Design concept and principle of the touch sensor based on the PI–ST method. The steps for designing the sensitive layer with an interconnected porous structure are shown in Fig. 2a. The ternary material system used in the PI–ST method consists of a polymer, solvent, and nonsolvent, i.e., thermoplastic polyurethane (TPU)/Ag nanowire (NW)/NaCl composite materials, N, N-dimethylformamide (DMF), and a Ag NW aqueous (Ag NW aq.) solution, respectively. The interconnected porous structure of the conductive TPU film is constructed during the PI–ST process by introducing Ag NWs as conductive fillers and soluble NaCl particles as sacrificial templates.
The one-step formation mechanism of the conductive TPU film is analyzed based on thermodynamics and kinetics. In terms of thermodynamics, the Flory–Huggins theory is used to draw the ternary phase diagram of this material system (Fig. 2b). P, S, and NS represent the polymer (TPU/Ag NW/NaCl composite materials), solvent (DMF), and nonsolvent (Ag NW aq. solution), respectively. A represents the composition of the initial casting liquid (DMF solution of TPU/Ag NW/NaCl). The casting liquid enters the phase separation zone under the action of DMF and the Ag NW aq. solution, thus forming two equilibrium liquid phases (B’ and B’’) to realize phase inversion. The expressions for the free energy and interaction parameters of this ternary material system are provided in Supplementary Eq. (1) and Supplementary Table 131. From the perspective of kinetics, when TPU is initially immersed in the Ag NW aq. solution, the overall movement and migration cannot occur because of its large molecular weight. However, the Ag NW aq. solution shows a low diffusion resistance to DMF. Therefore, the mass of DMF that is removed from the polymer solution is more than the mass of the Ag NW aq. solution that is added to the solution. This results in a rapid increase in the TPU concentration and the components of the polymer solution are rapidly separated through the metastable zone. The apparent diffusion coefficient is expressed in Supplementary Eq. (2)32.
Different pores are formed during phase inversion. However, the surface solvent continuously evaporates into the interior and pores of aggregates, resulting in smaller pores. Although this process can be controlled to modulate the micropore morphology, it is still difficult to obtain the interconnected porous structure when only phase inversion is used33. Therefore, NaCl particles are used as sacrificial templates to resolve this issue. The SEM images of the film formed using phase inversion (left) and the PI–ST method (right) are shown in Fig. 2a. It is noteworthy that the TPU film and interconnected porous structure are simultaneously created. NaCl particles are readily soluble in the Ag NW aq. solution, which is the nonsolvent in phase inversion. Thus, a conductive TPU film with a controllable interconnected porous structure is created using the PI–ST method. This structure acts as a sensitive layer of the touch sensor and is highly effective in improving the compressibility of TPU to significantly increase the sensitivity of the touch sensor.
Films based on various material systems are prepared using phase inversion to verify the generalizability (Supplementary Figs. 1 and 2). The interaction parameters and apparent diffusivities of these ternary material systems are listed in Supplementary Tables 1 and 231,32. The optical photographs of the TPU, polyvinylidene fluoride, polyacrylonitrile, cellulose acetate, polyvinyl chloride, and polystyrene micropore films and their corresponding cross-sectional SEM images are shown in Supplementary Fig. 2. These images demonstrate the feasibility of this method for preparing touch sensors with different polymers. Additionally, the phase inversion processes of these polymers are shown in Supplementary Movies 1 and 2.
Interface characteristics and performance optimization of sensitive layer. The electromechanical properties of sensitive layers are crucial for ultrasensitive touch sensors. The content of Ag NWs is an important factor that affects the electrical properties of the sensitive layer34. The effective injection amount of Ag NWs in TPU can be increased through the mechanical mixing of these materials as composite polymers and introducing the Ag NW aq. solution as a nonsolvent. According to the formula of the percolation threshold35, the electroosmotic phenomenon occurs when the volume percentage of Ag NWs in TPU is 0.15 vol% (the corresponding mass fraction is 0.012 wt%). However, if the nonsolvent in-phase inversion is deionized (DI) water, the volume resistance is still high when the Ag NW content is considerably larger than the theoretical value. In contrast, the volume resistance of the composite film decreases by 6 orders of magnitude when the Ag NW aq. solution is used as the nonsolvent (Supplementary Fig. 3). This proves that the Ag NW aq. solution increases the injection and entanglement of Ag NWs on the TPU skeleton with the interconnected porous structure during phase inversion. The LED emits light even when the film is connected in series to the circuit; this demonstrates the excellent conductivity of the composite film (Supplementary Fig. 4).
According to electroosmotic flow and the tunnel conduction theory, the uniform distribution of conductive fillers (DMF solution of Ag NWs) in a polymer solution (DMF solution of TPU) affects the conductivity and response sensitivity of a prefabricated composite film36. Thus, interface adhesion has been studied in the structure–activity relationship of touch sensors. Compared to the DMF solution of Ag NWs, there is no significant sedimentation in the composite solution (DMF solution of Ag NWs and TPU) within 48 h, indicating better stability (Supplementary Fig. 5). Ultraviolet–visible spectroscopy is performed to further analyze the homogeneity of the composite solution (Supplementary Fig. 6). The relationship between the Ag NW concentration and the absorbance of the composite solution at a wavelength of 409 nm (inset of Supplementary Fig. 6) is approximately linear. This verifies the homogeneity of the composite solution37. A detailed analysis is provided in the Supporting Information. The energy spectrum of the composite films is obtained via energy-dispersive X-ray spectroscopy. The results show that C, N, O, and Ag are uniformly distributed in the analysis area (Supplementary Fig. 7). Van der Waals and frictional forces affect the stability of the interface between Ag NWs and TPU, resulting in the stability of the composite solution38,39. The results of Fourier transform infrared spectroscopy (Supplementary Fig. 8), a wettability test (Supplementary Fig. 9), and a friction theory test (Supplementary Fig. 10) are presented in the Supplementary Information.
The compressibility of the Ag NW/TPU composite film can be adjusted by varying the NaCl particle content to optimize the sensitivity of the device. All composite films with different porosities show uniform porous structures with pore sizes of 100–200 µm (Supplementary Fig. 11). Tensile tests are carried out to investigate the mechanical properties (see the pressure-compressive strain relationship in Supplementary Fig. 12). The composite film with a porosity of 45.4% exhibits the highest compressive strain under equal pressure, with an ultra-low compressive modulus of 23.8 Pa (Supplementary Fig. 13). An increase in porosity causes structural collapse owing to the insufficient supporting force of the composite skeleton. Deformation under compressive stress is investigated for the three samples under identical parameters (Supplementary Fig. 14); the three curves overlap, indicating that the prepared films exhibit consistent mechanical properties.
Characterization of pressure sensing properties. The aforementioned results verify that the compressibility of soft materials is significantly improved when an interconnected porous structure is introduced because air is more easily squeezed due to its low compressive modulus. Mathematically, the rate of variation in the interfacial contact area (\(\varDelta A/{A}_{0}\)) increases with the porosity under compression. However, the initial contact area (\({A}_{0}\)) is inversely proportional to porosity. Therefore, high porosity is conducive to the high sensitivity of touch sensors. Supplementary Fig. 15 shows the electrical properties of sensors with different porosities, confirming that the sensitivities conform to the mechanical properties. The degree of deformation of the composite film under different pressures helped to identify two regions where the touch sensor’s sensing performance has a linear response to pressure. Supplementary Fig. 16 shows the SEM images of the composite film in different compressed states. The deformation becomes more limited as compressive stress increases. The composite film with 45.5% porosity has a sensitivity of up to 1167 kPa− 1 at pressures below 1179 Pa (Fig. 2c). The minimum pressure detection limit is as low as 1.34 Pa (Fig. 2d), which is a significant improvement compared with conventional touch sensors (Fig. 2e)2,13,14,20,34,40−57. The sensitivity reaches 25 kPa− 1 even in the sensing range of 1179–10240 Pa. Touch sensors can perceive minor and weak stimuli in daily life owing to their high sensitivity. As shown in Figs. 2f, 2g, and 2h, insect specimens (weights in milligrams) are placed on touch sensors with extremely low pressures of 35 Pa (55 mg), 60 Pa (80 mg), and 110 Pa (120 mg), respectively. Evident electrical output signals are generated even at these low pressures. This verifies that the proposed sensor provides a fast and stable response to minor static pressures. Furthermore, the dynamic response can be achieved, as in the case of continuously falling water droplets and weak airflow (Supplementary Figs. 17 and 18). This ultra-high sensitivity of the touch sensor makes it suitable for the detection and discrimination of tactile and slip signals.
The current–voltage (I–V) curves of the touch sensor at different pressures are shown in Supplementary Fig. 19. Ohmic contact is confirmed, and resistance decreases as pressure increases. The dynamic response curve shows that the sensor can provide a stable real-time response to a preset pressure gradient of 2–1000 Pa (Supplementary Fig. 20). The response time of loading and recovery time of releasing are 46.5 ms and 248.8 ms, respectively (Supplementary Fig. 21). This apparent hysteresis may be caused by the viscoelasticity of the flexible TPU backbone. The repeatability and durability of the touch sensor are presented in Supplementary Fig. 22. It can sustainably function over 2000 loading and unloading cycles under a pressure of 500 Pa at 0.5 V.
Verification of high sensitivity of the touch sensor. The high sensitivity of the touch sensor is verified from other perspectives by attaching it to the carotid artery of an adult female experimenter (Fig. 3a). The sensor can capture weak fluctuations in the carotid pulse and respiratory rate (Fig. 3b). The pulse and respiratory rates are approximately 84 beats/min and 12 cycles/min, respectively, which are consistent with the normal physiological indices of a healthy adult female in a calm state. It is worth noting that the three characteristic peaks of the pulse, namely P (percussion wave), T (tidal wave), and D (dicrotic wave), can be clearly observed (Fig. 3c). In addition, the touch sensor can detect audio vibrations when it is placed on a phone while it is ringing. Python programming language is used to extract the characteristic signals of the time domain for three different ringtones. Then, the frequency information is obtained using a short-time Fourier transform (Supplementary Fig. 23). The signals collected by the touch sensor coincide with the specific characteristic peaks of the three ringtones in the time domain (Figs. 3d–3f). These weak signals can be monitored because of the high sensitivity of the sensor. This shows that the sensor can capture weak signals for slip sensing.
Signal processing and application of slip sensing. Another key advantage of a highly sensitive touch sensor is its ability to discriminate between tactile and slip sensing using a single sensor unit. After signals are collected by the touch sensor, an appropriate wavelet transform analysis method is used to extract the basic information for estimating whether slip occurs58.
The schematics of the experimental procedure, data acquisition, and signal processing are illustrated in Fig. 4a, and the test apparatus is shown in Supplementary Fig. 24. The slipping process is simulated by pulling a weighted object on the surface of the touch sensor. The initial electric signal (current–time) and mechanical signal (tension–time) are collected and analyzed (Fig. 4b). The variation in the tensile force is consistent with the change in current in a complete slipping process. The resistance of the sensor decreases due to an increase in the conductive path. Thus, the tension force (tangential deformation) applied on the sensor increases with the static friction force, resulting in a decrease in resistance and an increase in current (Supplementary Fig. 25). Once slip occurs, the tensile force slightly decreases, accompanied by a slight decrease in current. At this moment, the interaction force between the sensor and object changes from static friction to kinetic friction.
In general, the stick–slip phenomenon exists between the target object and the touch sensor. This phenomenon can affect the contact area, resulting in a fluctuation in the output signal20. As mentioned earlier, a weak signal can be captured by an ultrasensitive sensor when an object undergoes a slipping process. Continuous wavelet transform (CWT) and discrete wavelet transform (DWT) are applied to the current signal to magnify the signal difference between the slip and non-slip states. The specific analysis methods are shown in Supplementary Figs. 26 and Fig. 2759.
The original electric curve, scale map, and intuitive time–frequency map are shown in Fig. 4c. The slip process is broadly divided into four stages: (i) In the first stage, the tangential deformation and the electrical output signal of the sensor remain unchanged in the presence of only the vertical normal force. Consequently, there are no high-frequency components at this stage. (ii) In the second stage, the mutual tensile force and tangential deformation gradually increase. A high-frequency component is observed before the object begins to slip. (iii) In the third stage, the tensile force continues to increase until it becomes equal to the maximum static friction force. At the moment of slipping, the tangential deformation of the sensor rapidly increases, leading to an increase in the conduction paths and corresponding high-frequency components. At this point, the highest frequency reaches approximately 600 Hz. (iv) The last stage shows the object in a slipping state. The tensile force is balanced by dynamic friction, and the number of conducting paths and frequency of electrical signals are slightly larger than those in the first stage.
The signal generated by normal pressure is analyzed and compared with the slipping signal to exclude the interference of the frequency component caused by the change in the normal pressure during the slipping process (Fig. 4d). Different frequency components are extracted from the signal by utilizing the DWT to process the electrical signal output. The maximum DWT detail coefficient of the sensor is obtained under different force states. An appropriate slipping threshold is used to determine the occurrence of slipping on the basis of the results of the existing experimental analysis. Slipping occurs when the maximum DWT detail coefficient exceeds the slipping threshold. When normal pressure is applied to the sensor, the DWT detail coefficient shows a small overall fluctuation, with a maximum value of approximately 0.3414. Under tension force, the DWT detail coefficient reaches its maximum value at the moment of slipping, with a value of approximately 1.6125. In addition, on the basis of the surface roughness of the object, the slipping threshold of the object is assumed to be ± 1 to carry out the subsequent experiments.
The slip signals of sandpapers with different values of surface roughness are measured using the touch sensor. The current output signal, DWT images, CWT images, and time–frequency graphs are shown in Supplementary Fig. 28. The maximum DWT detail coefficient increases with roughness. This implies that as roughness increases, the vibrations caused by the change in the motion state become stronger, and it becomes easier to detect the slipping state.
Next, slip-sensing tests are conducted along the device surface using different weights (with acrylic at the bottom). The initial normal loads (weights) are set as 50, 100, 150, and 200 g. The slip speed of the weights is set to 5 mm/s. Figure 5a shows the raw signal of the current response during slipping. The DWT detail coefficients of the touch sensor with different mass loadings exceed the set threshold (± 1) range before the instant when the object initially slips (Fig. 5b). Moreover, the maximum DWT detail coefficient when slip occurs increases with the mass loading. This implies that the slip signal becomes stronger as the pressure increases. The corresponding CWT and time–frequency diagrams are consistent with the results shown in Fig. 5c. This may be because the vibration of the sensor surface becomes stronger as the pressure increases. Therefore, slipping becomes easier to detect when the mass of the object increases.
Acrylic, cotton, and wood are selected as the contact surfaces between the weights and touch sensor, as shown in Figs. 5d–5g. The weights are 100 g each, and the slip speed is 5 mm/s. The DWT detail coefficient exceeds the set threshold (± 1) range for all contact surfaces. The tensile force and maximum DWT detail coefficient are the smallest for the acrylic surface. This may be because acrylic is smoother than the other materials, leading to a small coefficient of friction and low sensor vibration. The static and sliding friction coefficients of the three materials are calculated according to the tensile force change curve recorded by a peel-force testing machine. The values are consistent with the analysis results; this verifies the potential of the ultrasensitive touch sensor for slip detection.
The sensor can distinguish between the contact forces for tactile and slip perception using a machine learning algorithm (Supplementary Fig. 29). Data are collected for different contact forces and classified as tactile and slip forces. 120 sets of data are tested for each contact force. The contact force dataset consists of 240 samples with a measurement length of 5 s per sample and a sampling rate of 500 Sps; the total number of data points is 5 × 500. We built machine learning models using two popular algorithms, namely decision tree and multilayer perceptron. The confusion matrices of the two trained models show that the overall recognition accuracies for classifying tactile and slip forces are approximately 98.75% and 100%, respectively. Such high accuracies are crucial for automatic multimodal recognition required for intelligent sensing by manipulators.