IntroductionAdvancements in our understanding of brain disease mechanisms have spurred efforts to develop targeted pharmacological treatments for conditions such as brain tumors, neurodegenerative disorders (e.g., Alzheimer’s and Parkinson’s diseases), and epilepsy1,2,3. Despite this progress, the therapeutic efficacy of central nervous system (CNS)-targeted drugs remains limited by the difficulty of achieving and maintaining effective concentrations at the disease site. A major obstacle is the blood–brain barrier (BBB), which restricts the passage of most therapeutic agents, particularly macromolecules, from systemic circulation into brain tissue4,5,6,7.To circumvent this limitation, convection-enhanced delivery (CED) has emerged as a promising approach, involving the direct infusion of drugs into brain tissue through implanted probes8,9,10,11. By enabling localized drug delivery, CED minimizes systemic exposure and reduces off-target effects12,13,14. Microfluidic neural probes are increasingly used for this purpose, yet conventional designs typically depend on external pumps and tubing, limiting their utility in freely moving animals and hindering long-term in vivo studies. Tethered systems also compromise the fidelity of neuropharmacological assessments, highlighting the need for wireless, untethered platforms with integrated drug delivery capabilities10,15,16,17.In response to this challenge, various integrated micropump-based probes have been developed. However, conventional mechanical micropumps often rely on rigid materials and complex components for actuation, making them difficult to integrate with soft tissue environments18,19. Many of these systems, such as screw–piston-based micropumps, rely on external power supplies and bulky, rigid components, which limit their feasibility for fully implantable and wireless operation20. To address mechanical complexity, alternative actuation strategies have been explored, including diaphragm-based pumping mechanisms driven by piezoelectric21,22,23, electrostatic24,25, electromagnetic26,27 or thermo-pneumatic forces28,29,30. Although various micropump systems have been investigated, limitations related to size, material compatibility, and power supply continue to restrict their applicability in implantable drug delivery systems. Among these, thermo-pneumatic systems employ on-chip heaters to thermally expand sealed air chambers, deforming elastic diaphragms to induce fluid flow31. Microheater-based peristaltic pumps operate at low driving voltages and require only simple control circuitry, making them well suited for applications that demand miniaturization and implantability. This approach enables drug delivery without the use of rigid mechanical parts, and is thus more amenable to integration with soft, flexible platforms. Despite these developments, several technical hurdles persist, including miniaturization, precise flow rate control, and the prevention of backflow. To overcome the power delivery challenge, micropumps actuated by external stimuli such as RF, ultrasound, or alternating magnetic fields have also been studied, however, these approaches still face critical limitations for intracranial use, including the need for precise alignment and shallow effective penetration depth32,33. Addressing these challenges will require innovations in pump design that enhance integration with flexible neural probes, enable fine-tuned control of fluid dynamics, and minimize tissue trauma during operation34.Here, we report a wireless, miniaturized neural probe capable of on-demand drug delivery via a thermo-pneumatic peristaltic micropump integrated with asymmetrically tapered nozzle–diffuser microchannels. All structural and functional components of the pump are fabricated entirely from flexible materials, enabling seamless integration with soft brain tissue. The peristaltic actuation is driven by localized Joule heating of embedded microheaters, which deform elastomeric diaphragms to generate directional airflow without the need for mechanical valves31,35. This design not only minimizes backflow but also allows for programmable drug release through precise wireless control. The system is capable of delivering controlled drug volumes on demand, supporting spatially and temporally regulated infusion. Computational simulations and benchtop validation in a brain-mimicking phantom confirmed the device’s capability for consistent, real-time drug infusion. These results demonstrate the potential of this fully flexible peristaltic system for next-generation, untethered neuropharmacological interventions.ResultsFlexible neural interface design for wireless thermo-pneumatic drug deliveryFigure 1a schematically illustrates the proposed wireless thermo-pneumatic micropump implanted in the brain. The system comprises a wireless control module and an integrated microfluidic probe capable of targeting deep brain regions. Wireless actuation is achieved through localized Joule heating of microheater embedded within the system, eliminating the need for external connections. Figure 1b illustrates the architecture of the proposed thermo-pneumatic micropump, comprising three functional layers: the microheater-embedded pumping layer (bottom), the microfluidic layer (middle), and the drug reservoir layer (top).Fig. 1: Schematic illustrations and rheological analysis of flexible drug delivery system with valveless micropump.a Schematic illustration of valveless micropump implanted into the brain to generate directional airflow for drug delivery b Exploded view of the device architecture, consisting of a drug reservoir, microfluidic layer, pumping layer, microheaters, and a flexible substrate. c Working principle of the micropump showing airflow generation in supply (S, top) and pump (P, bottom) modes via diaphragm deformation controlled by microheater heating and cooling. d Top and Cross-sectional views of the micropump with three chambers for peristaltic motion. Fluid intake through the supply mode (S) and fluid pumping for drug delivery through the pump mode (P). e Finite element analysis (FEA) showing directional airflow induced during the supply mode. f Sequence of peristaltic pumping actuation. Velocity profiles during pump mode (g) and supply mode (h). Flow rate comparison during pump (i) and supply mode (j), illustrating forward flow (solid line), reverse flow (dotted line), and differential volume changes (colored in red or blue line).Full size imageThe cross-sectional architecture shown in Fig. 1c highlights the deformation of the blue-colored diaphragm, situated between the bottom pumping layer and the top microfluidic layer, during two distinct operational modes: supply mode (S) and pump mode (P). The pumping layer incorporates microheater array that modulate the diaphragm’s deformation in response to local temperature changes (Supplementary Fig. 1). The pump mode refers to the state in which the deformable diaphragm is deflected upward by the heating of the microheater, while the supply mode indicates the state in which the diaphragm returns to its original position after cooling. The upper microfluidic layer either intakes or expels air depending on the diaphragm’s state.Figure 1d shows the top and cross-sectional views of the microfluidic system, which consists of three microfluidic chambers. The rectilinear-shaped microheater can be toggled between a “cooling” state (no current flowing to the heater) and a “heating” state (current flowing through the heater). In the cooling state, the diaphragm deflates, expanding the local volume of the upper microfluidic layer and drawing in air. Conversely, in the heating state, the diaphragm inflates, increasing internal pressure and expelling air during the pump mode.The three microfluidic chambers are interconnected by asymmetrically tapered microchannels, which promote unidirectional flow—an advantage over conventional symmetrical microchannels. This design leverages a nozzle–diffuser structure, where fluid is directed through two distinct sections: a nozzle (converging channel) and a diffuser (diverging channel). In the forward direction, fluid passes easily through the nozzle, accelerating as the cross-sectional area narrows and reducing pressure. In contrast, in the reverse direction, fluid encounters greater resistance in the diffuser section, resulting in energy loss and restricted flow. This geometric asymmetry rectifies fluid movement, enabling unidirectional flow without the need for mechanical valves36,37,38. Figure 1e simulates the airflow through the asymmetric nozzle–diffuser channel structure, demonstrating directional airflow during the supply mode of the middle chamber, in contrast to the bidirectional behavior typically seen in symmetrical channels (Supplementary Fig. 2).The actuation mechanism is illustrated in Fig. 1f: alternating thermal expansion and contraction of air chambers via Joule heating drives the diaphragm’s oscillation, enabling peristaltic flow without mechanical valves. Sequential deactivation of the heaters from right to left causes localized cooling and chamber contraction in the pumping layer, which draws the diaphragm downward and inflates the corresponding chamber in the microfluidic layer. The rightmost and leftmost chambers in Fig. 1f serve as the inlet and outlet, respectively, while the middle chamber acts as an intermediary stage that separates the inlet and outlet in the peristaltic cycle. Unlike the isolated chambers in the pumping layer, the microfluidic chambers are interconnected via narrow microchannels. These channels constrain the flow path and enable directional movement of air from the inlet through the system to the outlet. During the final compression of the leftmost chamber, the pressurized air displaces fluid from the drug reservoir, enabling precise and targeted drug release into brain tissue. The full actuation cycle follows the sequence (P–P–S) → (P–S–P) → (S–P–P) → (P–P–P), corresponding to the state of the three chambers from left (outlet) to right (inlet) in Fig. 1f. This configuration allows the device size to be flexible and miniaturized for neural applications.Simulated flow dynamics of unidirectional transport in flexible peristaltic micropumpThe peristaltic mechanism enabled by the three-chamber design plays a crucial role in achieving unidirectional drug transport. Figure 1g and Supplementary Fig. 3a compare the simulated backflow of air during the pump mode for drug injection in both three-chamber (Fig. 1g) and single-chamber (Supplementary Fig. 3a) configurations. In these simulations, the inlet is positioned on the left and the outlet on the right. The red lines at the two ends of the channels of the third chamber in Fig. 1g represent the airflow velocity profiles when all three chambers are in pump mode to inject the drug from the reservoir by pushing the air to the outlet. Notably, the sequential design of the peristaltic pump effectively suppresses undesirable backflow from the third chamber to the middle chamber, demonstrating enhanced control over flow direction compared to the single-chamber layout. Similarly, the velocity profiles during the supply mode, illustrated by the blue lines in Fig. 1h and Supplementary Fig. 3b, indicate that the three-chamber sequential peristaltic configuration significantly reduces the backflow to the first chamber from the second chamber.Figure 1i and Supplementary Fig. 3c further illustrate the performance benefits of the peristaltic configuration. In the single-chamber model in Supplementary Fig. 3c, the pump transitions from supply to pump modes while adjacent chambers remain inactive. This configuration leads to limited net air delivery during the pump cycle with supplied air volume of 0.53 mm³. In contrast, in the peristaltic model where the third chamber transitions to pump mode while the others remain in pump mode, the net air displacement increases to 0.86 mm³ due to enhanced resistance to backflow. Figure 1j describes that a similar trend is observed in supply mode: when the first chamber is in supply mode with downstream chambers in pump mode, the intake air volume increases from 0.55 mm³ (Supplementary Fig. 3 d) to 0.74 mm³, demonstrating the effectiveness of coordinated peristaltic actuation.Flexible drug delivery probe design and operationFigure 2a shows the fabricated image of the micropump assembly, which includes integrated microheaters, a drug reservoir, and a long delivery probe mounted on a flexible polyimide (PI) substrate. Figure 2b provides a magnified cross-sectional view of the multilayered structure of the pump, composed of three primary layers: the microheater layer, the pumping layer, and the top microfluidic channel layer. Each chamber in the pumping layer is equipped with a serpentine-patterned Au (200 nm) microheater, designed for efficient localized heating via Joule heating (red box in Fig. 2b). These microheaters are patterned onto the PI substrate through photolithography and wet etching, enabling precise temperature control (Supplementary Fig. 4a–c). Heating or cooling of these chambers drives the deformation of diaphragms positioned between the bottom pumping layer and the top microfluidic layer, thereby generating airflow through the system.Fig. 2: Structure of the soft drug delivery device and optimization of design parameters based on FEA analysis.a Images of the fabricated drug delivery device (scale bar: 5 mm). b Cross-sectional SEM images, showing the multilayered structure of the microheater, pumping and microfluidic layers (scale bar: 1 mm). Schematic (c) and optical (d) images of the main design parameters affecting flow volume in the drug delivery device (scale bar: 500 µm). e Detailed design parameters and standards: I. Radius of the chamber (R) (scale bar: 500 µm), II. Width of the channel (W) (scale bar: 100 µm), III. Tapered angle (θ) (scale bar: 200 µm). Simulation results of flow per cycle according to R (f), W (g), θ (h).Full size imageThe top microfluidic and underneath pneumatic chambers are fabricated using PDMS (polydimethylsiloxane), which is molded with SU-8 templates that have been pre-treated with a self-assembled monolayer (SAM) coating to ensure smooth demolding (Supplementary Fig. 4d–f). The microfluidic layer is then plasma-bonded to a thin PDMS cover layer that serves as a deformable diaphragm. This composite structure is aligned and permanently bonded to the pumping layer using oxygen plasma treatment (Supplementary Fig. 4h). To further validate the mechanical robustness of the flexible probe under dynamic conditions, we performed cyclic mechanical deformation tests under both compression (10%) and tension (20%) while continuously monitoring the resistance of the embedded microheater. As shown in Supplementary Fig. 5a, b, the resistance remained within ±1% of the baseline over 100 loading cycles, confirming that the microheater maintains electrical stability without delamination or mechanical failure. These results experimentally demonstrate the device’s dynamic flexibility and structural integrity under physiologically relevant mechanical stress. Cross-sectional SEM images in Fig. 2b confirm the precise alignment of layers and well-defined channel formations. For the drug reservoir, a 3D-molded PDMS block is attached to the microfluidic layer using a thin film of uncured PDMS as an adhesive, followed by thermal curing to achieve leak-free sealing. To ensure sealing integrity during repeated heating cycles, the device employs two complementary strategies. The microfluidic and pneumatic layers are permanently bonded via oxygen plasma treatment, forming chemically robust siloxane bonds that withstand pressures exceeding 100 kPa. In addition, the interface between the PDMS pumping layer and the PI-based microheater substrate is sealed using a thin layer of uncured PDMS, which is thermally cured to form a strong mechanical bond. During actuation, the internal pressure rise due to thermal expansion is estimated to be approximately 2 kPa—well within the tolerance of both bonding interfaces.Asymmetric microchannel design optimizationA critical feature of the microfluidic system is the incorporation of asymmetrically tapered microchannels between chambers, as shown in the top-view schematic in Fig. 2c. These channels gradually widen from the inlet to the outlet, forming a nozzle-diffuser configuration that facilitates unidirectional airflow by exploiting differences in flow resistance and momentum. Several geometric parameters influence the drug delivery performance, most notably the chamber radius (R), the channel width (W), and the tapering angle of the diffuser (\({\rm{\theta }}\)) (Fig. 2d, e). Larger chamber radii correspond to increased internal air volume, enabling greater airflow per peristaltic cycle. This trend is confirmed by simulations in Fig. 2f, which show that the volume of displaced air per cycle increases with chamber size. However, under continuous operation, the influence of chamber size diminishes, potentially due to cumulative temperature elevation and system-wide thermal effects. This reduction may be attributed to thermal accumulation during repeated actuation, where insufficient cooling reduces the temperature gradient and limits diaphragm deformation, particularly in larger chambers with higher thermal inertia.Figure 2g examines the effect of the outlet channel width, showing that wider channels reduce flow resistance and allow greater volume displacement during each cycle. Meanwhile, Fig. 2h analyzes the influence of the tapering angle of the connecting channels (as defined in Region III of Fig. 2e). Larger tapering angles enhance forward flow by reducing flow resistance; however, angles that are too steep induce flow separation and vortex formation, leading to turbulence and reduced net flow. Together, these results demonstrate that careful tuning of channel geometry—specifically tapering angle, chamber radius, and channel width—is essential for optimizing directional flow, minimizing backflow, and enhancing drug delivery efficiency in neural applications. Based on simulation results, the optimized geometric parameters for neural probe were set to a chamber radius (R) of 1.5 mm, a channel width (W) of 0.5 mm, and a tapering angle (θ) of 5°.Mechanism and flow dynamics of the valveless thermo-pneumatic micropumpFigure 3a shows the thermal infrared imaging results obtained during actual peristaltic pumping operation. Based on this experimentally observed temperature range, the simulation was performed to evaluate the thermal behavior of the micropump when implanted on brain. As shown in Fig. 3b, the simulation confirms that the device maintains thermal safety under operational conditions. Combined with the thermal insulation properties of PI, this design ensures that the temperature rise at the tissue interface remains within the 2 °C limit recommended by the American Association of Medical Instrumentation (AAMI) for chronically implanted devices39. Compared to conventional Flame-Retardant Type 4 (FR4)-based drug delivery devices, the use of PI enhances thermal safety due to its lower thermal conductivity and mechanical flexibility (Supplementary Fig. 6a, b).Fig. 3: Mechanism of the valveless micropump and flow analysis results through FEA and in-vitro experiments.a Thermal IR imaging during peristaltic pumping operation (scale bar: 500 µm). b FEA results evaluating thermal safety at the device–brain interface, with temperature distribution relative to the Association for the Advancement of Medical Instrumentation (AAMI) safety standard. c FEA results for temperature, pressure, and injection volume according to digital input.Full size imageFigure 3c shows the pressure and temperature changes across actuation cycles obtained from FEA. The top plot shows the sequential cooling (‘0’) of the microheaters from inlet (yellow) to outlet (pink) chambers. The simulation results reveal strong correlations between digital control signals and resulting temperature profiles within the air chambers, pressure generation in the outlet, and net drug injection volume. These results confirm the effectiveness of sequential actuation for generating stable, directional flow and guided the optimization of pump parameters for minimal backflow and maximum delivery efficiency.To further evaluate long-term thermal stability, we conducted a 30-min continuous actuation test using the wireless system. As shown in Supplementary Fig. 7a, the heater temperature remains stably maintained within the 60–70 °C range during prolonged operation. Corresponding current measurements (Supplementary Fig. 7b, c) confirm consistent power delivery across different actuation modes, validating both thermal and electrical safety over extended periods.Wireless benchtop demonstration of the integrated drug delivery moduleThe feasibility of the proposed micropump-based neural probe was validated in a brain tissue phantom model. Figure 4a presents optical images of the probe embedded in 0.6% agarose gel, simulating brain mechanical properties. The close-up view reveals the compact integration of the microfluidic architecture. The distal end of the probe, which serves as the drug release interface, is designed for direct insertion into brain tissue, including deep regions such as the ventricular space. The 7 cm-long, 500 µm-wide probe balances structural compliance with sufficient channel dimensions for effective convection-enhanced delivery. To minimize drug backflow following infusion, the outlet geometry incorporates a narrow cross-section to increase Laplace pressure and support localized, stable drug retention. The corresponding control circuit, shown in Fig. 4b, enables fully wireless operation of the micropump through digital modulation of the microheaters. The gray-shaded region indicates the module responsible for the wireless control of the implanted drug delivery system. This is achieved through a Bluetooth Low Energy (BLE)-enabled System-on-Chip (SoC), which transmits real-time commands via a 2.4 GHz antenna. The General-Purpose Input/Output (GPIO) ports of the SoC are configured to control an array of analog switches, each connected to individual microheaters, enabling precise and programmable actuation sequences. We present Supplementary Fig. 8, which shows the fully assembled system combining the flexible PI-based circuit (20 × 12 × 0.1 mm) and a compact 65 mAh lithium polymer battery (15 × 15 × 2 mm), confirming its suitability for untethered in vivo applications. Power is supplied through a LiPo battery, regulated to a stable 3.3 V for the SoC, ensuring consistent and reliable operation.Fig. 4: Experimental demonstration of the fabricated drug delivery module in a benchtop setting.a Optical image of the micropump-based microfluidic probe (scale bar: 5 mm). b Circuit diagram showing wireless functionality and micropump operation. c Real-time GPIO on/off pattern of a microheater controlled wirelessly, which directly modulates the flow rate of the drug delivery module. d Flow analysis results according to frequency changes in in-vitro experiments. e Results of flow per cycle according to switching frequency in in-vitro experiments (n = 3). Error bar represents the standard deviation. f Optical images of drug delivery over time in in-vitro experiments (scale bar: 1 cm).Full size imageWireless heating cycles were precisely controlled in real time, as measured by the output voltage modulation plotted in Fig. 4c with various frequencies. The observed changes in the on/off pattern of a single GPIO channel among the three-microheater array confirm that the peristaltic pumping speed can be dynamically adjusted through wireless commands, enabling real-time control of drug delivery flow rate. Flow characterization is performed using 2 µm polystyrene microspheres suspended in distilled water. Microsphere motion is tracked using a high-speed camera, and flow velocity is extracted via MATLAB-based image analysis. Figure 4d demonstrates the dynamic change in injection speed corresponding to mode switching, as visualized using microsphere particles. High-speed camera was positioned at the midpoint of the microfluidic channel, where the flow is assumed to have reached near saturation following the onset of actuation. The observed displacement of microspheres confirms that the cumulative injected volume increases over successive actuation cycles, and that real-time modulation of switching frequency effectively alters the instantaneous injection speed, thereby enabling tunable control of total drug volume delivery.Figure 4e presents the injected volume per cycle as a function of actuation frequency, revealing that slower frequencies yield higher net injection per cycle due to more complete diaphragm recovery during the extended interval between successive actuations. However, when the data are converted to flow rate (nL/s), a nonlinear trend is observed. Notably, the flow rate initially decreases from 0.5 Hz to 1 Hz, then increases with further increases in switching frequency. This phenomenon likely arises from the thermal and mechanical dynamics of the thermo-pneumatic actuation mechanism. At very low frequencies (e.g., 0.5 Hz), although each cycle allows sufficient time for full diaphragm expansion and contraction, the long period results in a lower total volume delivered per unit time. At 1 Hz, the switching speed increases but may still be insufficient for optimal thermal buildup and pressure transfer, leading to a temporary dip in pumping efficiency. As the frequency increases further (2–3 Hz), thermal accumulation within the microchambers appears to improve pressure buildup, enhancing diaphragm deflection and restoring flow efficiency. This result highlights that the relative contributions of thermal accumulation and mechanical diaphragm dynamics vary depending on the switching frequency, with mechanical recovery dominating at lower frequencies and thermal effects becoming more prominent at higher frequencies. This results in a net increase in flow rate despite a slight compromise in per-cycle volume. From a functional standpoint, this frequency-dependent flow modulation highlights a key advantage of the proposed neural probe system. By adjusting the switching frequency, the device enables tunable drug delivery rates without altering device geometry or materials. In addition to frequency modulation, we also observe that the applied voltage to the microheaters significantly influences the flow rate. This is attributed to the voltage-dependent displacement of the diaphragm, which alters the pressure generated during actuation and consequently affects the net volume delivered per cycle (Supplementary Figs. 9, 10).A time-lapse optical image series (Fig. 4f) showing continuous dye infusion into a brain phantom over a 540-s period. At 0 s, the wireless system was activated using a battery-powered circuit supplying 3.3 V, and the device operated in Mode 1 continuously throughout the experiment. Under these conditions, the peristaltic micropump functioned stably, resulting in a gradual and sustained increase in the infused volume over time. Furthermore, repeatable infusion performance was confirmed across four consecutive refill cycles, demonstrating consistent dye propagation patterns without performance degradation (Supplementary Fig. 11). Notably, the infused dye exhibits a clear directional propagation along the central axis of the outlet channel without visible lateral leakage. This directional behavior may be influenced by the pre-formed insertion path created using a guide needle prior to neural probe implantation, which could reduce mechanical resistance and facilitate axial flow. While the potential clinical advantages of such spatially confined delivery remain to be further validated, this characteristic may be beneficial for minimizing off-target diffusion in applications requiring localized drug administration.DiscussionWe have developed a fully flexible, wireless neural probe capable of on-demand, directional drug delivery using a thermo-pneumatic peristaltic micropump integrated with asymmetrically tapered nozzle–diffuser microchannels. The system employs sequential Joule heating of embedded microheaters to generate unidirectional airflow without the need for mechanical valves or external tubing. All functional and structural components are fabricated from soft, biocompatible materials, ensuring mechanical compliance with neural tissue and supporting full implantability of the probe component. To accommodate the need for air input during thermo-pneumatic actuation, the drug reservoir and control circuitry are positioned externally in a head-mounted configuration, allowing ambient air access without requiring an open cranial incision. Fluidic performance was optimized through computational simulations, and device functionality was validated in brain-mimicking phantoms. The results confirm that the proposed design enables consistent, programmable drug infusion with minimal backflow. This work addresses several key limitations of conventional neural drug delivery systems, including mechanical rigidity, external tethering, and imprecise dosing. By combining wireless control, flexible integration, and precise fluid actuation, the platform offers a promising route toward untethered, closed-loop neuropharmacological therapies (Supplementary Table 1). To support potential future in vivo applications, several important factors must be considered. Long-term operation in complex biological environments may pose challenges such as channel blockage due to biofouling and heat accumulation near sensitive tissue. Biofouling can be mitigated through anti-fouling surface treatments that modify the inherently hydrophobic PDMS channel walls to be more hydrophilic, thereby reducing nonspecific adsorption of proteins and cells40. For thermal management, we envision placing the actuation components on the skull with an encapsulating package, which, according to our simulations, would confine temperature increases to within 2 °C within safe biological limits. In addition, the digitally programmable actuation with sub-100 nL resolution and wireless design enables integration with biosignal acquisition systems for closed-loop and individualized drug delivery. This supports real-time, patient-specific modulation of intracerebral dosing, as needed in conditions like epilepsy and Parkinson’s disease. Taking these factors into account, future efforts will focus on in vivo validation and integration with real-time biosensors to enable autonomous therapeutic feedback systems for chronic neurological disorders.MethodsDevice fabricationThe micropump device was fabricated using standard microelectromechanical systems (MEMS) techniques. Each layer—pumping, microfluidic, and drug reservoir—was constructed using PDMS. The microheater substrate was prepared by depositing thin layers of Cr/Au (7 nm/200 nm) onto a 100 μm-thick PI film using electron beam evaporation. Photolithography followed by wet etching defined serpentine-shaped heater patterns on the substrate. To fabricate the pumping layer, a SU-8 2150 epoxy photoresist (Kayaku Advanced Materials, USA) was spin-coated and patterned on a silicon wafer to form a mold with cavity structures (150 μm height), allowing for volume expansion under thermo-pneumatic actuation. Prior to PDMS casting, the SU-8 mold surface was treated with a vapor-phase anti-stiction layer using trichloro(1H,1H,2H,-perfluorooctyl) silane (Sigma-Aldrich, USA) at room temperature for 2 h. Degassed PDMS (10:1 base to curing agent) was poured onto the mold, covered with a glass slide (76 × 52 mm²) pre-treated with a Pt catalyst inhibitor solution (5% AEAPS in 95% methanol), and cured at 70 °C for 1 h After curing, the PDMS replica was peeled off from the mold. The microfluidic layer was fabricated through a similar soft-lithography process. SU-8 was patterned on a silicon wafer to define microchannel and micropump geometries. Following anti-stiction treatment, degassed PDMS (10:1) was cast and bonded to a pre-treated glass substrate, then cured and released. Inlet and outlet ports were created using a 1.2 mm diameter punch. A thin PDMS film was spin-cast on a 125 μm-thick PI substrate and cured, then bonded to the microfluidic layer using oxygen plasma treatment (100 W, 20 mtorr, 32 s, 22 sccm O₂; Femto Science, Korea) to form strong siloxane bonds. For the drug reservoir, degassed PDMS (5:1 base to curing agent) was cast into a custom 3D-printed mold (Project 3500, 3D Systems, USA) and cured at 70 °C for 1 h. The individual components were integrated in sequence: the pumping and microfluidic layers were aligned and permanently bonded using oxygen plasma treatment, and the drug reservoir was affixed to the outlet region of the micropump channel using silicone adhesive to prevent leakage.Measurement of device temperature during peristaltic pumpingThe device’s transient temperature profile during peristaltic pumping was monitored using an infrared imaging system (A655sc, FLIR Systems, Inc.). To assess the localized heating behavior of each microheater, circular regions of interest (ROIs) were manually defined over each heater element. During sequential thermal actuation, the maximum temperature within each ROI was continuously monitored in real time. This analysis enabled precise tracking of temperature fluctuations associated with each actuation cycle, providing insights into the thermal dynamics and spatial uniformity of the heating elements during pump operation.Finite element analysis of micropump’s directional airflowTo analyze whether the airflow, which drives drug delivery, flows directionally, numerical simulations were performed using COMSOL Multiphysics. Simulation parameters for the micropump material are provided in Supplementary Table 2. The micropump generates airflow through diaphragm displacement caused by heating of the microheater. The heat generated by the microheater was conducted to the air inside the cavity, leading to an increase in internal air pressure. This pressure was applied to the inner boundary walls containing the diaphragm. Following the calculation of diaphragm displacement induced by heating of the microheater (Cr/Au; thickness 7 nm/200 nm), the resulting airflow was analyzed based on the structural deformation within the sequential nozzle-diffuser configuration of the micropump. To verify the directional nature of the airflow, instantaneous flow rates perpendicular to both ends of the nozzle-diffuser structure were measured during both pump mode and supply mode. The difference in the perpendicular flow components at both ends was calculated to assess directional airflow. To enhance computational accuracy, a high-density mesh was applied to the cross-sectional areas where this analysis was conducted.Parametric optimization of micropump design for enhanced drug deliveryNumerical simulations were conducted using COMSOL Multiphysics to optimize the design of a micropump-based drug delivery mechanism for enhanced delivery efficiency. The simulation workflow was established to evaluate drug delivery performance based on the generated pressure. The temperature distribution induced by the applied digital input to the circuit was analyzed, followed by the calculation of diaphragm displacement and the resulting pressure at the outlet during sequential pumping. The corresponding drug injection volume was then determined from the calculated pressure. To further improve delivery efficiency, parametric optimization was performed on key design parameters influencing drug delivery, including the radius of the chamber (R), width of the channel (W), and tapered angle (θ) (Fig. 2e). Each parameter was systematically varied within a predefined range to identify optimal design conditions, considering both delivery efficiency and airflow stability.Finite element analysis of thermal safety at the device–brain interfaceTo evaluate the thermal impact on brain tissue during device operation, numerical simulations were conducted using COMSOL Multiphysics. The Heat Transfer in Solids and Fluids (ht) module was employed to model the transient temperature distribution within the device and the surrounding biological environment. The simulation domain included both the device structure and the brain tissue interface. Material parameters used in the thermal simulation are listed in Supplementary Table 2. The outer boundaries were fixed at ambient temperature (25 °C), while the brain tissue was initialized at physiological temperature (37 °C). A refined mesh was applied to regions with steep temperature gradients to ensure computational accuracy.Flow rate characterizationTo evaluate the performance of the peristaltic micropump, flow rate characterization was conducted using a high-speed imaging system. The drug reservoir was filled with distilled water containing 2 µm-diameter polystyrene microspheres (Sigma-Aldrich, USA) at a dilution ratio of 100:1. During pump operation, the motion of the microspheres within the microfluidic channel was recorded using a high-speed camera (Chronos 1.4, KRON Technologies) equipped with a microscope lens, capturing video at a frame rate of 100 frames per second. The recorded video was analyzed in MATLAB (MathWorks, USA) to determine the displacement of individual microspheres across consecutive frames. The pixel-wise displacement over time was converted into flow velocity by calibrating the spatial resolution of the imaging system. The volumetric flow rate was then computed by multiplying the measured flow speed with the cross-sectional area of the microchannel.Wireless drug delivery demonstrationThe wireless drug delivery system was designed by integrating several key electronic components to enable real-time remote control. A Bluetooth Low Energy System-on-Chip (BLE SoC; QN9080SIP, NXP Semiconductors) equipped with a built-in 2.4 GHz antenna was employed to wirelessly transmit operational commands, including start, stop, and mode switching, to a smartphone. A voltage regulator (NCV8161ASN330T1G) was incorporated to provide a stable 3.3 V power supply to both the microheaters and the microcontroller unit. An analog switch (MMBT2222ALT1G) controlled via GPIO pins was used to modulate the actuation sequences of the microheaters. The drug delivery neural probe fabricated on a PI substrate was electrically connected to the wireless circuit via an FPC connector (2328702-6, TE Connectivity. To mimic physiological insertion conditions, the neural probe was inserted into a 0.6% agarose gel phantom. Prior to insertion, a guide needle was used to create a track within the phantom to prevent blockage of the microfluidic channel. Wireless operation of the device was demonstrated in this benchtop setup, utilizing colored dye as a drug analog to visualize infusion dynamics. The wireless system allowed modulation of the drug delivery rate, and the spatial distribution of the dye was analyzed through cross-sectional imaging of the phantom. Additionally, the occurrence of backflow was monitored by placing marker particles near the outlet and tracking their displacement during device operation.Data availabilityAll data needed to evaluate the conclusion in the paper are present in source data. Additional data are available on request from the corresponding author on reasonable request.Code availabilityThe custom codes used in this study for wireless drug delivery control are available from the corresponding author upon request.ReferencesJamal, A. et al. Insights into infusion-based targeted drug delivery in the brain: perspectives, challenges and opportunities. Int. J. Mol. Sci. 23, 3139 (2022).PubMed PubMed Central Google Scholar Lee, H. J. et al. A multichannel neural probe with embedded microfluidic channels for simultaneous in vivo neural recording and drug delivery. Lab Chip 15, 1590–1597 (2015).PubMed Google Scholar Shin, H. et al. Neural probes with multi-drug delivery capability. Lab Chip 15, 3730–3737 (2015).PubMed Google Scholar Theodorakis, P. E., Müller, E. A., Craster, R. V. & Matar, O. K. Physical insights into the blood–brain barrier translocation mechanisms. Phys. Biol. 14, 041001 (2017).PubMed Google Scholar Pathan, S. A. et al. CNS drug delivery systems: novel approaches. Recent Pat. Drug Deliv. Formul. 3, 71–89 (2009).PubMed Google Scholar Nance, E., Pun, S. H., Saigal, R. & Sellers, D. L. Drug delivery to the central nervous system. Nat. Rev. Mater. 7, 314–331 (2022).PubMed Google Scholar Bors, L. A. & Erdő, F. Overcoming the blood–brain barrier. challenges and tricks for CNS drug delivery. Sci. Pharm. 87, 6 (2019).Google Scholar Vogelbaum, M. A. et al. First-in-human evaluation of the Cleveland Multiport Catheter for convection-enhanced delivery of topotecan in recurrent high-grade glioma: results of pilot trial 1. J. Neurosurg. 130, 476–485 (2018).PubMed Google Scholar Lidar, Z. et al. Convection-enhanced delivery of paclitaxel for the treatment of recurrent malignant glioma: a phase I/II clinical study. J. Neurosurg. 100, 472–479 (2004).PubMed Google Scholar Wu, Y. et al. Wireless multi-lateral optofluidic microsystems for real-time programmable optogenetics and photopharmacology. Nat. Commun. 13, 5571 (2022).PubMed PubMed Central Google Scholar Kaurav, H. & Kapoor, D. N. Implantable systems for drug delivery to the brain. Ther. Deliv. 8, 1097–1107 (2017).PubMed Google Scholar D’Amico, R. S., Aghi, M. K., Vogelbaum, M. A. & Bruce, J. N. Convection-enhanced drug delivery for glioblastoma: a review. J. Neuro-Oncol. 151, 415–427 (2021).Google Scholar Mehta, A., Sonabend, A. & Bruce, J. Convection-enhanced delivery. Neurotherapeutics 14, 358–371 (2017).PubMed PubMed Central Google Scholar Lonser, R. R., Sarntinoranont, M., Morrison, P. F. & Oldfield, E. H. Convection-enhanced delivery to the central nervous system. J. Neurosurg. 122, 697–706 (2015).PubMed Google Scholar Dagdeviren, C. et al. Miniaturized neural system for chronic, local intracerebral drug delivery. Sci. Transl. Med. 10, eaan2742 (2018).Qazi, R. et al. Wireless optofluidic brain probes for chronic neuropharmacology and photostimulation. Nat. Biomed. Eng. 3, 655–669 (2019).PubMed Google Scholar Zhang, Y. et al. Battery-free, fully implantable optofluidic cuff system for wireless optogenetic and pharmacological neuromodulation of peripheral nerves. Sci. Adv. 5, eaaw5296 (2019).PubMed PubMed Central Google Scholar Nisar, A., Afzulpurkar, N., Mahaisavariya, B. & Tuantranont, A. MEMS-based micropumps in drug delivery and biomedical applications. Sens. Actuators B Chem. 130, 917–942 (2008).Google Scholar Farra, R. et al. First-in-human testing of a wirelessly controlled drug delivery microchip. Sci. Transl. Med. 4, 122ra21–122ra21 (2012).PubMed Google Scholar Rhie, W. & Higuchi, T. Design and fabrication of a screw-driven multi-channel peristaltic pump for portable microfluidic devices. J. Micromech. Microeng. 20, 085036 (2010).Munas, F. R. et al. Development of PZT actuated valveless micropump. Sensors 18, 1302 (2018).PubMed PubMed Central Google Scholar Singh, S., Kumar, N., George, D. & Sen, A. Analytical modeling, simulations and experimental studies of a PZT actuated planar valveless PDMS micropump. Sens. Actuators A Phys. 225, 81–94 (2015).Google Scholar Hsu, Y.-C. & Le, N. B. Equivalent electrical network for performance characterization of piezoelectric peristaltic micropump. Microfluid Nanofluid 7, 237–248 (2009).Google Scholar Uhlig, S. et al. Electrostatically driven in-plane silicon micropump for modular configuration. Micromachines 9, 190 (2018).PubMed PubMed Central Google Scholar Machauf, A., Nemirovsky, Y. & Dinnar, U. A membrane micropump electrostatically actuated across the working fluid. J. Micromech. Microeng. 15, 2309 (2005).Google Scholar Zhou, Y. & Amirouche, F. An electromagnetically-actuated all-PDMS valveless micropump for drug delivery. Micromachines 2, 345–355 (2011).Google Scholar Gidde, R. R., Pawar, P. M., Ronge, B. P. & Dhamgaye, V. P. Design optimization of an electromagnetic actuation based valveless micropump for drug delivery application. Microsyst. Technol. 25, 509–519 (2019).Google Scholar Hamid, N. A. et al. A stack bonded thermo-pneumatic micro-pump utilizing polyimide based actuator membrane for biomedical applications. Microsyst. Technol. 23, 4037–4043 (2017).Google Scholar Chee, P. S., Minjal, M. N., Leow, P. L. & Ali, M. S. M. Wireless powered thermo-pneumatic micropump using frequency-controlled heater. Sens. Actuators A Phys. 233, 1–8 (2015).Google Scholar Van de Pol, F., Van Lintel, H., Elwenspoek, M. & Fluitman, J. A thermopneumatic micropump based on micro-engineering techniques. Sens. Actuators A Phys. 21, 198–202 (1990).Google Scholar Forouzandeh, F., Arevalo, A., Alfadhel, A. & Borkholder, D. A. A review of peristaltic micropumps. Sens. Actuators A Phys. 326, 112602 (2021).PubMed PubMed Central Google Scholar Armenia, I. et al. Photonic and magnetic materials for on-demand local drug delivery. Adv. drug Deliv. Rev. 191, 114584 (2022).PubMed Google Scholar Zhou, J. et al. An ultrasonically powered micropump for on-demand in-situ drug delivery. In IEEE 29th International Conference on Micro Electro Mechanical Systems (MEMS). 349–352 (IEEE, 2016).Lueshen, E., Tangen, K., Mehta, A. I. & Linninger, A. Backflow-free catheters for efficient and safe convection-enhanced delivery of therapeutics. Med. Eng. Phys. 45, 15–24 (2017).PubMed Google Scholar Husband, B., Bu, M., Evans, A. G. & Melvin, T. Investigation for the operation of an integrated peristaltic micropump. J. Micromech. Microeng. 14, S64 (2004).Google Scholar Kawun, P., Leahy, S. & Lai, Y. A thin PDMS nozzle/diffuser micropump for biomedical applications. Sens. Actuators A Phys. 249, 149–154 (2016).Google Scholar Singhal, V., Garimella, S. V. & Murthy, J. Y. Low Reynolds number flow through nozzle-diffuser elements in valveless micropumps. Sens. Actuators A Phys. 113, 226–235 (2004).Google Scholar Stemme, E. & Stemme, G. A valveless diffuser/nozzle-based fluid pump. Sens. Actuators A Phys. 39, 159–167 (1993).Google Scholar Reichert, W. M. Indwelling neural implants: strategies for contending with the in vivo environment. in Indwelling neural implants: strategies for contending with the in vivo environment (CRC Press, 2007).Sim, J. Y. et al. Microfluidic neural probes: in vivo tools for advancing neuroscience. Lab Chip 17, 1406–1435 (2017).PubMed Google Scholar Download referencesAcknowledgementsThis work was supported by the National Research Foundation of Korea (NRF) funded by Ministry of Science and ICT (RS-2023-00234581) and the Technology Innovation Program (RS-2025-08672969) funded by the Ministry of Trade Industry and Energy (MOTIE, Korea).Author informationAuthor notesThese authors contributed equally: Hyeokjun Lee, Soojeong Song.Authors and AffiliationsDepartment of Robotics and Mechatronics Engineering, Daegu Gyeongbuk Institute of Science and Technology (DGIST), Daegu, Republic of KoreaHyeokjun Lee, Soojeong Song, Jeongdae Ha & Kyung-In JangDepartment of Nano Convergence Engineering, Jeonbuk National University, Jeonju, Republic of KoreaYoon Kyeung LeeAuthorsHyeokjun LeeView author publicationsSearch author on:PubMed Google ScholarSoojeong SongView author publicationsSearch author on:PubMed Google ScholarJeongdae HaView author publicationsSearch author on:PubMed Google ScholarYoon Kyeung LeeView author publicationsSearch author on:PubMed Google ScholarKyung-In JangView author publicationsSearch author on:PubMed Google ScholarContributionsH.L. and S.S. equally contributed to the work. H.L., S.S., Y.K.L., and K.-I.J. designed the project and the detailed experimental protocols. H.L. designed and fabricated the devices for all experiments. S.S. and J.H. designed FEA models and performed the simulations. H.L. carried out experiments. H.L. and S.S. conducted the investigation and analyzed the data. H.L., S.S., and Y.K.L. wrote the paper. K.-I.J. acquired funding. Y.K.L. and K.-I.J. equally supervised the project. All authors discussed the results and contributed to revision of the manuscript.Corresponding authorsCorrespondence to Yoon Kyeung Lee or Kyung-In Jang.Ethics declarationsCompeting interestsThe authors declare no competing interests.Additional informationPublisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.Supplementary informationSupplementary informationRights and permissionsOpen Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.Reprints and permissionsAbout this article