Microfluidic pulse oscillator enabled by a shapeshifting liquid metal capacitor Authors: Yong Liu, Mingyi Liang, Ping Liu, Jialin Wu, Shanshan Xu, Minsu Liu, Aibing Yu, Sheng Yan Published: 2026 ABSTRACT Analogous to an inductor-capacitor oscillator circuit, the microfluidic oscillator employs a logic fluidic network to generate cascading and oscillatory flows, bringing the advanced function of flow control to the lab-on-chip platform. Despite notable advancements, current microfluidic oscillators are predominantly limited to fixed multilayer structures and cannot satisfy the requirements for flexibility, reconfigurability, and rapid-response characteristics. Herein, we developed a monolayer microfluidic pulse oscillator with a shapeshifting liquid metal (LM) capacitor for real-time control of oscillatory flows. This fluidic pulse oscillator exploits the interplay between LM capacitive energy storage and the surrounding Marangoni flow to convert stored interfacial energy into fluidic pulsed motion and realize fluid oscillation in alternating current fields. A theoretical model considering fluidic resistance, capacitance, and inductance is developed to decipher the fluid flow behavior. Furthermore, we employed both particle tracing to track transient flow field changes and numerical simulations to characterize the flow dynamics. The results reveal that vortex flow emerges in the near-field region adjacent to the LM capacitor, while pulse oscillatory flow prevails in the far-field region distant from the LM capacitor and occupies most of the microchannel. Axial oscillation and lateral lift-induced cumulative displacement are employed by oscillatory flow for frequency-tunable particle focusing and separation. By leveraging the principles of electronic circuits, the microfluidic pulse oscillator enables the advanced control of liquids and suspended particles for more lab-on-a-chip applications. Introduction Electronic oscillators generate precise timing signals through resonant energy exchange that serve as the heartbeat of modern electronics1. Extending this principle via the hydraulic-electric analogy, fluidic oscillators accomplish autonomous rhythm generation and fluid propulsion, being directly applied to soft robot actuation2,3 and lab-on-a-chip automation4,5. Conventional flow oscillators used electronically controlled valves and pumps for dynamic flow switching, which can limit convenience, scalability, and robustness6–8. To overcome these shortcomings, microfluidic networks with multilayer structures were developed for timed generation of flow and flow switching9–11. In these fluidic oscillators, deformable elastomer films as passive elements are embedded into the microchannel and act as fluidic capacitors for energy storage. The deformation of the elastomeric film converts the pressure into elastic potential, much like charging an electric capacitor. Since the fluidic capacitance is related to the elastic modulus and geometries of an elastomeric film, the fluidic capacitance is therefore pre-defined and fixed after the fluidic capacitor is fabricated, limiting its flexibility, reconfigurability, and tunability. The energy storage materials impart a degree of behavioral flexibility to the oscillators, allowing their output characteristics to be dynamically molded and optimized in response to real-time feedback. Phase-change energy storage materials enable fluid manipulation via phase transition-induced volume expansion/contraction or abrupt viscosity changes, but suffer from irreversible responses, single triggers, and limited dynamic tunability; fluidic energy storage media (e.g., smart fluids) achieve fluid control through external field-regulated viscosity adjustment, yet hinder miniaturization and integration due to complex packaging or inherently low energy density12–15. Liquid metal (LM), as an alternative fluidic energy storage medium, can store high-density energy via reversible shapeshifting16. This process is underpinned by LMs' ability to store a significant amount of energy in their deformed shapes against surface tension, which is then released to power fluidic functions. This energy-storage mechanism is physically analogous to an electrical capacitor. As an active element that can be modulated and rapidly respond to the excited electrical field, the presence of LM is a prerequisite for forming resonant fluidic structures like oscillators when coupled with the system’s inherent inductance. However, such a fluidic system with the active LM capacitor has not yet been validated or made available. In this work, we demonstrate a fluidic pulse oscillator implemented with an active LM capacitor, in which the LM capacitor stores and releases interfacial energy in synergy with the microchannel's fluidic inductance, autonomously generating robust pulsed oscillatory flows without any moving mechanical parts or multilayer structures. The core physics lies in the dynamic interplay between electrically controlled Marangoni stresses, reversible deformation of the liquid metal interface, and the resulting near-field vortex shedding and far-field pulsatile propagation. By modulating the AC frequency and voltage, we can control the behavior of the LM capacitor and establish the conditions required for pulsed oscillatory flow, enabling high flexibility, reconfigurability, and rapid-response characteristics. To show the power and utility of this active fluidic oscillator, we demonstrate the generation of oscillatory flows, frequency-tunable focusing of suspended particles, and size-based separation of particles. Results and discussion Equivalent circuit concept and theory for pulsed fluidic oscillator Electrical oscillations in an LC circuit are sustained by the periodic exchange of energy stored in the capacitor’s electric field and the inductor’s magnetic field (Fig. 1A). Based on the hydraulic-electric analogy, we propose the concept of a LM capacitor to achieve fluidic pulse oscillation (Fig. 1B). Under AC actuation, periodic charge redistribution at the LM/electrolyte interface generates a time-dependent surface-tension gradient according to the Lippmann relation:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma={\gamma }_{0}-\frac{1}{2}{C}_{s}{V}_{{{{\rm{EDL}}}}}^{2}$$\end{document}γ=γ0−12CsVEDL2producing a tangential Marangoni stress \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tau }_{M}={\nabla }_{s}\gamma$$\end{document}τM=∇sγ that drives reversible LM deformation. The fluidic volume displaced by the LM is2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{d}(t)=h\,{A}_{{{\rm{proj}}}}(t)$$\end{document}Vd(t)=hAproj(t)where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{{{\rm{proj}}}}}(t)$$\end{document}Aproj(t) is the instantaneous projected LM area and hh is the channel height. Periodic penetration and retraction of the LM produce oscillatory variation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{d}(t)$$\end{document}Vd(t), which can be mapped onto an equivalent R-C-L circuit. The governing pressure balance is:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${L}_{h}(a)\frac{{dQ}}{{dt}}+{R}_{h}(a)Q+\frac{1}{{C}_{h}(a)}\int {Q\,dt}=\Delta {P}_{M}(t)={K}_{M}(a)\Delta \gamma (t)$$\end{document}Lh(a)dQdt+Rh(a)Q+1Ch(a)∫Qdt=ΔPM(t)=KM(a)Δγ(t)where QQ is the volumetric flow rate. Here, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${L}_{h}(a)$$\end{document}Lh(a) represents the hydraulic inertance associated with fluid acceleration, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{h}(a)$$\end{document}Rh(a) the viscous resistance in the microchannel and thin electrolyte layer, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{h}(a)$$\end{document}Ch(a) the hydraulic capacitance reflecting LM deformation and transient fluid storage. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {P}_{M}(t)$$\end{document}ΔPM(t) is the Marangoni-induced driving pressure, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${K}_{M}(a)$$\end{document}KM(a) is a geometric factor depending on the LM shape aa. This formulation captures the three contributions of fluidic response: energy storage via LM deformation (capacitive), viscous dissipation in confined fluid layers (resistive), and transient acceleration of fluid mass (inertial). Detailed derivations and parameter definitions are provided in Supplementary Note S1.Fig. 1Schematic of the LM capacitor-based microfluidic oscillator.A Electrical oscillations in an LC circuit. B Fluidic pulse oscillation actuated by a LM capacitor. C Periodic flow of LM in the microchannel enabled by AC electrowetting. Near-field vortex flow and far-field oscillatory flow induced by the Marangoni effect. D Mechanism of the conductive pathway and electric double layer (EDL) capacitance-driven LM flow. E Operating range of the LM capacitor and application for particle manipulation via the far-field pulse oscillatory flow. Schematic of the LM capacitor-based microfluidic oscillator. A Electrical oscillations in an LC circuit. B Fluidic pulse oscillation actuated by a LM capacitor. C Periodic flow of LM in the microchannel enabled by AC electrowetting. Near-field vortex flow and far-field oscillatory flow induced by the Marangoni effect. D Mechanism of the conductive pathway and electric double layer (EDL) capacitance-driven LM flow. E Operating range of the LM capacitor and application for particle manipulation via the far-field pulse oscillatory flow. Mechanism of LM capacitor in fluidic pulse oscillator The formation of a gallate [Ga(OH)4]⁻ layer on the LM surface renders the interface negatively charged, attracting Na⁺ counterions from the NaOH solution and forming an EDL at the LM/electrolyte interface (Fig. 1C, D). This EDL can be regarded as a nanoscale interfacial charge-separation layer with polarization or dipolar characteristics. Before applying voltage, the charge distribution is approximately symmetric due to the high conductivity of LM. When an external voltage is applied, the potential gradient along the microchannel induces asymmetric redistribution of charges within the EDL. This field-induced interfacial polarization leads to a higher charge density on one side of the LM, thereby reducing the local interfacial tension on that side relative to the other. Through electrocapillary modulation, the resulting interfacial tension gradient generates a pressure difference that drives LM motion and surrounding electrolyte flow. Under an alternating electric field, this Marangoni-driven interfacial flow periodically entrains the adjacent electrolyte, generating transient vortices near the polarized side of the LM (Supplementary Note S2). Upon polarity reversal, the vortex decays on the previous side and reappears on the opposite side, consistent with the alternating redistribution of EDL charges and the switching of the interfacial-tension gradient. In this work, we investigate the near-field and far-field hydrodynamics of the LM capacitor to establish a complete flow profile (Fig. 1C and Supplementary Note S3), while characterizing its shapeshifting behavior under various operational conditions to enable precise manipulation of flow fields and particles (Fig. 1E). To verify that the chamber structure enables the charging-discharging behavior (Fig. 2A), we investigated the motion of the LM with and without structural constraints. We introduced a 0.3 mm-long axially constrained LM into the microchannel and employed 5 μm polystyrene (PS) microspheres to characterize the surrounding flow field. Under an AC field, the LM exhibits anode-directed oscillatory motion, with the surrounding fluid forming a Marangoni flow opposite to the LM’s movement direction and flowing from regions of high surface tension to low surface tension (Fig. 2B, C). On the side of the LM with high surface tension, a vortex flow emerges in the fluid. In the straight channel, the LM was laterally confined by the channel walls but remained free to move along the axial direction. Under AC actuation, periodic polarization of the LM/electrolyte interface generated surface-tension gradients and interfacial flow along the LM surface. Without axial confinement, the LM underwent pronounced whole-body axial oscillation rather than localized periodic deformation at a fixed actuation site (Fig. 2B and Supplementary Note S4). This large displacement disturbed the surrounding electrolyte and generated vortex-induced perturbations near the LM interface, which propagated into the microchannel and disrupted the local oscillatory flow field. By contrast, axial confinement suppressed whole-body LM translation while preserving localized interfacial deformation, thereby enabling a more stable and controllable oscillatory flow.Fig. 2Conceptual diagram of global flow in the microchannel driven by LM capacitor pumping, and the movement of LM in the microchannel under the CEW effect.A Charging and discharging of the LM capacitor. B Unconstrained LM and C Confined LM: movement in the microchannel and induced flow field. (LP low pressure, HP high pressure, LV low velocity, HV high velocity). D Construction of EDL capacitance. E LM dynamic movement. LM theoretical cyclic motion in microchannels under AC, including invasion and retreat. F Sequential snapshots of LM on the right side of the chip, with AC voltage and frequency set to 6 Vp-p and 3 Hz respectively. G Projected LM area of the droplet in the microchannel. H Rate of change of the projected LM area. I Influence of NaOH concentration and LM droplet size on the maximum projected LM area during LM invasion into the microchannel. Error bands indicate standard deviation from three independent measurements of the maximum projected LM area. Conceptual diagram of global flow in the microchannel driven by LM capacitor pumping, and the movement of LM in the microchannel under the CEW effect. A Charging and discharging of the LM capacitor. B Unconstrained LM and C Confined LM: movement in the microchannel and induced flow field. (LP low pressure, HP high pressure, LV low velocity, HV high velocity). D Construction of EDL capacitance. E LM dynamic movement. LM theoretical cyclic motion in microchannels under AC, including invasion and retreat. F Sequential snapshots of LM on the right side of the chip, with AC voltage and frequency set to 6 Vp-p and 3 Hz respectively. G Projected LM area of the droplet in the microchannel. H Rate of change of the projected LM area. I Influence of NaOH concentration and LM droplet size on the maximum projected LM area during LM invasion into the microchannel. Error bands indicate standard deviation from three independent measurements of the maximum projected LM area. We designed a symmetrically arranged microchannel, confining the LM at the center, aiming to generate oscillatory flow in the LM’s far field via the reciprocating Marangoni flow (Fig. 2C). The LM was injected into the microcavity via the microchannel to construct the oscillatory microfluidic platform as illustrated in Supplementary Note S5. 0.3 M NaOH was injected to construct the LM EDL capacitor (Fig. 2D). Leveraging the LM’s Continuous Electrowetting (CEW) effect and the device’s AC electric field, the LM achieves on-chip oscillatory motion, including microchannel invasion and retreat, as illustrated in Fig. 2E. The dynamics of a full invasion-retreat cycle of the LM at the right chamber, occurring over half a period under a 6 Vp-p, 3 Hz square wave, are shown in the evenly spaced sequential snapshots of Fig. 2F. Due to the system’s symmetry, the remaining half-cycle of the oscillation occurs at the left side of the microcavity. We observed that within one cycle, the invasion phase accounted for the vast majority of the period, while the retreat phase was rapid (Fig. 2F). These observations imply that the rapid reversal of electrode polarity is responsible for relocating the region of high charge density from the right side to the left. This asymmetric invasion-retreat motion was further reflected in the projected LM area inside the microchannel, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{{{\rm{proj}}}}}(t)$$\end{document}Aproj(t), extracted from the recorded image sequence (Fig. 2G). A periodic pulsed pattern appeared in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{dA}}_{{{{\rm{proj}}}}}(t)/{dt}$$\end{document}dAproj(t)/dt at the applied driving frequency (Fig. 2H), indicating cyclic LM displacement within the channel. The corresponding temporal variation, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{{{\rm{proj}}}},\max }$$\end{document}Aproj,max, showed that the major pulse occurred during the rapid retreat phase, whereas the signal during the invasion phase remained close to zero with only minor fluctuations. This result confirms that the LM pulse is mainly generated by the fast withdrawal of the LM, rather than by its gradual invasion into the microchannel. We further examined how NaOH concentration and LM diameter affected the pulse behavior (Fig. 2I). The maximum projected area of LM invasion increased with NaOH concentration before reaching a plateau, suggesting that ionic concentration facilitates interfacial charging until the response saturates. In addition, the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{{{\rm{proj}}}},\max }$$\end{document}Aproj,max increased with LM diameter, consistent with the larger interfacial area and deformation capacity of larger LM droplets. These results show that the pulsatile motion of the LM can be tuned by electrolyte concentration and LM geometry, and the transient capacitive behavior underlying the oscillation is examined later in the manuscript. Flow field generated by a LM capacitor We investigated the fluid flow in both the near and far fields to elucidate the underlying actuation mechanism of the LM capacitor (Fig. 3A–F). When the anode is positioned on the left side of the microfluidic chip, the LM is drawn into the left channel, while fluid flows along the slip layer between the LM and the PDMS wall into the right side of the chip (Fig. 3A). Numerical simulations show that the fluid converges into the main right channel in the near field (Fig. 3B), exhibiting a Poiseuille flow profile with a maximum velocity of 0.024 m/s at the channel center (Fig. 3C). Conversely, when the anode is positioned on the right side, the LM moves into the right channel, driving the surrounding fluid toward the left-side region of lower surface tension (Fig. 3D). This reverse actuation triggers a vortex flow in the near field, while the far field maintains a Poiseuille flow profile (Fig. 3E, F). It is noteworthy that the far-field region exhibits two superimposed oscillatory flows in opposite directions. Notably, two counter-directed oscillatory flows are observed in the far field, each corresponding to one state of the electrode shift.Fig. 3Characterization of the flow field in a LM capacitor-based oscillator.Near-field and far-field flows under the actuation of a LM capacitor: A Schematic, B near-field, and C far-field simulation of the flow as the LM enters the left channel and the fluid flows into the right channel. D Schematic, E near-field, and F far-field simulation of the flow as the LM enters the right channel and the fluid flows into the left channel. At 1 Hz and 4 Vp-p: G Sequential snapshots of LM motion and 5 μm particle-tracked flow fields, revealing near-field vortices around the LM interface and far-field oscillatory flow in the microchannel. H Schematic illustrations of fluid flow at three specific positions, with the initial particle-tracking positions x0 being 0.02 mm, 0.07 mm, and 0.81 mm, respectively. I Particle-tracked fluid flows corresponding to the three aforementioned positions. J Microscopic images of the particulate states for flow field labeling at near-field and far-field locations after 77 s. K Analysis of particle distribution uniformity across near and far fields. L Temporal evolution of particle focusing in the far-field. Characterization of the flow field in a LM capacitor-based oscillator. Near-field and far-field flows under the actuation of a LM capacitor: A Schematic, B near-field, and C far-field simulation of the flow as the LM enters the left channel and the fluid flows into the right channel. D Schematic, E near-field, and F far-field simulation of the flow as the LM enters the right channel and the fluid flows into the left channel. At 1 Hz and 4 Vp-p: G Sequential snapshots of LM motion and 5 μm particle-tracked flow fields, revealing near-field vortices around the LM interface and far-field oscillatory flow in the microchannel. H Schematic illustrations of fluid flow at three specific positions, with the initial particle-tracking positions x0 being 0.02 mm, 0.07 mm, and 0.81 mm, respectively. I Particle-tracked fluid flows corresponding to the three aforementioned positions. J Microscopic images of the particulate states for flow field labeling at near-field and far-field locations after 77 s. K Analysis of particle distribution uniformity across near and far fields. L Temporal evolution of particle focusing in the far-field. The flow field was tracked using 5 μm particles, and the near-field and far-field flows were observed by extending the exposure time to 5 ms (matched with a frame rate of 200 fps) to capture particle trails (Fig. 3G, Supplementary Note S6, and Supplementary Video 1). The chaotic nature of this vortex flow may inhibit the generation of oscillatory flow. To further explore the existence law of the near-field vortex flow, the time-series images of the near-field over one cycle under the electrical signal of 1 Hz and 4 Vp-p are shown in Fig. 3G. It can be observed that during the first half-cycle, the LM moves in the left microchannel: the fluid flows along the right side of the LM surface, where no vortex is present, and this phase corresponds to half a cycle of the oscillatory flow. The vortex flow occurs predominantly in the second half-cycle, starting when the LM enters the microchannel and lasting until it exits. We tracked the time-resolved 2D motion of three 5-μm tracer particles located in the near-field region at distances of x₁ = 0.02 mm, x₂ = 0.07 mm, and x₃ = 0.81 mm from the junction between the right channel and the chamber over one full cycle (1 s) of the electric signal (Fig. 3H). The particle closest to the LM (x₁ = 0.02 mm) was attracted toward the LM, entrained into the vortex, and eventually expelled through the slip layer (Fig. 3I). The slightly more distant particle (x₂ = 0.07 mm) was guided around the vortex and directed through the slip layer. In contrast, the particle at x₃ = 0.81 mm was not drawn into the slip layer but was instead driven into a symmetric oscillatory motion along the x-axis, with negligible displacement in the y-direction. However, after 77 s of fluid pumping driven by the LM capacitor, significant displacement in the y-direction was observed for particles in the far-field region, while the near-field flow remained in a chaotic state (Fig. 3J, K). The mixing effect induced by Marangoni vortex flow in the near-field region leads to a broad transverse distribution of particles across the microchannel (Fig. 3K). Such chaotic advection generated by LM-driven Marangoni vortices has been utilized for enhanced fluid mixing, as demonstrated in the work of Tang et al.17,18. In the present study, we do not extensively explore the near-field vortex dynamics, but instead concentrate on the far-field oscillatory flow, which promotes spatially ordered particle arrangement. Particle position analysis reveals that initially randomly distributed particles progressively organize into a single-line configuration at the channel center within 77 s (Fig. 3L). Oscillatory microfluidics enables continuous fluid transport in finite channels, allowing prolonged particle actuation and superior performance compared to unidirectional flow. This concept, pioneered by Mutlu et al.19, was advanced by Asghari et al.20 for nanoparticle focusing in non-Newtonian fluids. Our work extends this paradigm to pump-free oscillatory microfluidic systems21. In this work, within a PEO and NaOH solution, 5 μm particles are focused at the channel center under the time-averaged combined action of elastic lift forces in the far-field oscillatory flow driven by an LM capacitor. To leverage the unique advantages of LM capacitors in microfluidics, we employ far-field oscillatory flow, generated by LM capacitor pumping, to achieve precise particle manipulation. Dynamic LM oscillation and flow responses under AC actuation To understand AC-driven LM deformation and the resulting far-field flow, we analyzed the evolution of the projected LM area within the microchannel, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{{{\rm{proj}}}}}(t)$$\end{document}Aproj(t), and the far-field velocity, U(t)U(t), under different driving voltages and frequencies (Fig. 4). At a fixed frequency of 3 Hz, the voltage was increased stepwise from 1-10 Vp-p, with each voltage held for 1 s (Fig. 4A, C). Increasing the voltage amplifies the interfacial surface-tension difference, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \gamma (V)$$\end{document}Δγ(V), and thus strengthens the Marangoni stress at the LM/electrolyte interface. At low voltages (1-3 Vp-p), the weak Marangoni stress induces only slight LM deformation, resulting in shallow penetration into the microchannel, small \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{{{\rm{proj}}}}}(t)$$\end{document}Aproj(t), and weak far-field flow. At intermediate voltages (4-6 Vp-p), stronger Marangoni stress drives deeper LM penetration and a larger \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{{{\rm{proj}}}}}(t)$$\end{document}Aproj(t), thereby enhancing the far-field oscillatory velocity. At higher voltages (7-10 Vp-p), however, excessive LM penetration markedly increases fluidic resistance and viscous dissipation. These resistive losses progressively counteract the additional Marangoni driving pressure, causing the far-field flow to saturate and eventually become unstable, accompanied by LM necking and breakup.Fig. 4Morphological dynamics of LM and far-field fluidic response under AC actuation.A Instantaneous morphologies of the LM immediately before electric-field reversal under different driving voltages of 1–9 Vp-p at 3 Hz. B Instantaneous LM morphologies immediately before electric-field reversal at 4 Vp-p over driving frequencies of 1–10 Hz. C, D Time-resolved responses of the projected LM area \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{{{\rm{proj}}}}}(t)$$\end{document}Aproj(t), and far-field velocity, under varying voltages at 3 Hz and varying frequencies at 4 Vp-p. E Coupled LM deformation and flow response at 4 Vp-p and 2 Hz, highlighting the delayed response associated with LM extension and retraction. F Maximum projected LM area at different driving frequencies and the corresponding phase lag of LM response during potential step-up and step-down. G Frequency-dependent evolution of the effective fluidic capacitance. H Theoretically calculated pressure contributions from fluidic capacitance, resistance and inertance during LM-based fluidic oscillation at 4 Vp-p and 1 Hz. Morphological dynamics of LM and far-field fluidic response under AC actuation. A Instantaneous morphologies of the LM immediately before electric-field reversal under different driving voltages of 1–9 Vp-p at 3 Hz. B Instantaneous LM morphologies immediately before electric-field reversal at 4 Vp-p over driving frequencies of 1–10 Hz. C, D Time-resolved responses of the projected LM area \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{{{\rm{proj}}}}}(t)$$\end{document}Aproj(t), and far-field velocity, under varying voltages at 3 Hz and varying frequencies at 4 Vp-p. E Coupled LM deformation and flow response at 4 Vp-p and 2 Hz, highlighting the delayed response associated with LM extension and retraction. F Maximum projected LM area at different driving frequencies and the corresponding phase lag of LM response during potential step-up and step-down. G Frequency-dependent evolution of the effective fluidic capacitance. H Theoretically calculated pressure contributions from fluidic capacitance, resistance and inertance during LM-based fluidic oscillation at 4 Vp-p and 1 Hz. With the driving voltage fixed at 4 Vp-p, we next varied the frequency from 1 to 10 Hz to tune the available response time for LM deformation and far-field flow (Fig. 4B, D). Because the applied voltage was held constant, the interfacial surface-tension difference, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \gamma (V)$$\end{document}Δγ(V), remained approximately unchanged. The driving frequency, therefore, primarily determines the available response time for LM deformation and far-field flow within each half-cycle. At low frequencies of 1–3 Hz, the longer half-cycle permits more complete LM penetration and retraction within the microchannel, producing pronounced \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{{{\rm{proj}}}}}(t)$$\end{document}Aproj(t) modulation and strong far-field flow. Increasing the frequency shortens the half-cycle duration, thereby limiting LM penetration depth and progressively suppressing the peak oscillatory velocity to approximately 5 mm s-1. As shown in Fig. 4E, neither LM deformation nor far-field flow responds instantaneously to polarity reversal, but instead exhibits a finite phase lag. Immediately after the applied potential increases, the LM does not enter the microchannel at once, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{{{\rm{proj}}}}}(t)$$\end{document}Aproj(t) initially remains nearly unchanged. After several tens of milliseconds, the LM gradually penetrates into the microchannel. In contrast, upon potential decrease, the LM retracts over a much shorter timescale, indicating that the delayed establishment of LM penetration dominates the overall response asymmetry. This asymmetry reflects the time required for interfacial driving to be established via transient charge redistribution and surface-tension reconfiguration, whereas retraction is additionally assisted by capillary recovery of the LM interface. The far-field flow develops following the LM penetration, indicating that inertial effects are negligible (Fig. 4E). During the early stage after polarity reversal, the LM deforms locally within the chamber without substantially entering the microchannel, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{{{\rm{proj}}}}}(t)$$\end{document}Aproj(t) changes only slightly (Supplementary Note S7). As the LM gradually penetrates into the microchannel, the far-field flow begins to respond, showing oscillatory reversal. Figure 4F shows that the measured phase lag exhibits the same frequency-dependent trend as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{{{\rm{proj}}}}}(t)$$\end{document}Aproj(t), implying that deeper LM penetration requires more time for interfacial charge redistribution and consequently generates stronger far-field flow. Consistently, under 4 Vp-p and 1-3 Hz, the calculated fluidic capacitance increases with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{{{\rm{proj}}}}}(t)$$\end{document}Aproj(t) (Fig. 4G), supporting the conclusion that enhanced LM penetration promotes stronger far-field hydrodynamic response. To quantify the relative contributions of fluidic capacitance, resistance, and inertance during LM operation, we calculated the pressure terms in the equivalent RCL model, including \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${P}_{R}={R}_{h}Q$$\end{document}PR=RhQ, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${P}_{C}={C}_{f}^{-1}\int {Qdt}$$\end{document}PC=Cf−1∫Qdt, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${P}_{L}={L}_{h}\,{dQ}/{dt}$$\end{document}PL=LhdQ/dt (Fig. 4H). The resistive and capacitive terms dominate the response, both contributing at the 104 Pa scale, whereas the inertial contribution remains negligible at approximately 10-1 Pa. During the early stage of LM penetration, the capacitive term dominates, reflecting transient LM deformation and interfacial volume accommodation. As the LM penetrates deeper into the microchannel, local hydrodynamic resistance and viscous dissipation increase, and the response gradually shifts toward resistance-dominated behavior. The characteristic response time is therefore governed by the combined effect of fluidic capacitance and resistance, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tau }_{{RC}}={R}_{h}{C}_{h}$$\end{document}τRC=RhCh, which is comparable to the experimentally observed delay times (Supplementary Note S8). Continuous cycling tests were conducted to assess operational stability. At 10 Vp-p and 3 Hz, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{{{\rm{proj}}}}}$$\end{document}Aproj remained essentially unchanged over 2040 cycles (Supplementary Note S9). To probe long-term performance, the LM oscillator was operated continuously for 0.9 h, during which the oscillatory amplitude of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{{{\rm{proj}}}}}$$\end{document}Aproj declined by ~88.9%, and oscillation nearly ceased. Refreshing the NaOH electrolyte restored the oscillatory response, indicating that the decay was not due to irreversible LM degradation. We attribute this performance decay to the gradual depletion of local OH⁻ ions and the progressive accumulation of surface oxides at the LM interface, which reduces the effectiveness of EDL-mediated surface-tension modulation. This decrease in interfacial responsiveness likely contributes to the increased delay observed in LM penetration and far-field flow, highlighting the direct role of the electrolyte environment in controlling both the amplitude and timing of the Marangoni-driven oscillatory response. The electrolyte environment strongly influences oscillator performance because actuation relies on oxide regulation and EDL reconstruction at the LM/electrolyte interface. In 0.3 M NaOH, sustained interfacial deoxidation enabled efficient surface-tension modulation and produced a maximum flow rate of 6.58 mm/s (Supplementary Note S10). To evaluate compatibility with non-alkaline environments, neutral sodium chloride (NaCl) and phosphate-buffered saline (PBS) solutions were tested under the same driving conditions. The maximum flow rates in NaCl and PBS were 2.88 mm/s and 1.91 mm/s, respectively, showing that the oscillator can operate beyond alkaline media but with reduced efficiency. The weaker response arises from less effective oxide removal, which reduces modulation of interfacial tension and EDL reconstruction. Biological fluids in microfluidic applications often exhibit viscoelasticity, as in whole blood, plasma, or cell suspensions22,23. To assess operation in viscoelastic media, polyethylene oxide (PEO) was added to the NaOH electrolyte at different concentrations, and the resulting LM deformation and far-field flow were examined (Supplementary Note S11). Increasing the PEO concentration increased the viscosity and viscoelastic dissipation of the solution. According to the hydraulic RCL model, the microchannel resistance \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{h}$$\end{document}Rh scales positively with viscosity μμ. Therefore, increasing PEO concentration increases \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{h}$$\end{document}Rh and reduces the far-field flow under the same Marangoni drive. In addition, at a fixed geometric length scale, the near-field Marangoni velocity scales with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \gamma /\mu$$\end{document}Δγ/μ, so higher viscosity also reduces the efficiency of interfacial shear transfer to the surrounding fluid. The experiments were consistent with this analysis. As the PEO concentration increased, both LM deformation and far-field electrolyte flow were suppressed. At 8000 ppm PEO, the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{{{\rm{proj}}}}}(t)$$\end{document}Aproj(t) oscillation amplitude decreased to only 22.31% of that in the PEO-free electrolyte, and the far-field flow nearly ceased. To avoid potential strong viscoelastic stresses that could hinder gentle biological manipulation, PEO concentrations were typically limited to below 4000 ppm24–26. In 4000 ppm PEO-NaOH solution, the system still generated an oscillatory flow velocity amplitude of 4.37 mm/s, demonstrating effective actuation in moderately viscoelastic media. Lateral cross-streamline particle migration in oscillatory flow We turn our focus to the mechanism of particle lateral migration induced by oscillatory flow, aiming to elucidate the particle focusing behavior observed in the far-field region of the channel in Fig. 3K. In these experiments, 2000 ppm PEO in NaOH was used as the electrolyte for the LM-based oscillator. Under the influence of the far-field oscillatory flow generated by the LM, 5 μm particles are advected by the fluid while undergoing axial oscillations along the microchannel (Fig. 5A, B). The particle displacement exhibits highly reproducible oscillatory cycles (Fig. 5C). During the periods when the LM penetrates into or retracts from the microchannel following its delayed response to polarity reversal, the fluid exhibits a transient acceleration pulse, whereas outside these intervals the flow velocity remains approximately ±2.5 mm s⁻¹ (Fig. 5D).Fig. 5Particle motion and forces under oscillatory flow generated by a LM capacitive-based oscillator.A Schematic diagram of particle motion observation in a fluidic oscillator for oscillatory flow characterization. B 5 μm particle motion in microchannels before and after fluid pulse generation. Under 1 Hz, 3 Vp-p electrical control, C position variation and D velocity variation of 5 μm particles over three cycles. E Experimental lateral migration processes of 5 μm and 10 μm particles. F Pressure distribution on particle surfaces for 5 and 10 μm particles at transverse channel position (y, z) = (24 μm, 0 μm) during pulse instants. Particle motion and forces under oscillatory flow generated by a LM capacitive-based oscillator. A Schematic diagram of particle motion observation in a fluidic oscillator for oscillatory flow characterization. B 5 μm particle motion in microchannels before and after fluid pulse generation. Under 1 Hz, 3 Vp-p electrical control, C position variation and D velocity variation of 5 μm particles over three cycles. E Experimental lateral migration processes of 5 μm and 10 μm particles. F Pressure distribution on particle surfaces for 5 and 10 μm particles at transverse channel position (y, z) = (24 μm, 0 μm) during pulse instants. The lateral position changes of 5 and 10 μm particles in the microchannel are shown in Fig. 5E. A distinct migration pattern is observed where 5 μm particles move toward the channel centerline while 10 μm particles migrate toward the sidewalls. The presence of particle blockage (β) in viscoelastic fluids enhances normal stress differences, leading to destabilization of the channel’s central equilibrium position. This reverse migration phenomenon has been extensively observed and validated in our experiments27,28. Notably, 10 μm particles (β = 0.242) consistently migrate toward the sidewalls, defying the predicted elasto-inertial single-line focusing despite being below the conventional β = 0.25 threshold for flow-reversal migration in unidirectional flows. Transient high shear rates during pulsation amplify the first normal stress difference (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${N}_{1}\propto {\dot{\gamma }}^{2}$$\end{document}N1∝γ°2), generating strong elastic lift that drives 10 μm particles toward sidewalls. The unsteady polymer stretching prevents the system from reaching steady-state force equilibrium, invalidating the steady-flow-based β = 0.25 threshold. Our Oldroyd-B model simulations, evaluated at pulse instants (t = (n + 1)T/2, n = 0,1,2,3…), reveal that the stress distribution on particle surfaces yields a net force directed toward the channel center for 5 μm particles but toward the wall for 10 μm particles (Fig. 5F). However, the computed force magnitude for 10 μm particles significantly exceeds predictions from elastic (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${F}_{E}\propto {a}^{3}$$\end{document}FE∝a3) scaling laws. This discrepancy arises from the strongly non-uniform normal stress field induced by the pulsed flow, which cannot be described by conventional lift models derived under steady, fully developed flow conditions. The transient interaction between particle size and localized stress gradients during pulsation generates a nonlinear amplification of the lift force for larger particles. This explains why 5 μm particles migrate faster toward the central equilibrium position than 10 μm particles move toward the channel walls (Fig. 5E). Controllable particle focusing and separation Based on the understanding of particle motion in the LM capacitor-driven fluidic oscillator, we configured PEO-NaOH particle suspensions and built a testing platform to verify the proposed concept. Leveraging oscillatory flow, we manipulated the motion of a monodisperse particle population by adjusting the AC field (voltages: 3, 4, 5 Vp-p; frequencies: 1, 2, 3 Hz). Consistent with other oscillatory microfluidic platforms, the LM capacitor-driven system provides a distinct advantage in manipulating small particles (with a β < 0.07) compared to continuous-flow microfluidics20,21. Under the oscillatory flow induced by an LM capacitor with operating parameters of 3 Vp-p and 1 Hz, 1 μm particles in the microchannel are tightly aligned along the center of the channel (Fig. 6A). This final distribution results from the elastic lift force positioning particles at equilibrium points in the center and corners of the channel, which are regions of minimal normal stress difference, whereas the inertial lift force displaces them from the corners toward the centerline27. Despite achieving a higher flow rate, an increased voltage hindered centerline particle focusing owing to a strengthened inertial lift force that, under the Poiseuille shear gradient, directed particle migration from the center toward the channel wall. The increased frequency of the LM capacitor’s penetration into the channel proves detrimental, as it causes flow resistance fluctuations and instability, ultimately leading to the defocusing of particles in the channel. In this context, flow-stabilizing controllers are beneficial for particle focusing under high-frequency LM capacitor actuation29.Fig. 6Particle focusing and separation based on LM capacitor-driven oscillatory microfluidics.A Lateral distribution of 1 μm particles in the microchannel at voltages of 3, 4, 5 Vp-p and frequencies of 1, 2, 3 Hz. B Controlling the distribution of 1, 5, 10, and 15 μm particles in a microchannel by modulating the electric field frequency. C Separation of the 5 and 10 μm binary mixture based on their distribution in the microchannel. Error bars represent standard deviation from three independent measurements. Particle focusing and separation based on LM capacitor-driven oscillatory microfluidics. A Lateral distribution of 1 μm particles in the microchannel at voltages of 3, 4, 5 Vp-p and frequencies of 1, 2, 3 Hz. B Controlling the distribution of 1, 5, 10, and 15 μm particles in a microchannel by modulating the electric field frequency. C Separation of the 5 and 10 μm binary mixture based on their distribution in the microchannel. Error bars represent standard deviation from three independent measurements. Under LM-induced oscillatory flow, particles of different sizes exhibited distinct lateral distributions in the viscoelastic medium (Fig. 6B). As the driving frequency increased from 1 to 3 Hz, the amount of LM entering the microchannel decreased, leading to a reduced far-field flow velocity (Fig. 4D). At a driving frequency of 3 Hz, this lower-flow condition produced the most evident size-dependent particle distribution: 15 μm particles focused closest to the channel sidewalls, followed by 10 and 5 μm particles, while 1 μm particles exhibited a much broader and less confined focusing band. The broader distribution of 1 μm particles reflects their much weaker elastic lift, which is insufficient to produce a narrow and stable focused stream (Supplementary Video 2). This ordered lateral distribution is attributed to size-dependent elastic lift in the viscoelastic flow27. The first normal stress difference, together with the degree of particle confinement in the microchannel, generates an elastic lift force that scales strongly with particle size30,31. For smaller particles, the confinement effect is weak and the overall parabolic flow field dominates, causing them to remain closer to the channel center. For larger particles, the stronger confinement between the particle surface and the channel wall enhances the local shear and normal-stress imbalance, shifting their equilibrium positions toward the sidewalls. Therefore, increasing particle size progressively moves the focused particle stream from the channel center toward the channel sidewalls. As a practical application of the LM capacitor, we demonstrate frequency-modulated separation of particles based on their size (Fig. 6C). For randomly distributed 5 μm and 10 μm particles within the microchannel, we applied an electrical signal of 3 Vp-p at 1 Hz. Under the elastic lift forces, all particles were focused at the center of the channel. Following particle release via a DC electric field, the central outlet (OT1) achieved recovery exceeding 98% for both 5 μm and 10 μm particles originally present in the channel. In contrast, the side outlet (OT2) delivered solutions virtually free of particles. At an increased AC frequency of 3 Hz, elastic lift forces drive both 5 and 10 μm particles toward the walls, yet the confinement effect causes the larger 10 μm particles to equilibrate closer to the walls than the 5 μm particles (Supplementary Video 3). Under subsequent DC field actuation, particles are transported to the outlets, with most 10 μm particles exiting through OT2 and most 5 μm particles through OT1. By leveraging the pulsed oscillatory flow generated by the liquid metal capacitor, the system accomplishes precise size-dependent separation, as evidenced by the high purity levels of 95.5% (5 μm in OT1) and 96.4% (10 μm in OT2). To further examine the effect of microchannel geometry on size-dependent separation, we tested 5 and 15 μm particles in channels with different width-to-height ratios (W/H = 3.6, 4.8 and 6.0), while keeping the channel height fixed at 25 μm (Supplementary Note S12). In all channel geometries, 15 μm particles were tightly focused near the sidewalls, showing stable wall-directed migration. In contrast, 5 μm particles were more sensitive to the channel aspect ratio: they formed a broader distribution at W/H = 6.0, but became progressively more confined as W/H decreased. This narrowing of the 5 μm particle band increases the spatial separation from 15 μm particles, indicating that channel aspect ratio can be used to optimize LM-driven size-based particle separation. Because PS particles may also become polarized under an applied electric field, we further examined whether the 5 μm particles used for flow tracing were influenced by direct particle-field interactions. COOH- and NH2-functionalized particles, together with unmodified PS particles, were tested under the same LM actuation conditions (Supplementary Note S13). Despite their different surface chemistries, these particles showed similar lateral distributions during frequency-modulated focusing. This result indicates that the observed motion of 5 μm particles mainly reflects the LM-induced oscillatory flow, rather than direct electrophoretic or polarization-driven particle migration, supporting their use as tracers for visualizing the Marangoni-driven flow field. Functioning as an embedded active fluidic element, the LM capacitor outperforms pre-designed passive components through its exceptional flexibility, reconfigurability, and rapid response characteristics. This electromechanically coupled system eliminates the need for fixed geometric configurations, allowing real-time programming of fluidic functions for adaptive microfluidic processing. The transient pulsed flow demonstrates considerable potential for thermal management applications, as its oscillatory characteristics effectively disrupt thermal boundary layers and enhance convective heat transfer in microfluidic systems32. The LM capacitor creates new opportunities for chemical synthesis by enabling dynamic control over reaction interfaces and enhancing mass/charge transfer33,34. These capabilities could lead to novel pathways for fabricating sophisticated nanomaterials and molecules with high selectivity. Its parallel configuration enables the design of logical fluidic circuits, providing a foundation for microfluidic computational architectures35,36. The LM capacitor-driven pulse oscillator provides a highly controllable timing mechanism for rhythmic actuation in soft robotics2. Its unique electrically regulated pulsation characteristics enable precise programming of periodic deformation sequences in soft actuators, facilitating biomimetic coordinated movements. Methods Design and fabrication of microfluidic devices Microfluidic devices were fabricated by soft lithography. An SU-8 photoresist master with a height of 25 μm was patterned on a silicon wafer using standard photolithography. Polydimethylsiloxane (PDMS, Dow Corning) prepolymer and curing agent were mixed at a mass ratio of 10:1, thoroughly stirred and degassed under vacuum. The degassed PDMS mixture was poured onto the SU-8 master and cured at 60 °C, 2 h. After curing, the PDMS replica was peeled from the master. Holes were punched at both ends of the microchannel to form terminal reservoirs for electrolyte filling and bubble removal, and additional electrode-access ports were punched adjacent to the terminal reservoirs near the reservoir–channel junctions for electrical connection. An independent liquid metal (LM) injection port was also punched above the central circular chamber to allow direct injection of the eutectic gallium-indium LM (EGaIn LM, 75.5 wt% Ga and 24.5 wt% In) into the chamber. The microfluidic device used for LM pulse oscillation consisted of a symmetric straight-channel geometry. The two terminal reservoirs served only as passive reservoirs and were not operated as pressure-controlled inlets and outlets. Unless otherwise specified, the straight microchannel was 120 μm wide and 25 μm high. For aspect-ratio studies, channel widths of 90, 120, and 150 μm were used while the channel height was maintained at 25 μm. A circular chamber with a diameter of approximately 3 mm was placed at the center of the channel and connected to the left and right straight channels. Each straight channel was 8 mm in length. Since the circular chamber and straight channels were patterned in the same photolithographic step, they had the same height of 25 μm. The PDMS channel layer was bonded to a glass substrate after oxygen plasma treatment to form a sealed microfluidic device. After bonding, the device was connected to the external environment only through the two terminal reservoirs and the central LM injection port. For particle-sorting experiments, the microfluidic device was modified by introducing a trifurcated outlet structure at one end of the straight channel. The trifurcated outlet was used to collect particles located at different transverse positions during the subsequent transport step. Except for the outlet region, the central circular chamber, channel height, channel width and LM injection port were identical to those of the pulse-oscillator device. All devices were rinsed with ethanol and dried before use. Preparation of electrolyte-based particle suspensions Electrolyte-based particle suspensions were prepared by dispersing polystyrene (PS) microspheres in NaOH or PEO-NaOH electrolyte solutions. NaOH aqueous solutions were first prepared by dissolving NaOH in deionized water at concentrations of 0.1, 0.2, 0.3, 0.4, 0.5, 0.6 M. For polymer-containing electrolytes, polyethylene oxide (PEO; 2 MDa; Macklin) was added to NaOH solutions of specified concentrations and mixed at 3000 rpm for 15 min to allow dispersion and dissolution, yielding PEO-NaOH electrolyte solutions. The final PEO concentrations were 1000, 2000, 4000, 8000 and 12000 ppm (w/w). All electrolyte solutions were degassed under vacuum before particle addition to reduce dissolved and trapped air bubbles. Polystyrene microspheres were used for flow tracing, particle focusing and particle sorting experiments. Particles with diameters of 1, 5, 10, and 15 μm were used. Microspheres of the specified diameter were added to NaOH or PEO–NaOH electrolytes to prepare particle suspensions. For particle focusing and sorting experiments, the particle mass fraction in the suspension was 0.1–0.2 wt%. Tween 20 (Macklin) was then added to the suspension at a final concentration of 0.1% (w/w) to reduce particle aggregation and improve dispersion stability. The suspensions were thoroughly mixed before use. For flow visualization and velocity measurements, 5 μm particles were used as tracer particles at a concentration of 0.20–0.25 wt% to provide sufficient seeding density for PIV analysis. All particle suspensions were thoroughly mixed before use to ensure a homogeneous particle distribution. Assembly and electrical actuation of the LM pulse oscillator Before each experiment, LM was injected into the central circular chamber through the central LM injection port. The injected LM droplet was positioned at the center of the circular chamber and used as the LM actuation element. The prepared NaOH or PEO-NaOH electrolyte-based particle suspension was introduced into the device through the terminal reservoirs to fill the straight channels, and the gap around the LM droplet. Visible air bubbles were removed during filling. After loading, external injection or pressure-driven flow was stopped, and the two terminal reservoirs were kept open to the atmosphere. Copper electrodes were inserted through the electrode-access ports adjacent to the terminal reservoirs and connected to a function generator, ensuring contact with the electrolyte without touching the glass substrate or obstructing the reservoir–microchannel connection. During oscillation experiments, the LM droplet was driven only by the applied electrical signal. A square-wave AC signal with no DC offset was applied, with a voltage amplitude of 1–10 Vp-p, a frequency of 1–10 Hz, and a duty cycle of 50%. Under AC actuation, the LM droplet underwent periodic deformation and induced reciprocating fluid motion in the microchannel. Therefore, the observed flow was oscillatory, with negligible net volumetric pumping between the two terminal reservoirs. For particle sorting experiments, the AC signal was first applied to generate oscillatory flow and induce size-dependent transverse redistribution of the particles. The electrical signal was then switched to an 8 V DC voltage to transport the particle suspension toward the trifurcated outlet, where particles were collected from different outlet branches. Optical imaging and flow velocity analysis LM droplet deformation and PS particle motion were recorded in real time using an inverted microscope equipped with a high-speed camera (Ti2-U, Nikon). A 4× objective was used for imaging. Images were acquired at a frame rate of 200 frames s⁻¹, corresponding to a temporal sampling interval of 5 ms. The phase lag was determined from the temporal delay (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {t}_{{{{\rm{LM}}}}}$$\end{document}ΔtLM) between the electric-field switching event and the LM deformation response, using \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi=2\pi f\Delta {t}_{{{{\rm{LM}}}}}$$\end{document}ϕ=2πfΔtLM, where ff denotes the electric-field switching frequency. Fluid motion was quantified from the displacement of 5 μm polystyrene tracer particles. Raw image sequences were first processed by background subtraction and contrast enhancement to remove stationary structures and uneven illumination and to improve the signal-to-noise ratio of tracer particles. The LM droplet, channel walls and other non-fluid regions were masked before analysis. Velocity fields were extracted from pairs of consecutive image frames using the cross-correlation algorithm in PIVlab 3.12. The velocity fields were calibrated using the pixel size and the time interval between consecutive frames. For each PIV frame pair, all valid velocity vectors within a predefined region of interest were extracted, and the axial component of the velocity along the channel was analyzed. The spatial mean of the axial velocity within the region of interest was calculated and used as the instantaneous average axial velocity at each time point. Numerical simulation To elucidate the flow field generated by the LM capacitor, a numerical model of the microchannel with identical dimensions to the experimental setup was developed. The simulation was implemented in COMSOL Multiphysics 6.3 using the Two-Phase Flow, Phase Field interface to model fluid motion and track the interface between the LM and the NaOH electrolyte. The Electrostatics interface was employed to solve for the electric potential V, governed by the equation:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-\nabla \cdot \left({\varepsilon }_{0}{\varepsilon }_{r}\nabla V\right)=0$$\end{document}−∇⋅ε0εr∇V=0where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\varepsilon }_{0}$$\end{document}ε0 is the vacuum permittivity and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\varepsilon }_{r}$$\end{document}εr is the relative permittivity. The resulting electric stress was incorporated as a body force via the divergence of the Maxwell stress tensor:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\bf{F}}}}=\nabla \cdot {{{\bf{T}}}}=\nabla \cdot \left[{{{\bf{E}}}}{{{{\bf{D}}}}}^{T}-\frac{1}{2}\left({{{\bf{E}}}}\cdot {{{\bf{D}}}}\right){{{\bf{I}}}}\right]$$\end{document}F=∇⋅T=∇⋅EDT−12E⋅DI Here, the electric field is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\bf{E}}}}=-\nabla V$$\end{document}E=−∇V, and the electric displacement field is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\bf{D}}}}={\varepsilon }_{0}{\varepsilon }_{r}{{{\bf{E}}}}$$\end{document}D=ε0εrE. In the present two-dimensional model, the stress tensor takes the form:6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\bf{T}}}}=\left[\begin{array}{cc}{T}_{{xx}} & {T}_{{xy}}\\ {T}_{{yx}} & {T}_{{yy}}\end{array}\right]=\left[\begin{array}{cc}{\varepsilon }_{0}{\varepsilon }_{r}{E}_{x}^{2}-\frac{1}{2}{\varepsilon }_{0}{\varepsilon }_{r}\left({E}_{x}^{2}+{E}_{y}^{2}\right) & {\varepsilon }_{0}{\varepsilon }_{r}{E}_{x}{E}_{y}\\ {\varepsilon }_{0}{\varepsilon }_{r}{E}_{y}{E}_{x} & {\varepsilon }_{0}{\varepsilon }_{r}{E}_{y}^{2}-\frac{1}{2}{\varepsilon }_{0}{\varepsilon }_{r}\left({E}_{x}^{2}+{E}_{y}^{2}\right)\end{array}\right]$$\end{document}T=TxxTxyTyxTyy=ε0εrEx2−12ε0εrEx2+Ey2ε0εrExEyε0εrEyExε0εrEy2−12ε0εrEx2+Ey2 Supplementary information Supplementary information Description of Additional Supplementary Files Video 1: Near-field and far-field flow Video 2: Migration of 1 um particles Video3: Separation of 5 um and 10 umparticles Transparent Peer Review file Supplementary information Description of Additional Supplementary Files Video 1: Near-field and far-field flow Video 2: Migration of 1 um particles Video3: Separation of 5 um and 10 umparticles Transparent Peer Review file Source data Source Data Source Data