Highly pixelated, untethered tactile interfaces for an on-skin telehaptic system

For highly immersive telehaptic applications, skin-integrated, untethered, and highly pixelated transducer devices that can record and generate tactile stimuli are required. Here, we propose a skin-conformable tactile sensor and actuator array with high spatial resolution of 1.8 mm for realising untethered tactile communication on human skin. The tactile sensors are designed to exhibit ultra-exibility and bimodal sensitivity to static and dynamic pressure. The actuators are miniaturised to sub-millimetre scale to provide sophisticated, high spatiotemporal resolution tactile feedback over a centimetre square area of the �ngertip with the capacity to generate vibrotactile feedback under an external load of up to 529 kPa. Short time Fourier transform analysis showed that our telehaptic system can transmit various types of tactile stimuli, such as the shape of objects and letters, textures of fabrics, and vibration patterns with high �delity.


Introduction
Interest in virtual interaction has grown rapidly in various elds, including education, healthcare, retail, robotics, and manufacturing [1][2][3][4][5][6][7][8][9] . Performing tasks in a virtual environment can provide indirect but realistic experiences in a low-cost, contactless, and convenient manner 10,11 . Conventional virtual/augmented reality (VR/AR) systems rely mainly on visual and auditory stimuli. The degree of immersion and accuracy in virtual interactions can be signi cantly enhanced by providing tactile feedback [12][13][14][15] . However, the tactile feedback system is at a preliminary stage compared to audio and video technologies. Further improvements in functionalities, such as high-resolution, multimodal feedback, and real-time communication, are required to enable sophisticated tangible interactions in a distant place and/or virtual place. We refer to a tactile feedback system with these features as a telehaptic system.
The ultimate goal of the telehaptic system is to transmit sophisticated and near-realistic tactile sensations from one user to another. The key ability of a telehaptic transducer relies on precisely monitoring the spatiotemporal physical deformations occurring at the soft skin surface and reproducing them in remote places. To impart the capability to sense and actuate tactile information on and to the skin without disrupting natural hand functionality, mechanically exible, lightweight, and untethered designs are required. Although telehaptic transducer devices based on rigid sensors and actuators, such as desktop haptic devices 16,17 , tablet-type surface haptic devices [17][18][19] , and wearable gloves [21][22][23] have already been demonstrated, they still suffer from the constraints of user motion originating from rigid and tethered designs.
To address the mechanical compliance and untethering issue of haptic transducers, tremendous efforts have been made to develop materials and novel structural designs for skin-attachable haptic interfaces.
Improving the spatial resolution of a pixelated actuator array is another major challenge in terms of achieving immersive tactile feedback. Miniaturised coil and magnet-based 32 actuators have been embedded in silicone elastomer and wirelessly operated and powered by near-eld communication (NFC), enabling untethered and skin conformable designs 3 . However, the pixel size of these devices is of the order of centimetres, which is too large for application onto narrow and sensitive areas such as the hand and ngertips. Electrostatic and pneumatic type high-resolution actuators have also been reported.
Flexible hydraulically ampli ed electrostatic actuators have achieved a 6 mm pitch resolution to t at least one actuator on a ngertip 31 . A 4 × 4 pneumatic tactile display with Braille standard resolution (2.5 mm) using a bistable electroactive polymer (BSEP) thin lm has also been reported 33 . However, these devices are limited in their operating speed, originating from the slow hydraulic, pneumatic ow and thermal change.
Piezoelectric actuators are an alternative method to generate high spatiotemporal tactile patterns with low operational burden. High-performance piezoceramic-based actuators have been studied to provide fast response and dynamic movements with accelerations as high as 500 m/s 2 and slew rates in the microsecond range at micrometre displacement. This is due to the very fast piezoelectric ionic shift in and the high stiffness of the materials. 34 Wide frequency range up to several 10 kHz, and capability of miniaturisation are also advantageous for generating high-resolution and crisp haptic response necessary for emulated buttons and textures. 35 However, a major drawback of piezoceramic actuators for on-skin applications is their planar and rigid format. To overcome these limitations, ultrathin Pb[Zr x Ti 1x ]O (PZT) membranes have been utilised as mechanical sensors and actuators with high spatial resolution 36 . However, the thin membrane of PZT leads to insu cient output force for haptic applications, although it was effective for modulus sensing in medical applications. Hence, transmitting tactile stimuli using high-resolution on-skin piezoelectric actuators still remains a key challenge in terms of achieving both mechanical compliance and su cient performance.
Here, we demonstrate a skin-attachable, untethered telehaptic system that consists of an ultra-exible bimodal tactile sensor and a exible tactile actuator array that can acquire and regenerate tactile stimuli patterns with 1.8 mm pitch and 1.55 ms high spatiotemporal resolution. The actuator array is designed to be wirelessly coupled with an ultra-exible bimodal sensor array that consists of a fast adaptive (FA) and slow adaptive (SA) mechanoreceptor-like pressure sensor array that can simultaneously monitor static and dynamic pressures with high spatial resolution. The actuator array can generate sinusoidal vibration patterns as well as near-realistic high-resolution tactile stimuli patterns corresponding to the signals acquired from each sensor pixel. The sensor-actuator interworking system is mechanically exible, lightweight, and wirelessly operated for immersive telehaptics. Finally, the developed system enables realtime transmission of high-de nition tactile stimuli. Various types of spatiotemporal tactile patterns, including textures of fabrics, and pressing, clicking, and apping of butter y wings generated by a load cell and a commercial haptic device (PowerHap TM , TDK Corp.), were acquired using the high-resolution bimodal sensor array. They were then regenerated by the actuator array with high delity, as proven by short-time Fourier transform spectral analysis.

Results And Discussion
Design and fabrication of skin-attachable tactile sensor and actuator array The telehaptic system is designed to transmit tactile stimuli in real time using a wirelessly coupled on-skin tactile sensor and actuator, as shown in Fig. 1. The basic principle of the on-skin telehaptic interface is shown in Fig. 1a. The skin-attached bimodal sensor array detects friction-induced vibrations, impacts, and normal pressure generated during contact between the ngertip and an object. The acquired tactile signals were ampli ed and transmitted to the actuator array using Bluetooth communication. The actuator array generates vibration patterns for rendering the delivered tactile information composed of various frequencies, amplitudes, and spatial distributions. The sensor array is designed to have selective sensitivity to static and dynamic pressure for recording complex tactile signals by mimicking human mechanoreceptors, Merkel's discs and Pacinian corpuscles. For the static pressure sensor, chemically grafted Polypyrrole (Ppy) on elastomeric micro-pyramids (made with PDMS, polydimethylsiloxane) was exploited because of its reproducibility and high sensitivity based on previous studies 37,38 . Unlike many other conductive materials, Ppy can chemically react with the hydroxyl groups on the PDMS surface; therefore, higher durability and repeatability are expected when using Ppy. 37 Using this material, a piezoresistive pressure sensor array is designed into miniaturised sensing pixels with individual pixel area of 1.045 mm 2 .
To monitor the dynamic pressure, the piezoelectric polymer PVDF-TrFE was chosen because of its superior mechanical exibility 39 and process compatibility (Supplementary Figure 1) for a highresolution array. The annealing process was important for the crystallinity and piezoelectric characteristics of PVDF-TrFE thin lm. A 3 μm PVDF-TrFE thin lm was annealed in a vacuum oven at an optimized annealing temperature of 130 ℃ (Supplementary Figure 2). The sensor array contains static and dynamic pressure sensors arranged in a chessboard pattern densely packed into an area of a centimetre square to resolve and monitor the spatial distribution of ne tactile stimuli (Fig. 1b, c). Unlike the previously reported exible multi-layered bimodal sensor arrays 40,41 that have relatively high thickness of 120 μm, our sensor pixels are integrated on a single layer of ultrathin polyimide substrate (~4 μm) to achieve a higher bending compliance, which is bene cial for making a conformal contact with the skin surface 42 . Owing to the ultra-exibility of the sensor array, it can be conformally laminated on a curvilinear ngertip surface, as shown in Fig. 1d, which is bene cial for acquiring reliable tactile information without mechanical distortion at the interface between the skin and the object. The collected tactile information is ampli ed and wirelessly transmitted by the signal processing module located at the back of the hand, realising the untethered operation of the telehaptic system (Fig. 1e).
In the high-resolution tactile actuator array, the actuating pixels are fabricated using miniaturized piezoelectric ceramics with a sub-millimetre length scale (0.98 mm), and these actuators are assembled into a 32 channel-array on a exible polyimide substrate. Because piezoelectric actuators are processed into 32 layers using standard piezo-ceramic manufacturing techniques, the actuators are rigid. However, since they are mounted on a exible substrate via a surface mounting process (i.e. using anisotropic conductive lm (ACF) bonding process (Supplementary Figure 3)), a exible and skin-conformable actuator array can be fabricated. Thereafter, actuator array can be embedded into elastomer by pouring PDMS on top, which acts as a protection layer and as vibration isolator. Owing to its high piezoelectric coe cient (47525 pC/N in average) and small pixel size (< 1 mm 2 ), piezo-ceramic actuators can generate spatially distributed tactile stimuli patterns with high resolution (1.8 mm pitch) over a wide dynamic frequency range (1 Hz-1 kHz) (Fig. 1f, g). The tactile actuator array is designed to have mechanical exibility to directly adhere onto a ngertip and deliver spatially segmented actuation on the curved surface (Fig. 1h). The acquired tactile signals from the sensor array are wirelessly transmitted to the actuator array's drive module located on the back of the hand (Fig. 1i) that supplies up to 40 V pp for each actuator pixel.
Detection of static/dynamic pressure for acquiring tactile stimuli The ultra-exible bimodal (UFB) tactile sensor array can detect static and dynamic pressures simultaneously and selectively. Fig. 2a, b show the working mechanisms of the piezoresistive static pressure sensor and the piezoelectric dynamic pressure sensor, respectively. For the piezoresistive pressure sensor, the Ppy-coated micropyramids are placed on top of the interdigitated electrodes. When pressure is applied to the sensor, the tip of the pyramid is deformed from a triangular shape to that of a trapezoid, and no signi cant cracking was observed on the surface of Ppy even under 40 kPa of pressure, as shown in Supplementary Figure 1c. The contact area between the electrode and the conductive pyramid increases as the pressure increases, resulting in a decrease in the contact resistance (R c ) ( Fig.   2a). As applied pressure increases from 3 kPa to 80 kPa, the contact resistance was observed to continuously decrease from 1 MΩ to 100 kΩ. The highest value of sensitivity is 13.93 kPa -1 in the lowpressure range of 3-10 kPa, and the sensitivity decreases to 4.65 kPa -1 (10-30 kPa) and 0.32 kPa -1 (30-80 kPa) in the high-pressure range, as shown in Fig. 2c. Over the entire sensing range of 3-80 kPa (which is the recognisable pressure range of human touch 43 ), our sensor is capable of differentiating a pressure change of <1 kPa.
For the piezoelectric dynamic pressure sensors, the PVDF-TrFE lm was patterned and the mechanism of electric charge generation by the piezoelectric effect was exploited. The sensor provides a fast response and a wide range of frequency detection when dynamic vibration is applied. It is shown that the rising time of the patterned dynamic pressure sensor is 250 µs as shown in Fig. 2d. The human tactile response time is 385 µs 44 , which implies that the sensor has a su ciently fast response and is capable of precisely capturing tactile stimuli. The fast response is important because unwanted effects similar to motion sickness may occur if the time interval between the visual and tactile feedback exceeds 1 ms 43 . Furthermore, vibration can be detected with a su cient average SNR of 15.4 dB (=10logS/N) in the frequency range from 1 Hz to 1 kHz, as shown in Fig. 2e and Supplementary Figure 4. The decrease in the output current in the high frequency range is attributed to the inherent frequency dependence of the vibrator that is applied to the bimodal sensor, showing a reduced force with increasing frequency.
To evaluate the simultaneous sensing of static and dynamic pressure, a static pressure change and a 200 Hz vibration were applied to the neighbouring static and dynamic sensor pixels by repeatedly pressing the vibrator mounted on the sensor as shown in Fig. 2f. Static pressure changes were measured in terms of the voltage across the reference resistor in a voltage divider circuit as shown by the red and orange lines in Figs. 2g, h. The vibrations detected by the dynamic sensor pixel is indicated by the green line in Figs. 2g, h. Since the centre of the vibrator is closer to the static-L, the amplitude of static-L is larger than that of static-R. There is a limitation in detection of static pressure and vibration simultaneously with a unimodal sensor. For instance, static pressure can be estimated from the signal envelop of dynamic sensor. But it is hard to precisely detect the absolute magnitude of the static pressure as the amplitude of dynamic sensor can be distorted by the in uence of the triboelectric effect and capacitance coupling. Therefore, a bimodal sensor capable of both static and dynamic pressure sensing is required to detect a wide range of complex tactile stimuli.
The mechanical compliance of the ultra-exible and ultra-thin bimodal sensor enables conformable attachment onto the soft skin surface, as shown in Fig. 2i. The ultra-thin sensor is also advantageous for enabling natural deformation of skin upon application of small area pressure (i.e. thick sensors have higher stiffness preventing natural deformation of skin). To verify the mechanical compliance of the ultra-exible bimodal sensor, the compressive strain-stress curve of a exible and ultra-exible substrate laminated on a silicone elastomer was measured. The exible substrate is a 125 μm thick PI lm used for industrial standard FPCB, and the ultra-exible substrate is 4 μm-thick PI lm. The strain-stress characteristics of the silicone elastomer (Eco-ex 00-30, modulus: 210 kPa (measured) and 100 kPa (reference) 45 ), which has a Young's modulus similar to that of human skin (epidermis modulus: 140 to 600 kPa 46 ), was measured for reference, and each sample loaded on the elastomer was pressed by a load cell, as shown in Fig. 2j. The stresses of the exible sensor, ultra-exible sensor, and silicone elastomer were respectively 44, 24 and 21 kPa at 10% strain, as shown in the compressive strain-stress curve of Fig. 2k. This result con rms that the ultra-exible sensor would only marginally affect the modulus of skin-like soft elastomers.
Sub-millimetre scale, high-resolution actuator array for generating vibrotactile stimuli A single sub-mm actuator has a size of 0.980.980.8 mm (widthlengthheight) and average weight of 6.7 mg. It is a multi-layered piezoceramic actuator in which actuation displacement and output force is proportional to the piezoelectric coe cient and number of stacked ceramic layers ( Fig. 3c and Supplementary Figure 5). As shown in Fig. 3a and b, our actuator can stimulate an area of less than 1 mm 2 on the ngertip. The actuators are placed on the contact pad of a 125 μm-thick exible and 4 μmthick ultra-exible substrate, and the electrical and mechanical connections are made by the re ow soldering and ACF transfer process, respectively, in a 1.8 mm pitch chessboard array (Supplementary Figure 6). Because the actuator pixels are embedded with PDMS, the vibration generated in each pixel can be isolated so that it is hardly transmitted to adjacent pixels (Supplementary Figure 7).
The characteristics of the actuator pixels were evaluated under mass and force loading in the operating voltage and frequency range of 5-60 V pp and 100-1000 Hz, respectively. For practical applications, tactile actuators must endure external loads applied by human nger interactions. Our results con rm that our ceramic-based piezoelectric actuator is robust under an external load or contact force, which is a challenging issue for other soft or exible actuators 47 . is the threshold acceleration that humans can feel 49,50 . At 40 and 60 V pp , the acceleration of the actuator with 0.92 g mass loading was 1.19 m/s 2 and 2.04 m/s 2 , respectively, which is 10 times and 20 times higher than the absolute limen. The maximum input voltage was limited to 60 V pp because larger voltages can cause degradation of the actuator performance.  Fig. 3e. Interestingly, the acceleration and displacement were found to increase with the mass and force loading. This phenomenon can be well described by an in uence of applied preload on piezoelectric stack actuator, where strain output is enhanced with an increase in mechanical preload and the maximum strain values can be obtained at an optimum preload 48 . Above 10 V pp , the displacement is larger than the Pacinian threshold (150 nm at 200 Hz) 51,52 which is the displacement threshold. At 40 and 60 V pp operation, the displacement of the actuator under 592 kPa of pressure was 755 nm and 1315 nm, respectively. Since the output displacement characteristics of the actuator was saturated at pressures larger than 592 kPa, external pressures were applied up to 592 kPa which means that it is the optimum preload for the actuator stack and a higher preload starts to impede the overall piezoelectric effect. 48 The sub-mm actuator showed a displacement larger than 970 nm in the frequency range from 100 Hz to 1 kHz. The displacement was found to slightly decrease as the operation frequency increased, as seen in Fig. 3f. The displacement at 1 kHz decreased by 4% from the displacement at 100 Hz, which implies the actuator system can cover up to 1 kHz without signi cant performance degradation.
Generating su cient displacement to exceed the threshold for sensation at frequencies below 10 Hz is important to accurately deliver the static pressure tactile information. However, in this frequency range, since the mechanoreceptors require a large displacement of at least 40 μm for recognition 52 , this is unrealistic for piezoelectric actuators to achieve. To address this fundamental challenge, we introduce amplitude modulation in the form of envelops as depicted in Fig. 3g. Here, carrier frequency of 300 Hz and envelope frequency ranging from 1 Hz to 10 Hz was applied to the actuator since human can easily recognize 300 Hz vibration with threshold displacement of 150 nm, while our actuator can generate much larger displacement of up to 1.3 μm. Fig. 3h shows the FFT (fast Fourier transform) result of modulated signals for imparting the sensation of static pressure. Along with the carrier frequency peak (300 Hz), low frequency peaks at 1, 2, 3, 5, 7, and 10 Hz, are clearly visible in the FFT result.
Wireless tactile acquisition and rendering system for telehaptic communication For realising immersive telehaptic communication, proper designing of signal processing units and wireless connections are crucial for transmitting the complex tactile stimuli with high similarity and low delay. In the proposed telehaptic system, the tactile sensor and actuator were wirelessly coupled by signal processing modules, as shown in Fig. 4. The module for processing and transmitting the sensor signals is composed of a pre-ampli er, microcontroller, and Bluetooth low-energy (BLE), as shown in Fig.  4a, b. Pre-ampli ers for bimodal pressure sensor array have been designed as two different circuits for each sensing type. The static pressure was measured by a voltage divider and a voltage follower circuit to cover wide range of sensor output impedance, as shown in Supplementary Figure 8b. The dynamic pressure was measured by using a high gain voltage-mode piezoelectric ampli er which can amplify pAorder charge generation, as shown in Supplementary Figure 8c. The module for the tactile actuator array is composed of an actuator driver, a microcontroller, and a BLE for generating vibrations in accordance with the transmitted sensor signals. The actuator driver distributed voltage inputs to each actuator pixel with maximum peak-to-peak voltage of 90 V. A LED driver is added to visualise the generated tactile patterns during telehaptic communication (Fig. 4c, d).
The telehaptic system has small delay time (< 1.55 ms), which is advantageous for preventing virtual reality sickness, and high similarity (> 80%) between detected and regenerated complex tactile stimuli, which implies the capability of accurate tactile communication. To measure the delay time of the wireless telehaptic system, a 50 Hz sine wave was applied to the dynamic sensor pixel using a vibrator, and the velocity of the vibration on the actuator was monitored using a 3D laser scanning vibrometer (LSV), during which the sensor signal was being wirelessly transmitted across a distance of 3 m, as shown in Fig. 4e. The delay time was 1.55 ms, which was measured by comparing the signal applied to the sensor with the signal regenerated by the actuator (Fig. 4f, g).
To verify the similarity of the acquired complex tactile signals and the regenerated tactile stimuli, a vibration signal pattern similar with a vibration of motorcycle was transmitted through the telehaptic system, as shown in the blue plot depicting the sensor signal ( Fig. 4 (h)) and the red plot depicting the actuator's vibrational displacement (Fig. 4i). The correlation between the sensor signals and the actuator vibration signals (orange windows of Fig 4h, i, with the area name of S1, A1, S2 and A2) was calculated and shown in the correlation matrix correlogram from the Pearson correlation coe cient (PCC) (Fig. 4j).
The correlation coe cients between S1-A1 and S2-A2 were both 0.8 and 0.81, and the correlation coe cients of signal envelop for S1-A1 and S2-A2 were 0.97 and 0.91, respectively. In other words, when both amplitude and frequency considered, the wireless telehaptic system shows signal similarity of 80 % and 81 % between the acquired and generated tactile stimuli. On the other hand, when considering the amplitude only, our system show higher signal similarity of 97% and 91%. We have also tested the similarity of sine waves with various frequencies (100, 200, 300, 400, and 500 Hz) (Supplementary Figure   12). The correlation coe cient was 0.99, 0.96, 0.99, 0.92, and 0.82 at 100, 200, 300, 400, and 500 Hz, respectively. In other words, the signals reproduced by the actuator shows similarities of more than 90% to the sensor signals, except for 500 Hz. At 500 Hz, the actuator signals included a beat wave with a beat frequency of 7 Hz, which may have originated from the noise in the optical measurement system, i.e., the doppler signal and speckle noise of the laser vibrometer. 53 Demonstration of a wireless telehaptic system Fig. 5a shows a skin-attachable wireless telehaptic system demonstrating the transmission of spatial pressure distribution. Volunteer 1 (left) attaches the UFB tactile sensor on the ngertip of the index nger and rolls a plastic stick on it. The UFB tactile sensor acquires the spatial distribution of the static pressure changes and transmits the signals to the actuator processing module. The tactile information is reproduced by a tactile actuator attached on the ngertip of volunteer 2 (right) and visualised by an LED display placed on the back side of the hand of volunteer 2 (Supplementary movie 1). Fig. 5b shows a magni ed image of the pressed position of the UFB sensor on the nger and the illuminated LED (inset of Fig. 5b). The UFB sensor can detect more complex shapes due to its highly pixelated static pressure sensor array, such as alphabet letters (I, V, U), as shown in Fig. 5c, and the actuator array can regenerate acquired tactile signals, as shown in the LSV scanning image (Fig. 5d, Supplementary video 4). The red colour in Fig. 5d indicates a larger vibrational amplitude.
The UFB tactile sensor can also detect dynamic vibrations that are relevant to texture and motion. The volunteer who attached the UFB tactile sensor rubbed the surface of three different fabrics (Fig. 5e, f,  Supplementary movie 2, 3). To amplify the surface vibration, the PDMS grating pattern was laminated on the surface of the sensor (Supplementary Figure 9). The roughness of the fabric increases with fabric number; for example, fabric 3 is the roughest among the fabrics used in the study (Supplementary Figure  10). To further investigate the correlation between the sensor and actuator signals after transmitting the texture information of the fabrics to the actuator, short-term Fourier transform (STFT) analysis was conducted. Fig. 5g and 5h show the STFT results of the sensor signal and the actuator vibration. The main frequencies (f main ) of the sensor signals were 403, 326, and 229 Hz for fabrics 1, 2, and 3, respectively. The main frequency decreased when the sensor rubbed the rougher surface. 54 Similar to the sensor tendency, the main frequencies of the actuator signals were 400 Hz, 326 Hz, and 229 Hz (Supplementary Figure 11). The STFT patterns of the sensor and actuator signals show strong similarities, which indicates that the vibration feature has been successfully transmitted. Other complicated tactile stimuli such as press, click, and butter y wing ap have been successfully transmitted by the telehaptic system. Here, the pattern of press was generated by a pressure applying system with load cell, while tactile patterns of click and butter y wing ap were generated by using sample waveforms of the vibration actuator (PowerHap TM , TDK Corp, and BOS1901, Boreas technologies). The correlation coe cients of the envelop signals of the sensor and actuator were respectively 0.89, 0.98 and 0.93 for pressing, clicking and butter y wing apping patterns ( Supplementary   Figure 12, 13).

Conclusion
In this study, we demonstrated telehaptic communication with an on-skin platform. Ultra-exible sensor array and miniaturised actuator array enable real-time tangible interactions with various tactile stimuli.
The piezoresistive and piezoelectric hybridised pressure sensor array was used to detect the spatial distribution of static pressure levels and frequency information of vibrotactile stimuli at high resolution. A miniaturised 32-channel actuator array successfully regenerated the detected tactile stimuli signals. Although the miniaturized piezoelectric actuator can su ciently generate high spatiotemporal vibration patterns above the detection limit of the skin, the displacement should be further improved for greater enhancement of tactile feelings, through designing displacement ampli cation mechanisms. The proposed sensor and actuator system can transmit and share tactile sensations remotely, which may open new avenues for tactile-based human-machine interface. The telehaptic platform can enable assisting the users with manual tasks, making people feel physically connected even over long distances, and enriching the feeling of presence in virtual environments like augmented/virtual reality (AR/VR).

Experimental Section
Fabrication of ultra-exible bimodal (UFB) tactile sensor array Ultra-exible backplane The polyimide precursor was spin-coated onto a carrier glass and cured to form a 2 µm-thick base layer. And Ti/Au (10 nm/200 nm) electrode and transmission line were fabricated on the base layer by photolithography, with an interdigitated (IDE) pattern with minimum line and space width of 50 and 100 μm, respectively. And 2 µm-thick PI was fabricated on the device for encapsulation layer and windows of contact electrode are opened by PI wet etching. After the fabrication of ultra-exible backplane, it was peeled off from the carrier glass by laser lift off process with energy density of 180 mJ/cm 2 (KORONA TM LLO, AP systems). The 32-channel sensor pixel array for static pressure sensor is arranged in 1.8 mm pitch and 4 additional pixels for dynamic pressure sensor are added between the pixels. The width and length of the IDE electrode pattern for the static pressure sensor are 0.25 and 1.1 mm, respectively. The gap between the two counter electrodes is 0.45 mm. The electrode size and gap were determined by considering pixel size, sensor sensitivity, and processability. Resistance change is proportional to area of the electrode pad, such that larger width and length are desirable for sensitivity, but the area is limited for high-resolution device fabrication. The gap was chosen to be 0.45 mm for sharing the backplane with actuator array, which is the suitable size for the mounting process. For the dynamic pressure sensor, the gap between the counter electrode was set at 0.1 mm. The electrode pattern for the piezoelectric sensor has a narrower gap than that of the piezoresistive sensor because smaller gap is advantageous for detecting changes in electric eld originated from piezoelectric material under pressure.
Static pressure sensor A microstructured PDMS lm was fabricated using a replica moulding process, i.e., soft photolithographic moulding. A Si wafer was etched using potassium hydroxide (KOH) solution to form micro-pyramid structures with a side wall angle of 54.7°. The PDMS pyramid soft mould was prepared by mechanically mixing the PDMS (Sylgard 184, Dow Corning) base solution and the curing agent (10:1 wt.%). It was cured at 80 °C for 3 h and lifted off the replica. The width and the gap of the micropyramid structure are both 100 μm. The pyramid sheet was prepared by spin-coating on a soft replica at 800 rpm for 30 s and heating in a convection oven at 80 °C. Polypyrrole (Pyrrole (98%), Sigma Aldrich) was coated on the cured 100-µm-thick micro pyramid lm by depositing pyrrole monomer silane (Trideca uoro-1,1,2,2,tetrahydrooctyl trichlorosilane, Sigma Aldrich). The pyrrole was then polymerised in a pyrrole and oxidant solution (iron (III) chloride hexahydrate, Sigma Aldrich) at 25 °C.

Dynamic pressure sensor
The PVDF-TrFE solutions (20 wt.%) were prepared by dissolving 2 g of PVDF-TrFE copolymer pellets (75:25 (mol%), Piezotech) in 2-butanone (MEK, 99 +%, Sigma Aldrich) and acetone (1:1 (wt.%)). The PVDF-TrFE solution was spin-coated at 4000 rpm for 40 s and onto the IDE metal electrode substrate. The solution was sequentially degassed overnight under vacuum to eliminate the bubbles. The PVDF-TrFE lms were crystalline, followed by annealing at 130 ℃ for 3h. After preparation of the PVDF-TrFE lm, a 2-nm/200-nm-thick Ti/Au lm was deposited by E-beam evaporation (KVET-C500200, Korea Vacuum) under a DC eld of 10 kV at 9 × 10 -7 Torr as an etch stop mask for reactive ion etching (RIE). The PVDF-TrFE lm was etched under 10 sccm O 2 and 20 sccm CF 4 gas environments for 30 min with 20 W RF power. It was polarised under a DC eld of 50 kV/mm at 25 °C for 3 h. Finally, the micro-pattern for enhanced texture perception was prepared by the PDMS micromolding process with line width, height, and spacing, each equal to 1 mm.
Fabrication of ultra-exible sub-mm piezoelectric (UFP) tactile actuator array Flexible Sub-mm piezoelectric tactile actuator Array The sub-mm tactile actuator is a piezoelectric ceramic multilayer which is composed of 32 layers of lead zirconate titanate (PZT) ceramic with a thickness of 50 µm. Fabrication process of sub-mm tactile actuator is industrially standardized process of piezoelectric ceramic actuator. Firstly, green sheet of PZT slurry with top Ag electrode is laminated into 32-layer-stack and sintered in high temperature (~1200 ˚C). After the sintering process, the piezoelectric ceramic multilayer is cut into a die of dimensions 0.98 mm 0.98 mm 0.8 mm. Finally, contact electrodes (Ag) were sputtered on both sides of the ceramic block for connecting inner electrodes and on bottom for contact pad (designed by our group and manufactured by WiSOL Co., Ltd). The tactile actuator pixel was bonded together for double stack actuator and electrically connected by silver epoxy. Then, 32 actuator pixels are rmly anchored above the exible printed circuit board (FPCB) by re ow soldering process for exible actuator array. For the ultra-exible actuator array, actuators are mounted on 3 µm-thick PI substrate with carrier glass by using adhesive conductive lm (ACF) transfer process and peeled off from the carrier glass by using above mentioned laser lift off process (LLO).

Electrical characteristics
The static sensitivities of the bimodal pressure sensor were measured by a load cell pressure measurement station, which has a power supply (E3631A, Keysight) with a bias voltage of 2 V, a lownoise current preampli er (SR570, Stanford Research Systems), and a force gauge (PI V-275.431, force range of 0.01 N to 10 N and resolution of 0.001 N (0.01 gf)).
The characteristics of the dynamic pressure sensors were acquired using a low-noise current preampli er (SR570, Stanford Research Systems) and oscilloscope (DPO5054, Tektronics). We used a function generator (33220A, Agilent) and a piezo ampli er (E663, PI) to apply the dynamic frequency.

Mechanical characteristics
The strain-stress measurement setup consisted of a load cell pressure station with a force gauge and the signals were captured using a data acquisition system (DAQ).

Wireless tactile acquisition and rendering system
The tactile sensor and actuator array were operated by a wireless tactile acquisition circuit board (Transmit Data, TX) and a tactile rendering circuit board (Receive Data, RX), respectively. The tactile TX circuit comprises of a pre-ampli er circuit, a microcontroller with Bluetooth communication system, a power management circuit to provide constant voltage, and a battery.
Signal processing and pre-ampli er circuit The sensor signal acquired by the bimodal tactile sensor was ampli ed using a pre-ampli er circuit. A voltage mode ampli er (AD8643, Analog Devices) circuit was used for the dynamic pressure sensor, and a voltage divider and follower circuit were used for the static pressure sensor.

Microcontrollers and Bluetooth communication system
A microcontroller (nRF528321, Nordic semi) was mounted on a PCB board for data sampling (12-bit builtin ADC). The sampled data was transmitted to the RX board using a 2.4 GHz RF built-in Bluetooth module. The actuator driver (BOS1901, Boreas Technologies) for driving piezoelectric actuators is controlled by elevating the voltage from 3.3 V to a maximum voltage of 95 V pp . The LED driver and the LED matrix (PIM442, PIMRONI Ltd) were used for the visualisation of tactile feedback. The brightness of the LED was controlled based on the spatial distribution and pressure level data received from the static pressure sensor array.

Power supply and regulation
The TX board was powered by a Li-polymer battery (TW401215, 15124 mm 2 , 40 mAh), and the power was regulated by a boost and inverting converter. The RX board is composed of a 300 mAh Li-polymer battery, and a buck converter is used for power regulation.
Demonstration of wireless telehaptic system

Spatial recognition
Haptic communication was demonstrated with an ultra-exible static/dynamic pressure sensor, a customised 32-channel ampli er, and a multi-morph piezoelectric actuator. The bimodal pressure sensor was connected to an ampli er. The noise of the signal is ltered, and the gain of the signal is controlled. The tactile stimuli information (sensor output) was recorded simultaneously using an oscilloscope (DPO-5054, Tektronics). The recognised information was transmitted to the actuator. The vibration of the actuator was monitored using the LSV system. The monitored vibration information included vibration displacement for each point within the scanning area.

Texture recognition
To quantitatively analyse the average roughness of the three fabrics, the average maximum height of the 1.6 mm pro le was measured using a confocal microscope (DCM-8, Leica).
The signals from the UFS attached to the ngertip were acquired while rubbing the three different fabrics in the same direction and transmitted to the FST.

Data availability
The data that support the plots within this paper and other ndings of this study are available from the corresponding author upon reasonable request. Ultra-exible bimodal (UFB) tactile sensor and sub-mm tactile actuator array for telehaptic system. a, Schematic of telehaptic system composed of tactile acquisition (left) and rendering (right). b, 3D schematic of ultra-exible bimodal (UFB) tactile sensor array. c, Optical microscopy image of the UFB tactile sensor array. d, Image of the UFB tactile sensor laminated on the ngertip. e, UFB tactile sensor with a wireless system attached to the hand for tactile acquisition. f, 3D schematic of ultra-exible submm piezoelectric (UFP) tactile actuator array. g, Image of the actuator array assembled on an ultraexible substrate. h, the ultra-exible sub-mm tactile actuator array laminated on the ngertip. i, Sub-mm piezoelectric tactile actuator array with a wireless system attached to the hand for tactile rendering.    vibrator. g, The delay time while the sensor signal was transmitted to the actuator board, and the vibration of the actuator was measured by LSV. h, The acquired tactile sensor output voltage, while random tactile stimuli patterns were applied by the vibrator. i, Vibrational displacement of the tactile actuator rendered by transmitted sensor signals. j, Correlation matrix correlogram of vibration signal pattern of motorcycle describing the correlation coe cient between the signals acquired by the sensor (S1, S2) and the rendered actuator signals (A1, A2). The S1(e), S2(e) and A1(e), A2(e) were envelope of sensor and actuator signals, respectively.

Figure 5
Demonstration of a wireless telehaptic system. a, Spatial information of tactile stimuli (static pressure) was acquired by UFB tactile sensor array and the signals were wirelessly transmitted to the actuator by