Introduction

Over the last few decades, artificial intelligence (AI) systems have been replicating biological functions realistically. Researchers globally have developed artificial organs such as bionic eyes1 that can provide sensory capabilities to robots or restore sensory feelings to the disabled. The development of a self-powered electronic skin (e-skin) inspired by human mechanoreceptors is a prime need to restore sensory functions. The advancement of AI systems has facilitated the scale-up of e-skins that can emulate biological sensory systems and introduce a sense of touch for humanoid robots and those who wear prosthetic devices. In the human body, the skin is the largest biological sensory system containing complex arrays of mechanoreceptors. When an external stimulus is applied, the sensory organs of the skin generate a receptive potential that is transmitted to the brain through axons2,3. Subsequently, the brain examines the type of the external stimulus using rapidly adaptive (RA) and slowly adaptive (SA) mechanoreceptors. SA-mechanoreceptors respond to static stimuli by showing continuous response to the maintained skin deformation, whereas RA-mechanoreceptors respond to dynamic stimuli by showing on and off responses with respect to the changes in skin deformation4. They differ with regard to the sizes and structures of their receptive fields and exhibit different sensitivities to static and dynamic stimuli5,6.

Generally, mechanoreceptors are spread over the entire area of the human skin. In the glabrous skin of one hand, there are ~17,000 cutaneous mechanoreceptors5, which are categorized into four major types: Meissner’s corpuscles, Pacinian corpuscles, Merkel cells, and Ruffini endings (Fig. 1a)7. The number and densities of these mechanoreceptors vary within the separate subregions of the glabrous skin and exhibit different resolutions. With regards to the positioning of these mechanoreceptors in the glabrous skin, Merkel receptors and Meissner corpuscles are located close to the surface of the skin, whereas Ruffini endings and Pacinian corpuscles are located deeper in the skin. Merkel receptors and Meissner corpuscles are representative of SA-I and RA-I; they have small receptive fields and respond to minor pressures and low vibration frequencies. Similarly, Ruffini endings and Pacinian corpuscles are representative of SA-II and RA-II; however, they have large receptive fields and respond to large pressures and high vibration frequencies.

Fig. 1: Types of mechanoreceptors and their dynamic and static responses.
figure 1

a Mechanoreceptors in human skin7, and (b) dynamic and (c) static responses of RA- and SA-mechanoreceptors under the same mechanical stimuli.

Full size image

Based on human tactile perception principles, artificial tactile perception systems can be advanced similarly to perceive complex mechanical stimuli, enabling robots or artificial prosthetics to interact with the surrounding environment efficiently. The interest in the sensory system of the human skin began in the late 1970s when Knibestöl and Vallbo analyzed 61 mechanoreceptor units in the glabrous skin8. Based on physical feedback principles, Clippinger et al. demonstrated the first prosthetic hand in 19749. Subsequently, the tactile sensation has been extensively explored by several researchers, and different strategies have been employed to mimic each type of mechanoreceptor. An e-skin simultaneously mimicking SA- and RA- mechanoreceptors can analyze objects based on their physical properties, such as the surface texture, sense of shape, pressure, and dynamic or static strain10. To mimic these functionalities of the human skin, several sensing mechanisms have been employed for the development of an e-skin such as ion-channel systems11,12, transistors13,14, capacitive15,16, piezoresistive17,18, triboelectric19,20, and piezoelectric-based sensors21,22.

At the current stage, tactile sensors have several interesting features such as device scalability, mechanical flexibility, and biocompatibility; however, none of them possess self-power capability. Tactile sensors based on triboelectric and piezoelectric effects are self-powered because they scavenge electrical energy from mechanical stimuli. However, owing to the fast decay of their voltage response, they are limited to represent a single sensing mechanism by mimicking RA-mechanoreceptors only23,24,25,26,27. In contrast, to mimic SA-mechanoreceptors, generally piezoresistive, capacitive, and ion-channel methods are used23,28,29,30,31, which require a separate powering source to operate them.

In the emerging fields of wearable healthcare devices, humanoid robotics, and artificial prosthetics, flexible and self-powered sensors simultaneously mimicking SA- and RA-mechanoreceptors are in high demand. In recent years, researchers worldwide have integrated different sensing elements into one platform to achieve this goal. In 2017, Chen et al. proposed e-skins fabricated by spatially integrating two sensing mechanisms based on triboelectric and piezoresistive effects; the e-skins mimicked SA- and RA-mechanoreceptors simultaneously32. Similarly, Chun et al. proposed a self-powered sensor based on an ion-channel system spatially combined with a piezoelectric film that could simultaneously mimic SA- and RA- mechanoreceptors23. However, fabricating an e-skin by spatial integration of required sensing elements is a complex process and makes the device bulky. Park et al. applied another strategy in which they merged piezoresistive and ferroelectric devices to detect static and dynamic pressures simultaneously33. Similarly, Ha et al. proposed an e-skin by merging piezoelectric and piezoresistive sensing mechanisms that simultaneously mimicked SA- and RA-mechanoreceptors34. The strategy of merging multimode sensors is relatively easy compared with device fabrication with spatial integration of sensors. However, the main problem is that such devices show high crosstalk between the multiple sensing modes, which makes them incapable of simultaneous detection of static and dynamic stresses.

In addition to the simultaneous detection of static and dynamic stresses, an e-skin must demonstrate self-power capability, rapid response, and high sensitivity. As mentioned above, an e-skin based on piezoelectric or triboelectric effects possesses self-power capability. Compared to sensors based on the triboelectric effect, piezoelectric sensors are lesser affected by extreme environmental conditions such as humidity or high temperature. Among inorganic piezoelectric materials, such as lead zirconate titanate35, MoS236,37, and BaTiO338, ZnO and GaN exhibit high sensitivity owing to their high piezoelectric coefficient d33 along the [0001] direction39,40,41. However, the existence of free carriers in ZnO and GaN screens the stress-induced piezoelectric potential from decaying the voltage response immediately, which limits them to mimicking RA-mechanoreceptors only. In intrinsic semiconductors, the pressure-induced piezoelectric polarization remains stable as long as a stimulus is applied42, which is the key requirement for mimicking SA-mechanoreceptors.

The successful growth of semi-insulating piezoelectric semiconductors is a key factor in suppressing the free-carrier screening effect40,43,44. Compared with ZnO, GaN has a well-developed growth process, and its p-type doping can easily be controlled. The growth of Mg-doped GaN NWs in a hydrogen-rich environment, without postgrowth annealing, produces semi-insulating GaN NWs because of the formation of inactive Mg–H complexes at a deep level in the energy bandgap45,46. In this study, we established a concept for fabricating a self-powered and flexible e-skin based on unintentionally doped GaN (uGaN) and Mg-doped semi-insulating GaN (GaN:Mg) nanowire arrays (NW arrays) that simultaneously mimics SA- and RA-mechanoreceptors, respectively. The mechanoreceptors fabricated with the uGaN NW arrays are represented as RA-mechanoreceptors and those with the semi-insulating GaN:Mg NW arrays as SA-mechanoreceptors.

Results and discussion

Material characterization

The overall fabrication process of the self-powered and flexible e-skin based on the GaN NW arrays is illustrated in Fig. 2. The GaN NW arrays were grown using a vapor–liquid–solid (VLS) growth technique, and the details are explained in the experimental procedure. Supplementary Fig. 2a and Fig. 3a show scanning electron microscopy (SEM) images of the uGaN and GaN:Mg NW arrays, respectively. The magnified SEM images of uGaN and GaN:Mg NWs can be seen in Supplementary Fig. 2b, c, respectively. The average length of the NWs was ~5 µm, whereas the diameter of the NWs varied from 30 to 70 nm. As mentioned in the Introduction section, the sizes and densities of SA- and RA-mechanoreceptors vary within the separate subregion of the glabrous skin and present different resolutions, which can be achieved artificially by synthesizing the NW arrays on a required pattern size. However, this study is more focused on the detection of static and dynamic stresses by the e-skin based on the GaN NW arrays, which has not been reported previously. Therefore, in this study, only one resolution of the NW arrays is investigated. The NWs were grown in circular-patterned arrays with a diameter of 20 µm; the pitch size between the circular patterns was 20 µm. The transmission electron microscopy (TEM) images, as shown in Supplementary Figs. 2d and Fig. 3d, confirm the single crystallinity of the uGaN and GaN:Mg NWs, respectively; both the NWs are grown along the c-axis. Figure 3e–h shows the TEM-energy-dispersive X-ray spectroscopy (EDS) elemental mapping of a single Mg-doped GaN:Mg NW, and the successful incorporation of Mg can be seen in Fig. 3h. The Mg-doped GaN NW without postgrowth annealing contains electrically neutral Mg–H complexes, which act as deep acceptors making the GaN semi-insulating or low conducting.

Fig. 2: Fabrication steps.
figure 2

Schematic flow diagram of growth of GaN NW arrays and fabrication of flexible e-skin.

Full size image
Fig. 3: Analysis of GaN NW.
figure 3

a SEM image of GaN NW arrays, b SEM image of PDMS-encapsulated GaN NW arrays, c PDMS etching from top surface of GaN NW arrays, d TEM image of single GaN:Mg NW; inset shows SAED pattern and lattice fringes, and e–h EDS elemental mapping of single GaN:Mg NW.

Full size image

Investigation of piezoelectric responses

Figure 4 shows the overall measured output voltage signals of RA- and SA-mechanoreceptors based on the uGaN and GaN:Mg NW arrays, and a mechanical stress was applied using a probe station tip (diameter 10 µm). Figure 4a, b illustrates the general process of the generation of piezoelectric polarization in the uGaN NWs and the screening effect. The stress-induced piezoelectric polarization is generated at the top and bottom sides of the NWs, whereas the high density of free charge carriers (electrons) immediately screens the piezoelectric polarization resulting in a sharp decay of the voltage peaks (Fig. 4c). It is well known that uGaN behaves as n-type material and its Fermi level exists near the conduction band, as illustrated in Fig. 4b; the high density of free electrons can be seen in the conduction band. Because of the significant free-carrier screening effect, the uGaN NWs exhibit dynamic characteristics by instantaneous pressure signals and can be used to mimic RA-mechanoreceptors. As mentioned before, to mimic SA-mechanoreceptors, the free-carrier screening effect must be suppressed in GaN. The successful suppression of free-carrier screening enables GaN to measure static pressure signals. Therefore, the Mg-doped GaN:Mg NW arrays were synthesized without postgrowth annealing. During the growth, hydrogen atoms passivate the Mg acceptors electrically by forming Mg–H complexes. The formation of electrically neutral Mg–H complexes at a deep level shifts the Fermi level toward the valence band, reducing the electron density and making the GaN:Mg NWs semi-insulating. A schematic (Fig. 4d) shows the semi-insulating GaN:Mg NWs under mechanical stress, and its electron density is comparatively lower than that of the uGaN NWs. This reduces the screening effect and enables the GaN:Mg NWs to exhibit static voltage responses, which reveal static characteristics such as the magnitude and duration of the applied stimulus; the respective change in energy band because of Mg-doping can be seen in Fig. 4e. Based on the static voltage responses exhibited by the GaN:Mg NWs (Fig. 4f), they can be used to mimic SA-mechanoreceptors.

Fig. 4: Schematic diagram of piezoelectric operation.
figure 4

a Schematic representing piezoelectric polarization inside uGaN NWs, b corresponding voltage responses of uGaN NWs under different weights applied by probe station tip, c schematic of semi-insulating GaN:Mg NWs, and d their voltage response under different weights.

Full size image

Depending on the static and dynamic voltage responses of the GaN-based mechanoreceptors measured by the probe station tip, we fabricated two separate mechanoreceptor patches (RA- and SA-mechanoreceptor patches) based on the uGaN and GaN:Mg NW arrays, respectively. The RA- and SA-mechanoreceptor patches were affixed on an index fingertip to test them for realistic tactile sensing applications; the size of each mechanoreceptor patch was 0.5 × 0.5 cm2. Figure 5a shows the voltage responses when pressure is applied on the RA-mechanoreceptor patch; the voltage magnitude is small (7 V), and the peak behavior is dynamic because of the high screening effect caused by the high electron density in the uGaN NW arrays. Although free-carrier screening is always considered as a negative factor that degrades the piezoelectric performance of piezoelectric nanogenerators, it exhibits dynamic voltage responses, which are meaningful for mimicking RA-mechanoreceptors. As discussed earlier, to mimic SA-mechanoreceptors, the voltage signal was tuned from dynamic to static by producing the semi-insulating GaN:Mg NW arrays. When the same pressure was applied on the SA-mechanoreceptor patch, a static voltage peak with a larger magnitude was observed (relative to RA-mechanoreceptor) (Fig. 5b) because of the semi-insulating nature of the GaN:Mg NW arrays. The response and reset times of the NW-array SA-mechanoreceptor was also investigated under an approximately ~1 Hz frequency (Fig. 5c); the mechanoreceptor exhibited very rapid response and reset times of 11 ms and 18 ms, respectively (Fig. 5c, inset). The response and reset times changed with respect to the actuation frequencies. Supplementary Figs. 4a and b show the response and reset times of SA-mechanoreceptor measured at an actuation frequency of ~ 5 Hz; the response time was reduced to 7 ms while the reset time was reduced to 12 ms. These are state-of-the-art values, which have not been reported before for GaN:Mg NW-based pressure or tactile sensors. The response and reset time of RA-mechanoreceptor was also measured at 1 Hz frequency (Supplementary Fig. 5), however, it showed quite different behavior. The piezoelectric voltage started to decay even when the stress was not removed and it was followed by a long trail due to a strong internal screening effect in uGaN NW array. The response time was 8 ms, but a very long trail (>820 ms) made it difficult to evaluate accurate reset time.

Fig. 5: Response of RA- and SA-mechanoreceptor patches.
figure 5

a Dynamic voltage responses of RA-mechanoreceptor-patch, the scale bar shows 5 mm, b static voltage responses of SA-mechanoreceptors patch, the scale bar shows 5 mm, and c response and reset times of SA-mechanoreceptors.

Full size image

Object grasping and surface detection by mechanoreceptors

To investigate the NW-array mechanoreceptors for realistic applications, the SA- and RA-mechanoreceptor patches, affixed on an index fingertip (Fig. 6a), were used for grasping and lifting a small beaker (half-filled with water). Subsequently, the beaker was grasped between the index fingertip and the thumb, lifted off the table, held in the air for a few seconds, and placed back on the table; the process was conducted in three steps: A–C. Step A is the process of grasping the beaker and gradual tightening of the grip on the beaker (A-i). This was followed by the grip pressure being maintained for a few seconds (A-ii) before lifting the beaker up from the table to investigate the consequent voltage responses of SA- and RA-mechanoreceptors. Step B is the process of lifting the beaker (B-i) up from the table and holding it for a few seconds (B-ii). Step C is the process of placing the beaker back on the table (C-i) and releasing the grip (C-ii), as shown in Fig. 6b. Figure 6c, d shows the corresponding voltage responses exhibited by SA- and RA-mechanoreceptors. For step A-i, the SA-mechanoreceptors showed a gradual increase in the voltage peak (Fig. 6c, A-i), which indicates an increase in the pressure on the sensors because of the gradual tightening of the grip on the beaker. Under the same condition, a similar voltage peak behavior was exhibited by the RA-mechanoreceptors with a smaller voltage magnitude (Fig. 6d, A-i). For step A-ii, when the grip pressure was maintained for a few seconds, the SA-mechanoreceptors showed a sustained voltage response (Fig. 6c, A-ii), which confirmed the semi-insulating property of the GaN:Mg NWs, as discussed before. However, for the same step, the voltage response exhibited by the RA-mechanoreceptors decayed to zero (Fig. 6d, A-ii) because of the high density of the free electrons in the uGaN NWs, which screened the piezoelectric potential. In steps B-i and B-ii, the beaker was lifted off the table and held in air for a few seconds. As shown in Fig. 6c, d, the SA- and RA-mechanoreceptors produce the resultant voltage peaks (B-i, B-ii) with high magnitudes, confirming that the pressure on the mechanoreceptors is increased when the beaker is lifted off the table and held in the air. When the beaker was held in the air for a few seconds, the SA-mechanoreceptors exhibited a static voltage behavior (Fig. 6b, B-ii), whereas the voltage decayed in the RA-mechanoreceptors (Fig. 6d, B-ii), showing a dynamic behavior. Finally, when the beaker was placed back on the table in step C-i and the grip was released (C-ii), the voltage dropped to zero accordingly. In this study, we established an overall concept to utilize the uGaN and semi-insulating GaN:Mg NW arrays to mimic RA- and SA-mechanoreceptors, respectively. However, the transfer process of the mechanoreceptors patches on the fingertip can further be advanced and their number can also be increased.

Fig. 6: Grasping, lifting, and releasing responses of RA- and SA-mechanoreceptors.
figure 6

a SA- and RA-mechanoreceptor patches affixed on index fingertip, b grasping of beaker between index fingertip and thumb (A-i, A-ii), lifting of beaker off the table (B-i, B-ii), and releasing beaker back on the table (C-i, C-ii), c corresponding responses of SA-mechanoreceptors, and d RA-mechanoreceptors.

Full size image

The NW-array mechanoreceptors were further investigated for another realistic application: surface sensation. For this purpose, an index fingertip (affixed with the mechanoreceptor patches) was rubbed on the top surfaces of three different objects S1, S2, and S3; this is shown in Fig. 7a–i. The S1 (Fig. 7a) and S2 (Fig. 7d) are manually fabricated grids on a piece of regular thermocol (made of polystyrene), while S3 (Fig. 7g) is a plastic box having smooth grids on its surface. When the finger touched the surface of object S1, the SA-mechanoreceptors presented a positive voltage signal that did not drop to zero unless the finger pressure was released; several small signals were also observed during finger sliding (Fig. 7b). The RA-mechanoreceptors, for the same surface, presented an instantaneous positive voltage signal, which immediately dropped to zero. Several small signals were also observed while the finger was sliding on the surface, followed by a negative peak after the finger was released (Fig. 7c). The appearance of small signals during the finger sliding indicated that the surface had a texture. For further investigation, the same process was repeated for objects S2 and S3, which had lower and higher grid densities on their surfaces than objects S1. Figure 7e, f shows the respective voltage responses of the SA- and RA-mechanoreceptors for surface S2. When the finger was slid from one edge to the other edge on the surface of S2, the SA- and RA-mechanoreceptors presented five voltage peaks in ~4 s, which suggested that the grids were narrower than S1. As seen in Fig. 7a, d, objects S1 and S2 have rough surfaces, which can also be analyzed from the shapes of the corresponding voltage peaks exhibited by the SA- and RA-mechanoreceptors. For further investigation of the surface roughness, we selected object S3, which had a comparatively smoother surface (Fig. 7g). During the sliding of the fingertip on object S3, the SA- and RA-mechanoreceptors exhibited very smooth voltage peaks (Fig. 7h, i), which confirmed that object S3 had a smoother surface.

Fig. 7: Analysis of surface texture.
figure 7

a Photograph of S1, and voltage responses of (b) SA-mechanoreceptors and (c) RA-mechanoreceptors from S1, (d) photograph of S2, and voltage responses of (e) SA-mechanoreceptors and (f) RA-mechanoreceptors from S2, (g) photograph of S3, and voltage responses of (h) SA-mechanoreceptors, and (i) RA-mechanoreceptors from S3.

Full size image

RA-I, SA-I, RA-II, SA-II under different stimuli

As mentioned in the Introduction section, the human skin has separate mechanoreceptors for detecting various stimuli. The RA-I and SA-I mechanoreceptors respond to low pressures and low vibration rates and are responsible for controlling hand grip and sensing fine details. In comparison, the RA-II and SA-II mechanoreceptors respond to high pressures and high vibration rates and are responsible for recognizing fine textures and stretching of the skin. In the human skin, the RA-I and SA-I mechanoreceptors are located close to the skin surface, whereas the RA-II and SA-II ones are located deeper in the skin (Fig. 8a). However, the GaN NWs are sensitive to several pressures and low- and high frequencies43,47. Depending on these characteristics, the mechanoreceptors based on the uGaN NW arrays can simultaneously work as the RA-I and RA-II mechanoreceptors, and those based on the semi-insulating GaN:Mg NW arrays can simultaneously act as the SA-I and SA-II mechanoreceptors. These NW-array mechanoreceptors can be affixed alongside on the e-skin, as proposed in Fig. 8b. Depending on the above-mentioned characteristics of mechanoreceptors, the mechanoreceptors based on the uGaN and GaN:Mg NWs are further investigated under different pressures and frequencies. Figure 8c, d shows the dynamic voltage responses under a low weight (10 g) and a slow actuation frequency (0.3 Hz), which are consistent with the responses of the RA-I mechanoreceptors. In contrast, Fig. 8e, f shows the static voltage responses under the same measurement conditions as above, which are consistent with the responses of the SA-I mechanoreceptors; this static behavior is due to the semi-insulating properties of the GaN:Mg NW arrays. Similarly, the voltage responses were further investigated under high pressures and high vibration frequencies (Fig. 8g–j), and accordingly, the results are consistent with the dynamic responses of the RA-II mechanoreceptors and the static responses of the SA-II mechanoreceptors.

Fig. 8: Responses under low and high weight and frequencies.
figure 8

a Schematic representing positioning of mechanoreceptors in human skin, (b) proposed e-skin composed of artificial SA- and RA-mechanoreceptors; voltage responses of RA-I mechanoreceptors under (c) low weight and (d) low frequency, voltage responses of SA-I mechanoreceptors under (e) low weight and (f) low frequency, and voltage responses of RA-II mechanoreceptors under (g) high weight and (h) high frequency and voltage responses of SA-II mechanoreceptors under (i) high weight and (j) high frequency.

Full size image

As mentioned in the Introduction section, tactile sensors configured in multisensory modes present high crosstalk due to the parasitic capacitance between multimode sensor arrays, which makes them incapable of distinguishing different types of stimuli. In this study, the crosstalk between the NW arrays was confirmed by applying pressure on several uGaN and GaN:Mg NW arrays using probe station tips (Fig. 9). Initially, a pressure was applied by tip-1 and the corresponding voltage peaks from the uGaN and GaN:Mg NW arrays were observed (Fig. 9a, b, Zone A). Subsequently, a pressure was applied on adjacent NW arrays sequentially using tip-2, and when the pressure of tip-1 was released, no parasitic conduction was observed (Fig. 9a, b, Zones B and C), which confirmed zero crosstalk between the NW arrays of the SA- and RA-mechanoreceptors. The self-power capability of the mechanoreceptor was evaluated by charging a 4.7 nF capacitor with RA-mechanoreceptor. Supplementary Fig. 3a shows the circuit diagram designed to charge a capacitor. The RA-mechanoreceptor was connected with the linear motor which can bend the mechanoreceptor at different angles and frequency. Supplementary Fig. 3b shows the successful charging of a capacitor by RA-mechanoreceptor by bending it at 155° with 6 Hz frequency. The cyclability of uGaN and GaN:Mg has already been reported several times in our preliminary studies41,43,47,48,49,50,51.

Fig. 9: Crosstalk measurement.
figure 9

Crosstalk measurements between (a) semi-insulating GaN:Mg NW arrays (SA-mechanoreceptors) and (b) uGaN NW arrays (RA-mechanoreceptors).

Full size image

In summary, a state-of-the-art self-powered e-skin is proposed that can simultaneously mimic SA- and RA-mechanoreceptors using piezoelectric sensing principles. Initially, separate SA- and RA-mechanoreceptors were fabricated using semi-insulating GaN:Mg and uGaN NW arrays, and subsequently, their patches were transferred to a 30-µm-thick silicone rubber for fabricating the self-powered e-skin. The proposed e-skin successfully demonstrated the grasping and lifting of a beaker off a table and detected surface textures of three different objects. Because of the properties of the uGaN and GaN:Mg NW arrays, the SA- and RA-mechanoreceptors interpreted diverse mechanical characteristics well, including change in the pressure, low- and high frequencies, and durations and magnitudes of the applied stimuli. The results were consistent with the responses of the RA-I, RA-II, SA-I, and SA-II mechanoreceptors in the human skin. The proposed e-skin possesses an interesting and simple structure because it operates on the piezoelectric effect only. In comparison, the conventional e-skins and mechanoreceptors have hybrid structures and operate on several principles.

Methods

Growth of uGaN and GaN:Mg NW arrays

We performed a VLS growth of Au-catalyzed GaN NW arrays on a GaN thin film by metal–organic chemical vapor deposition (MOCVD). Prior to the growth of the GaN NW arrays, photoresist (PR) patterning was performed on a GaN thin film for selective area deposition of a Au film, which acts as a metal catalyst for the VLS growth of the GaN NW arrays (Fig. 2). A chrome mask patterned into equidistant circular and square arrays was used for the PR patterning. For the patterning, a positive photoresist (AZ GXR 601) diluted with a thinner (AZ 1500) in a 1:1 ratio was spin-coated on the GaN thin film followed by soft baking of the PR at 115 °C. The ultraviolet light exposure time and the PR developing time were optimized to 6 s and 5 s, respectively; MIF 500 was used as the PR developer. Subsequent to the successful patterning of the PR, the Au film (0.5 nm) was deposited using an e-beam evaporator, followed by the removal of the residual PR by immersing the sample in acetone for 5 s (Fig. 2).

After deposition of the patterned Au arrays, the GaN thin film was loaded into the MOCVD reactor for the VLS growth of the GaN NW arrays. Prior to the growth step, a low-temperature in situ deposition of indium- and gallium layers was performed on the Au-patterned GaN thin film by introducing 5.7 μmol min−1 of trimethylindium and 98 μmol min−1 of trimethylgallium (TMGa) as the metal–organic precursors. Subsequently, the reactor temperature was increased to 830 °C and the agglomeration of In–Ga–Au layers was performed for 300 s at 35 torr to produce nano-sized spherical-shaped In–Ga–Au metal alloy particles. After the agglomeration step, an ~4 mmol min−1 flow of NH3 was introduced with a 55 µmol min−1 flow of TMGa in a H2 environment. Growth of the NW arrays was conducted for 6000 s at a reactor pressure of 35 torr. As mentioned in the Introduction section, we synthesized the uGaN and semi-insulating GaN:Mg NW arrays. For the growth of the semi-insulating GaN:Mg NW arrays, ~1 µmol min−1 flow of bis(cyclopentadienyl) magnesium (Cp2Mg) was introduced during the growth process, and postgrowth annealing was not performed to allow the formation of neutral Mg–H complexes in the GaN:Mg NW arrays.

Device fabrication and characterization

For the fabrication of the flexible e-skin, the uGaN and GaN:Mg NW arrays were transferred to an indium tin oxide (ITO)-coated flexible and stretchable 30-µm-thick silicone rubber. The ITO was deposited on the silicone rubber in an argon (Ar) environment using a radiofrequency magnetron sputter. Using a 99.999% pure ITO target, the deposition was conducted for 9000 s by maintaining the cathode power at 250 W under an Ar flow of 20 sccm. To ease the transferring process, the NW arrays grown on the GaN thin film were encapsulated in polydimethylsiloxane (PDMS) prior to their transfer (Fig. 2). However, to confirm the direct contact of the NW arrays with a metal, PDMS was etched from the top surface of the NW arrays by the ICP-RIE. The ICP-RIE conditions for the PDMS etching can be found in our preliminary study47. After the etching of the PDMS from the top surface, PDMS-encapsulated NW arrays (NW-array-matrix) were transferred to the ITO-coated silicone rubber using the doctor blading technique (Fig. 2), followed by the deposition of Ni/Au (10/150 nm) as the top metal contact. The photographs of the mechanoreceptor patches on silicone before connecting the wires can be seen in Supplementary Fig. 1. To affix the mechanoreceptors on the fingertip, the mechanoreceptor patches of 0.5 × 0.5 cm2 size were fabricated. The mechanoreceptor patches fabricated with that of uGaN NWs are referred as RA-Mechanoreceptor patches while the mechanoreceptor patches fabricated with that of GaN:Mg NW arrays are referred as SA-mechanoreceptor patches.

Characterization of the e-skin based on the uGaN and GaN:Mg NW arrays was conducted using a high-speed current/voltage measurement unit potentiostat/galvanostat/EIS (PARSTAT 3000, Princeton Applied Research). To confirm the specific voltage response of each NW array of the e-skin, mechanical stimuli were applied using a probe station tip. However, the e-skin was also tested by applying different weights and bending it at low and high frequencies using homemade actuators. At the current stage, it was not possible to characterize the RA- and SA-mechanoreceptors simultaneously. All the mechanoreceptors were characterized individually, and their responses were compared manually, because of which a minor delay in their responses is observed.