Introduction

Neural interfaces for neuromodulation or recordings are an advancing field that requires the development of better technologies and devices for improving safety, volumetric distribution, cellular/sub-cellular resolution, and lowering the implant foreign body reaction and rejection. The literature in the field suggests a clear trend for neural interfaces - also called brain-machine interfaces - to possess essential features such as softness (decreasing the mechanical mismatch between the synthetic material and the neural tissue), reduced dimension (reaching cellular resolution and volumetric distribution, reducing invasiveness, improving membrane coupling), freestanding or wireless capabilities (improving resolution and volumetric distribution, reducing invasiveness), and optimal material and coating composition (improving biocompatibility and electrodes impedance)1,2,3,4,5,6. Ideally, a neural implant would interrogate and modulate thousands of cells over a large volume and seamlessly integrate among them without interfering with the brain’s natural structure and activity. This can only be achieved with devices that are as small as the cells (or smaller) and freestanding to minimize any disruption to the surrounding tissue.

Current neural implants can be categorized into three groups: wired implants, millimeter-scale wireless devices, and nanoparticles. A list of various state-of-the-art implants7 as well as their reachable brain depth and potential volumetric distribution is depicted in Fig. 1, and some of their characteristics and references are specified in Table 1. While there have been developments in miniaturizing wired neural implants, conventional bioelectronic interfaces like deep brain stimulation (DBS) leads8,9 and the Utah array10,11,12 are intrinsically rigid and non-compliant with the brain’s mechanics and motions, prompting the destruction of the neural tissue and an extensive foreign body reaction. To address the mechanical mismatch between the implants and the brain, new substrates based on polymers, such as polyimide and parylene-C-based Michigan-style arrays, have been developed to improve the softness and flexibility of the implants13,14,15. However, these penetrating substrates are still very invasive and prone to inflammatory responses. Another class of polymeric materials is silicone elastomers which can be used for achieving better mechanical compliance with the neural tissue thanks to their stretchability. Stretchable implants like electrocorticography (ECoG) arrays16,17,18 can conform well to the very curvilinear surface of the cortex and cover a large area but are limited to the cortex surface, carries challenges for integrating stretchable electronics, and their resolution is usually poor. Other approaches to reducing invasiveness and achieving minimal damage to the surrounding tissue include fibers-based19,20 and mesh electronics21,22,23, as they minimize the footprint of the implants. Nevertheless, fibers have a very limited number of active sites.

Fig. 1: State-of-the-art neural interfaces and the concept of freestanding micro-nanodevices.

Schematic representation of a human brain sagittal plane decorated with current neural implants—grouped into three categories: wired implants, wireless devices, and nanoparticles—and with the novel class of freestanding micro-nanodevices introduced in this work (i). Enlarged views of the implants surround the figure: (a) Utah array, (b) Mesh electronics, (c) Multifunctional fibers, (d) DBS lead, (e) Michigan-style array, (f) Stretchable ECoG electrodes, (g) nanoparticles, (h) mm-scale wireless devices, and (i) freestanding micro-nanodevices

Wireless devices at the scale of a few millimeters or several 100 s of micrometers have also been developed to remove the inconvenience of physical connections to external (and usually bulky) electronics24,25,26,27. However, they still occupy a significant tissue volume and are much larger than a single neuron, resulting in lower spatial resolution than some wired options.

Nanomaterials that can transduce a wireless stimulus (optical, magnetic, acoustic) into electrical signals and vice versa, including nanoparticles, -wires or -rods, and -flakes, have the advantage of further reduced dimensions, increasing the overall mechanical compliance and volumetric distribution, and possibly decreasing the inflammatory response of the tissue28,29,30,31,32. However, they have limited freedom in design and do not permit multimodality integration, mostly due to their incompatibility with planar manufacturing processes.

Despite these advancements, there is still no ideal neural or bioelectronic interface that combines wireless (sub)cellular modulation with great mechanical compliance, spatial resolution, volumetric distribution obtained through single or multiple injections, compatibility with planar manufacturing technologies, and design freedom. In particular, compatibility with planar processes is essential not only to enable very large-scale integration (VLSI) processes and leverage Moore’s law scaling but also to permit enhanced design freedom compared to nanoparticles and achieve more sophisticated functions, such as computation and multimodal integration. In this context, freestanding micro-nanodevices—which differ from nanoparticles by their compatibility with planar manufacturing technologies—emerge as promising candidates for the next generation of neural interfaces (Fig. 1). Nevertheless, practical implementations of these devices are hindered by technical challenges. Specifically, fabricating and harvesting devices at subcellular sizes that are free-floating poses significant hurdles, and characterizing them—crucial for ensuring efficient energy transfer and functionality at such small scales—remains exceedingly challenging.

To address this, our team proposes fabrication, harvesting, and characterization process flows for freestanding micro-nanodevices that can potentially be used as bioelectronic neural interfaces. We developed three microfabrication methods that emphasize (i) high-throughput, (ii) quick and cell-friendly, and (iii) nanoscale patterning and precision harvesting, discussing the advantages and disadvantages of each method. Developing these ideal, freestanding micro-nanodevices involves not only challenging fabrication and harvesting but also tricky characterization and validation before bringing them to in vitro and in vivo studies. Conventional substrate-bound or wired measurements, or even larger and more accessible versions of the devices, do not often replicate the final condition. For instance, the electrical coupling between the device and the membrane is an essential feature that dictates the leaking currents and, therefore, the ability to stimulate or record effectively, which cannot be captured by measurements on substrate-bound, wired, or larger devices. To address these issues, we developed an efficient way of characterizing freestanding devices using bilayer lipid membranes (BLMs). This approach can uncover currents generated by the devices and injected through a lipid membrane like the plasma membrane. Finally, we propose an analytical model to scale the measured currents to single cells as a framework for freestanding device optimization for neuromodulation.

Through our research, we hope to contribute to the development of freestanding and effective neural interfaces with improved bio-integration that can unlock the potential of neuromodulation technologies as neuroprosthetics and as treatment options for psychiatric and neurodegenerative disorders.

Results

Fabrication, release, collection, and cell culture delivery of freestanding micro-nanodevices: process flows

Figure 2 illustrates the three process flows developed in this work to manufacture freestanding devices with diverse characteristics. Process flow I (Fig. 2a–c) relies on conventional VLSI microfabrication techniques to manufacture freestanding microdevices with high throughput. Photolithography and etching enable device patterning with design flexibility and resolution as fine as 2 µm on substrates as large as 200 mm in diameter. Following this approach, it is feasible to manufacture hundreds of millions of devices with a diameter of 10 µm on a single 200 mm wafer. Because sacrificial layers like aluminum can be used for this process, the device release step can be efficiently performed by immersing the substrates structured with devices in aluminum etchant until the sacrificial layer is fully etched. Following this step, freestanding microdevices form a suspension within the etchant solution. Using a centrifugal filter connected to a vacuum line, the hazardous etchant solution can be quickly exchanged with any physiological solution. The final volume of the suspension can be accurately controlled to define a device concentration suitable for the targeted application.

Fig. 2

Process flows: fabrication, release, collection, and cell culture delivery of freestanding micro-nanodevices. Three different protocols to manufacture freestanding devices. a–c illustrate the procedure to fabricate microdevices with high throughput (process flow I). a Devices are patterned on aluminum (Al)-coated Si wafers using conventional microfabrication techniques, such as photolithography and etching. The wafer is then diced into individual chips. b Devices are detached from the substrate by sacrificial layer wet etching. The aluminum etchant is subsequently exchanged with cell medium using centrifugal filters. c The resulting device suspension is injected directly into the cell culture. d–f describe the fabrication of freestanding microdevices using cell-friendly materials only (process flow II). d Devices are patterned by soft lithography and replica molding on a dextran-coated substrate. e Cell medium can be used directly to dissolve the dextran sacrificial layer and to collect the devices in suspension. f Devices are injected into the culture plate. g–i illustrate the manufacturing of freestanding nanodevices at a small scale (process flow III). g Focused ion beam lithography is used to pattern the devices on a parylene-C-coated substrate. h The polymeric sacrificial layer is etched by dry etching to release the nanodevices from the substrate. The devices are subsequently collected in cell medium using a custom micro vacuum aspiration system. i The resulting suspension is dispensed directly into the cell culture

Process flow II (Fig. 2d–f) introduces a quick and cell-friendly microfabrication process employing non-hazardous chemicals and biocompatible materials throughout its entirety to manufacture freestanding microdevices with dimensions between 2 µm to 50 µm and with moderate throughput. Soft lithography and replica molding enable the patterning of millions of devices over a substrate of typically 1–5 cm side length and coated with a sacrificial layer, in this case, dextran. A larger patterning area might lead to inhomogeneity in the device thicknesses. Devices can then be sterilized on substrate and, thanks to the water-solubility and biocompatibility of dextran, be easily and rapidly released and collected in any aqueous-based solution by pipetting the desired volume back and forth directly on the array. The resulting device suspension can be used with cells straightaway, avoiding loss of devices during solvent exchanges and facilitating seamless integration in biological applications.

Process flow III presented in Fig. 2g–i was designed to fabricate and manipulate nanoscale freestanding devices of limited quantity. Focused ion beam (FIB) lithography can be used to pattern devices with dimensions typically ranging from 200 nm to 3 µm and a resolution as high as 10 nm. FIB tools integrating a laser interferometric stage, such as the Velion FIB (Raith Nanofabrication, Germany), have the capability to create nanostructures at the wafer scale. However, the pattering time typically limits the number of devices to a few ten to hundred thousand across a patterning area of a few hundred micrometers by a few hundred micrometers. Parylene-C is used as a sacrificial layer to transfer the smoothness of the polished silicon substrate to the device layer, thus allowing device thicknesses as low as ten nanometers. The removal of the perylene-C is performed by dry etching, which makes the released devices tightly adhere to the substrate and cannot be re-suspended by simple immersion in a solution. Instead, the devices are locally harvested in cell medium by raster scanning the array with a glass micropipette connected to a suction line. This method permits the collection of most nanodevices in a well-controlled and small volume (typically 10 µL), resulting in concentrations well-suited for biological applications. Note that due to the limited number of nanodevices, the culture dish dimensions should be chosen wisely to maximize the probability of devices interacting with the cells.

According to the process flow and chosen materials, the micro-nanodevices can be sterilized before or after release using conventional methods (e.g., ultraviolet exposure, oxygen plasma, or autoclave) prior to being dispensed into cell cultures. Indeed, once the necessary rinsing and sterilization steps are implemented, all process flows presented can be considered cell friendly.

Fabrication, release, collection, and cell culture delivery of freestanding micro-nanodevices: experimental results

Figure 3 provides a comprehensive summary of the experimental results obtained following process flows I to III. In Fig. 3a, a scanning electron microscopy (SEM) image of an array of microdevices manufactured by conventional microfabrication techniques is shown. In this particular case, the devices are composed of a mixed stack of metallic and polymeric layers, but the fabrication process can be extended to any type of material. Figure 3b depicts the same devices following release, collection, solvent exchange to deionized water, and drop-casting on a silicon substrate. The harvesting yield, defined as the ratio of the number of devices in suspension after collection to the number of devices on the chip before release (2.8 × 106 devices in this case), was estimated to be 89.3% ± 6.96% (mean ± standard deviation, N = 3 experiments). The number of devices in the final solution was experimentally measured by drop-casting a small volume of the suspension, allowing it to dry, and then counting the number of devices under an optical microscope to determine the device concentration in the suspension. To assess the interaction between freestanding devices and biological cells, the devices were injected into primary embryonic rat hippocampal neuron cultures after 15 days in vitro (DIV) and a few hours before fixation. Figure 3c shows an SEM image of a neuron decorated with 10 µm in diameter freestanding devices.

Fig. 3

Experimental results: fabrication, release, collection, and cell culture delivery of freestanding micro-nanodevices. SEM and optical microscopy images of the micro-nanodevices at the different stages of process flows I to III depicted in Fig. 2. a–c present the microdevices resulting from the high-throughput process flow I. The fluorescent signal in the widefield fluorescent microscope image in (b) comes from the autofluorescence of the material composing the device. d–f show the experimental outcome of the cell-friendly process flow II for polymeric materials. g–i present the nanoengineered metallic device resulting from process flow III

Figure 3d, e show optical and SEM images of 6 × 6 µm2 polymer (poly(disperse red 1 methacrylate)—pDR1M) sheets manufactured by soft lithography before and after the harvesting procedure, respectively. Note that this method coupled with dextran sacrificial layer can be extended to all organic solvent-processed polymers. For the specific design tested (1.2·106 devices per cm2), the typical resulting concentration of the device suspension is 8000 devices/µL when using 40 µL of collection solution. The collection solution volume and device concentration can be adjusted based on the specific requirements of the targeted application. Similarly to process flow I, the freestanding polymer microsheets were injected into primary neuronal cell cultures at DIV 5, and fixed and imaged with SEM after two additional days of culture to demonstrate their compatibility with biological cells (Fig. 3f). Compared to metallic-based freestanding devices, the polymer sheets gently and accurately conform to the neuron’s shape. It should be noted that our group systematically tested Process Flow II for biocompatibility using an extended set of in vitro assays—including live/dead staining, (3-[4,5-dimethylthiazol-2-yl]-2,5 diphenyl-tetrazolium bromide) (MTT) assay, α-tubulin immunostaining, and patch-clamp electrophysiological recordings—on neurons cultured with the devices for up to 27 days. These tests revealed no adverse effects compared to control conditions, confirming the compatibility of both the fabrication process and materials used33.

Using FIB lithography, freestanding metallic micro-nanodevices with dimensions ranging from 200 nm to 3 µm were successfully manufactured. Note that this method can be extended to most compositions of metal and ceramic materials. Figure 3g depicts a representative SEM image of 3 µm hexagonal devices following FIB patterning with the letters MIT patterned on half of them. In Fig. 3h, two variations of the released devices can be seen: devices showing a plain design on the left and others decorated with a 200 nm diameter nanopore on the right. These images, together with Fig. 3h, illustrate the design versatility of the FIB process. More details regarding the collection of freestanding nanodevices using the custom micro vacuum aspiration system can be found in Section S1 of Supplementary Information. Figure S1 suggests that most of the released devices can be effectively collected by the glass micropipette in a volume as small as 10 µL. For a design featuring 10,000 devices, this corresponds to a concentration of 1000 devices/µL in the resulting suspension. Free-standing devices decorated with the MIT logo were dispensed into neuronal cell cultures at DIV 7 and left for 1 day before fixation (Fig. 3i). Considering the limited number of devices, 2 × 4 mm2 culture inserts were used to reduce the culture size and enhance the likelihood of device-cell interactions. It is interesting to note how the neural processes encompass freestanding devices, indicating a significant affinity with the nanodevice.

Experimental characterization of freestanding microdevices with bilayer lipid membranes

Once released from their substrates, freestanding micro-nanodevices become challenging to characterize as no direct electrical or mechanical contact can be established. In this section, we introduce BLMs as a platform to characterize freestanding micro-nanodevices designed for wireless neural interfaces, including stimulation and recording of electrophysiological activity. Compared to biological cells, BLMs provide a simpler and more controlled model to study the device interaction and coupling with phospholipid membranes34. Furthermore, the direct electrical and fluidic access to both sides of the BLM provides an easy means of controlling and recording the transmembrane potential, making it a notable advantage compared to biological cells. To illustrate this approach, we studied the photo-modulation of a Mueller-Rudin BLM35, measuring the transmembrane potential generated by freestanding poly[2,6-(4,4-bis-(2-ethylhexyl)-4H-cyclopenta [2,1-b;3,4-b′]dithiophene)-alt-4,7(2,1,3-benzothiadiazole)]:[6,6]-phenyl-C61-butyric acid methyl ester (PCPDTBT:PCBM) microsheets fabricated with the process flow II and exposed to near-infrared light. We then estimated how these experimental findings might translate to biological cells.

The characterization setup is illustrated in Fig. 4a. Following the BLM formation (see Supplementary Information Section S5 for more details about the BLM formation and validation), 6 × 6 µm2 PCPDTBT:PCBM sheets of about 200 nm in thickness were microinjected onto the freshly established membrane. It is important to be aware that the devices might overlap at this stage and affect the measurements. Hence, optimizing the surface chemistry of the freestanding micro-nanodevices devices to reduce aggregation is essential. After the sheets settled down (Fig. 4b), pulses of near-infrared (NIR) light were used to illuminate the sample through the 20× objective of an inverted microscope. The BLM transmembrane potential VBLM was synchronously measured during this process. Figure 4c shows the temporal evolution of VBLM in response to light pulses of various durations (20, 50, 100, and 200 ms) and intensities when decorated with PCPDTBT:PCBM microsheets. Note that control experiments performed without any devices showed no modulation of VBLM and no photoelectric effect, as the recording electrode was hidden from the light. As expected, longer pulse durations and higher light intensities resulted in larger voltage peaks (Fig. 4d). The voltage peaks plateaued at just above 700 µV for 200 ms long light pulses, with intensities of 4.98 mW/mm2 and higher. As demonstrated in Fig. 4e, greater coverage of the BLM by the freestanding microdevices led to a larger modulation of the BLM transmembrane potential. Finally, experimental data presented in Fig. 4f confirmed that NIR light pulses resulted in larger photo-modulation of VBLM compared to green light, a likely direct consequence of PCPDTBT:PCBM absorption spectrum and wavelength-dependent external quantum efficiency36.

Fig. 4

Characterization of the photo-modulation properties of freestanding PCPDTBT:PCBM microdevices using BLMs. a Schematic representation of the experimental setup used to evaluate the photo-modulation of the bilayer lipid membrane transmembrane potential VBLM induced by PCPDTBT:PCBM microdevices. b Optical microscopy image of freestanding devices sitting on top of a BLM. c VBLM modulation produced by the microdevices shown in (b) in response to NIR pulsed illumination of various durations and intensities (wavelength = 750 nm). d Plot illustrating the influence of illumination intensity and pulse duration on the photovoltage peak amplitude (ΔVBLM). e Graph showing the impact of the BLM relative device coverage on ΔVBLM for various light intensities (NIR, pulse duration = 200 ms). f Curve displaying ΔVBLM for different wavelengths and light intensities (pulse duration = 200 ms, device coverage = 3.69%). For all the graphs, mean ± standard deviation of N = 3 data points is plotted

Theoretical modeling to translate experimental results from BLMs to single cells

To facilitate the interpretation of the experimental data obtained using the BLM characterization platform and to explore how the modulation of VBLM may translate into the modulation of a single cell’s transmembrane potential, we established an electrical equivalent circuit representing the devices-BLM interface (Fig. 5a). For simplification purposes, each device is considered a current source with a seal resistance in parallel. Upon illumination, each device delivers a current ID,n. The portion of ID,n (IBLM,n) going through the BLM capacitance (CBLM) and contributing to VBLM is defined by the device coupling with the BLM, represented by its seal resistance RSeal,n. The overall current going through CBLM can easily be expressed in terms of the experimentally measured potential VBLM using the following equation:

$${I}_{\mathrm{BLM}}={C}_{\mathrm{BLM}}\cdot \frac{d{V}_{\mathrm{BLM}}}{{dt}}$$

(1)

Fig. 5

Theoretical modeling of the device-BLM and device-cell interface. a Electrical equivalent circuit of n devices in close proximity of a BLM. ID,n represents the current generated by the device n, and RSeal,n depicts the current leaks between the device n and the BLM. IBLM,n symbolizes the current generated by the device n that goes into the BLM. CBLM, IBLM, and VBLM represent the BLM capacitance, the overall current going through the BLM, and the voltage across it, respectively. b Electrical equivalent circuit of configuration where q devices sit on a cell. The cell membrane capacitance, current, and voltage are denoted Cm, Im, and Vm, respectively. c Plot showing the predicted photo-modulation of a cell transmembrane potential induced by a single 6 × 6 µm2 PCPDTBT:PCBM device with current characteristics derived from BLM experiments (device efficiency η = ηdevice, red), or improved by a factor ten (η = 10 ηdevice, magenta), twenty (η = 20 ηdevice, blue), or thirty (η = 30 ηdevice, cyan). The black curve represents the experimentally measured BLM transmembrane potential VBLM modulated by several devices and used as input to the theoretical model

Assuming that each device delivers a similar current ID,n and forms a comparable coupling with the BLM RSeal,n, the density of current going to CBLM per device can be expressed as:

$${J}_{\mathrm{BLM}/\mathrm{device}}=\frac{{I}_{\mathrm{BLM}}}{N}$$

(2)

Where N is the number of devices on the BLM.

Similarly, the density of current going through CBLM per unit area can be defined as:

$${J}_{\mathrm{BLM}/\mathrm{unit}\mathrm{area}}=\frac{{I}_{\mathrm{BLM}}}{N\cdot A}=\frac{{I}_{\mathrm{BLM}}}{{A}_{\mathrm{Devicecoverage}}}$$

(3)

Where A is the area of a single device, and ADevice coverage is the overall BLM area covered with devices. N or ADevice coverage can easily be estimated experimentally by optical microscopy.

JBLM can now be used to estimate the effect of a device (JBLM/device) or coverage (JBLM/unit area) on the transmembrane potential Vm of a single cell of capacitance Cm (Fig. 5b).

$${V}_{{\rm{m}}}=\frac{1}{{C}_{{\rm{m}}}}\int {I}_{{\rm{m}}}\cdot {dt}$$

(4)

And

$${I}_{{\rm{m}}}={J}_{\mathrm{BLM}/\mathrm{unit}\mathrm{area}}\cdot {A}_{\mathrm{Device}\mathrm{coverage}}={,J}_{\mathrm{BLM}/\mathrm{device},}\cdot q,$$

(5)

Where q is the number of devices on the single cell. Note that Cm accounts for the total membrane capacitance, not just the junctional patch, to provide a conservative estimate.

To illustrate the method, we simulated the effect of a 6 × 6 µm2 PCPDTBT:PCBM device on a single spherical neuron of 10 µm radius upon NIR light illumination with a 200 ms long and 4.98 mW/mm2 intense pulse, starting from the BLM experimental characterization results (Fig. 5c). Simulation results suggest that devices with higher efficiency should be employed to depolarize the cell transmembrane potential above its threshold and enable neuronal stimulation at the single-device level. The threshold was measured using the patch-clamp method to be 8.1 mV ± 3.5 mV (mean ± standard deviation, N = 3) for primary embryonic rat hippocampal neurons in vitro (see Supplementary Information Section S7 for details). We estimated that the current generated by a single device should be 27-fold larger to achieve stimulation (Fig. 5c). Overall, the proposed characterization method combined with theoretical modeling enables the optimization of device performance with ease and provides valuable insights about the required device characteristics to achieve stimulation under representative conditions. The numerical values for the parameters used in Eqs. 1–5, which led to the simulated transmembrane potentials Vm shown in Fig. 5c, are summarized in Table S1 of the Supplementary Information.

Discussion

The protocols presented in this study serve as non-exhaustive examples of effective strategies for producing and collecting freestanding devices with distinct characteristics. These protocols can be tuned, combined, and interchanged based on individual requirements and fabrication compatibility. However, certain considerations should be taken into account. For instance, the sacrificial layer material should be compatible with the targeted patterning strategy and desired device characteristics. Metallic sacrificial layers are resistant to water and organic solvents and thus compatible with high-throughput photolithography processes. Nonetheless, they often show noticeable roughness due to the minimum thickness required to facilitate their etching (100 nm in this work). This roughness can negatively affect the device’s geometry and characteristics, especially when working with thin layers (<50 nm, see Supplementary Information Fig. S2A, B for more details). Furthermore, in the case of titanium-based devices (process flow III), it has been observed that aluminum can react with titanium and form an alloy that cannot be etched with Al etchants, subsequently affecting both the releasing process and the properties of the device (Supplementary Information Fig. S2C, D). Additionally, using hazardous etchant during the release process implies that a thorough rinsing step is necessary to remove any trace that can be problematic for biological applications. Finally, the centrifugal filter pore size limits the minimum dimensions of the devices compatible with this approach.

Conversely, polymer sacrificial layers are smooth and do not form alloys with metallic or semiconducting device layers. In particular, water-soluble and biocompatible polymers (e.g., dextran) do not necessitate hazardous solutions for etching or solvent exchange processes, thus ensuring the biocompatibility and practicality of the release step. Moreover, the water-insoluble calcium alginate can be employed as a sacrificial layer for (1) manufacturing polymeric structures in aqueous solutions following process flow II and (2) to achieve in situ release of devices even after cells have been cultured on the substrate when ethylenediaminetetraacetic acid (EDTA) is added to the physiological solution (Supplementary Information, Section S3).

Process flow II also offers an alternative patterning strategy when the materials composing either the sacrificial layer or the devices are incompatible with conventional photolithography or electron-beam lithography processes as they can be damaged by the harsh organic solvents, strong bases, or water during resist spin coating, development, and rinsing steps. However, this manufacturing method results in limited freedom in design and choice of device materials.

Other polymers, such as parylene-C, have exceptional resistance to chemicals and can serve as a reliable alternative for smooth and photo/electron-beam lithography-compatible sacrificial layers. Yet, the release process based on dry etching necessitates local harvesting of the devices, which limits the throughput of the process. In process flow III, we used FIB instead of electron-beam lithography for simplicity. When working with gold double-plus ions (Au++)-based FIB lithography—known to exhibit higher sputtering yields and, therefore, faster patterning times—the implantation of gold ions in the sacrificial layer is an important aspect that should be carefully considered as it can result in the formation of a residual porous gold network surrounding and interconnecting each device together (Supplementary Information, Section S4). To ensure clean device sidewalls and efficient release, the dose should be increased so that the implantation happens in the substrate and not in the sacrificial layer.

Given the diversity in materials and fabrication strategies discussed above, the long-term mechanical, chemical, and electrical stability of the resulting devices should be assessed on a case-by-case basis, as it strongly depends on the specific device design, material composition, and intended application.

Freestanding devices come with an additional challenge in terms of characterization and validation prior in vitro studies with neurons. BLMs form a convenient intermediate platform to study and optimize the functionality of micro-nanodevices in a freestanding configuration. BLMs are routinely used in a wide variety of applications ranging from the study of transmembrane protein electrophysiological characteristics to the characterization of nanoparticle translocation by impedance monitoring37,38. Furthermore, the proposed electrical-equivalent circuit of the BLM-devices interface enables a simple analytical estimation of the freestanding device modulation capabilities on biological cells from experimental characterization results with BLMs. Interestingly, the proposed characterization method intrinsically considers the current leaks at the membrane-device interface—the seal resistance—providing a more representative estimation of the studied device modulation capabilities compared to conventional electrical characterization techniques. While the example showcased in this manuscript focuses on using the BLM platform to characterize devices designed for wireless stimulation, the platform is equally suitable for characterizing devices intended to wirelessly probe transmembrane potential fluctuations. For instance, a device capable of converting voltage fluctuations into optical signals39 could be characterized by applying an external voltage to modulate the BLM transmembrane potential while simultaneously monitoring the optical signal emitted by the device. This approach enables a systematic and controlled evaluation of the device’s recording capabilities before transitioning to experiments with biological cells.

Nonetheless, it is important to comprehend the limitations of this approach. First, BLMs exhibit capacitances similar to but not identical to biological cell membranes. In this study, the BLM capacitance was measured to be 0.53 µF/cm2 (Supplementary Information, Section S5), nearly two times lower than that of a neuronal cell membrane (1 µF/cm2)4. Consequently, measuring CBLM is essential to account for this discrepancy. Furthermore, the composition of the cellular membrane is highly complex and varies depending on the cell type. Within a single cell, the inner leaflet shows different phospholipid types and relative concentrations compared to the outer leaflet. More importantly, the cell membrane is adorned with countless types of membrane proteins40. On the other hand, BLMs are usually symmetric and mostly composed of phospholipids, often with residual solvents in the interstitial space between their inner and outer leaflets, which can affect their physical and structural properties41. The BLM composition can certainly be adjusted to more accurately represent a specific cell type42, but a difference in coupling between the BLM-device interface and the cell-device interface should still be expected. Finally, the equivalent electrical circuit of the cell-device interface presented in Supplementary Information Fig. S6B represents the oversimplified case of a spherical cell whose membrane is only modeled by a single capacitance. While the presented model is well-suited for demonstration purposes, more sophisticated developments considering the complex geometry of a cell, as well as its junctional and non-junctional capacitances43, will undoubtedly lead to more accurate estimations.

Building on the technical developments presented in this work, the next progression involves incorporating specific functionalities into freestanding platforms to address long-lasting challenges in the field of neuroscience and beyond. To reach this goal, one approach would consist of scaling down transistor-based integrated circuit (IC) chips to a subcellular scale by limiting the circuit complexity to a minimum, therefore leveraging decades of research in nanoelectronics. Currently, a typical transistor gate pitch is ~50 nm in commercial chips44. It is conceivable to integrate several of them within a 10 µm platform footprint to perform basic processes. However, the amount of energy that can be wirelessly transferred to such small devices will limit the number of transistors in the circuit and, thus, the application spectrum. Instead, simpler designs based on nanotransducers should be employed, leveraging the remarkable characteristics of the nanoworld to perform transduction efficiently. For example, creating a device that converts the electrophysiological activity of a single cell into light for wireless electrophysiology following a conventional IC approach would require a considerable amount of complexity. It would entail incorporating an electrode, a processing unit, a light-emitting module, an energy harvesting module, and more, likely exceeding the 10 µm footprint. In contrast, sub-micrometer organic electro-scattering nanoantennas39 or quantum dots can efficiently transduce voltage variations into light intensity fluctuations with a much simpler and confined design45. Similarly, organic semiconductors[46](https://www.nature.com/articles/s41378-025-01117-9#ref-CR46 “Medagoda, D. I. & Ghezzi, D.