Recent progress in flexible tactile sensor systems: from design to application

With the rapid development of artificial intelligence, human-machine interaction, and healthcare systems, flexible tactile sensors have huge market potentials and research needs, so that both fundamental research and application demonstrations are evolving rapidly to push the potential to reality. In this review, we briefly summarize the recent progress of the flexible tactile sensor system, including the common sensing mechanisms, the important performance evaluation parameters, the device design trend, and the main applications. Moreover, the current device design trend towards flexible tactile sensor systems is discussed, including novel structures for outstanding performance, sensor arrays for large-area information acquisition, multi-mode information acquisition, and integration of tactile sensors with transistors. Various emerging applications enabled with these sensors are also exemplified in this review to show the potentials of the tactile sensors. Finally, we also discuss the technical demands and the future perspectives of flexible tactile sensor systems.


INTRODUCTION
Tactile sensation is the basic perception ability of human skin, which is a kind of feeling produced by tactile receptors in response to external stimuli. There is a tactile cone in the deep layer of the skin, where the various tactile mechanoreceptors respond to the pressure exerted on the skin surface and generate a small current signal, which is immediately transported to the brain through the nerve fibers. The sense of touch is enabled by distinguishing the magnitude and position of the signal [1] . The touch of the finger is the most sensitive as a result of the maximum contact density distributed on the fingertip, providing one of the most direct resourceful means for the human to obtain interactive information from the environment. Humans can feel the temperature, humidity, material type, shape, surface texture, hardness, vibration, and other information of the target object through touch. However, the tactile sensation of human skin cannot quantitatively gain the specific sensing parameters. Mimicking the functionality of human skin, a tactile sensor is a sensing system that can detect pressure, strain, temperature, humidity, and other information about an object. The measured information is converted into electrical signals and analyzed by an external processing circuit to obtain a more accurate tactile detection capability than human skin [2] . The tactile sensor needs to be flexible to adapt to objects with different surface morphologies and shapes, which is an important difference between the tactile sensor and the traditional sensor.
In recent years, with the rapid development of flexible electronics and the rapid popularization of intelligent terminals, the flexible tactile sensing technology has been developed rapidly and widely used in intelligent robots, human-machine interfaces, virtual and augmented reality, medical treatment, health monitoring, and other emerging applications [ Figure 1]. Meanwhile, to satisfy the requirements of numerous emerging applications, flexibility, sensitivity, sensing range, response time, stability, and other performance parameters of flexible tactile sensors are also continuously improved [3][4][5][6][7][8][9][10][11][12][13][14][15][16][17][18][19] . Through the optimization of sensing material and device design, an ultra-thin sensor was realized without any sensory interference [20] . A multi-mode integrated tactile sensor system that can detect strain, temperature, and gas simultaneously was reported [21] . Wireless tactile sensors were enabled with the development of Bluetooth transmission and near field communication technology [22][23][24] . To extend the application fields, a sensor array was realized to decode the facial strains through machine learning [25] , and a sweat sensor that can detect blood glucose, nicotine, and other health information was developed [26][27][28] . Thus, flexible tactile sensing has sufficient momentum of development and huge market potentials. Therefore, both fundamental research and application demonstration are evolving rapidly to push the potential to reality.
In this review, we summarize the recent progress of the flexible tactile sensor systems. After a short introduction about the common sensing mechanisms, the important parameters to evaluate the sensor performance are summarized, with typical data reported for different mechanisms listed for comparison. The development trend of tactile sensor systems is covered in Section 3. Starting from the performance optimization for a single tactile sensor emphasizing certain performance indicators, we further discuss the multi-mode tactile sensor aiming to obtain multiple sensing capabilities and the integration of processing circuits with the tactile sensor systems. The outstanding properties of the tactile sensors cover ultra-high sensitivity, self-powered operation, wireless connection, self-repairing capability, degradable material system, good biocompatibility, and nontoxicity. These properties enable wide applications including virtual and augmented reality, intelligent robots, human-machine interfaces, medical treatments, and wearable health monitoring, which are discussed in Section 4. Finally, we address the potential development trends and remaining challenges for the flexible tactile sensor systems in future research.

Sensing mechanisms
Tactile sensors usually convert tactile information into electrical signals. They can be classified as piezoresistive, piezoelectric, capacitive, inductive, electromagnetic, photoelectric, and triboelectric types depending on the transducing mechanisms. The most widely used four types of tactile sensors are piezoresistive, capacitive, piezoelectric, and triboelectric, which are briefly reviewed as follows. The basic The first two terms describe the resistance change due to the differential geometrical deformation effect, and the last term describes the resistance change due to the piezoresistive effect. For metallic conductors, the geometrical deformation effect usually dominates, while some semiconducting materials could exhibit a large piezoresistive effect, with which the order of magnitude for the signal is larger than that induced by the pure geometrical deformation. While the above equation can be used to explain the resistance change of a resistor formed by a single type of material, other factors also play an important effect for resistive tactile sensors, as the composition of the resistor includes both polymeric and inorganic nanomaterials. The change of the contact resistance between two conductive materials, and the change of conductive paths in conductive composites vary with the deformation of the flexible structures. For example, Xie et al. [29] embedded vertical aligned multi-walled carbon nanotube (CNT) arrays into the polydimethylsiloxane (PDMS) and bonded two CNT/PDMS films face to face to realize a tunable range piezoresistive tactile sensor by adjusting the thickness of the soft spacer between the two CNT/PDMS films. The resistance variation mainly results from increasing CNT contact area with applied pressure [29] .
The piezoresistive tactile sensor has been widely studied and applied due to its simple structure design and readout mechanism. The piezoresistive tactile sensors have the advantages of simple structure, easy fabrication, low cost, high sensitivity [30,31] , and large measurement range, while there are also several drawbacks, such as obvious hysteresis effect, poor long-term stability, and requirement for external power supply, which are worthy of further investigating. The piezoresistive materials commonly used in piezoresistive tactile sensors include monocrystalline silicon [32] , carbon nanotubes [29,[33][34][35] , carbon black [36] , graphene [37,38] , MoS 2 [39] , and conductive polymers [40,41] .

Capacitive mechanism
The capacitive tactile sensor is based on the principle of capacitance change when the sensing structure responds to a mechanical stimulus, as shown in Figure 2B. The capacitance (C) of the device can be calculated by using the parallel-plate capacitor model: C = ε 0 ε r A/s, where ε 0 is the vacuum dielectric constant, ε r is the relative dielectric constant of the material, A is the overlap area of upper and lower parallel plates, and s is the spacing between the upper and lower electrodes. Typical structures adopted in tactile sensors include the planar interdigital electrode/dielectric material and electrode/dielectric material/electrode sandwich structures. The differential capacitance of the sensing capacitor is given by: Similarly, the first two terms describe the contribution by the geometrical deformation, while the last term describes the contribution by the material dielectric constant. As flexible dielectric materials usually exhibit a small Young's modulus, a large deformation can be induced by a gentle touch operation, resulting in a large differential capacitance.
Compared with the piezoresistive sensor, the sensitivity of the traditional capacitance sensor is relatively low, but it can be enhanced by using different micro/nanostructures. As-fabricated, the nanostructured capacitor features an air gap between pillars sandwiched between the upper and lower electrodes, so that both the overlapped area and distance, as well as the effective dielectric constant, are changed with applied pressure [42] . Beyond the traditional capacitive sensor, the ionic sensor based on ion-interfacial sensing has been widely investigated in recent years because of its unique double electrode layer structure [43][44][45] . For instance, Bai et al. [44] reported an ionic sensor based on a graded intrafillable architecture with a high sensitivity (> 220 kPa -1 ) in an ultra-broad sensing range (0.08 Pa-360 kPa). The advantages of capacitive tactile sensors include low energy consumption, independence of the operating temperature, and good stability. However, its initial capacitance is very small so that it is very easy to be interfered with by the parasitic capacitance of the testing circuit, resulting in a poor signal-to-noise ratio. The dielectric materials commonly used in capacitance tactile sensors include air gap [42] , ionic conductor [43][44][45] , TiO 2 [46] , polyimide [47] , elastomers [48,49] , and 3D fabrics [50,51] .

Piezoelectric mechanism
The piezoelectric tactile sensor is based on the piezoelectric effect, that is, generating electric displacement in response to the applied mechanical stimulus, as shown in Figure 2C. Considering an application scenario that both normal and shear stress are applied to the piezoelectric tactile sensor, the differential electric displacement is given by dD = -d 31 T 1 -d 32 T 2 + d 33 T 3 , where d is the piezoelectric coefficients and T is the applied stress, with the subscripts 1 and 2 describing the contribution from shear stress and 3 from the normal stress.
The piezoelectric tactile sensor has the advantages of high sensitivity, fast response time, and self-powered operation. As its operation does not have to involve the direct contact of the sensing object, it is widely used in the detection of acoustic vibration, sliding vibration, and other dynamic pressures. On the other hand, if an external force remains unchanged, the charge will be neutralized and cannot be measured statically. However, considering the unique energy collection characteristics of the piezoelectric materials, they have very important advantages in the development of fast dynamic response, low energy consumption, and selfpowered flexible tactile sensors. The piezoelectric materials commonly used in piezoelectric sensors include lead zirconate titanate (PZT) [52][53][54][55][56] , ZnO [57] , BaTiO 3 [58] , LiTaO 3 [59,60] , Group III-nitride (III-N) materials (AlN and GaN) [61][62][63] , and intrinsic flexible polyvinylidene fluoride (PVDF) [64][65][66] .

Triboelectric mechanism
The triboelectric effect is a common physical phenomenon in nature. When two substances with different electron binding energies come into contact with each other, one substance loses electrons to the other. This phenomenon is called the triboelectric effect [67] . Triboelectric tactile sensors are built based on the triboelectric effect. By contacting and then separating two materials with different binding energies in sequence, an electrical-potential difference is generated between the two materials. The resulting electric current flows through the external circuit and converts mechanical energy into electrical energy, as shown in Figure 2D. The research group of Wang Zhong-Lin at the Georgia Institute of Technology developed a triboelectric nanogenerator (TENG) using the triboelectric effect for the first time, which created a breakthrough way for energy collection technology [68] . As scientists continue to explore and study the triboelectric phenomenon, the triboelectric effect has been utilized for environmental energy harvesting [69] .
The triboelectric tactile sensor has the advantages of simple structure, low cost, and lightweight, thus it is an effective method for energy acquisition and dynamic information acquisition. Therefore, it is very suitable for building self-powered sensors to monitor and measure surrounding information, such as speed, acceleration, and displacement. The materials commonly selected for triboelectric tactile sensors include polydimethylsiloxane, polyvinyl alcohol (PVA) [70] , poly(3,4-ethylene dioxythiophene)-poly(styrene sulfonate) (PEDOT:PSS), polyethylene terephthalate (PET) [71] , polyacrylamide (PAM) [72] , and graphene [73] .

Performance evaluation parameters
One of the main advantages of a flexible tactile sensor over human skin is that it is a quantitative rather than a qualitative "feeling", as its electrical output signal corresponding to stimulation can be read out with an exact value. From this point of view, the performance of the flexible tactile sensor is evaluable and comparable. The main performance evaluation parameters of a flexible tactile sensor include sensitivity, sensing range, response time, repeatability, stability, etc.
The sensitivity parameters and units for tactile sensors with different mechanisms are different. Generally, the sensitivity (S) can be defined as the ratio of the increment of the output signal (M) for the tactile sensors to the increment of the input pressure (P), described as S = ∂M/∂P. The sensitivity may vary with the pressure, thus exhibiting nonlinearity over the measured pressure range. For most tactile sensors, the sensitivity decreases with the increase of pressure. It is difficult to compare the sensitivity performance of tactile sensors with different mechanisms because the output physical quantities of them are different. To solve this problem, the relative change of the output signal, defined as ΔM/M 0 , can be introduced to define the sensitivity as S = ∂(ΔM/M 0 )/∂P, and the unit of the sensitivity is unified as Pa -1 . In some studies, the unit of sensitivity is V·Pa -1 or A·Pa -1 for certain, piezoelectric tactile sensors, corresponding to an output signal of voltage (V) or current (A), respectively. Besides, when the input stimulus is strain, the sensitivity is unitless and is denoted as gauge factor.
The sensing range refers to the pressure range that a tactile sensor can measure normally, and it can be represented by the minimum and maximum pressures that the sensor can detect. High sensitivity tactile sensors often have a very low minimum sensing range, and the maximum sensing pressure is relatively small [74] . Meanwhile, a device that can sense a high external stimulus cannot sensitively respond to a small one [75] . Tactile sensors with different response ranges can be applied in different situations. Pressure sensors with a large pressure detection range and low detection limit can be used for more tasks.
The response time describes how fast the tactile sensor changes its output signal in response to an external stimulus, which is also related to viscoelasticity and interface adhesion. There are different ways to calculate the response time. One is the time difference between the input time of external stimulus and the stable time of the response output, and the other is the time difference between the input time of input stimulus and the output response reach 90% or (1 -1/e) position. Besides, the frequency response is also used to represent the response speed. Periodic dynamic pressure can be applied to the tactile sensor to obtain the dynamic response and extract the time lag between the output response and input stimulus.
Repeatability refers to the coincidence degree of response curves between different cycles of measurement under the same measurement conditions. The stability of tactile sensors is evaluated by the drift of the response values after thousands or tens of thousands of cycles. The repeatability and stability are related to the irreversible deformation of the constituent materials in the process of material pressure deformation. Generally speaking, the lower is the irreversible deformation, the better are the repeatability and stability.
In addition, flexibility is also a main evaluation parameter for the tactile sensors, and it can be realized by adopting intrinsically stretchable materials and flexible structure designs. Nowadays, most flexible tactile sensors have achieved satisfactory flexibility [40,44,46] . Tactile sensors are expected to have high sensitivity to detect tiny pressure and hold a high sensitivity in a broad pressure range, simultaneously, which can be achieved by further exploring functional materials and proposing novel device designs.
A perfect flexible tactile sensor that is impeccable in all sensing performance is certainly required, but most flexible tactile sensors cannot be perfect. We can design flexible tactile sensors with different performance highlights according to practical application requirements. Of course, the design of tactile sensors with an increasing number of high-level performance indicators simultaneously is an important research direction. Here, we list some typical tactile sensors with different sensing mechanisms and functional materials and compare their sensing performance including sensitivity, sensing range, response time, and cycling numbers [ Table 1].

DEVICE DESIGN TOWARDS THE FLEXIBLE TACTILE SENSOR SYSTEM
In Section 2, we introduce some typical tactile sensors with outstanding sensing performance such as ultrahigh sensitivity up to 10 7 kPa -1 [ Figure 3A], a fast response time of 0.03 ms [ Figure 3B] [42] , broad sensing range of 10 Pa to 800 kPa [ Figure 3C] [36] , and good stable performance after 50,000 cycles of measurement [ Figure 3D] [46] . In other more specific applications, the corresponding performance parameter needs to be considered, including low detect limit (1.2 Pa) for response to tiny pressure [76] , large stretchability (workable strain range of up to 187%) for applications in flexible electronics [77] , biodegradable [78] , anisotropic [79] , recyclable [80] , and self-powered [81] . Based on the performance constraints, the design of tactile sensors involves the choice of working principle; the device structure design including the functional material, flexible substrate, and electrode materials; and the fabrication technology. It may also entail further integration of information acquisition, processing technology, and multi-physical quantity fusion [82,83] .

From single sensor to tactile sensor arrays
A single sensor has to be expanded to sensor arrays to gain more information and adapt to the requirements of a large detection area and spatial mapping of the measurand. The adoption of a sensor array not only enables a large-area sensor but also increases the service lifetime of the sensor. Even though one or more sensor units malfunction, the overall functionality of the sensor remains. Meanwhile, information acquired from each independent single sensing unit in a sensor array can be combined for pressure mapping. For example, Kang et al. [84] reported an 8 × 8 transparent capacitive sensor array with 84.6% optical transmittance, as shown in Figure 4A. The sensor array shows overall good sensing performance in the < 10 kPa pressure range and remains stable within 200 cycles of measurement [84] . Zeng et al. [85] developed a 4 × 4 pressure sensor array based on a PDMS microstructure [ Figure 4B]. The sensor array exhibits a high sensitivity of 14.268 kPa -1 , very low detectable pressure limit (1.5 Pa), a fast response time (< 50 ms), excellent cycling stability (more than 10,000 cycles at 0.15 kPa), and an excellent pressure mapping ability [85] . Gong et al. [86] reported a 5 × 5 pressure sensor array based on the PDMS/ultrathin gold nanowires/PDMS sandwiching assembly with high stability (> 50,000 cycles) and demonstrated the current mapping ability under pressure, as shown in Figure 4C. Liu et al. [87] reported a 4 × 4 tactile sensor array based on PZT arrays on a PDMS substrate, as shown in Figure 4D. The sensor array exhibits a high flexibility under continuous stretching (~8%), bending (180° at a radius of 3 mm), and twisting (~90°) [87] . The rapid development of tactile sensor arrays has laid a solid foundation for realizing a large area "E-skin", mimicking the human skin's tactile functions.

From single-mode sensor to multi-mode tactile sensor
Similar to real skin, future flexible tactile sensors should be able to provide real-time feedback of various external stimuli and respond intelligently to the complex changes of the external environment. Besides simulating the basic physiological functions of human skin, such as realizing the sensing functions of pressure, strain, temperature, and humidity, other characteristics of the intelligent tactile sensor should be explored, including the detection of sound, light, heat, magnetic field, and chemical information. In this situation, a single-mode sensor cannot fulfill all of the requirements, and researchers have developed dualmode or even multi-mode sensors to realize various sensing functions. For example, Qiu et al. [88] reported a dual-mode sensor based on interlocked piezoresistive and piezoelectric films, realizing the complementary performance of the two sensing mechanisms, and thus improving the sensitivity and broadening the sensing range [ Figure 5A]. Hwang et al. [89] reported a transparent sensor combining piezoresistive and capacitive transduction mechanisms, which can distinguish touch and pressure, as shown in Figure 5B. Besides the direct combination of two sensing mechanisms in a single device design, a single sensor can respond to a variety of stimuli by selecting specific functional materials. For example, Yang et al. [90] reported a selfpowered elastic conductor based on PEDOT:PSS natural rubber films and realized both temperature-and tensile strain-sensing abilities, as shown in Figure 5C. You et al. [91] presented multimodal ion-electronic skin based on a deformable artificial ionic receptor, which can differentiate thermal and mechanical information without signal interference, as shown in Figure 5D. Wang et al. [21] used the blow-spinning method to prepare integrated wearable electronics based on 3D inorganic nanofiber IGZO networks, which can detect and differentiate multiple stimuli including test gas, strain, temperature, light, body movement, and respiratory functions [ Figure 5E].  [42] . Copyright © 2020. Springer Nature. (C) Broad sensing range (10 Pa-800 kPa). Reproduced with permission from Ref. [36] . Copyright © 2018. John Wiley and Sons. (D) Stable performance after 50,000 cycles of measurement. Reproduced with permission from Ref. [46] . Copyright © 2020. John Wiley and Sons.

Tactile sensors integrated with transistors
The small output signal of a tactile sensor has to be filtered and amplified for information interpretation. However, many demonstrations usually connect the sensor to high-precision equipment to demonstrate the functionality of the sensor, which is not practical in real-life applications. The interconnections required to connect the sensor with the processing circuit bring extra noise as well as increase the overall area. To alleviate this problem, researchers have explored novel device structures to directly integrate tactile sensors with transistors. For example, Huang et al. [42] reported an ultra-high sensitivity capacitive sensor based on the vertical integration of air-gap gates (capacitor sensor) with a two-dimensional semiconductor transistor, as shown in Figure 3A. The integrated 2D transistor is used to amplify the output signal, resulting in an ultra-high sensitivity up to 10 7 kPa -1 , compared to 44 kPa -1 for the traditional air-gap pressure sensor [42] .
Wang et al. [92] reported a piezoelectric tactile sensor integrated with an organic field-effect transistor, as shown in Figure 6A. External mechanical force is converted into a piezoelectric voltage to drive the transistor, and the piezoelectric voltage can be effectively amplified by the transistors to improve the  Figure 6B [92] . Dagdeviren et al. [52] reported an 8 × 8 PZT piezoelectric sensor array, in which the total signal output is horizontally integrated with a metal oxide semiconductor field-effect transistor that transforms the output voltage signal into the drain current [ Figure 6C]. Oh et al. [93] reported scalable tactile sensor arrays based on dual-gate piezoelectric zinc oxide thin-film transistors, which can detect compressive and shear stress, as shown in Figure 6D. This ability has application value in achieving the slip and grip function of closed-loop robotics [93] . Guo et al. [94] reported an integrated contact lens sensor system, integrating temperature sensor, glucose sensor, photodetector, and multiple MoS 2 transistors on the same platform, as shown in Figure 6E, providing a possibility for realizing a multi-functional sensor system. Triboelectric sensors, as promising self-powered sensors, can also be integrated with transistors, showing the feasibility of novel self-powered sensor systems [95] . For example, Zhao et al. [96] reported a triboelectric sensor integrated with a stretchable organic transistor based on PDMS substrate for smart tactile interaction [ Figure 6F].

APPLICATIONS
With the rapid performance improvement of tactile sensors and sensor arrays, various emerging application scenarios have been demonstrated in the fields of artificial intelligence, human-machine interaction, virtual and augmented reality, and healthcare systems. For example, the tactile sensor system in a humanoid robot plays an important role in evaluating the overall performance of the robot [97] . The tactile sensor system is an indispensable part of artificial arms and knee joints, flexible exoskeletons, intelligent prostheses, and medical robotics. Quantification of the pressure on the joint during moving extraordinarily relies upon the feedback signal from the tactile sensor system [98][99][100] . A flexible piezoelectric sensor array based on the

CONCLUSIONS AND PERSPECTIVES
In this review, we briefly summarize the recent progress of flexible tactile sensor systems, including the sensing mechanisms, the common functional materials, the important parameters to evaluate the performance, the device design, and the main applications. In particular, the development trend of tactile sensor systems is discussed. Besides the performance optimization for a single tactile sensor emphasizing certain performance indicators, the tactile sensor array for large-area pressure detection and the multi-mode tactile sensor aiming to obtain multiple sensing capabilities and tactile sensors integrated with transistors will be the development trends of integrated tactile sensor systems to satisfy the requirements of various applications in virtual and augmented reality, intelligent robots, human-machine interfaces, medical treatment, and wearable health monitoring. [81] . Copyright The device performance of the tactile sensor remains a key challenge for realizing advanced tactile sensing systems. Highly sensitive tactile sensors with a broad pressure range are always a research focus. At the same time, reducing detection area, improving detecting accuracy, and distinguishing the stimulations of various force [115] coming from different directions are also challenging for advanced sensing systems. For the multimode tactile sensor, a critical problem is how to solve the crosstalk when more than two physical parameters are changing together, especially the changing of temperature, which can easily influence the other physical parameters. In the aspect of device design towards flexible tactile sensor systems, the realization of scalable sensor arrays is the foundation of large area tactile sensor systems.
To satisfy the requirements of various applications in virtual and augmented reality, intelligent robots, human-machine interfaces, medical treatment, and wearable health monitoring, the future trend of the flexible tactile sensor system is summarized as follows. First, besides the basic requirement of flexibility and conformality, an ideal flexible tactile sensor is expected to exhibit a high sensitivity and multi-function integration to realize multi-dimensional synchronous sensing. Second, recent strategies toward low-power or even self-powered tactile sensors could overcome the energy constraints for longtime monitoring in wearable electronics. Third, various machine learning algorithms are adopted to interpret or enhance the information acquired from the tactile sensor array, so that the artificial intelligence auxiliary detection for human activity monitoring, medical services, human-machine interfaces, artificial intelligence robots, and other applications can be realized. Similar to real skin, future flexible tactile sensors are expected to provide real-time feedback of various external stimuli and respond intelligently to the complex changes of the external environment. In addition, besides simulating the basic physiological functions of human skin, such as the sensing functions of pressure, strain, temperature, and humidity, and obtaining quantitative data, other characteristics of the intelligent tactile sensors should also be explored, including the detection of sound, light, magnetic field, and chemical information. Moreover, emerging smart contact lenses based on integrated flexible tactile sensor systems are also expected for virtual and augmented reality applications. Therefore, research on flexible tactile sensors and the integration systems is an important subject in the science community and its practical applications in human activity monitoring, medical services, humanmachine interfaces, and artificial intelligence robots will promote the development of human civilization.

Authors' contributions
Writing -original draft: Zhu J Writing -review & editing: Zhou C Supervision, writing -review & editing, project administration: Zhang M

Availability of data and materials
Not applicable.

Financial support and sponsorship
This work was supported by Shenzhen Fundamental Research Program (No. GXWD20201231165807007-20200827122756001), Guangdong Science and Technology Plan Project 2019A050510011, and National Natural Science Foundation of China 62074008.