All-printed nanomembrane wireless bioelectronics using a biocompatible solderable graphene for multimodal human-machine interfaces

The study addresses the challenge of manufacturing flexible, wearable hybrid electronics without relying on conventional multi-step cleanroom microfabrication. Prior state-of-the-art systems, though mechanically compliant and capable of integrating commercial components, require high-vacuum equipment, photolithography, and skilled maintenance. The authors aim to create an all-printed, multilayered nanomembrane hybrid electronic platform (p-NHE) using additive nanomanufacturing that reduces material waste, cost, and complexity while enabling scalable, high-precision multilayer alignment. The research targets key limitations in existing printed wearables: dependence on rigid printed circuit boards for active components, potential cytotoxicity of Ag/Cu nanoparticle inks due to ion release, and the relatively low conductivity of CNT conductors. The central hypothesis is that a functionalized conductive graphene (FCG) can simultaneously provide biocompatible, gel-free skin interfaces for high-fidelity EMG sensing and act as an oxidation barrier to enable solderable, reliable Ag-based printed interconnects, thereby realizing fully printed, wireless wearable bioelectronics. The importance lies in enabling comfortable, skin-conformal systems for health monitoring and human-machine interfaces with reduced manufacturing barriers and improved performance.

The paper positions its contribution against prior approaches to flexible/wearable electronics that relied on microfabrication for sensors and circuits or hybrid strategies combining printed passive sensors with rigid PCBs. Recent advances have included multilayer flexible systems using CMOS processes, laser ablation, casting, and microfabrication to achieve stretchability and integration of commercial components. Printed systems have often been limited to passive electrodes, with active modules remaining rigid. Material advances in Ag, Cu, and CNT inks and annealing have improved conductivity and skin-electrode impedance, but Ag/Cu nanoparticles can leach ions causing cytotoxicity and oxidation, while CNT-based conductors can be insufficiently conductive for robust circuit operation. Graphene approaches such as graphene oxide or epitaxial growth have been explored but present challenges for simple, high-resolution patterning. The authors’ selective edge-oxidized FCG ink supports aqueous dispersion without surfactants and is compatible with aerosol jet printing, aiming to overcome these gaps by enabling multilayer, solderable, biocompatible all-printed systems.

Materials and ink preparation: Functionalized conductive graphene (FCG) ink was synthesized by electrochemical exfoliation of graphite in ammonium sulfate electrolyte under 10 V DC, followed by DI water purification, vacuum filtration, and dispersion to 15% in water. Ag nanoparticle ink (Ag40XL) was diluted with m-xylene to 20% Ag concentration. Polyimide (PI) ink was prepared from PI-2545 precursor with NMP at 4:1.

Printing platform and process: All structures were fabricated via aerosol jet printing (AJP) using ultrasonic and pneumatic atomizers to accommodate viscosities spanning ~1–1000 cP. A sacrificial PMMA layer on glass served as the release substrate.

Printed FCG electrode fabrication: PMMA was spin-coated (1000 rpm, 30 s; bake 200 °C, 2 min). A PI base layer was AJP printed (pneumatic, 300 µm nozzle) and cured at 250 °C for 1 h. FCG ink was deposited (ultrasonic, 200–300 µm nozzle) and dried at 100 °C for 1 h. The elastomeric substrate (Ecoflex 00-30 and Ecoflex Gel, mixed 1:1 by systems and combined 10 g) was cast and cured at room temperature overnight. The printed electrode was released by dissolving PMMA in acetone, picked up with water-soluble tape, and transferred to ~500 µm Ecoflex.

Printed multilayer circuit fabrication: On PMMA/glass, PI was spin-coated; the first Ag layer was AJP printed (ultrasonic, 200 µm nozzle) and photonic-sintered (IPL: 2 kV, 2 ms, 5 pulses). A printed PI dielectric (pneumatic, 300 µm nozzle) defined VIAs (50 µm circular openings). A second Ag layer was printed and similarly sintered. A thin FCG layer (~0.1 µm) was printed atop Ag to provide an oxidation barrier and solderable surface (dry at 100 °C, 1 h). A final PI encapsulation was printed and cured (250 °C, 1 h). The stack was released by dissolving PMMA in acetone and transferred to ~500 µm Ecoflex. Layer thicknesses included (typical): electrode PI 10.5 µm + FCG 0.8 µm; circuit: Ag (1st) 0.5 µm, middle PI 2.0 µm, Ag (2nd) 2.0 µm, FCG 0.1 µm, final PI 1.3 µm.

Integration and packaging: Solder paste (Sn/Bi/Ag 42/57.6/0.4 wt%) was stencil-printed. Chip components (BLE microcontroller, amplifier/ADC, voltage regulator, passives, antenna) were placed and reflowed per manufacturer’s profile. Firmware for BLE was flashed. A 40 mAh Li-polymer battery was connected via soldered neodymium magnet terminals. The assembly was encapsulated with low-modulus silicone elastomer (E ≈ 8.5 kPa).

Characterization: Morphology and structure were assessed by AFM (flake lateral size 1–5 µm; mean thickness ~3.1 nm), TEM/HRTEM (bilayer edges ~1.5 nm), SEM/FIB-SEM (cross-sections of stacks), profilometry (layer thickness). XRD and XPS analyzed Ag oxidation with/without FCG. Electrical/mechanical testing included resistance change under 180° bending (up to 100 cycles), and FEA (ABAQUS) using EPI = 2.5 GPa (ν=0.34), EAg = 40 GPa (ν=0.37) to validate strain <1% in interconnects. Bluetooth RSSI versus distance was compared to a rigid PCB system. Fabrication yield was tracked over printing, soldering, and integration steps.

Biocompatibility: Human primary keratinocyte cultures (5000 cells/cm²) were seeded on substrates (control polystyrene, elastomer, graphene, Au, Ag) and cultured 7 days (37 °C, 5% CO₂). Viability was quantified via absorbance and fluorescence (calcein blue AM staining) using a microplate reader.

EMG mapping and in vivo studies: With IRB approval (H17212), adult volunteers performed seven gestures while the forearm was tiled with 1"×1" hydrogel electrodes. Signals were recorded (BioRadio) in a three-electrode setup, bandpass-filtered 30–150 Hz with 60 Hz notch, and RMS-mapped to identify muscles with strongest activations for gesture-specific placement (palmaris longus, brachioradialis, flexor carpi ulnaris).

HMI and machine learning: Single-device classification used BLE streaming to an Android app. A data buffer processed 0.512 s windows (128 points), updated every 10 packets (0.240 s). A safety buffer of five consecutive identical outputs yielded a command, introducing ~1.2 s delay; commands halted on mismatches. Preprocessing used a 3rd-order high-pass Butterworth at 1 Hz and linear scaling. A 2-layer CNN (trained on 25 min, tested on 8.3 min; 5 subjects; 4 gestures) achieved mean accuracy 99.0 ± 1.7% offline; real-time validation on a 5th subject achieved 99.2 ± 1.0% over 10 trials. Arm rotation (z-axis accelerometer) provided auxiliary controls for drone/RC car altitude or motion.

Three-device classification for seven gestures used three synchronized EMG channels (three separate devices). Sessions were synchronized per session and retrained with a short protocol (5 s per command, two repeats; total ~140 s). KNN was used due to the small, well-separated dataset. Real-time session results and confusion matrices were generated. A 3D-printed robotic hand was controlled in real time using classified outputs.

• All-printed multilayer nanomembrane hybrid electronics (p-NHE) achieved fully printed electrodes and wireless circuit stacks with precise VIA formation and alignment using aerosol jet printing.
• Device form factor: exceptionally light (<5 g) and thin (<2 mm), enabling conformal skin lamination via elastomer adhesion alone; powered by a 40 mAh Li-polymer battery.
• Layer architecture: electrodes (PI 10.5 µm / FCG 0.8 µm); circuits (Ag 0.5 µm / PI 2.0 µm / Ag 2.0 µm / FCG 0.1 µm / PI 1.3 µm) on elastomer.
• FCG properties and structure: FCG flakes 1–5 µm lateral size; mean thickness ~3.1 nm (AFM); stacked multilayer morphology in prints (SEM/TEM). High-aspect ratio stacks improved skin contact.
• Solderability and anti-oxidation: FCG capping prevented Ag pad dissolution during reflow soldering; XRD showed Ag2O (32.7°) only in samples without FCG, while FCG/Ag showed pure Ag peaks at 38.0°, 44.1°, 54.8°. XPS: Ag(0) peaks at 368.6 and 374.6 eV with FCG; without FCG, Ag+ peaks at 368.1 and 374.1 eV. Ag+/Ag(0) ratio reduced from 0.34 (Ag without FCG) to 0.16 (with FCG), indicating suppressed oxidation.
• Mechanical reliability: negligible electrical resistance change over 100 cycles of 180° bending; FEA predicted interconnect strains <1%. Electrodes tolerated bending to 180° (radius ~1.5 mm) and stretching up to ~60%, with cyclic resistance stability (R² ≈ 0.98–0.99 over 100 cycles).
• Wireless performance: printed circuit BLE RSSI comparable to rigid PCB; consistent transmission up to ~15 m, with communication distance up to ~20 m indicated in Fig. 3h.
• Biocompatibility: Keratinocyte viability on graphene and elastomer comparable to control; printed Ag showed cytotoxicity. Data support FCG for skin-contact electrodes and as Ag oxidation barrier.
• EMG signal quality: FCG electrodes produced higher SNR than printed Au/Ag and were comparable to commercial gel electrodes (slightly lower but within error). Open-mesh FCG conformed well under skin stretch/compression.
• HMI performance: • Single-device (palmaris longus): 2-layer CNN achieved mean offline accuracy 99.0 ± 1.7% (4 classes); real-time 99.2 ± 1.0% over 10 trials; enabled control of a drone, RC car, and presentation software; auxiliary accelerometer gestures expanded command set. • Three-device (palmaris longus, brachioradialis, flexor carpi ulnaris): 7-class classification achieved overall accuracy 98.6% across ten trials, enabling real-time control of a robotic hand. 3D RMS plots showed clear separability of gestures.

The results demonstrate that additive nanomanufacturing alone can produce a complete, wireless, multilayer flexible bioelectronic system integrating both sensors and active circuitry. Using FCG as a functional material addresses multiple constraints simultaneously: it forms biocompatible, conformal, gel-free EMG electrodes with high-fidelity recordings, and it passivates printed Ag interconnects against oxidation, thereby enabling reliable solder attachment of surface-mount components on printed circuits. Mechanical testing verifies robustness under bending and stretching representative of skin deformation, while wireless performance matches rigid benchmarks at practical distances. By combining EMG sensing with machine learning, the platform supports precise, real-time multi-class HMIs, reducing the number of required electrodes by anatomically optimizing placement and employing synchronized multi-device acquisition. These findings collectively validate the proposed approach to overcoming limitations of traditional cleanroom-dependent fabrication and earlier printed systems that relied on rigid modules, advancing the field toward scalable, cost-effective, and comfortable bio-integrated electronics.

This work presents the first fully all-printed nanomembrane hybrid electronics that integrate biocompatible FCG electrodes and solderable, oxidation-resistant Ag interconnects into a wireless, skin-conformal platform. The additive process yields thin, lightweight devices with robust mechanical reliability and long-range BLE communication. High-quality EMG acquisition paired with machine learning enables real-time, multi-class human-machine interfaces, including accurate seven-gesture control with only three strategically placed devices. Future research will extend these technologies to clinical applications such as biofeedback-enabled prosthetics and enhanced rehabilitation training, and further optimize automated integration to improve overall fabrication yield and scalability.

• Final system integration (sensor-circuit assembly) reduced overall fabrication yield compared with >90% yields for individual printing and soldering steps; automated assembly is anticipated to improve this.
• The single-device HMI utilized a safety buffer that introduced an approximate 1.2 s command latency, affecting responsiveness though improving safety.
• Three-device classification required per-session synchronization and retraining with a brief calibration protocol, indicating session/subject dependence.
• Although FCG electrode SNR was comparable to gel electrodes, it was slightly lower (within error), which may matter in low-amplitude signals.
• For seven-class control, three devices (channels) were required, which may limit minimalistic configurations for complex gesture sets.