trapped 1-µm PSs at varying laser powers and trajectory of a single 1-µm polystyrene sphere (inset, 18 mW, −5°C). All data were obtained using a 20× objective (NA = 0.45). Notably, the melting zone can be repositioned by steering the laser focus. Inside this microscale volume, particles are directed by thermophoresis, local flow fields, or the receding ice–water interface. Subsequently, rapid refreezing upon laser shutoff enables precise printing onto the substrate. Particle manipulation operates in two distinct modes when the XY stage is translated with the laser fixed: suspension mode (Fig. 1a) and surface mode (Fig. 1b). For the suspension mode, the particle disengages from the substrate and traces a vortex-like trajectory in the XZ plane at an average height of 3.7 μm above the surface (See Fig. 1a and Supplementary Fig. 2), realizing a non-contact, three-dimensional motion within the melting zone. When the laser focus is displaced, melting occurs at the leading edge of the melting zone while solidification takes place at the trailing edge. The density variations associated with these phase transitions generate a flow from the rear toward the front of the melting zone40, thereby transporting the particle along with the advancing molten region. In addition, the optothermally induced temperature gradient in the flow cell drives a thermophoretic drift velocity \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u$$\end{document}u of the particles41. These local thermal perturbations at the solid–liquid interface further generate a surface-parallel creep flow, known as thermo-osmotic flow v‖23,42,43, as described by the following equations:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u=-{D}_{{{{\rm{T}}}}}\nabla T$$\end{document}u=−DT∇T2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${v}_{\parallel }=\chi \frac{{\nabla }_{\parallel }T}{T}$$\end{document}v∥=χ∇∥TTWhere \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{{{\rm{T}}}}}$$\end{document}DT is the thermophoretic mobility and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi$$\end{document}χ is the thermo-osmotic coefficient. When \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi$$\end{document}χ is positive, the liquid flows toward the hotter region. Conversely, when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi$$\end{document}χ is negative, the direction of thermo-osmotic flow is from hot to cold. To understand this behavior, we simulated the spatial extent of the stable optothermal melting zone, along with its temperature distribution and thermo-osmotic flow field (Fig. 1c; see also Supplementary Note 2 and Supplementary Figs. 3–4). Since the height of the melting zone is approximately 30 µm, the convection velocity is significantly lower than that of the thermo-osmotic flow and, therefore, can be neglected in the analysis (Supplementary Fig. 5). The combined effects of the melting zone border, thermophoresis, and thermo-osmotic flow allow particles to be trapped and transported along with the melting zone. In most optothermal tweezers, particles are trapped at the hottest center. By contrast, in our system, trapping at the laser focus was rarely observed, regardless of particle surface properties. At room temperature, most particles and molecules are thermophobic (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{{{\rm{T}}}}}$$\end{document}DT > 0), and cooling to near-freezing (0–4 °C) can reverse \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{{{\rm{T}}}}}$$\end{document}DT to thermophilic41,44. In our system, however, the manipulation is not only governed by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{{{\rm{T}}}}}$$\end{document}DT but also by the confined liquid region of the melting zone, in which thermo-osmotic flow and phase-boundary dynamics cooperatively trap and transport particles. Therefore, the thermo-osmotic flow in the melting zone drives vortex-like particle motion in the XZ plane (Fig. 1g, h). As shown in Fig. 1h, the vortex center radius (dashed blue line) is ~12 µm and remains nearly constant across different laser powers. The flow field within the melting zone drives particles in tandem with the zone, enabling precise, dynamic manipulation. Additionally, groups of particles can be trapped and transported in this mode, forming a circular distribution in the XY plane (see Fig. 2a, Supplementary Movie 1).Fig. 2Versatile particle manipulation.a Trapping of 1-µm PS particles along circular vortex trajectories in the XY plane under suspension mode, co-moving with the melting zone. b Manipulation of a single 300-nm PS in surface mode, with the dashed circle marking the laser spot and the solid circle indicating the reference particle. c Collective manipulation of 1-µm carboxylated PSs (PS–COOH) in surface mode, merging with another particle group, with the solid circle marking the reference particle. d Critical velocity (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${v}_{c}$$\end{document}vc) of particles with different surface functional groups (27 mW, −20°C). e \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${v}_{c}$$\end{document}vc at different laser powers (−20°C). f The ratio of the melting zone height to radius (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H/R}_{m}$$\end{document}H/Rm) as a function of laser power, obtained through simulation at −20 °C. g–i ETS of 1-µm plain PSs, amine-functionalized PSs (PS–NR³⁺), and PS–COOH under varying laser powers and ambient temperatures, respectively. Error bars are shown only on the upper side due to the logarithmic scale, as the lower-side error bars would extend too far to be properly displayed in the figure. Experiments in (a and d–i) were performed with a 20× objective (NA = 0.45); b and c were conducted with a 100× objective (NA = 1.3). The experiments in a–f were performed at an ambient temperature of −20 °C. Error bars represent standard deviation (SD). Versatile particle manipulation. a Trapping of 1-µm PS particles along circular vortex trajectories in the XY plane under suspension mode, co-moving with the melting zone. b Manipulation of a single 300-nm PS in surface mode, with the dashed circle marking the laser spot and the solid circle indicating the reference particle. c Collective manipulation of 1-µm carboxylated PSs (PS–COOH) in surface mode, merging with another particle group, with the solid circle marking the reference particle. d Critical velocity (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${v}_{c}$$\end{document}vc) of particles with different surface functional groups (27 mW, −20°C). e \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${v}_{c}$$\end{document}vc at different laser powers (−20°C). f The ratio of the melting zone height to radius (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H/R}_{m}$$\end{document}H/Rm) as a function of laser power, obtained through simulation at −20 °C. g–i ETS of 1-µm plain PSs, amine-functionalized PSs (PS–NR³⁺), and PS–COOH under varying laser powers and ambient temperatures, respectively. Error bars are shown only on the upper side due to the logarithmic scale, as the lower-side error bars would extend too far to be properly displayed in the figure. Experiments in (a and d–i) were performed with a 20× objective (NA = 0.45); b and c were conducted with a 100× objective (NA = 1.3). The experiments in a–f were performed at an ambient temperature of −20 °C. Error bars represent standard deviation (SD). The use of flow fields within this femtoliter-scale, ice-encapsulated liquid region is particularly significant, as it enables a synergistic trapping mechanism that addresses key limitations of conventional optothermal approaches. Within this confined micro-environment, a low-power laser generates a steep thermal gradient (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nabla T$$\end{document}∇T), giving rise to both a thermophoretic force and thermo-osmotic flow. These flows act in concert with the rigid boundary of the surrounding ice–water interface, which completely suppresses particle escape. This dual-confinement system, combining a soft potential well with a hard physical trap, ensures robust particle capture irrespective of their intrinsic thermophoretic properties (i.e., thermophobic or thermophilic), thereby offering broad material versatility without the need for specific solutes16,17,45. In addition, the bulk ice functions as an efficient heat sink, confining thermal effects to the microscale and substantially reducing the risk of damage to sensitive biological specimens. For the surface mode, as shown in Fig. 1b, by translating the laser focus, the ice–water interface is displaced, pushing particles located at the ice–water–substrate boundary. The particle is pinned at the narrow junction where the tilted ice–water interface (in ZX plane) meets the substrate. Consequently, particle manipulation in this mode is governed by two coupled factors, i.e., the particle–substrate interaction and the influence of the tilted ice–water interface. In this mode, particles are propelled by the advancing ice–water interface. As shown in Fig. 2b (and Supplementary Movie 2), translating the laser spot displaces a single PS particle within the melting zone, akin to repositioning cargo, with the particle consistently residing at its trailing edge. A well-known limitation of optical tweezers is that particles adhered to the substrate cannot be confined within the optical potential, rendering conventional manipulation ineffective. Our method readily overcomes this barrier. As demonstrated in Fig. 2c (Supplementary Movie 3), a population of 1 μm PS particles can be precisely guided toward an existing aggregate, enabling intentional merging and subsequent collective manipulation. This functionality resembles a microscale sweeping microrobot operating within ice, capable of selectively gathering or removing target particles such as microplastics46–48, within a defined micrometer-scale region. Such capability provides a powerful tool for studying environmental microcontaminants in frozen systems and opens a pathway to explore swarm-like nanorobotic behaviors under cryogenic conditions. During the pushing process, a thin liquid film remains between the particles and the advancing ice–water interface, sustaining the transport of solute molecules toward the freezing front49. Consequently, the particles do not directly contact the interface but instead move ahead of it in a premelted film (thickness of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$d$$\end{document}d). They are subjected to a repulsive force (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${F}_{\sigma }$$\end{document}Fσ) from the ice–water interface and a viscous drag force (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${F}_{\mu }$$\end{document}Fμ) from the surrounding liquid, which can be expressed as49–51:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${F}_{\sigma }=2\pi R\Delta {\sigma }_{0}{\left(\frac{{a}_{0}}{d}\right)}^{n}$$\end{document}Fσ=2πRΔσ0a0dn4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${F}_{\mu }=6\pi \mu v{R}^{2}/d$$\end{document}Fμ=6πμvR2/dWhere \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R$$\end{document}R is the particle radius, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {\sigma }_{0}$$\end{document}Δσ0 is the difference in interfacial free energy between the particle solid and the sum of the surface free energies, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu$$\end{document}μ is the viscosity of water, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v$$\end{document}v is the particle velocity, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${a}_{0}$$\end{document}a0 is the average intermolecular distance, and the expression in parentheses is a correction factor, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n$$\end{document}n is normally in a range of 4–550. Notably, increasing the laser scanning speed beyond a certain threshold causes the particle to become engulfed by the advancing ice, indicating the existence of a critical velocity \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${v}_{c}$$\end{document}vc for sustained pushing. As the velocity \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v$$\end{document}v increases, the particle approaches the ice–water interface, leading to a reduction in the premelted distance \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$d$$\end{document}d. When \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$d$$\end{document}d becomes too small to sustain ice growth ahead of the particle (where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$d$$\end{document}d is still much larger than the molecular diameter of water)52, the interface engulfs it. Thus, we further examined the dependence of the critical velocity \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${v}_{c}$$\end{document}vc on laser power and different surface functional groups. As shown in Fig. 2d, we measured the critical velocity (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${v}_{c}$$\end{document}vc) for plain 1 μm-PS particles and those with different surface functional groups (carboxylated- and trialkylammonium-modified). Due to the different zeta potential of PSs with varying surface groups (Supplementary Fig. 6), their particle–substrate interactions also differ. Carboxylated PSs with the strongest negative zeta potential (−87.09 mV) may experience weaker attraction to the gold film substrate compared to other particles, resulting in the highest critical velocity. Moreover, Fig. 2e shows the critical velocity \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${v}_{c}$$\end{document}vc of 1 μm carboxylated PSs at ambient temperature (−20°C) under different laser powers. As the power increases from 9 to 27 mW, the ice–water interface tilting angle becomes larger in the ZX plane, as reflected by the rising H/Rm ratio (Fig. 2f). This tilting angle increment reduces the downward normal load exerted by the ice–water interface on the particle at the pinning neck, thereby lowering the frictional/lubrication resistance and facilitating particle advection with the moving melting zone. As a result, the critical velocity \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${v}_{c}$$\end{document}vc initially increases with laser power. Beyond a threshold, however, the melting zone grows, and the in-plane (XY and XZ) curvature flattens. According to the Gibbs–Thomson relation, a flatter interface provides less curvature undercooling and forces the premelted film to thin (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$d$$\end{document}d↓)53,54, which raises hydrodynamic resistance and promotes side-encapsulation around the particle. This curvature-controlled thinning counteracts the load-reduction benefit of the ZX tilt, so \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${v}_{c}$$\end{document}vc decreases at higher powers, yielding the observed non-monotonic \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${v}_{c}(P)$$\end{document}vc(P). Additionally, we measured the critical velocity of carboxylated PS particles with different sizes (Supplementary Fig. 7). The results show that the critical velocity decreases with increasing particle size, indicating that larger particles require lower manipulation speeds for stable transport and confinement by IOT (see Supplementary Note 4 for details). Therefore, by harnessing the interfacial dynamics of the ice–water boundary, our method enables remote transport of a broad spectrum of particle types. For PSs, particles with diverse surface chemistries, including carboxylated, aminated, and unmodified surfaces, can all be manipulated. We further demonstrated precise control of particles of varying compositions, including PS, silica, and gold, across a size range from 100 nm to 5 μm in diameter. This unique ability to manipulate particles irrespective of size, surface charge, or material composition greatly expands the operational scope of optical manipulation, paving the way for applications in heterogeneous colloidal assembly, surface chemistry studies, and bio-interface engineering, as further illustrated in Supplementary Figs. 8–10. Manipulation ability analysis We have demonstrated manipulation of both individual and ensembles of polystyrene particles (PSs) using IOT. In suspension mode (Fig. 2a), particles are entrained by local flow fields to form a vortex ring, rolling inward and outward along the loop (Fig. 1g). The entire group can be transported together with the melting zone, and although particles cluster near its trailing edge due to inertial effects, they remain fully suspended without attaching to the ice–water interface. This mode provides a non-contact strategy for collective transport in confined femtoliter-scale aqueous environments, enabling dynamic particle transportation without substrate adhesion. In surface mode (Fig. 2b), by contrast, particles are immobilized at the substrate by the advancing ice–water interface and propelled by its motion, enabling stable printing into defined structures. Both suspended and surface-bound particles can thus be manipulated within IOT. Suspension mode is suited for dynamic, floating transport, whereas surface mode provides reconfigurable anchoring and substrate-level control. Together, these complementary modes expand the operational scope of IOT, allowing flexible transitions between reconfigurable manipulation and stable nanostructure fabrication. Notably, IOT is not limited to manipulating individual colloids. As shown in Fig. 2c and Supplementary Fig. 8, entire particle ensembles can be arranged into symmetric architectures with sub-micrometer precision. While the laser remains on, the melting zone allows dynamic reconfiguration; upon laser deactivation, the rapid refreezing step locks the particles into the desired configuration without distortion. Particle confinement in IOT is dynamically stabilized by interface-driven transport and localized flow fields rather than a static conservative trapping potential (Details are shown in Supplementary Note 3). To assess manipulation capability, we quantified the effective trapping stiffness (ETS) of 1-µm PS particles with different surface functional groups under varying laser powers and ambient temperatures. For particles with stronger attraction to the gold film (typically PS–NR³⁺ and plain PS at temperatures below −10 °C), which are normally confined on the substrate, the ETS reaches ~10 pN/µm (Fig. 2g, h), nearly an order of magnitude greater than that of conventional optical tweezers. In contrast, particles with weaker attraction to the gold film (e.g., plain PS or PS–COOH) typically remain suspended and exhibit circular, radially oscillatory trajectories in the ZX plane (Fig. 1g) together with tangential circular motion in the XY plane (Fig. 1h); under these conditions, the ETS is weak(~10 fN µm⁻¹; Fig. 2g at −5 °C and Fig. 2i), resulting in kx and ky values smaller than kr and approximately two orders of magnitude lower than those of conventional optical tweezers. This indicates a low-force, non-contact transport regime that causes minimal perturbation, making it suitable for fragile specimens. Additionally, as shown in Fig. 2g–i, tuning the laser power and ambient temperature also determines whether particles remain suspended or attach to the substrate. At low laser power, the melting zone contracts and favors substrate printing, corresponding to the high-ETS regime. Notably, control measurements on firmly deposited particles (subject to stage and atmospheric vibrations) produced ETS values identical to those in surface mode (shaded regions in Fig. 2g–i), confirming that our method applies forces sufficient to manipulate particles rigidly attached to the substrate. Furthermore, it is worth noting that the IOT can trap and manipulate particles at high speeds (10–100 μm/s) regardless of whether they are printed onto the substrate or floating in the melting zone. These results demonstrate that IOT provides a highly programmable force landscape, from non-invasive transport to robust substrate-level manipulation, enabling both gentle handling of delicate assemblies and active control of surface-bound particles for applications in micro-assembly, surface chemistry, and biological interface studies. Versatile manipulation of particles and microbubbles IOT enables universal sorting55, trapping, and reconfigurable assembly of diverse particles spanning inorganic/organic colloids to biological specimens. Crucially, the technique exploits a laser-power-tunable melting zone, i.e., by modulating incident power, we can precisely control the melting zone dimensions (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{m}$$\end{document}Rm: 5–50 µm; H: 5–30 µm), establishing a size-selective manipulation platform. As demonstrated in Fig. 3a, this facilitates efficient sorting of polystyrene microparticles (1 μm vs. 2 μm PSs). Lower-power heating generates smaller melting zones capable of mobilizing smaller particles, while larger counterparts remain immobilized due to partial embedding in the solid ice matrix.Fig. 3Manipulation of diverse target objects in surface mode.a Sorting of 1-µm PS from 2-µm PS, under 3 mW laser power. b Manipulation of Escherichia coli (E. coli), under 0.2 mW laser power. Dashed circles are superimposed onto the images to guide the visualization of the ice-water interface. c Manipulation of fluorescein isothiocyanate (FITC)–labeled anti-human IgG, under 3.8 mW laser power. d Manipulation of FITC–labeled single-stranded DNA (ssDNA, 41 bp), under 18 mW laser power. e Manipulation of a single microbubble, under 18 mW laser power. Experiments in (a and c–e) were performed with a 20× objective (NA = 0.45); b was performed with a 100× objective (NA = 1.3). Manipulation of diverse target objects in surface mode. a Sorting of 1-µm PS from 2-µm PS, under 3 mW laser power. b Manipulation of Escherichia coli (E. coli), under 0.2 mW laser power. Dashed circles are superimposed onto the images to guide the visualization of the ice-water interface. c Manipulation of fluorescein isothiocyanate (FITC)–labeled anti-human IgG, under 3.8 mW laser power. d Manipulation of FITC–labeled single-stranded DNA (ssDNA, 41 bp), under 18 mW laser power. e Manipulation of a single microbubble, under 18 mW laser power. Experiments in (a and c–e) were performed with a 20× objective (NA = 0.45); b was performed with a 100× objective (NA = 1.3). In addition, IOT can manipulate spherical colloidal particles of different compositions, such as Au and SiO₂ (Supplementary Fig. 9). Beyond spherical colloids, it enables orientation-independent control of anisotropic biological particles; for example, as shown in Fig. 3b, IOT selectively extracts E. coli from dense populations by overcoming adhesion forces through localized ice-melting dynamics. Moreover, it can manipulate nanoscale biomolecules such as anti-human IgG, inducing in situ aggregation and directed transport (Fig. 3c), which can be steered along predefined trajectories (Supplementary Movie 4). Parallel manipulation of DNA further corroborates this capability (Fig. 3d). Therefore, these results demonstrate the broad versatility of IOT, spanning scales from nanoscale biomolecules to whole bacteria and across materials from inorganic colloids to complex biological systems. Remarkably, IOT is not limited to solid particles but can also manipulate microbubbles within ice. Previous studies have shown that air bubbles trapped in ice undergo only passive deformation, with their final shape dictated by freezing rate, capillarity, and mass diffusion, processes entirely governed by the natural dynamics of solidification34,56. In contrast, our IOT provides versatile, active manipulation of bubble dynamics within ice. In surface mode, a microbubble can be guided along a square trajectory (Fig. 3e; Supplementary Movie 5). By steering the laser focus, bubble merging (Fig. 4a), splitting (Fig. 4b), and elongation (Fig. 4c) can also be achieved with micrometer-scale precision (Supplementary Movies 6–8), thereby overcoming the intrinsic passivity of conventional freezing processes. Moreover, as shown in Fig. 4d, the length of a deformed bubble can be reversibly and precisely tuned by adjusting the speed of melting zone translation, i.e., faster laser movement produces shorter bubbles, consistent with conventional theory34. The active process exhibits high reversibility and reconfigurability (See Supplementary Fig. 11, Supplementary Movie 9), demonstrating quantitative control over bubble geometry that was previously unattainable with conventional freezing methods.Fig. 4Bubble merging, splitting, deformation, and programmable sculpting enabled by IOT.a Controlled merging of two bubbles into a single bubble. b Splitting of one bubble into two smaller ones. c Reversible elongation of a bubble to user-defined lengths, demonstrating geometric programmability. d Variation of bubble length with the speed of melting zone movement after deformation. The experiment was performed using a 20× objective (NA = 0.45) at 6 mW and –10 °C. e Printing of multiple bubbles into an SZU pattern. f Printing of bubbles into a 4×4 ordered array, demonstrating the capability to construct complex architectures. g The process of bubble sculpting into a four-leaf clover–like shape, with the arrow indicating the direction of melting zone movement. Experiments in (a–d and g) were performed with a 20× objective (NA = 0.45), and e, f with a 40× objective (NA = 0.6). Dashed circles are superimposed onto the images to guide the visualization of the ice-water interface. Bubble merging, splitting, deformation, and programmable sculpting enabled by IOT. a Controlled merging of two bubbles into a single bubble. b Splitting of one bubble into two smaller ones. c Reversible elongation of a bubble to user-defined lengths, demonstrating geometric programmability. d Variation of bubble length with the speed of melting zone movement after deformation. The experiment was performed using a 20× objective (NA = 0.45) at 6 mW and –10 °C. e Printing of multiple bubbles into an SZU pattern. f Printing of bubbles into a 4×4 ordered array, demonstrating the capability to construct complex architectures. g The process of bubble sculpting into a four-leaf clover–like shape, with the arrow indicating the direction of melting zone movement. Experiments in (a–d and g) were performed with a 20× objective (NA = 0.45), and e, f with a 40× objective (NA = 0.6). Dashed circles are superimposed onto the images to guide the visualization of the ice-water interface. Beyond this, our approach transforms frozen microbubbles from passive markers of freezing conditions into programmable building blocks. As shown in Fig. 4e, f, bubbles can be assembled into arrays or predefined shapes. Additionally, microbubbles can be sculpted via the melting zone into a micro four-leaf clover morphology (Fig. 4g; Supplementary Movie 10). Other shapes, such as L-shaped bubbles, can likewise be generated (Supplementary Fig. 12). These results establish IOT as a unique tool for sculpting bubbles into complex, non-spherical geometries, marking a breakthrough beyond conventional freezing techniques. Actively tuning bubble deformation and manipulation in situ opens opportunities to design reconfigurable porous architectures in ice-templated materials, enable information storage in ice57, and explore bio-inspired phenomena such as embolism repair or microbubble-mediated cryopreservation58,59. Reconfigurable particle assembly and spectral detection Remarkably, in the surface mode, IOT enables reconfigurable and high-precision assembly of colloidal particles directly on solid substrates. By simply toggling the laser on or off, we can rapidly (~10 ms, Fig. 1d) switch particles between a fixed state, i.e., printed onto the substrate, and a controllable state where they can be repositioned at will. Furthermore, fine modulation of the laser power allows precise control of the melting zone size on the heat-absorbing substrate, thereby dictating the minimum achievable inter-particle spacing and enabling tunable printing accuracy. By combining high-NA objective lenses with reduced laser power, we achieve nanometer-scale printing resolution. Firstly, we demonstrated the highly reconfigurable patterning capability of IOT. As illustrated in Fig. 5a, we first assembled 1 μm PS particles into a square configuration. By selectively moving the particles at the upper-left and lower-right corners to the left and right edges, respectively, the square was transformed into a straight line. Subsequent repositioning of the left-edge particles to the bottom-right corner and the right-edge particles to the top-left corner restored the square geometry, but with the diagonal particles interchanged. Besides, to assess the post-printing stability of IOT-fabricated structures, we examined 400 nm Au particles arranged into a square pattern on the substrate. As shown in Fig. 5b, the printed structures retain their geometry both in dark-field optical images after ice melting and in SEM images following drying, confirming that the assemblies remain firmly anchored to the substrate. We also quantified the patterning accuracy of IOT. As shown in Fig. 5c(i), a 3 × 3 array of 400 nm Au particles was assembled with the target positions defined at the array nodes. Using grayscale centroid analysis (Fig. 5c(ii)), we determined that the mean positional deviation from the targets was ~330 nm. For one-dimensional patterns, we assembled 400 nm Au particles into a straight line (SEM image in Fig. 5d) and measured the vertical displacement of each particle relative to the predefined line, yielding an average deviation of ~80 nm. These results demonstrate that IOT achieves sub-wavelength positioning precision in both 2D and 1D assemblies, enabling the real-time fabrication of highly controllable micro- and nanostructures. Further improvements in assembly precision can be achieved by employing high-precision displacement stages and by reducing the melting zone size, for example through low-thermal-conductivity absorbing substrates or pulsed optothermal excitation to enhance thermal localization.Fig. 5Reconfigurability and printing accuracy of IOT.a Reconfigurable patterning of five 400-nm gold (Au) particles, transformed from a square to a straight line and back to a square. Arrows indicate particle movement (40× objective, laser power = 6.5 mW). b Assembly of 400-nm Au particles; patterns formed in ice remain intact after melting and drying. c Bright-field images showing the printing of a 3 × 3 matrix with 400-nm Au particles, where deviations of individual particles from the target positions are indicated. The sequence number of each particle corresponds to that in (ii), and the color scale denotes the magnitude of deviation (nm). d Scanning electron microscopy (SEM) image of a straight line assembled from five 400-nm Au particles, with position deviations (h) of individual particles relative to the target line. The shaded region highlights the mean deviation of ~80 nm. Reconfigurability and printing accuracy of IOT. a Reconfigurable patterning of five 400-nm gold (Au) particles, transformed from a square to a straight line and back to a square. Arrows indicate particle movement (40× objective, laser power = 6.5 mW). b Assembly of 400-nm Au particles; patterns formed in ice remain intact after melting and drying. c Bright-field images showing the printing of a 3 × 3 matrix with 400-nm Au particles, where deviations of individual particles from the target positions are indicated. The sequence number of each particle corresponds to that in (ii), and the color scale denotes the magnitude of deviation (nm). d Scanning electron microscopy (SEM) image of a straight line assembled from five 400-nm Au particles, with position deviations (h) of individual particles relative to the target line. The shaded region highlights the mean deviation of ~80 nm. Additionally, as shown in Fig. 6a, we demonstrate the capability of IOT to construct user-defined micro-patterns with single-particle precision. Using 1 µm carboxylated PS particles, we printed the letters SZU, and with 500 nm PS particles, we fabricated finer patterns such as 40 and PS. Beyond simple planar arrangements, the system allows in situ freezing immediately after laser deactivation, thereby enhancing interparticle coupling through van der Waals forces. This capability facilitates the formation of assemblies ranging from perfectly symmetric clusters to heterogeneous composites. As illustrated in Fig. 6b, within a binary colloidal mixture of 500 nm and 1 µm PS particles, we selectively manipulated target particles into prescribed configurations. Symmetric heptamers composed of seven 1 µm PS particles were reliably constructed [Fig. 6b(i)], while heterostructures, such as dimers, trimers, and tetramers, were fabricated by combining particles of distinct sizes [Fig. 6b(ii–iv), Supplementary Movie 11].Fig. 6Printing and assembly of particles using IOT.a Schematic (yellow background) and experimental demonstrations of particle printing into specific words and patterns: (i) SZU composed of 1-µm PSs; (ii) 40 composed of 500-nm PSs; (iii) PS composed of 1-µm PSs; and (iv) a symmetric pattern composed of 1-µm and 500-nm PSs. b Illustration of the assembly-printing process, showing sequential positioning, attachment, and integration of particles into more complex structures: (i) heptamer assembled from seven 1-µm PSs; (ii) dimer of one 1-µm and one 500-nm PS; (iii) trimer of one 1-µm and two 500-nm PSs; (iv) tetramer of one 1-µm and three 500-nm PSs. Scale bars in (a, b), 5 µm. c A trimer consisting of one 400-nm gold nanoparticle and two 500-nm SiO₂ particles under polarized beam illumination. Parallel and vertical denote the incident polarization relative to the trimer: aligned with the long axis and perpendicular to it (short-axis–aligned), respectively. d A pentamer consisting of one 400-nm gold nanoparticle and four 500-nm SiO₂ particles under circularly polarized beam illumination (LCP and RCP). Panels (i–iii) in (c, d) show dark-field image (with white circles marking the particle positions), SEM image, and dark-field normalized scattering spectrum, respectively. Experiments in a, b were performed with a 100× objective (NA = 1.3); c, d were performed with a 40× objective (NA = 0.6). Printing and assembly of particles using IOT. a Schematic (yellow background) and experimental demonstrations of particle printing into specific words and patterns: (i) SZU composed of 1-µm PSs; (ii) 40 composed of 500-nm PSs; (iii) PS composed of 1-µm PSs; and (iv) a symmetric pattern composed of 1-µm and 500-nm PSs. b Illustration of the assembly-printing process, showing sequential positioning, attachment, and integration of particles into more complex structures: (i) heptamer assembled from seven 1-µm PSs; (ii) dimer of one 1-µm and one 500-nm PS; (iii) trimer of one 1-µm and two 500-nm PSs; (iv) tetramer of one 1-µm and three 500-nm PSs. Scale bars in (a, b), 5 µm. c A trimer consisting of one 400-nm gold nanoparticle and two 500-nm SiO₂ particles under polarized beam illumination. Parallel and vertical denote the incident polarization relative to the trimer: aligned with the long axis and perpendicular to it (short-axis–aligned), respectively. d A pentamer consisting of one 400-nm gold nanoparticle and four 500-nm SiO₂ particles under circularly polarized beam illumination (LCP and RCP). Panels (i–iii) in (c, d) show dark-field image (with white circles marking the particle positions), SEM image, and dark-field normalized scattering spectrum, respectively. Experiments in a, b were performed with a 100× objective (NA = 1.3); c, d were performed with a 40× objective (NA = 0.6). Furthermore, as shown in Fig. 6c, d, we demonstrate the fabrication of multi-material heterostructures composed of 400-nm Au nanoparticles (AuNPs) and 500-nm SiO₂ particles, which exhibit pronounced anisotropic, polarization-dependent optical responses, with spectra normalized to the incident light intensity (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S={I}_{{sca}}/{I}_{{{\mathrm{int}}}}$$\end{document}S=Isca/Iint). In Fig. 6c, the AuNP–SiO₂–SiO₂ trimer shows polarization-dependent scattering, in which linearly polarized excitation parallel to the long axis and perpendicular to it produce distinct spectra, indicating anisotropic coupling in the heterostructure. In Fig. 6d, the pentamer (AuNP–4 SiO₂) displays distinct spectral shifts under left- and right-handed circularly polarized light (LCP and RCP), highlighting its chiral response arising from near-field coupling. By contrast, isolated Au and SiO₂ particles exhibit nearly identical spectra (Supplementary Fig. 13), confirming that the observed effects originate from collective interactions in the assembled heterostructures. Our method thus provides a versatile route for constructing directionally anisotropic nanostructures with tailored optical responses, opening opportunities for applications in chiral photonics, metasurfaces, and polarization-controlled nano-optical devices. Discussion In this work, we introduce ice-phase optothermal tweezers (IOT), a paradigm for nanoscale manipulation that harnesses a controllable solid–liquid phase transition to extend optical manipulation into frozen environments. IOT resolves the trade-off between manipulation flexibility and structural stability by seamlessly integrating dynamic, reconfigurable control within a localized melting zone with the ability to statically print structures upon refreezing. Freezing fixes the assembly in place, effectively quenching Brownian motion and allowing structures to be written, erased, and rewritten. Compared with conventional optical tweezers, IOT enables direct manipulation of particles on solid substrates, where optical trapping typically fails, while simultaneously achieving precise particle-level control without relying on high-NA objectives. This distinction arises because particle motion in IOT is governed by interface-driven actuation and displacement-stage precision rather than optical confinement, thereby overcoming these fundamental constraints of optical tweezers. The versatility of IOT, which can assemble building blocks ranging from nanoparticles and proteins to microbubbles, arises from a manipulation mechanism governed by universal physical principles rather than material-specific properties, achieving a fabrication resolution of 80 nm. This broad applicability is crucial for constructing heterogeneous structures that underpin advanced functional materials and bio-interfaces. In particular, achieving site-selective and asymmetric assembly between biomacromolecules and inorganic nanoparticles remains challenging due to the difficulty of deterministic co-localization and controlled interfacial binding while suppressing nonspecific adsorption. Leveraging the cross-material manipulation capability of IOT, a promising future direction is to combine IOT with the freezing-induced concentration effect to co-confine DNA and gold nanoparticles within a localized melting zone60–62, followed by controlled refreezing to promote proximity-driven, site-selective binding at nanoparticle interfaces. While we demonstrate controlled transport of DNA ensembles, deterministic manipulation of free single biomolecules remains challenging due to biomolecule–substrate interactions and limitations in optical imaging for tracking individual biomolecules. Nevertheless, experiments (Supplementary Fig. 10) with 40 nm and 100 nm fluorescent nanoparticles indicate that IOT remains effective well below the 100 nm scale. IOT also offers a versatile platform for assembling directionally anisotropic nanostructures from diverse material combinations with tunable polarization responses, enabling applications in chiral photonics, metasurfaces, and polarization-controlled nano-optical devices. Beyond nanoparticle assembly, IOT transforms microbubbles from passive byproducts of freezing into actively programmable elements34,56. We have achieved reversible and precise merging, splitting, elongation, and even sculpting microbubbles within ice, resolving a fundamental challenge and unlocking opportunities for bio-inspired cryopreservation, reconfigurable ice-templated materials, and bubble-based photonic architectures. While powerful, we acknowledge the current limitations of IOT, which also define clear avenues for future development. The reliance on a cryogenic ice matrix, essential for its unique stability, presently restricts applications requiring physiological temperatures. Furthermore, our demonstrations are primarily confined to quasi-2D assembly on a substrate. A significant future challenge is to extend this method to true layer-by-layer 3D fabrication, creating complex volumetric architectures within the ice. Looking forward, we envision several key advancements: exploring alternative phase-change, biocompatible materials to enable operation at or near room temperature; integrating IOT with nanofluidic systems for high-precision biomaterial delivery and sensing63; and implementing super-resolved photonic force microscopy under cryogenic conditions64. Such developments could transform IOT from a specialized tool into a general-purpose platform for nanoscale manufacturing and analysis. Methods Optical setup Our system integrates a Nikon inverted microscope (Ti2-E) with a 785 nm solid-state laser. The laser beam undergoes 5× expansion before passing through a non-polarizing beam splitter (reflection ratio 90:10; transmission ratio in the 700–1100 nm wavelength range, from Thorlabs). Three objectives were employed: a 100× oil dark-field objective (Nikon, NA = 1.3), a 40× objective (Nikon, NA = 0.6), and a 20× objective (Nikon, NA = 0.45). Optical images were recorded using a color sCMOS camera (Dhyana 400DC, Tucsen Ltd.) with a short-pass filter (bandwidth 4, cutoff wavelength 675 nm, Edmund) to mitigate the impact of reflected 785 nm laser light. FITC fluorescence is excited by a 475 nm LED light source (Nikon). Microfluidic cell setup The microfluidic cell consisted of a glass slide and a gold-coated (10 nm) coverslip joined by parafilm, with a thickness of approximately 30 μm. The cooling stage features a glass aperture at both the top and bottom to ensure the objective lens receives images and allows illumination light to enter normally. The detailed structure of the microfluidic cell and the cooling stage is shown in Supplementary Note 1. When using the 100× oil dark-field objective, the cover beneath the cooling stage must be opened to maintain an appropriate working distance for the objective. Under the configuration of the experiment in this article, the cooling water temperature was set at 3 °C. The cooling stage temperature is set to -5, -10, or −20°C, with the cooling start temperature set to 25 °C. Substrate and material preparation The fabrication process for the optothermal substrate is as follows: First, a 5nm-thick chrome adhesion layer is deposited onto the glass cover slip (20 × 20 mm), followed by a 10 nm-thick gold layer. These deposition procedures were performed by Suzhou Hengxin Microelectronics CO., Ltd. The cooling stage (TES120E-IM) was procured from Wuhan Congtical Technology Co., Ltd. Reagents such as anti-human IgG, and ssDNA were procured from Sangon Biotech (Shanghai) Co., Ltd. Carboxylated PS particles (spheres, 1 μm, 10.09 wt %) were sourced from Bangs Laboratories, Inc. Plain and trialkylammonium-modified PS particles (spheres, 50 mg/ml) were sourced from micromod particle technology, GmbH. Carboxylated Polystyrene particles (spheres, 2, 3, 4 μm, 2.5 wt %) were purchased from Jiangsu Zhichuan Technology CO., Ltd. Green-fluorescent polystyrene particles (spheres, 0.04, 0.10 μm, 1 wt %) were purchased from Thermo Fisher Scientific, Inc. Supplementary information supplementary information Description of Additional Supplementary Files Supplementary Movie 1 Supplementary Movie 2 Supplementary Movie 3 Supplementary Movie 4 Supplementary Movie 5 Supplementary Movie 6 Supplementary Movie 7 Supplementary Movie 8 Supplementary Movie 9 Supplementary Movie 10 Supplementary Movie 11 Transparent Peer Review file supplementary information Description of Additional Supplementary Files Supplementary Movie 1 Supplementary Movie 2 Supplementary Movie 3 Supplementary Movie 4 Supplementary Movie 5 Supplementary Movie 6 Supplementary Movie 7 Supplementary Movie 8 Supplementary Movie 9 Supplementary Movie 10 Supplementary Movie 11 Transparent Peer Review file