Letter

Cite This: Nano Lett. 2019, 19, 3305−3312 pubs.acs.org/NanoLett

Self-Powered Pressure- and Vibration-Sensitive Tactile Sensors for

Learning Technique-Based Neural Finger Skin
Sungwoo Chun,†,‡ Wonkyeong Son,§ Haeyeon Kim,∥ Sang Kyoo Lim,§ Changhyun Pang,†,‡
and Changsoon Choi*,§
†
Department of SKKU Advanced Institute of Nanotechnology (SAINT), Sungkyunkwan University, Suwon, Gyeonggi-do 16419,
Republic of Korea
‡
School of Chemical Engineering, Sungkyunkwan University, Suwon, Gyeonggi-do 16419, Republic of Korea
§
Department of Smart Textile Convergence Research, Daegu Gyeongbuk Institute of Science and Technology (DGIST), Daegu
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

42988, Republic of Korea

∥
Advanced Process and Materials R&D Group, Korea Institute of Industrial Technology, Incheon 21999, Republic of Korea
Downloaded via NATL UNIV OF SINGAPORE on March 2, 2021 at 07:07:12 (UTC).

*
S Supporting Information

ABSTRACT: Finger skin electronics are essential for realizing humanoid soft robots and/or medical applications that are very
similar to human appendages. A selective sensitivity to pressure and vibration that are indispensable for tactile sensing is highly
desirable for mimicking sensory mechanoreceptors in skin. Additionally, for a human−machine interaction, output signals of a
skin sensor should be highly correlated to human neural spike signals. As a demonstration of fully mimicking the skin of a
human finger, we propose a self-powered flexible neural tactile sensor (NTS) that mimics all the functions of human finger skin
and that is selectively and sensitively activated by either pressure or vibration stimuli with laminated independent sensor
elements. A sensor array of ultrahigh-density pressure (20 × 20 pixels on 4 cm2) of interlocked percolative graphene films is
fabricated to detect pressure and its distribution by mimicking slow adaptive (SA) mechanoreceptors in human skin. A
triboelectric nanogenerator (TENG) was laminated on the sensor array to detect high-frequency vibrations like fast adaptive
(FA) mechanoreceptors, as well as produce electric power by itself. Importantly, each output signal for the SA- and FA-
mimicking sensors was very similar to real neural spike signals produced by SA and FA mechanoreceptors in human skin, thus
making it easy to convert the sensor signals into neural signals that can be perceived by humans. By introducing microline
patterns on the top surface of the NTS to mimic structural and functional properties of a human fingerprint, the integrated NTS
device was capable of classifying 12 fabrics possessing complex patterns with 99.1% classification accuracy.
KEYWORDS: Self-power, mechanoreceptors, skin electronics, sensors, triboelectric nanogenerator, finger skin

uman finger skin has the most effective sensory system

H and senses tactile modalities (e.g., touch, pressure,
vibration, warm and cold, pain) with spatiotemporal
inspires the development of efficient and adaptive tactile
sensing systems. Some attempts have been made to achieve
sensory functions of human skin by developing either a single
perception of externally applied stimuli through cutaneous
sensory receptors. Then the spatiotemporal tactile signals device sensitive to pressure, strain, vibration, and tactile
transmitted to the somatosensory cortex through neural sensors or their multistacked architectures.3−13
afferents are encoded as voltage spikes of action potential to
be transmitted to the brain.1 The brain then comprehensively Received: March 4, 2019
recognizes the types and intensity of the tactile stimuli.2 Revised: April 16, 2019
Mimicry of the finger skin functions and its sensory system Published: April 25, 2019

© 2019 American Chemical Society 3305 DOI: 10.1021/acs.nanolett.9b00922

Nano Lett. 2019, 19, 3305−3312
Nano Letters Letter

Figure 1. Flexible electronic mechanoreceptors mimicking human finger skin. (a) Human finger skin and fingerprint structure of epidermis and
mechanoreceptors of dermis. (b) Schematic illustration of an NTS device. The NTS device is composed of a bottom panel with a SA-mimicking
sensor, a top panel with an FA-mimicking sensor, and an artificial fingerprint structure with microlines. (c) Correlation between neural spikes of
human skin and electrical output signals of an NTS device for slow and fast adaptation in response to physical stimuli, respectively. (d) Interlocked
percolative graphene sensor array. (d(i)) Schematic of individual interlocked top and bottom GNP films. (d(ii)) SEM image showing graphene
morphology of the interlocked region. With natural force, top and bottom graphene films are slightly interlocking with distance. (d(iii)) Optical
image of the bottom graphene film array and top-view GNP film. (e) Photograph of completed SA-mimicking sensor array as bottom panel.

To realize the sense of touch by natural skin, it is important tactile sensors by emulating the material, structural, and
to mimic the sensing properties of sensory receptors in human functional properties of human skin.1,8−10,13−18 However, the
skin, wherein the receptors encode tactile information as a time reported sensors are mostly focused on sensitive pressure
interval between voltage spikes of action potentials. Specifi- sensing. Meanwhile, the capability to detect high-frequency
cally, there are four types of mechanoreceptors that perceive vibrations has been limited because of the inherently slow
innocuous mechanical stimuli (pressure and vibration) by the responses of the sensors as mostly polymeric materials5,8 have
rate of adaptation, respectively: two types of slow adaptive been adopted to achieve high sensitivity to pressure. Pressure
receptors (SA-I and SA-II) and two of fast adaptive receptors sensing only is not sufficient for achieving the function of
(FA-I and FA-II).2 The SA receptors sensitively respond to human tactile perception because texture recognition in human
static pressure to detect responses to sustained physical stimuli, perception is a complex process including the sensitive
perceiving high-resolution pressure or pressure distribution detection of both pressure and vibration.
information and skin stretching. In contrast, the FA receptors Human beings mainly perceive surface texture by detecting
preferentially respond to dynamic pressures or vibrations; thus, an interacting vibration induced by the roughness of the object
it is crucial to recognize their texture discrimination. Efforts being touched on the skin through FA mechanoreceptors
were made to achieve artificial mechanoreceptor-mimicking (Pacinian and Meissner) in the skin that are sensitively
3306 DOI: 10.1021/acs.nanolett.9b00922
Nano Lett. 2019, 19, 3305−3312
Nano Letters Letter

Figure 2. Pressure- and vibration-sensing performances. (a) Pressure sensitivity with statistically applied pressures in the range of 100 to 100 000
Pa. The pressure sensitivity (Sp = (ΔI/I0)/ΔP) was estimated as Sp1 = 1.63 kPa−1 from 0.1 to 6 kPa and as Sp2 = 0.04 kPa−1 from 6 to 100 kPa. (b)
The operating mechanism of the SA-mimicking sensor in the pressure range of 0 to 100 kPa. (c) Minimum pressure detection limit with application
of a weight of 4.8 mN on an area of 1 cm2. (d) Response time (<3 ms) for restoration with pressure-releasing profile. (e) Pressure distribution
detection. (e(i)) Photograph of a SA sensor array with a doughnut-shaped magnetic toy of ∼6.2 g and ∼25.45 mm2 (weight and area, respectively)
corresponding to a vertical pressure of ∼2.4 kPa. (e(ii)) The two-dimensional spatial distribution profile of pressure. (f) Open-circuit voltage and
(g) short-circuit current outputs of the TENG bottom panel with a dynamical vertical pressure of 51 kPa on a contact area of ∼4 cm2. (h) Turn on
of 100 LEDs with press-and-release operations of the TENG device.

triggered by dynamical pressure or vibration, although SA1 Here, we propose a self-powered flexible neural tactile sensor
afferents, which densely innervate the fingertip skin, respond (NTS) that mimics all the functions of human finger skin and
strongly to coarse textures by spatial patterns of activation that is selectively and sensitively activated to either pressure or
across its population but only weakly.19 Piezoelectric materials vibration stimulus with laminated-independent sensor ele-
have been proposed as sensors for recognizing textures with ments. A bottom panel consisting of a high-density pressure
vibration-sensitive characteristics because of a change in sensor array (20 × 20 pixels on 4 cm2) of interlocked
magnitude or in the number of dipoles at the molecular level percolative graphene films was fabricated to detect pressure
by introducing certain compounds such as ZnO and BaTiO3, and its distribution, like SA receptors do. As the top panel, for
a FA receptors-mimicking sensor, a triboelectric nanogenerator
and organic materials such as polyvinylidene fluoride
(TENG) was laminated on the bottom panel to detect high-
(PVDF).20−22 Moreover, their output signals are very similar
frequency vibration as well as produce electric power by itself.
to neural action potential signals for FA mechanoreceptors.
Importantly, each output signal for the SA- and FA-mimicking
However, they are limited for applications as skin-mimicking sensors was very similar to the real output signals produced by
sensors because of restricted material versatility. To realize the SA and FA mechanoreceptors in human skin, thus making it
self-powered artificial mechanoreceptor sensor, their output easy to convert the sensor signals into neural signals that can
power is also low to drive electronic sensor devices. Therefore, be perceived by humans. In addition, to mimic structural and
it would be very advantageous for texture perception devices if functional properties of a human fingerprint, interacting
the sensor device showed higher output signals and could be pressure or vibration signals from the contacted surface texture
applied for all materials while maintaining both vibration were amplified by introducing microline patterns on the top
detection ability and a neural spike-like waveform. surface of the NTS. Consequently, the integrated NTS device
3307 DOI: 10.1021/acs.nanolett.9b00922
Nano Lett. 2019, 19, 3305−3312
Nano Letters Letter

was capable of classifying 12 fabrics possessing complex individual GNP sheets inside the film spatially interacted well
patterns with 99.1% classification accuracy. at an appropriate distance (Figure 1d(iii)). The conductivity of
Figure 1a shows human finger skin, which is composed of the resulting GNP film was ∼119 S/m with 3 min spray
two layers, epidermis and dermis. The epidermis is the outer coating. After placing the upper panel onto the lower panel
layer of skin and provides a barrier against damage from without applying an external force, the interlocked array of
external environments. In particular, the frictional ridge of the GNP film was completed. With such a structure, a tiny
epidermis, called the fingerprint, serves to amplify vibrations displacement between the spatially distributed interlocked
triggered when the finger rubs on an even surface, thus GNP sheet films in response to the position- and time-
allowing transmission of the sensory signals involved in fine dependent external pressure could be sensitively detected as a
texture recognition.23 The dermis is occupied by cutaneous change in the electrical current with a response time delay
mechanoreceptors that can detect a variety of tactile stimuli. during deformation and recovery. In addition, the highly
Specifically, SA mechanoreceptors sensitively react to static hierarchical GNP sheets showed hydrophobicity at high
pressure in a general pressure range (0.1 to 100 kPa), and FA contact angles (>140°) (inset images of Figure 1d(iii)). Figure
mechanoreceptors are activated by dynamic pressure and 1e shows the completed flexible SA-mimicking sensor array
vibrations (50−400 Hz).2 Meanwhile, the mechanoreceptors fabricated by a simple solution process of GNP suspension.
encode tactile information as the time between voltage spikes The TENG sensor was fabricated as the top panel by using
of action potential. The SA receptors exhibit the maintained PEN and polytetrafluoroethylene (PTFE) surfaces for friction
firing of action potentials during sustained indentation, with an acryl bumper (∼1 mm height) to isolate the surfaces
whereas the FA receptors show the firing of action potentials without applied force and was laminated on the top surface of
only at the initial and final contacts of a mechanical stimulus the graphene sensor array.
(upper side of Figure 1c).2 Inspired by the sensory system and The piezoresistive response of the SA-mimicking graphene
fingerprint of human finger skin, we here developed artificial force sensor array was first investigated for vertical pressures.
NTSs by assembling a SA-mimicking interlocked percolative Figure 2a shows the response of the sensor under a statically
graphene sensor (bottom panel) and an FA-mimicking TENG applied pressure in the range of 100−100 000 Pa, which
sensor (top panel) with fingerprint-inspired periodic micro- corresponds to the general pressure detection range of human
lines (Figure 1b). The bottom graphene film sensor array of tactile perception.24,25 The applied pressure induces an
400 pixels can measure not only static pressure with a high enhancement in the interlocking contact between the upper
sensitivity in the pressure range of human tactile perception and lower GNP sheet films, resulting in an increase in the
(1.63 kPa−1 in the range of 0−6 kPa) but also the pressure current (I) from the initial current (I0). The pressure
distribution with a high spatial resolution (100 pixels per 1 sensitivity (Sp) is defined as Sp = (ΔI/I0)/ΔP, where ΔI is
cm2). The top TENG sensor can sensitively detect dynamic the change in current (I − I0) in response to the change in the
pressure or vibration with single-sensor architecture. In applied vertical pressure (ΔP). The sensitivity curve does not
particular, the TENG sensor can perceive surface texture change linearly with applied pressure (Figure 2a). In the
through a combination with the microline patterns. In pressure range below 6 kPa, the sensitivity (Sp1) was estimated
addition, the TENG can supply electric power at the NTS to be 1.63 kPa−1 through a linear fit. When the vertical pressure
system itself, thus enabling a self-power system to be achieved. increased from 6 to 100 kPa, the sensitivity (Sp2) rapidly
Importantly, the SA-mimicking graphene-based piezoresistive decreased to 0.04 kPa−1, a much smaller difference than that
sensor and the FA-mimicking TENG devices have firing observed in the lower pressure range. The change in sensitivity
outputs that emulate SA and FA mechanoreceptors of human with applied pressures can be attributed to different operating
skin, respectively (lower side of Figure 1c). In a graphene mechanisms of the interlocked percolative GNP film sensor
sensor, different levels of resistance in response to pressure are (Figure 2b). In the low-pressure range, the upper and the
easily converted to frequency rate of action potentials like SA lower GNP sheet films undergo a mechanical and electrical
receptors do. Likewise, the TENG sensor outputs an electrical contact stage with initial interlocking contact, leading to a
voltage/current only at the initial and final contacts of a rapid decrease in resistance. Once the electrical contact is
mechanical stimulus like FA receptors do. This allows our NTS stabilized at a certain pressure close to 6 kPa (thus forming a
device to be applied to skin electronics, including a human− single GNP film with the combination of the upper and lower
machine interface, prosthetics, and soft robotics, without a films), the value of Sp2 increases slowly because of the
complex conversion system. increased electrical conductivity of the GNP percolation
To fabricate an interlocked percolative graphene sensor networks. The sensor exhibits a pressure detection limit
array (20 × 20 pixels on 4 cm2), we used a facile spray coating (∼48 Pa), which was applied by a weight (4.8 mN) on an area
method using a graphene suspension of nanoplatelets (GNPs) of 1 cm2 (Figure 2c). This detection limit is attributed to the
on a flexible polyethylene naphthalate (PEN) substrate, as force needed to achieve the initial interlocking contact between
shown in Figure 1d. To construct the interlocked architecture the upper and lower films.
of the GNP film, we fabricated two independent films by The response time was confirmed through the change in the
coating the GNPs with an aligned stencil mask on patterned current response of the interlocked sensor accompanying a
platinum electrodes (Figure 1d(i)). The coating density and releasing operation where the measurement interval was 1 ms
thickness of the GNP film was controlled by the processing (Figure 2d). The response time was estimated to be less than 3
time of the GNP suspension (1.5 mg/mL) with a constant flow ms for restoration, indicating that the sensor can detect
rate (∼8 mL/min) (see Supporting Information (SI) Figure vibrating pressures up to ∼330 Hz with negligible signal loss.
S1). The facing GNP films interacted slightly electrically and This is comparable to the frequency range (<400 Hz) of
morphemically (Figure 1d(ii)). The GNP film was coated on human perception for detection of vibrations by Pacinian
the electrode pad (0.4 mm × 0.4 mm) through a fine corpuscles, which is known as the most sensitive detection of
alignment with uniform distribution of film morphology, and high-frequency vibration.1,2 SI Figure S2 shows the sensitive
3308 DOI: 10.1021/acs.nanolett.9b00922
Nano Lett. 2019, 19, 3305−3312
Nano Letters Letter

Figure 3. Selective and sensitive perception of surface texture. (a) Schematic illustration showing detection of interacting pressure or vibration
induced by physical interaction between the surface texture and the microline on a NTS device during rubbing motion. Texture-dependent (i)
electrical and (ii) frequency response signals of (b) the SA-mimicking sensor and (c) the FA mimicking sensor, respectively. The scanning speed is
24 mm/s in both forward and backward directions. Frequency responses in SNRs with different (d) FPS wavelengths and (e) scan velocities for
SA- and FA-mimicking sensors. (f) Schematic illustration showing interaction between sensor and fabric that has complex and fine roughness
patterns to perceive textures. (g) SEM images of the testing fabrics with different surface textures. (h) Results of classification for textures of 12
fabrics in 3 different classification cases performed using only SA sensor signals, only FA sensor signals, and combined SA and FA signals.

and reliable piezoresistive responses to dynamic vertical 100/cm2) of SA mechanoreceptors in the human finger skin.26
pressures of 17−85 kPa. In addition, the sensor showed highly The spatial recognition caused by a doughnut-shaped magnetic
consistent piezoresistive responses for repetitive operation of toy placed at the center of a sensor array is shown in Figure 2e.
10 000 loading−unloading cycles with the applied vertical The weight and area of contact on the array were ∼6.2 g and
pressure of 10 and 100 kPa (SI Figure S3). For bending strain, ∼25.45 mm2, respectively, corresponding to a vertical pressure
the sensor presented rapid current increases up to a bending of ∼2.4 kPa. Interestingly, the sensor array can distinguish the
curvature (κ) of 0.04 but a saturated behavior after that, pressure distribution with relative accuracy.
indicating a similar operating mechanism with operation under The self-powered electrical output performance of the FA-
vertical pressure (SI Figure S4). mimicking TENG sensor using PEN and PTFE as frictional
The fabrication method of the SA-mimicking graphene surfaces was investigated. The vertical pressure (51 kPa) was
sensor is simple, reproducible, and compatible with conven- commonly applied on the TENG with a contact area of ∼4
tional photolithography, and integration of the sensors can be cm2 and was instantly released with insignificant time delay.
easily accomplished. In contrast, it is very difficult to achieve a The open-circuit voltage and short-circuit current outputs of
highly integrated sensor array in polymeric materials-based the TENG are shown in Figure 2f,g, respectively. The TENG
pressure sensors. A flexible sensor matrix (20 × 20 pixels) was shows an average output voltage and current of ∼240 V and
produced on an area of 2 × 2 cm2, indicating an integration ∼37 μA (corresponding to a current density of 9.3 μA/cm2),
density of 100 sensors per cm.2 This highly integrated flexible which is comparable to previously reported values.27,28 The SI
pressure sensor array was hard to achieve until recently and is Figure S5 shows the output voltages with different pressure
comparable to the spatial density (1 mm corresponding to inputs ranging from 17 to 85 kPa. The output signals showed
3309 DOI: 10.1021/acs.nanolett.9b00922
Nano Lett. 2019, 19, 3305−3312
Nano Letters Letter

some pressure dependence but it was not large when compared sensor induces enhancement of the GNP interaction in the
with the SA-mimicking graphene sensor. Note that the TENG film, and the FA-mimicking sensor drives the contact of the
was selectively sensitive not to pressure but to vibration inputs, facing dielectric surfaces when they interact with roughness
mimicking the FA mechanoreceptors. The frequency responses texture on the sensor. With fast Fourier transfer (FFT) by
of the TENG sensor present a specific response curve which is Matlab to obtain the frequency response ( f = v/L), the
caused by the resonance frequency (∼20 Hz) of the TENG (SI periodicity information on the roughness patterns, that is, the
Figure S6).29 The SI Figure S7 presents the voltage versus time surface texture, is recognized. The NTS device was rubbed on
curve obtained by connecting the TENG to a power charging the single-ridge texture to measure the interacting pressure and
capacitor with a rectifying bride circuit. A pressure of 51 kPa vibration with a constant speed (24 mm/s). The changes in
was applied on the TENG with an operating frequency of 4 Hz electrical response because of contact with the sensors at the
to charge the capacitor during 50 s. The result indicates scanning speed of 24 mm/s (v) in both directions on the single
excellent charging performance with a low charging leakage texture are clearly identified by the periodic wave packet over
under repeated loading−unloading forces. In addition, the time (see Figure 3b(i) and c(i)). This periodic response is
electric power delivered from the TENG during the press-and- sufficient for obtaining geometrical information about the FPS
release operations was sufficient to directly turn on 100 light- through FFT analysis (Figure 3b(ii) and c(ii)). The periodic
emitting diodes (LEDs) brightly without charging capacitors pattern of the FPS (L = 600 μm) is represented at the peak
(Figure 2h). position (f 0 ∼ 40 Hz) as L = v/f 0. The complete periodic
Using the integrated NTS device, we finally demonstrated output is attributed to the ability of our NTS device to detect
surface texture recognition. Human beings perceive surface the interacting pressure and vibration produced by a local,
texture using the SA1 receptor (Merkel disk) and FA receptors subtle pressure change with a high signal-to-noise ratio (SNR).
(Meissner and Pacinian corpuscles). Specifically, touch by For a sensible perception of surface texture, geometrical
fingertip generates the most sensitive tactile feeling because features such as the wavelength and the amplitude of the
mechanoreceptors are more densely distributed in this area contact surface roughness are critical factors. The NTS sensor
(∼240 per cm2) than in the palm area (∼60 per cm2 for can perceive periodic distances of the ridge patterns up to 100
adults).30 It has been proven that SA1 afferents, which densely μm in the FPS while maintaining relatively high SNRs (>3 dB)
innervate the fingertip skin, respond strongly to coarse textures in the power spectrum (Figure 3d). Specifically, the SA-
by spatial patterns of activation across the population, but mimicking sensor showed a reduction in the SNR of signals
respond only weakly or not at all to finer patterns, whereas FA with decreasing roughness wavelength under a fine texture
corpuscles discern fine textures through features mediated by level (<1 mm) because of its selective and sensitive response to
transduction and processing of vibrations produced on the skin coarse textures (>1 mm), whereas the FA-mimicking sensor
during scanning.19 This fact indicates that a complex can detect fine textures without insignificant degradation in the
combination of output signals from SA and FA sensors SNR for signals over 8 dB. The frequency response of SNR
provides texture-dependent characteristic clues: coarse textural versus scan speed was investigated (Figure 3e). The frequency
features by spatial encoding arising from geometrical properties response of the SA-mimicking sensor strongly depended on
with induced pressure and fine textural features by vibrotactile scan speed because at a faster speed, the sensor experiences
encoding with interacting vibrations. incomplete restoration of the electrical response. In contrast,
The NTS device with periodic microlines (300 μm width the FA-mimicking sensor maintained high frequency responses
and intervals, 300−500 μm height) mimics fingerprint in the range of 1 to 100 mm/s, which corresponds to the
structure (FPS) scans on single-ridge texture (100 μm width general range of rubbing speed for human tactile perception.
and height) to detect interacting pressure or vibration, as Such sensing capability contributed to the vibration-sensitive
shown in the schematic in Figure 3a. The pressure and characteristic of the FA-mimicking sensor (TENG), wherein
vibration are directly transferred to the SA and FA sensors the sensor outputs electrical voltage/current only at the initial
underneath the individual FPS by a frictional shear force and final contacts of a mechanical stimulus. The SA and FA
induced by rubbing contact in the in-plane direction between sensors with a higher aspect ratio (AR) of microlines exhibited
the single ridge and the microlines on the sensor with a higher output signals in SNR, indicating that a higher AR
slipping motion.31 As demonstrated previously (Figure 2 and sensor is more effective for texture distinction (SI Figure S8).
Figure 3), our NTS device is designed so that the SA and FA Unlike periodic roughness patterns, a new tool is required
sensors respond selectively and sensitively to pressure and for quantifying the surface texture of intricately textured
vibration, respectively. Moreover, the FA sensors can provide materials, because these mostly contain irregular patterns
the electrical power in the sensor system. The fingerprint at the including roughness that is aperiodic in amplitude and spatial
fingertips effectively generates and amplifies the pressure and intervals. To analyze and classify the waveform of the
vibration induced by the interacting surface during physical interacting vibrations induced by the physical interaction
interaction with the touched object. The operation of the between our NTS and complex texture patterns (Figure 3f), we
sensors combined with FPS for recognition of a surface texture used a deep learning technique (MatLab 2017b version, Neural
with rubbing motion is analogous to human tactile perception Network Pattern Recognition Toolbox) that has been widely
by touching and rubbing the surface of an object (Figure 3b,c). adopted for the recognition of various patterns in fields
The rubbing of a textured object having a roughness pattern (L including speech and visual image recognition.32,33 Twelve
= width + pitch) with a rubbing velocity (v) on a periodic FPS different fabrics were tested to classify the texture patterns. The
pattern on the sensors generates time-dependent electrical SEM images of the testing fabrics with different surface
output signals by the interacting pressure and vibration textures show they are delicate enough to be distinguishable
between the FPS and the surface roughness. There is no (Figure 3g). With consecutive back-and-forth scanning with
electrical crosstalk between the SA and FA sensors because gentle contact of the testing fabrics on the sensor (∼10 kPa),
they are connected by dielectric film. The SA-mimicking electrical output signals were collected during 1 h with a
3310 DOI: 10.1021/acs.nanolett.9b00922
Nano Lett. 2019, 19, 3305−3312
Nano Letters Letter

sampling frequency of 1 kHz by rubbing the fabrics on the Author Contributions

NTS device with a scanning velocity of 24 mm/s. The textures S.C. and C.C. conceived this work and developed the design of
of fabrics were classified in three different cases: (1) using only the NTS device. S.C. and W.S. fabricated the NTS devices and
SA sensor signals, (2) using only FA sensor signals, and (3) S.C. and H.K. measured their properties. S.K.L. and C.P.
using combined SA and FA signals (Figure 3h). The results analyzed the experimental results. S.C. and C.C. wrote the
indicated that using only the electrical texture information paper. All authors reviewed the manuscript and provided
obtained by our device, our NTS sensor generally classified the feedback.
fabrics with an accuracy of over 92%. This was enabled by rich Notes
amplitude information on roughness with pressure responsivity The authors declare no competing financial interest.

■
from the SA sensor and a wealth of spatial interval information
on roughness with vibration responsivity from the FA sensor. ACKNOWLEDGMENTS
This implies that our sensor elements are sensitive enough to This work was supported by the Basic Science Research
recognize textures including fine and complex patterns, Program through the National Research Foundation of Korea
because of the ability to sensitively detect fine pressure or funded by the Ministry of Education (NRF-
vibration differences. With a combination of SA and FA sensor 2017R1A6A3A04004987 and 2018R1A6A3A01011866). This
signals, our NTS device successfully classified the fabrics with work was also supported by the DGIST R&D Program (18-
99.1% classification accuracy, quite comparable to human NT-02) of the Ministry of Science, ICT and Future Planning.

■
tactile perception ability.
In conclusion, we report a self-powered flexible neural tactile REFERENCES
sensor (NTS) mimicking human finger skin obtained by
(1) Chortos, A.; Liu, J.; Bao, Z. Nat. Mater. 2016, 15, 937−950.
complementing a high-density (100 pixels per cm2) pressure (2) Abraira, V. E.; Ginty, D. D. Neuron 2013, 79, 618−639.
sensor array of interlocked percolative graphene films and a (3) Someya, T.; Sekitani, T.; Iba, S.; Kato, Y.; Kawaguchi, H.;
self-powered triboelectric nanogenerator with a thin-film Sakurai, T. Proc. Natl. Acad. Sci. U. S. A. 2004, 101, 9966−9970.
structure. Our NTS device selectively enables sensitive (4) Yokota, T.; Zalar, P.; Kaltenbrunner, M.; Jinno, H.; Matsuhisa,
detection of pressure and vibration like real human skin N.; Kitanosako, H.; Tachibana, Y.; Yukita, W.; Koizumi, M.; Someya,
does. All signal outputs respectively produced by the SA and T. Sci. Adv. 2016, 2, e1501856.
FA mechanoreceptors are very similar to human neural spike (5) Gong, S.; Schwalb, W.; Wang, Y.; Chen, Y.; Tang, Y.; Si, J.;
signals. Using this characteristic, we successfully demonstrated Shirinzadeh, B.; Cheng, W. Nat. Commun. 2014, 5, 3132.
an excellent texture recognition capability of 12 fine and (6) Jason, N. N.; Wang, S. J.; Bhanushali, S.; Cheng, W. Nanoscale
complex fabrics. Our approach has enabled new possibilities 2016, 8, 16596−16605.
(7) Bao, Z.; Chen, X. Adv. Mater. 2016, 28, 4177−4179.
for skin electronics mimicking human finger skin for humanoid (8) Zhu, B.; Niu, Z.; Wang, H.; Leow, W. R.; Wang, H.; Li, Y.;
soft robots, artificial prosthetics, human−machine interactions, Zheng, L.; Wei, J.; Huo, F.; Chen, X. Small 2014, 10, 3625−3631.
and medical applications. (9) Jin, M. L.; Park, S.; Lee, Y.; Lee, J. H.; Chung, J.; Kim, J. S.; Kim,

■
*
ASSOCIATED CONTENT
S Supporting Information
J.-S.; Kim, S. Y.; Jee, E.; Kim, D. W.; Chung, J. W.; Lee, S. G.; Choi,
D.; Jung, H.-T.; Kim, D. H. Adv. Mater. 2017, 29, 1605973.
(10) Chun, K.-Y.; Son, Y. J.; Jeon, E.-S.; Lee, S.; Han, C.-S. Adv.
Mater. 2018, 30, 1706299.
The Supporting Information is available free of charge on the (11) Hua, Q.; Sun, J.; Liu, H.; Bao, R.; Yu, R.; Zhai, J.; Pan, C.;
ACS Publications website at DOI: 10.1021/acs.nano- Wang, Z. L. Nat. Commun. 2018, 9, 244.
lett.9b00922. (12) Lee, Y.; Park, J.; Cho, S.; Shin, Y.-E.; Lee, H.; Kim, J.; Myoung,
J.; Cho, S.; Kang, S.; Baig, C.; Ko, H. ACS Nano 2018, 12, 4045−
Fabrication of NTS devices, deep learning for tactile 4054.
classification, preparation of GNP suspension, electrical (13) Park, J.; Kim, M.; Lee, Y.; Lee, H. S.; Ko, H. Sci. Adv. 2015, 1,
response measurements of an interlocked percolative e1500661.
graphene sensor array and a TENG device, measure- (14) Tee, B. C.-K.; Chortos, A.; Berndt, A.; Nguyen, A. K.; Tom, A.;
ments of interacting pressure or vibration by rubbing McGuire, A.; Lin, Z. C.; Tien, K.; Bae, W.-G.; Wang, H.; Mei, P.;
motion, contact angle measurements, thickness of the Chou, H.-H.; Cui, B.; Deisseroth, K.; Ng, T. N.; Bao, Z. Science 2015,
GNP film, sensitive and reliable piezoresistive responses 350, 313−316.
(15) Darabi, M. A.; Khosrozadeh, A.; Mbeleck, R.; Liu, Y.; Chang,
to dynamic pressures, consistent piezoresistive re- Q.; Jiang, J.; Cai, J.; Wang, Q.; Luo, G.; Xing, M. Adv. Mater. 2017,
sponses, electrical responses as a function of bending 29, 1700533.
curvature, output voltages with different pressure inputs, (16) Zhao, S.; Zhu, R. Adv. Mater. 2017, 29, 1606151.
voltage versus time curve, SNR in frequency responses (17) Ho, D. H.; Sun, Q.; Kim, S. Y.; Han, J. T.; Kim, D. H.; Cho, J.
depending on the aspect ratio of microlines for SA- and H. Adv. Mater. 2016, 28, 2601−2608.
FA-mimicking sensors (PDF) (18) Lei, Z.; Wang, Q.; Wu, P. Mater. Horiz. 2017, 4, 694−700.
(19) Weber, A. I.; Saal, H. P.; Lieber, J. D.; Cheng, J.-W.; Manfredi,

■ AUTHOR INFORMATION
Corresponding Author
L. R.; Dammann, J. F., III; Bensmaia, S. J. Proc. Natl. Acad. Sci. U. S. A.
2013, 110, 17107−17112.
(20) Liao, Z.; Liu, W.; Wu, Y.; Zhang, C.; Zhang, Y.; Wang, X.; Li, X.
Nanoscale 2015, 7, 10801−10806.
*E-mail: [email protected]. (21) Zhuang, Y.; Xu, Z.; Fu, X.; Li, F.; Li, J.; Liao, Z.; Liu, W.
ORCID Semicond. Sci. Technol. 2017, 32, 074002.
(22) Hu, H.; Han, Y.; Song, A.; Chen, S.; Wang, C.; Wang, Z.
Sang Kyoo Lim: 0000-0002-8971-8232 Sensors 2014, 14, 4899−4913.
Changhyun Pang: 0000-0001-8339-7880 (23) Scheibert, J.; Leurent, S.; Prevost, A.; Debregeas, G. Science
Changsoon Choi: 0000-0003-4456-4548 2009, 323, 1503−1506.

3311 DOI: 10.1021/acs.nanolett.9b00922

Nano Lett. 2019, 19, 3305−3312
Nano Letters Letter

(24) Mannsfeld, S. C. B.; Tee, B. C-K.; Stoltenberg, R. M.; Chen, C.

V. H-H.; Barman, S.; Muir, B. V. O.; Sokolov, A. N.; Reese, C.; Bao,
Z. Nat. Mater. 2010, 9, 859−864.
(25) Chun, S.; Kim, Y.; Jung, H.; Park, W. Appl. Phys. Lett. 2014,
105, 041907.
(26) Dahiya, R. S.; Metta, G.; Valle, M.; Sandini, G. IEEE Trans.
Robot. 2010, 26, 1−20.
(27) Zhu, G.; Lin, Z.-H.; Jing, Q.; Bai, P.; Pan, C.; Yang, Y.; Zhou,
Y.; Wang, Z. L. Nano Lett. 2013, 13, 847−853.
(28) Wang, Z. L.; Jiang, T.; Xu, L. Nano Energy 2017, 39, 9−23.
(29) Chen, J.; Zhu, G.; Yang, W.; Jing, Q.; Bai, P.; Yang, Y.; Hou, T.-
H.; Wang, Z. L. Adv. Mater. 2013, 25, 6094−6099.
(30) Johansson, R. S.; Vallbo, A. B. J. Physiol. 1979, 286, 283−300.
(31) Chun, S.; Choi, Y.; Suh, D. I.; Bae, G. Y.; Hyun, S.; Park, W.
Nanoscale 2017, 9, 10248−10255.
(32) LeCun, Y.; Bengio, Y.; Hinton, G. Nature 2015, 521, 436−444.
(33) Jordan, M. I.; Mitchell, T. M. Science 2015, 349, 255−260.

3312 DOI: 10.1021/acs.nanolett.9b00922

Nano Lett. 2019, 19, 3305−3312