Dry-printed electrodes for transcutaneous electrical stimulation of innervated muscles: Towards wearable and closed-loop stimulation

Abstract

Background

While transcutaneous electrical stimulation is an established clinical technique, applied in clinical settings, the need for precise electrode placement, limited long-term stability, and user inconvenience hinders widespread adoption in wearable and home-based applications. To address these issues, soft screen-printed electrode arrays, previously validated for biopotential recording and bioimpedance applications, were evaluated in two new contexts: (1) Systematic comparison against gel electrodes for stimulation in the neck region, and (2) Real-time closed-loop facial stimulation in healthy subjects.

Methods

Contact impedance was measured and modeled for both printed (dry) and conventional gel electrodes. Accessory nerve stimulation efficacy was evaluated in 12 healthy subjects using surface electromyography and mechanical movement recorded with an inertial measurement unit.

Results

Despite higher contact impedance, dry electrodes achieved comparable activation thresholds and perceived discomfort to gel electrodes. Finite element modeling confirmed a similar electric field distribution for both electrodes. In addition to validating their effectiveness for transcutaneous stimulation of innervated muscles, we demonstrated the feasibility of closed-loop facial recording and stimulation using dry electrodes in one subject as proof-of-concept.

Conclusions

Our results pave the way for wearable neuromodulation systems for clinical and home use, particularly in restoring facial symmetry for individuals with neuromuscular impairments. Broader validation with larger sample sizes and testing in patients with denervated muscles will be necessary for clinical translation.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12984-025-01763-0.

Introduction

Skin-adhered, non-invasive electrical stimulation techniques such as transcutaneous electrical stimulation (TES), transcranial direct current stimulation (tDCS), and transcranial alternating current stimulation (tACS) are extensively studied for therapeutic purposes [1]. tDCS delivers a constant low-intensity direct current to modulate cortical activity, while tACS uses alternating current to influence brain wave oscillations. TES, which includes transcutaneous electrical nerve stimulation (TENS) and electrical muscle stimulation (EMS), targets peripheral nerves and muscles for pain relief and rehabilitation. When EMS is specifically applied to elicit functional movements by directly stimulating the muscle or its motor nerve, it is referred to as functional electrical stimulation (FES) [2]. Non-invasive electrical stimulation techniques are well-established in clinical practice, with applications ranging from chronic pain management to motor recovery after stroke or paralysis [3, 4]. Additionally, neuromodulation has also been explored for improving sleep quality, which could benefit individuals with sleep disorders or those seeking to enhance cognitive performance [2, 5–7].

Despite increasing interest in transcutaneous and transcranial stimulation, several significant challenges remain. These include unpleasant sensations such as tingling and pain, the need for precise electrode placement, difficulties maintaining electrode stability, and optimizing stimulation waveforms [8]. While motor nerves (A  efferent nerve fibers) are the intended target for functional stimulation, the tingling and pain sensations primarily arise from activation of cutaneous sensory nerves (A  and C afferent nerve fibers) in the skin and subcutaneous tissue. The targeted motor nerve fibers cannot convey pain [9]. Stimulation current limits for tingling and pain sensations using gel electrodes have been well documented and studied extensively over the past decades, yet they vary between individuals and depend on multiple factors. Different body regions exhibit varying sensitivities primarily due to differences in nerve fiber types and density. Sensory nerves typically have a lower rheobase than motor nerves, meaning they have a lower activation threshold. This threshold varies across different body regions, depending on the distribution and diameter of sensory fibers. Additionally, areas with thicker skin or more adipose tissue often require higher currents to achieve similar effects [4]. Despite nerve structure, function, and activation threshold, waveform characteristics also play a key role. Pulsed or random noise currents are often perceived as more comfortable than continuous waveforms [10]. Higher frequencies tend to reduce pain perception due to the accommodation effect, i.e., the neuronal membrane’s ability to adapt to slowly rising or prolonged stimuli, thereby failing to fire an action potential [4]. Electrode geometry further affects both stimulation efficiency and user comfort [11]. Generally, the current threshold for tingling sensation, or paresthesia, starts with 0.4 mA for direct current (DC) or low-frequency alternating current (AC), while pain thresholds begin above 3 mA [12, 13]. However, these values vary substantially based on the factors mentioned above, including individual sensitivity, pain tolerance, skin condition, and even psychological factors [9]. To ensure comfort and safety in therapeutic applications, stimulation is typically initiated at low current levels and gradually increased. This subject-tailored approach allows for parameter optimization based on individual sensitivity and response [9, 12–14].

A second major challenge in non-invasive stimulation lies in electrode positioning, stability, and usability. Conventional gel electrodes, the gold standard in skin electrophysiology and neuromodulation, are cumbersome, inconvenient to the user and operator (requiring individual placement procedures), and often suffer from poor stability due to gel dehydration. Thus, maintaining consistent skin contact is particularly difficult, compromising stimulation efficacy. These issues are especially pronounced in the facial region, where constant skin deformation and movement often lead to electrode detachment. The need for a strong physical coupling between the electrode and skin can further interfere with voluntary movement and is particularly disruptive during functional electrical stimulation (FES) training. Additionally, achieving precise electrode placement, essential for effective stimulation, is a complex and time-consuming process requiring trained personnel. While hydrogel improvements have been reported [15], spatial resolution remains limited due to transverse current flow between electrodes. Furthermore, conventional systems require wired connections to stimulation equipment, which restrict patient mobility and interfere with natural movement patterns. Altogether, these limitations highlight the clinical relevance of wireless operation with soft electrodes, as it may improve user comfort and support more natural movement during therapy [16–18].

Textile-embedded electrodes have been investigated as an alternative for long-term stimulation. However, they require continuous hydration to maintain effective stimulation and minimize discomfort [11, 19–22]. A more compelling technology is the use of printed electrodes, which can be fabricated on flexible and soft surfaces and easily placed on clean skin surfaces. In a dry form, such electrode arrays may reduce skin irritation and offer greater convenience for users and operators. In applications requiring prolonged stimulation sessions, soft and wireless electrodes can offer major benefits over existing approaches. In particular, such wearable systems can benefit applications necessitating closed-loop stimulation, while allowing movement (i.e., functional facial expressions). Dry electrodes have been studied in recent years as a promising alternative for non-invasive recording of biopotential and bioimpedance signals [23, 24]. However, their application in stimulation has been limited owing to their high impedance values. Despite this, their potential remains unexplored, especially regarding their ability to reach effective stimulation thresholds without causing discomfort. Notably, Marquez-Chin et al. [25] demonstrated the feasibility of dry polymer nanocomposite electrodes for transcutaneous electrical stimulation, achieving muscle activation comparable to conventional gel and rubber carbon electrodes.

In this study, we systematically evaluated previously reported printed dry electrodes [26] for transcutaneous stimulation of innervated muscles. Our focus was on examining the electrode impedance to establish a comprehensive model of the skin-electrode interface, including the equivalent circuit and finite element model. Following impedance characterization, we quantified stimulation thresholds for accessory nerve activation in the neck region using surface electromyography (sEMG) of the upper trapezius muscle and an inertial measurement unit (IMU) placed on the shoulder. Our goal was to assess whether dry electrodes, despite their relatively higher impedance, can effectively deliver stimulation in the neck region. We further demonstrate the feasibility of using these electrodes in a closed-loop neuromodulation setup aimed at restoring smile symmetry. We implemented a system with two printed dry electrode arrays positioned on the right (stimulation) and left (sEMG) sides of the face in one healthy subject. Real-time stimulation was triggered by the onset of muscle activation detected on the contralateral side. This preliminary demonstration highlights the feasibility of dry electrodes for both stimulation and real-time feedback.

Methods

Participants

12 healthy individuals (six females, six males; three left-handed, nine right-handed; age 29.42 ± 6.73 yr) participated in this study for the neck protocol, one female subject participated in closed-loop facial stimulation. All participants were screened using a questionnaire to ensure their eligibility for electrical stimulation experiments. Exclusion criteria included a history of neurological or psychiatric conditions, epilepsy (personal or family history), cardiac conditions, electronic implants, dermatological issues, pregnancy, hearing problems, recent use (within 24 hr) of medications, drugs, or alcohol, and a history of fainting. Written informed consent was obtained prior to participation. The study was approved by the Institutional Ethics Committee Review Board at Tel Aviv University (IRB, no 0,009,184–2). The authors have obtained written consent to publish the images and videos.

sEMG wearable system

The dry carbon electrode arrays, previously validated for biopotential and bioimpedance measurements [23, 24], were purchased from X-trodes Ltd. sEMG signals were recorded using a wireless data acquisition unit (DAU; X-trodes Ltd., Herzliya, Israel) connected to the multi-electrode array. The system handled dual-function data: storing recordings on a micro-SD card while simultaneously streaming data via Bluetooth to an Android tablet application. The DAU featured 16 unipolar channels with a sampling rate of 4000 S/s and 16-bit resolution. The specifications included an input range of  mV, input impedance of , and a noise floor of 2 µV root-mean-square (RMS) across a bandwidth of 0.32–700 Hz. The unit included an integrated inertial measurement unit (IMU) providing continuous 6-axis motion tracking with a sampling rate of 1000 S/s, comprising a three-axis angular velocity (range ± 2000 °/s) and three-axis linear accelerometer (range ± 16 g). To correct for a delay between the sEMG and IMU signals, a manual mechanical tap on the DAU casing was performed before each session. The tap generated a mechanical artifact detectable in both signal modalities, allowing offline precise alignment of the signals. The unit was powered by a 620 mAh battery, enabling continuous operation for up to 16 h.

Stimulation electrodes

The fabrication process of the dry carbon electrode arrays was described previously [24]. Briefly, silver traces (Creative Materials, 125–13 T) were screen-printed on polyurethane (PU) films (3 M 1524, 60 µm thickness). Carbon electrodes (Creative Materials, C 124–50T) were then screen-printed in alignment with the silver traces. Each printing step was carried out with a pre-patterned mesh stencil, followed by curing at 60 °C for 10 min. The arrays were passivated with a double-sided adhesive PU film (Delstar EU94DS), leaving the electrode sites exposed (Fig. 1A).

Fig. 1.

Dry electrode fabrication, surface morphology, and structural characterization. A Schematic illustration of the electrode fabrication process. B AFM 3D topography of the carbon electrode showing surface roughness. C FIB-SEM cross-sectional imaging showing the layered structure. D Stimulation setup to connect the electrodes with the stimulus generator via a customized PCB

Surface characterization

To assess the roughness of the exposed carbon layer, atomic force microscopy (AFM NX12 from Park Systems) was performed on three randomly selected areas of 30 x 30 µm. Three-dimensional topography was extracted, and RMS roughness (Rq) values were calculated using Gwyddion software [27]. In addition, cross-sectional imaging using focused ion beam scanning electron microscopy (FIB-SEM, Helios 5 UC DualBeam, ThermoFisher Scientific) was performed to estimate the thickness of the layers. Samples were manually cut with scissors, and cross-sectional images were acquired after gallium ion beam milling.

Neck stimulation experimental protocol

Each participant underwent a measurement of maximum voluntary contraction (MVC) and two electrical muscle stimulation sessions using two types of stimulation electrodes: dry electrodes (4-contact linear array, 0.8 cm diameter, 1.3 cm center-to-center spacing, using the two outermost contacts) and gel electrodes (Spes Medica, 1.52.0 cm disposable Ag/AgCl electrodes). The 8 mm electrode diameter represents an optimal balance between contact area and skin conformability for printed electrode arrays, as larger contacts suffer from poor skin attachment on curved surfaces, particularly for the neck region. Stimulation was delivered using an external stimulus generator (STG5, Multi Channel Systems, Reutlingen, Germany). The conductive traces of the stimulation electrodes were connected to a 1 mm pitch printed circuit board (PCB) card using a z-axis conductive adhesive. The 1 mm pitch PCB board was designed to fit a SAMTEC (MB1–120-01-F-S-01-SL-N) socket positioned on a second custom-made PCB board that can accommodate contact with an external stimulator and/or a DAU. In this arrangement, an operator can manually select the stimulation channels (see Fig. 1D). The stimulator used has a voltage compliance limit of 70 V. During the sessions, the device did not indicate any instance where the voltage required to deliver the programmed current in current mode exceeded this limit.

MVC measurement

For sEMG recordings, a dry electrode array was placed on the left upper trapezius muscle (LUT) following SENIAM recommendations, [28] the superior border of the array was positioned halfway between the C7 vertebra and the acromion. Following a 15 min stabilization period, participants performed the MVC protocol of manual muscle testing (MMT) in a horizontal position with bilateral resistance. The protocol consisted of three repetitions of 5 s baseline, 5 s shoulder elevation, 10 s maximum contraction, and 60 s recovery period.

Electrical accessory nerve stimulation

The accessory nerve, cranial nerve XI, stimulation site was first identified using two gel electrodes placed on the skin according to Melo et al. [29] Once the stimulation location was determined, the electrodes were removed, and the area was marked to ensure consistent placement between the stimulation sessions. The skin was then cleaned with an alcohol pad prior to electrode application. For comparison, both dry electrode arrays and two gel electrodes were positioned at the identical marked location (Fig. 2A). This procedure, including site identification, skin preparation, and electrode placement, was applied identically for all participants. Stimulation was delivered in current-controlled mode using charge-balanced biphasic pulses (cathode-first) with a pulse width of 1 ms (a train of 10 pulses at 1 Hz frequency). The stimulation current was initially set to a minimum and gradually increased in increments of 0.1 mA (0.1 µC/phase) until either the participant reported their maximum discomfort level or visible muscle activity was induced. The current amplitude used in this protocol ranged from 0.2 to 5.4 mA (0.2 to 5.4 µC/phase). The experiment was implemented in Python using PsychoPy (version 2024.2.1 [30]), which controlled both the data acquisition software and stimulator generator.

Fig. 2.

Experimental scheme, typical skin-electrode impedance, and accessory nerve stimulation responses using dry and gel electrodes. (A) Schematic of the experimental paradigm in posterior view: transcutaneous electrical stimulation of the accessory nerve to evoke trapezius muscle contractions, while recording the electromyography (EMG) (B) Illustration of the electrode types used for recording (dry electrodes) and stimulation (dry or gel electrodes) in lateral view. (C) Electrode placement configurations on the neck in lateral view. (D) Typical sEMG response after stimulation (4 mA, 4 C/phase), the shaded gray area represents the response window. The red asterisk indicates the stimulation artifact; the red arrow indicates the M-wave onset. (E) sEMG response curves showing normalized muscle activation (%MVC) as a function of stimulation current (mA) for both electrode types. Dashed and solid lines for fitted curves of dry and gel electrodes, respectively. Black arrows indicate activation thresholds (; see Methods). (F) Nyquist plots showing the impedance characteristics of dry (blue; dashed) and gel (green; solid) electrodes normalized to contact area (). Dots represent experimental data points, and lines represent corresponding model fits. (G) Typical angular velocity () of the shoulder over time, the gray area represents the window for the movement response. (H) Angular velocity response curves () as a function of stimulation current for both electrode types. Black arrows indicate the thresholds (; see Methods)

To assess pain, participants rated discomfort after each stimulation intensity using a visual analog scale (VAS, where: 0 = no pain, 1–3 = mild pain, 4–6 = moderate pain, 7–10 = severe pain). On random trials, the stimulation amplitude was slightly decreased to evaluate pain rating consistency. Contact impedance was measured with skin electrochemical impedance spectroscopy performed after each stimulation session using an impedance analyzer (MFIA, Zurich Instruments) under 2-terminal (2T) measurement at a frequency range of 1–1000 Hz. Each impedance measurement was performed twice, and data were averaged to improve the signal-to-noise ratio and minimize effects associated with environmental conditions (i.e., movement). The protocol was repeated with the two electrode types. This involved removing the first (dry or gel) electrodes, cleaning the skin with an alcohol pad, placing the other electrode type at the same location, and repeating an identical stimulation procedure. This systematic comparison enabled a direct evaluation of the performance and user comfort of the dry and the gel electrodes.

Closed-loop functional electrical stimulation of the face

To assess the feasibility of closed-loop facial neuromodulation, a single-subject experiment was conducted. The experiment utilized dry electrode arrays both for sEMG recording and stimulation (Fig. 4A). For recording, a 14-channel array (Fig. 4B), a mirrored configuration of the original right-side layout described by Funk et al. [16], was placed on the left side of the face. For stimulation, two channels from a 7-channel array were used (1 cm diameter, 4.9 cm center-to-center spacing, purchased from X-trodes Ltd, Fig. 4C), with the anode and cathode positioned over the motor point of the zygomaticus muscle (right side) as reported in Efthimiou et al. [31]. A charge-balanced biphasic pulse (cathodic-first) with a pulse width of 200 s (at 100 Hz frequency) with 7 mA (1.4 C/phase). The stimulation current was determined through a pre-session calibration where the current was increased until visible muscle contraction was observed. Facial sEMG signals were recorded from the left side using the real-time DAU system as described in Luxembourg et al. [32], sampled at 500 S/s. A moving RMS signal (window size of 40 ms) was computed continuously. The participant was instructed to maintain a neutral facial expression, followed by a unilateral smile (left side only) for 2 s. Smile onset events were detected via a threshold-crossing of 200 µV and a positive-slope detection method. Upon detection on channel 6, a trigger signal was sent to the right side via the STG stimulator to evoke a mirrored facial response. The stimulation stopped automatically once smile activity fell below the threshold. The closed-loop latency, defined as the delay between the detected EMG onset and the initiation of stimulation, was approximately 500 ms.

Fig. 4.

Experimental setup and typical closed-loop facial muscle recording and stimulation results. In this setup, the stimulation was triggered automatically by the onset of muscle activity of the left side of the face. A Schematic of the experimental setup showing simultaneous recording and stimulation using multi-channel facial arrays. B Facial electrode placement for stimulation, showing the 7-channel array (two channels used for stimulation; + for anode and - for cathode) and stimulator adapter positioned above the eyebrow region (adapter detailed in Fig. 1B). C Facial electrode placement for sEMG recording, showing the 16-channel dry electrode array connected to the wireless data acquisition unit (DAU). (D– F) Representative demonstration of facial neuromuscular activity during three conditions: natural rest (Neutral), left side voluntary facial activation (Voluntary asymmetric), and evoked response via electrical stimulation when unilateral activation is detected (Stimulation). (D) The corresponding facial expressions; (E) The sEMG signal from one channel; (F) The stimulation trigger

Data processing

EMG analysis

Surface EMG signals were first processed using a bandpass filter (10–500 Hz). Additionally, a comb filter was applied at 50 Hz and its harmonics to remove power line interference. For each stimulation trial, ten signal epochs were extracted and time-aligned to the stimulation onset. A 200–500 ms post-stimulus window was used to extract the response. The stimulation artifact, visible at stimulus onset (red asterisk in Fig. 2D), occurred before this analysis window, making additional artifact suppression unnecessary. The response magnitude was quantified using the median RMS values from the 10 epochs of this window. To standardize the responses between subjects, they were normalized to the MVC values for each channel. MVC values were computed by averaging the RMS values from three repetitions, with each repetition calculated over a 0.5 s window centered around the peak activation.

IMU analysis

Shoulder elicited movement responses were quantified using IMU data, with similar epochs extracted as in EMG analysis. The angular velocity signal was analyzed for each orthogonal axis (X, Y, Z), and the response magnitude was extracted using the RMS value over a 0.3 s window centered at the peak response of the X-axis. The overall movement response magnitude was then calculated as the Euclidean norm of these three RMS values.

Response curve analysis

The response curves were generated for both sEMG and IMU data as a function of current. The sEMG response curve was computed as the mean normalized response across all 16 channels for each current. Both curves were smoothed using a Savitzky–Golay filter and fitted with sigmoid functions. Activation thresholds for muscle contraction () and overt shoulder movement () were extracted from the fitted curves as the lowest current at which the response reached at least 10% of the total range from baseline, with positive derivative. If it failed, thresholds were determined directly from the smoothed response data by detecting changes in the first derivative and verifying a consistent increase above baseline variability. For pain perception analysis, VAS scores corresponding to determined thresholds were extracted for comparison between electrode types.

Skin-electrode interface and contact impedance

The contact-impedance of the skin-electrode interface was characterized by fitting measured impedance values with an equivalent circuit (Fig. 3A) [33, 34]. The equivalent circuit incorporates several key components: (i) the resistance of the tissue beneath the electrodes ; (ii) constant phase element (CPE) for the double layer capacitance where is magnitude, is exponent, and f - frequency; (iii) resistance characterizing electron transfer kinetics at the interface, and (iv) a Warburg impedance element, with the coefficient . This element accounts for diffusion processes that are dominant at low frequencies. At low frequencies, ion transport at the skin-electrode interface significantly contributes to the overall impedance characteristics. The total impedance of the equivalent circuit can be expressed as:

1

Model parameters (, , , , and ) were extracted by fitting the measured impedance data to this equivalent circuit using the impedance.py package by Murbach et al. [35].    

Fig. 3.

Electrode-skin interface characterization and stimulation efficacy comparison between dry (blue) and gel (green) electrodes across all subjects (N=12). A The equivalent circuit model used for fitting, consisting of tissue resistance (), a constant phase element () with magnitude () and exponent (), charge transfer resistance (), and a Warburg diffusion element () with coefficient (). Boxplots show the distribution of the circuit model parameters for comparison between electrode types across subjects; statistical significance is indicated by asterisks (** denotes ). B sEMG activation thresholds () and corresponding pain levels at threshold for each electrode type. C Angular velocity thresholds () and corresponding pain levels at the threshold. D 3D head model showing the electrode configuration over the neck region. The gray plane and black rectangle indicate the clipping plane and zoomed region, respectively, used for the electric field presented in (D1+D2). Black dots represent the stimulation electrodes. E–F Electric field magnitude across the clipping plane for dry ( S/m, D1) and gel ( S/m, D2) interface conductivity values. Black rectangles represent the electrodes. G Maximum electric field in the skin layer as a function of interface conductivity for different electrode diameters: 8 mm (), 10 mm (square), 20 mm (triangle), and 30 mm (lozenge). The dashed line indicates the conductivity of the skin layer (0.465 S/m)

Finite element model of electrical stimulation

The electric field distribution generated by transcutaneous electrical stimulation was modeled using SimNIBS [36], a specialized finite element modeling tool for bioelectric applications. To account for the skin-electrode interface properties not natively supported by SimNIBS, we implemented a custom approach by adding an artificial conductive layer with interface impedance similar to charge transfer resistance described in Sect. "Skin-electrode interface and contact impedance". A sensitivity analysis was performed to characterize how variability in skin-electrode interface properties influences electric field distribution. Conductivity values were swept across to 10 S/m , spanning several orders of magnitude beyond the modeled values of (Fig. 3G). The model incorporated realistic electrode properties used in the neck experiment Sect. "Electrical accessory nerve stimulation" with median conductivity values from the impedance model (88 mm,  S/m for gel electrodes and 1520 mm,  S/m for dry electrodes). Electrode positions (Fig. 3D, black dots) were matched to those used in the experimental protocol targeting the accessory nerve described in Sect. "Electrical accessory nerve stimulation". The model solved the quasi-static approximation of Maxwell’s equations to compute the electric field magnitude and vectors throughout the tissue volume. For quantitative analysis, electric field distributions were calculated across the entire model with particular focus on the skin layer where nerve stimulation occurs. Various electrode diameters (8, 10, 20, and 30 mm) were simulated to investigate size effects on field distribution patterns (Fig. 3G).

Statistical analysis

Statistical comparisons between gel and dry electrode parameters were performed using Wilcoxon signed-rank tests due to the non-normal distribution of the data (Shapiro-Wilk test, p < 0.05). All statistical analyses were conducted using Python (scipy.stats, version 1.14.1). Data are presented as mean ± standard deviation unless otherwise specified.

Results

Two distinct experimental setups were employed in this study. In the first, electrical stimulation of the accessory nerve (in the neck) was applied to elicit shoulder movement, using dry and gel electrodes. This setup enabled the evaluation of stimulation efficacy, perceived pain level, and skin–electrode interface impedance (contact impedance). In the second setup, a closed-loop system was implemented for facial stimulation in one subject, wherein sEMG signal recorded from the left side of the face was used to trigger stimulation on the contralateral side. Both recording and stimulation in this configuration were performed using dry electrodes.

The dry electrodes used consisted of screen-printed carbon on a thin polyurethane (PU) substrate, with conductive silver traces passivated to prevent direct skin contact (see Fig. 1A). Owing to their soft nature, the electrode arrays conform well to the skin. AFM analysis showed a carbon surface roughness of 615 ± 51 nm (see Fig. 1B). Additionally, cross-sectional FIB-SEM imaging revealed distinct electrode array layers (Fig. 1C): Carbon (16.7 µm), silver traces (8.8 µm), and PU layer (54.3 µm).

Neuromuscular stimulation of the neck

The accessory nerve stimulation paradigm (Fig. 2A) provides a well-defined and direct innervation pathway to the trapezius muscle, enabling straightforward quantification of stimulation responses. In addition, the low density of nociceptive fibers in this nerve supports well-tolerated stimulation with minimal discomfort [37]. The neck region is particularly suitable for testing due to its flat, relatively hairless, and anatomically consistent surface. The dry electrodes used for both recording and stimulation are shown in Figs. 2B. A four-electrode array was used to study contact impedance and stimulation thresholds. To evaluate stimulation efficacy, a 16-channel dry electrode array was placed on the upper trapezius muscle to record evoked sEMG activity, normalized to each subject’s maximal voluntary contraction (MVC). MVC values were measured separately prior to stimulation sessions. In parallel, the angular velocity was recorded using a six-axis inertial measurement unit (IMU) positioned on the acromion bone, near the muscle, to measure shoulder movement.

Figure 2F shows the Nyquist plots of contact impedance (normalized to contact area) of one subject. Dots represent the experimental data, while lines indicate fitted curves based on the equivalent circuit (Fig. 3A; see Methods section "Skin-electrode interface and contact impedance"). As demonstrated previously [23], dry electrodes typically exhibited higher overall impedance compared to gel electrodes. Figure 2D shows a representative sEMG signal from one channel, demonstrating the temporal profile of muscle activation following a stimulation pulse. Two responses were observed: Synchronous muscle fiber activation (M-wave, red arrow), with a latency of 3 ms and duration of 27 ms, and asynchronous muscle activity with a latency of  200 ms. The latter response corresponds to shoulder movement and intensifies as the stimulation current increases. An example of the response at varying stimulation amplitudes is shown in Supplementary Fig. A1. Similarly, Fig. 2G shows the corresponding angular velocity, illustrating shoulder movement. The gray areas indicate the time window used to estimate the responses at each current for sEMG (normalized to %MVC) and angular velocity (Figs. 2 E and H, respectively). Stimulation thresholds were calculated for each curve ( and , respectively). Both electrode types achieved comparable thresholds and reached similar plateaus at high stimulation currents.

To examine group-level trends, impedance and stimulation efficacy were quantified and compared across 12 subjects (Fig. 3). An equivalent circuit was used to characterize the electrical properties of the skin-electrode interface, with parameters fitted to the experimental data (Fig. 3A; see Methods section 2.8). The charge transfer resistance () and Warburg coefficient () tended to be higher for dry electrodes, although these differences did not reach statistical significance (Wilcoxon signed-rank test p-value range, 0.0520.910; effect size range, 0.0320.400). Notably, the only parameter showing a significant difference was the constant phase element exponent (), which was higher for gel electrodes (p-value , effect size 0.560). Stimulation thresholds and corresponding pain perception levels are presented in Figs. 3B and 3C for and , respectively. Pain perception was assessed immediately after each stimulation intensity using a visual analog scale (VAS, 0–10; 0 = no pain, 10 = severe pain). Pain ratings at corresponding stimulation threshold were then compared between the electrodes. No significant differences were observed between electrode types in either stimulation thresholds or perceived pain levels (p-value range, 0.2040.445; effect size range, 0.2560.472). No skin irritation was observed with either electrode type following electrode removal. Across participants, values were consistently equal to or lower than , indicating the onset of isotonic muscle activity prior to observable shoulder movement (concentric muscle activity) [38].

Finite element simulations were performed to examine electric field distribution. A head model from the SimNIBS package was used to simulate transcutaneous stimulation in the neck region (Fig. 3D). An additional layer between the electrodes and skin was added to model the interface. Because of the quasi-static approximation in SimNIBS, capacitance could not be modeled directly; instead, charge transfer resistance was used to approximate the interface behavior. In the sensitivity analysis, conductivity varied over a wide range ( to 10 S/m) using circular electrodes of different diameters (8, 10, 20, and 30 mm) under a constant 2 mA DC current. The maximum electric field was extracted as a function of conductivity (Fig. 3G). At very low conductivities (below that of skin 0.465 S/m), the electric field magnitude was strongly dependent on electrode size, with smaller electrodes producing stronger fields. As conductivity increased, these size-dependent differences diminished. Overall, the analysis shows that electric field distributions remain robust across the variability observed in skin-electrode interface properties. Figs. 3E and 3F show the electric field distribution in the cross-sectional plane indicated by a gray line (Fig. 3D), focusing on the zoomed-in region marked by the black rectangle. These simulations used the electrode geometries and their corresponding modeled values (see Methods section Neck stimulation experimental protocol and Fig. 3A). Figures 3E and 3F correspond to dry electrodes (8 mm diameter,  S/m) and gel electrodes (elliptical, 15 x 20 mm,  S/m). In both cases, a strong electric field was observed around the interface, with the dry electrode generating a more intense and spatially condensed field. Yet the field distribution was similar across deeper tissue regions. These results suggest that, despite clear differences in contact impedance, dry electrodes may demonstrate similar behavior when used for stimulation.

Closed-loop functional electrical stimulation of the face

Having established the stimulation efficacy and comfort of dry electrodes in the neck region, we next demonstrate their utility in a closed-loop facial functional stimulation (Fig. 4A) as a pilot experiment. Two electrode arrays were placed bilaterally on the face of a healthy subject: a 7-channel array (stimulation) on the right side (Fig. 4B) and a 16-channel array (sEMG) on the left side (Fig. 4C). Figures 4D-F illustrate a representative trial, segmented into three phases: Resting baseline (Neutral), voluntary asymmetric facial activation (Voluntary asymmetric), and stimulation-evoked response (Stimulation). During the trial, sEMG signals were continuously monitored, and stimulation was triggered in real-time upon detecting muscle activation on the left side. This demonstration highlights the feasibility of using printed dry electrode arrays for real-time, closed-loop facial stimulation and their potential for integration into conformal and wearable neuromodulation platforms.

Discussion

This study applied established printed dry carbon electrodes for transcutaneous electrical stimulation in healthy subjects in two contexts. First, their performance was benchmarked against conventional gel electrodes during accessory nerve stimulation in the neck. Second, their practical application was demonstrated in a closed-loop facial stimulation paradigm, where sEMG from the left side of the face triggered stimulation on the contralateral side.

The skin-electrode interface at the stimulation site showed high impedance magnitudes for dry electrodes, especially at low frequencies (< 1 kHz), consistent with previous reports [23, 24]. This has important implications for wearable stimulation applications, as it increases the voltage compliance required to deliver the desired current. Moreover, at high impedance, much of the applied voltage drops across the stratum corneum, the outermost skin layer and primary barrier to current flow. This can lead to structural changes such as electroporation and lipid reorganization, which may increase skin permeability but also raise the risk of irritation, pain, or even irreversible tissue damage [14, 39]. However, impedance decreases at higher frequencies, and values for dry and gel electrodes converge, reducing the voltage drop across the skin. To better interpret this behavior, data were fitted to an equivalent circuit model. Unlike previous studies that relied on RC circuits [40, 41], here we used a circuit incorporating a constant phase element (CPE) and a Warburg diffusion component to accurately model the complex interface dynamics [33, 39, 42]. This model can help characterize the stimulation outcome and improve the prediction of voltage drop across different stimulation waveforms. The finite element model also indicates that despite the high values of dry electrodes, this parameter is not the primary determinant of electric field distribution. Rather, as shown in Fig. 3G, electrode size exerts a greater impact on field patterns, particularly within the typical conductivity range of our electrodes.

Stimulation efficacy was assessed using sEMG and mechanical movement (IMU) response curves. Both electrode types showed a sigmoidal increase in activation with rising current intensity, consistent with typical neuromuscular recruitment dynamics. Stimulation thresholds were comparable between electrode types, with no significant differences. Pain perception, a key consideration for the clinical adoption of functional electrical stimulation, was also evaluated. In our study, reported pain levels were within a tolerable range (average score of 3 out of 10), with no significant difference. Although the accessory nerve is a motor nerve, the tingling and pain sensations experienced during stimulation are primarily due to activation of cutaneous sensory nerves in the skin and subcutaneous tissue. Specifically, branches of the cervical plexus, such as the lesser occipital, greater auricular, and transverse cervical nerves, innervate the skin overlying the neck and shoulder region and carry A  and C fibers responsible for transmitting pain and thermal sensations [9]. The targeted motor nerve fibers cannot convey pain. In some cases, strong muscle contractions (e.g., trapezius activation) may also contribute to the perception of discomfort or deep aching pain. The similarity in thresholds results demonstrates that higher skin-electrode impedance does not inherently lead to elevated stimulation thresholds or increased discomfort when using current-controlled stimulation. While higher impedance requires greater voltage compliance from the stimulator, the delivered current remains constant, which explains the similar thresholds observed. Although higher voltage at high-impedance interfaces could theoretically increase pain perception, this was not observed in our study, possibly due to the low density of nociceptive fibers in the neck region. While dry electrodes do not inherently eliminate these sensations, they may offer advantages in parameter optimization through their customizable design. Electric field distribution rather than impedance alone can explain this apparent contradiction. Finite element modeling showed that, despite higher impedance owing to their smaller diameter, the electric field of dry electrodes is more concentrated, potentially improving neural activation at similar current levels. Petrofsky et al. [43] demonstrated that current density is non-uniform across the electrode surface, with peak density at the center and a 50% reduction at the edges. This may explain the similarity of stimulation effects, with different electrode sizes and impedance values. Indeed, our modeling results (Fig. 3E-G) support this interpretation, that current density is influenced more by electrode geometry than impedance. Electrode size further contributes to comfort and stimulation efficacy. Lyons et al. [44] found that smaller electrodes provided more effective stimulation and greater comfort. Similarly, Keller and Kuhn’s [45] reported that electrode design parameters, such as geometry and material homogeneity, also influenced current distribution. Taken together, the relationship between electrode properties, placement, and stimulation parameters is complex and multifactorial. Our results suggest that the performance of dry electrodes can match that of gel electrodes in effective neuromuscular stimulation, offering comparable thresholds and comfort. Additionally, multi-electrode arrays offer advantages in complex applications through pre-designed layouts that provide reproducible electrode positioning and allow post-placement selection of optimal electrode pairs, eliminating the need for precise individual electrode placement. Given their ease of use and minimal preparation time, dry electrodes present a promising alternative for wearable stimulation applications. Table 1 summarizes the comparison between electrodes.

Table 1.

Dry and gel electrode specifications and characteristics

Parameter	Dry electrodes	Gel electrodes
Material	Carbon	Ag/AgCl
Size (diameter/area)	8 mm / 50.3	1520 mm / 300
Contact impedance (neck at 100 Hz)
Stimulation threshold	1.36 ± 1.10 mA	1.83 ± 1.05 mA
Adhesion method	Adhesive film	Conductive hydrogel
Application time (multi-channels)	Fast	Slow [16]
Ease of customizability	High	Limited
Placement	Easy	Manual positioning
Adjustment after placement	Not applicable	Possible
Hairy regions	Poor	Better
Long-term stability	Good [24, 46, 47]	Gel dehydration [41]
Home-based use	Suitable [47]	Limited [17]
Technology maturity	New [26]	Well-established

The preliminary implementation of a closed-loop facial stimulation system using dry electrodes for both recording and stimulation demonstrates the feasibility of practical neuromodulation and therapeutic applications. This approach is particularly relevant for facial rehabilitation, where synchronized and natural-looking muscle activation patterns are essential. In our demonstration in one subject, voluntary muscle activity on one side of the face was used to trigger real-time stimulation on the contralateral side, enabling symmetric activation. This configuration offers promising applications for conditions such as facial palsy, where restoring symmetrical facial expressions is a key rehabilitation goal [48]. Extensive validation in multiple subjects and clinical populations will be required for such applications.Using the same type of dry electrode for both recording and stimulation not only simplifies system integration but also improves user comfort and convenience [18].

Several limitations of this study should be acknowledged. First, the experiments were conducted in a controlled laboratory environment with healthy participants. Responses in clinical populations may differ, particularly in individuals with altered skin properties (e.g., elderly patients with thinner skin or dermatological conditions). Moreover, this study focused on the stimulation of innervated muscles. In clinical cases involving denervated muscles, such as muscle paralysis or facial palsy, direct muscle activation is required. This is typically characterized by a right shift of the strength-duration curve, meaning higher current amplitudes and longer pulse durations are needed. Therefore, the stimulation thresholds observed in our study, as well as the closed-loop stimulation paradigm we demonstrated, should not be considered directly applicable to patients with neuromuscular impairments. Further investigation is necessary to adapt and validate the system for diverse clinical populations. Second, while this study employed a well-defined stimulation protocol targeting the accessory nerve, the findings might not directly translate to other stimulation paradigms or anatomical locations. Different nerves and muscle groups may respond differently to electrical stimulation depending on their depth, pain sensations, and electrode placement to evoke the desired movement. Further exploration of stimulation protocols and target sites is needed to better understand the applicability of the electrodes across diverse anatomical regions. Third, this study focused on short-term stimulation sessions. Previous studies have demonstrated long-term stability of these electrodes in recording applications, including sleep monitoring [47], dynamic activities such as cycling [46], and maintained impedance after 13 h of wear [24]. The electrodes have also passed standardized skin irritation testing. To assess durability under stimulation, a pilot test was conducted on one subject, showing a stable impedance after 3,000 pulses (Supplementary Fig. B1). To contextualize this finding within clinical requirements, clinical protocols range from hundreds to tens of thousands of pulses weekly (Supplementary Table B1). Our pilot test (3,000 pulses over 10 min) represents only a small fraction of typical clinical requirements. While encouraging, extensive validation across the full range of clinical pulse counts and durations will be necessary to confirm long-term reliability for therapeutic use. Fourth, commercial stimulators may exhibit variability in actual current delivery even with identical parameters, which can affect activation thresholds across different systems [25, 49]. While both electrode types were tested using the same stimulator, ensuring a valid relative comparison, characterizing the actual delivered waveforms would require specialized equipment not available in our setup. Lastly, the latency of the closed-loop setup was approximately 0.5 s, which requires further optimization. This latency primarily stems from Bluetooth data transmission, with minor contributions from activation detection and stimulus triggering. While this delay was acceptable for the present proof-of-concept demonstration, reducing it is critical for real-time facial symmetry correction. Ongoing optimization efforts are focused on improving the real-time data acquisition pipeline.

For clinical applications such as pain relief, stroke rehabilitation, restoring breathing, treatment for epilepsy and depression, and sleep enhancement [2], reliable dry electrodes offer distinct advantages. By eliminating issues commonly associated with gel electrodes, such as degradation of stimulation quality over time, dry electrodes can improve patient comfort, compliance, and long-term therapeutic outcomes. Additionally, the ability to customize the shape and configuration of electrodes through printing technologies opens new possibilities for personalized neurostimulation approaches.

Conclusion

This study demonstrated that printed dry carbon electrodes perform similarly to conventional gel electrodes for neuromuscular stimulation, despite differences in skin-electrode interface characteristics. Similar activation thresholds and perceived comfort levels highlight their viability as an alternative for neurostimulation applications. Their advantages in terms of easy placement, high resolution, and customizability make them particularly promising for clinical and research applications.

Abbreviations

AC

Alternating current

AFM

Atomic force microscopy

CPE

Constant phase element

DAU

Data acquisition unit

DC

Direct current

EMS

Electrical muscle stimulation

FES

Functional electrical stimulation

FIB-SEM

Focused ion beam scanning electron microscopy

IMU

Inertial measurement unit

IRB

Institutional review board

LUT

Left upper trapezius muscle

MMT

Manual muscle testing

MVC

Maximum voluntary contraction

PCB

Printed circuit board

PU

Polyurethane

RMS

Root-mean-square

sEMG

Surface electromyography

tACS

Transcranial alternating current stimulation

tDCS

Transcranial direct current stimulation

TENS

Transcutaneous electrical nerve stimulation

TES

Transcutaneous electrical stimulation

VAS

Visual analog scale

Author contributions

R.I. developed and performed the data analysis. R.I and Y.H. steered the analysis process and result presentation. R.I. and D.R. developed the experimental setup and modeling. Y.H. acquired funding, conceptualized the study, and supervised the project and the data analysis. R.I. wrote the first manuscript draft. I.V. contributed to the data collection, interpretation of results, and manuscript revision. P.F.F., G.F.F., and O.G.L. contributed to the interpretation of results, clinical contextualization, and manuscript revision. All authors contributed to the final version of the manuscript.

Funding

This work was supported by the Israel Science Foundation (ISF, Grant No. 1355/17), the European Research Council (ERC, Grant Outer-Ret–101,053,186), and the Deutsche Forschungsgemeinschaft (DFG, Grant No. GU-463/12–1). The funding agencies were not involved in the design, acquisition, or interpretation of the findings or the writing of this manuscript.

Data availability

The data that support the findings of this study are available upon request from the corresponding author.

Declarations

Ethics approval and consent to participate

The study was approved by the Institutional Ethics Committee Review Board at Tel Aviv University (IRB, no 0,009,184–2). All participants provided written informed consent to participate in the study. The study has been conducted in accordance with relevant guidelines and regulations.

Consent for publication

The authors have obtained written consent to publish the images and videos.

Competing interests

The authors declare no Conflict of interest.