Three-dimensional micro strain gauges as flexible, modular tactile sensors for versatile integration with micro- and macroelectronics

Abstract

Flexible tactile sensors play important roles in many areas, like human-machine interface, robotic manipulation, and biomedicine. However, their flexible form factor poses challenges in their integration with wafer-based devices, commercial chips, or circuit boards. Here, we introduce manufacturing approaches, device designs, integration strategies, and biomedical applications of a set of flexible, modular tactile sensors, which overcome the above challenges and achieve cooperation with commercial electronics. The sensors exploit lithographically defined thin wires of metal or alloy as the sensing elements. Arranging these elements across three-dimensional space enables accurate, hysteresis-free, and decoupled measurements of temperature, normal force, and shear force. Assembly of such sensors on flexible printed circuit boards together with commercial electronics forms various flexible electronic systems with capabilities in wireless measurements at the skin interface, continuous monitoring of biomechanical signals, and spatial mapping of tactile information. The flexible, modular tactile sensors expand the portfolio of functional components in both microelectronics and macroelectronics.

Flexible tactile sensors based on three-dimensional micro strain gauges can measure triaxial forces and other parameters.

INTRODUCTION

Flexible tactile sensors mimic the physical properties and sensing capabilities of the human skin (1–3) to underpin a variety of applications ranging from industrial automation and metaverse to medical diagnosis and robotic surgery (4–6). These sensors exploit advanced materials [e.g., hydrogels, two-dimensional (2D) materials, conductive composites, and organic conductors/semiconductors, etc. (7–11)] or engineered structures [e.g., textile, micropyramids, and three-dimensional (3D) structures (12–15)] to achieve flexible, stretchable, or self-healing properties and to respond to a wide range of external thermal and mechanical stimuli (16–18). Recent progress in material science, advanced manufacturing, and data analytics brings a series of flexible tactile sensors, with capabilities in object recognition, sensing and decoupling various stimuli, spatiotemporal mapping at the tissue interface, and supporting human-machine interface (19–22).

However, the materials and manufacturing approaches of these flexible tactile sensors show some discordances with conventional microelectronics and macroelectronics (table S1) (23, 24). Such incompatibilities affect the performances of flexible tactile sensors in the following aspects. First, some elastic or composite materials used in flexible tactile sensors cannot be patterned using conventional photolithography and wet/dry etching, thereby limiting the feature size and spatial resolution of tactile sensors (1, 25). Second, the processes in constructing flexible tactile sensors usually involve transfer, bonding, and other alternative steps that hinder the monolithic integration with other wafer-based devices and integrated circuits (ICs) (26, 27). Third, flexible tactile sensor arrays in large areas typically have a fixed design, while customizable spatial distribution and overall shape of the array are necessary to meet the diverse application requirements (28, 29).

Here, we address the above limitations by introducing a set of flexible, modular tactile sensors fabricated with a microelectromechanical system (MEMS) technique. Key to this success is the incorporation of a layer of silicon dioxide (SiO₂) with internal stress that enables the construction of arrays of 3D micro strain gauges (μSGs) for measuring the amplitude and direction of mechanical stimuli. The compatibility with the microelectronic process allows such flexible tactile sensors to be monolithically integrated with other sensors on silicon (Si) wafer or configured into arrays with high spatial density. The flexible tactile sensors exhibit a modular feature, making them compatible with pick and place, flexible printed circuit boards (FPCBs), and other techniques used in macroelectronics, with possibilities in assembling into large-area arrays and cooperating with commercial devices. Demonstrations in high-density mapping of pressure, wireless monitoring of biomechanical signals, and decoupled measurement of normal force and shear force prove the versatility of the flexible, modular tactile sensors.

RESULTS

Design and manufacturing of the flexible, modular tactile sensor

As shown in Fig. 1 (A to C), each tactile sensor comprises four interconnected 3D μSGs in nickel chromium alloy (Ni₈₀Cr₂₀; thickness: 100 nm) for decoupled measurement of normal force and shear force and planar thin wires in chromium/gold (Cr/Au; thickness: 10/150 nm) positioned beneath the 3D μSGs for temperature sensing. Multiple layers of polyimide (PI) serve as the insulation of the μSGs and temperature module. The fabrication process of the tactile sensor is compatible with MEMS technology, thereby enabling the construction of high-density, wafer-scale tactile sensors in a parallel fashion. Specifically, the fabrication adopts lithography, film deposition, etching, and other conventional techniques to stack and pattern multiple layers of metal, alloy, polymer, and oxide on a Si wafer (note S1 and figs. S1 and S2). Controlling the deposition conditions allows for the formation of a SiO₂ (thickness: 150 nm) layer on the aluminum (Al; thickness: 100 nm) sacrificial layer, with an average internal compressive stress of ~152 MPa (SD < 15%) across a circular area with a diameter of 7 cm (fig. S3). After dissolving the sacrificial layer, the compressive stress in the SiO₂ layer drives the upper layers to bend upward from the substrate, thereby transforming the μSGs into 3D geometries (Fig. 1A). Here, we etch the sacrificial layer in phosphoric acid (H₃PO₄; dilute at a mass ratio of 10:1, 80°C) and replace the aqueous solution with ethanol to reduce the surface tension and to accelerate the evaporation of the liquid (movie S1). The radius of curvature of the 3D μSG depends on the internal stress, modulus, and thickness of each layer (30–32), as illustrated in note S2. Microscopic imaging confirms the uniformity of the geometric shapes of different 3D μSGs (feature sizes: 12 and 4.8 μm; figs. S4 to S7). The compatibility with MEMS technology ensures a high success rate in manufacturing and enables mass production of tactile sensors at the wafer scale. Figure 1D and movie S2 display an array of tactile sensors processed on a 4-inch Si wafer with a success rate of 100%. 2D microscopic photos in Fig. 1 (E to G) and the schematic diagrams in figs. S8 and S9 present the overall structure and different layers of the tactile sensor. It is noteworthy that the construction of our tactile sensors can operate in a parallel fashion. In comparison to some advanced 3D microfabrication technologies, such as 3D printing, our manufacturing approaches exhibit advantages in efficiency, especially when used in large-scale production.

Fig. 1. Design and manufacturing of the flexible, modular tactile sensors.

(A) Schematic diagram of the formation of 3D μSGs. (B) Microscopic image of a tactile sensor without encapsulation. (C) Magnified microscopic image of a single 3D μSG. (D) Optical image of 184 tactile sensors (corresponding to 736 μSGs) processed on a 4-inch Si wafer with a yield of 100%. (E to G) Microscopic images of the planar patterns of the tactile sensor (E) and magnified views of the μSGs [(F) and (G)]. (H) Optical image of the encapsulated tactile sensors on a 4-inch Si wafer. (I) Optical image of a flexible, modular tactile sensor with an encapsulation thickness of 0.5 mm after being separated from the wafer. (J) Optical images of tactile sensors with encapsulation thicknesses of 500 and 300 μm placed together with a commercial chip and a coin. (K and L) Optical images of a 3D tactile sensor processed simultaneously with various planar thin film devices on a 4-inch Si wafer (K) and on a flexible substrate after being separated from the wafer (L). The inset in (L) shows the magnified microscopic image of the tactile sensor on a flexible substrate (scale bar: 1 mm). (M) Tilt-view microscopic images of a tactile sensor fabricated on top of the ICs on a wafer.

These tactile sensors can be encapsulated in a silicone elastomer to improve the robustness or released from the Si wafer to exhibit a flexible form factor. The encapsulation process of the tactile sensor exploits an Al alloy mold placed onto the wafer, with hollow areas (lateral dimension: 3.4 mm by 3.4 mm; height: 1 to 2 mm) aligned to the tactile sensors. Curing silicone elastomers in the hollow areas, followed by demolding, yields an array of encapsulated tactile sensors (Fig. 1H). Microscopic images in fig. S10 and statistical analysis results in fig. S11 validate that the polymer encapsulation does not affect the geometric uniformity of the 3D μSGs. A capillary-assisted electrochemical delamination method (33) separates the bottom PI layer of the tactile sensor from the rigid Si substrate. The flexible PI substrate and the stretchable silicone elastomer together render a flexible feature of the tactile sensor, with thicknesses down to 500 and 300 μm (Fig. 1I and fig. S12). In each flexible tactile sensor, the encapsulation only covers the four μSGs and the planar temperature module but leaves the contact pads exposed on the PI substrate (Fig. 1J). This design allows the flexible tactile sensors to connect with circuits, cables, and/or other types of electronic components in a way similar to commercial chips.

The processes of encapsulation and release have no effect on the fabrication of other types of planar electronic devices. Figure 1K shows a photo of a 4-inch wafer containing a tactile sensor and various thin film devices, such as humidity sensors, electrode arrays, wireless coils, heaters, etc. (figs. S13 to S17), processed simultaneously on the same wafer. The capillary-assisted electrochemical delamination method enables separation of these devices from the Si wafer, resulting in a large-scale multifunctional flexible electronic system, as demonstrated in Fig. 1L. The inset in Fig. 1L shows the magnified microscopic image of the tactile sensor on a flexible substrate. The incorporation of the tactile sensor, especially the 3D μSGs, to these planar devices can expand the sensing modalities of the flexible electronic systems (34, 35). In addition, the fabrication process allows for the manufacturing of 3D μSGs directly on bare dies of ICs, thereby offering a potential avenue for the integration of flexible electronics, MEMS, and ICs (Fig. 1M and fig. S18).

Characterizations of the tactile sensor

As shown in the exploded diagram in Fig. 2A and schematic diagram in Fig. 2B, the flexible, modular tactile sensors reported here respond to external stimuli (i.e., normal force, shear force, and temperature) by changing the resistances of thin wires of Ni₈₀Cr₂₀ and Cr/Au. When subjected to normal force, the elastomer encapsulation of the tactile sensor deforms together with the four μSGs (middle frame in Fig. 2B). As a result, all of the μSGs exhibit a resistance increase (fig. S19). The change in resistance is linear to normal force, with a sensitivity of 8.16 × 10⁻³ N⁻¹ (R² > 0.995, nonlinear error < ±6%; green curve in Fig. 2C). By contrast, the planar configuration of the temperature module, along with the strain isolation effect of the unstretchable PI substrate, minimizes the strain changes in the Cr/Au thin wires under normal force (black curve in Fig. 2C).

Fig. 2. Characterization of the tactile sensor.

(A) Exploded view diagram of a tactile sensor with silicone encapsulation. (B) Schematic diagram of the responses of the tactile sensor under normal force and shear force. (C) ΔR/R₀ of the μSG and the temperature module in response to normal force. (D) ΔR/R₀ of four μSGs in a tactile sensor under shear force. The inset illustrates the positions of R₁, R₂, R₃, and R₄. (E) ΔR/R₀ of the μSG and the temperature module in response to temperature. (F) Response time of the tactile sensor during loading and unloading processes. (G) Magnified view of the response time curves. (H) ΔR/R₀ during 10,000 cycles of loading and unloading. The insets show the magnified view of the curves from three consecutive cycles. (I) Response of the tactile sensor with different encapsulation thicknesses under normal pressure. (J) Characterization of the drift of the tactile sensor.

Under shear forces, the four μSGs experience distinct deformations, thereby exhibiting different changes in resistance (bottom frame in Fig. 2B). For example, a shear force from left to right deforms the two μSGs parallel to the direction of the shear force (i.e., R₁ and R₃ in the inset of Fig. 2D). The symmetric geometry of R₁ and R₃ leads to an opposite response, where R₁ decreases and R₃ increases under the shear force from left to right (green and black curves in Fig. 2D). The fractional change in resistance (ΔR/R₀) as a function of shear force is also linear from 0 to 0.4 N (R² > 0.999, nonlinear error < ±4%), with sensitivities of 1.09 × 10⁻² N⁻¹ for R₃ and −0.79 × 10⁻² N⁻¹ for R₁. The two μSGs perpendicular to the shear force direction (i.e., R₂ and R₄ in the inset of Fig. 2D) show minimal changes in strain and resistance (ΔR/R₀ < ±0.02%; orange and gray curves in Fig. 2D). When applying normal force and shear force concurrently to the tactile sensor, the response can be seen as a linear superposition of the two (fig. S20) (36). Therefore, analyzing the responses of the four orthogonally distributed μSGs can decouple the normal force and shear force, thereby providing effective means to simultaneously measure the direction and amplitude of the external force (note S3). The calibration test results confirm that the decoupling deviations in the x, y, and z directions are all below 8% (figs. S21 to S23). Similar to the response under normal force, the temperature module located beneath the elastomer shows almost no resistance variation under shear force (fig. S24).

Under temperature fluctuation, both the μSGs and the temperature module exhibit variations in resistance. Hence, the temperature module not only provides important temperature information of the object being touched but also serves as a calibration unit for other sensors. The μSGs adopt Ni₈₀Cr₂₀ with a relatively small temperature coefficient of resistance (37), rather than Cr/Au, to mitigate the resistance variation under temperature fluctuation (fig. S25), whereas the temperature module exploits Cr/Au as the sensitive material to improve the sensitivity to temperature. As illustrated in Fig. 2E, both the μSGs and the temperature module exhibit linear responses (R² > 0.999, nonlinear error < ±2%) to temperature, with sensitivities of 0.028 × 10⁻² and 0.138 × 10⁻²/°C, respectively. Figure S26 shows the time-domain response of the μSGs and the temperature module under temperature variations, confirming their synchronous response to thermal changes. Therefore, the temperature module can effectively calibrate the temperature drift of the μSGs. These results indicate that the tactile sensor can decouple three distinct external stimuli (i.e., normal force, shear force, and temperature).

Other key metrics of the tactile sensor appear in Fig. 2 (F to J) and figs. S27 to S33. First, Fig. 2 (F and G) demonstrates the response time of the tactile sensor under normal force. The sensor exhibits response times of 69 and 63 ms during the fast loading and unloading of normal force. Second, the tactile sensor maintains stable performance (Fig. 2H and fig. S27) during 10,000 cycles of loading/unloading of normal force at full scale (ΔR/R₀ up to 0.3%). Third, the tactile sensor exhibits small hysteresis effects in response to normal force (Fig. 2I; hysteresis error: ~6%) and shear force (fig. S28; hysteresis error: ~6%). In five consecutive cycles of overload testing (~2.3 N with the ΔR/R₀ of ~2%), the sensor also demonstrates high linearity (R² > 0.990, nonlinear error < ±7%) and minimal hysteresis (hysteresis error: ~6%), as shown in fig. S29. Fourth, the performance and function of the tactile sensors can be adjusted on demand due to the compatibility of the fabrication method with traditional microfabrication techniques. For example, during the spin coating of the PI layers, it is possible to adjust the radius of curvature of the μSGs by varying the thickness of the PI layers, thereby modifying their sensitivity to normal force (figs. S30 and S31). During the photolithography, altering design patterns allows for precise control of the orientation of μSGs, thereby enabling an alternative way to control the sensitivity (fig. S32). These alterations in fabrication parameters and layout design all provide effective means for customizing the sensor performance. Furthermore, increasing the thickness of the elastomer from 1 to 2 mm reduces the sensitivity from 8.27 × 10⁻³ to 2.12 × 10⁻³ N⁻¹. In all these cases, the linearity and hysteresis remain good (Fig. 2I; R² > 0.994, nonlinear error < ±6%, and hysteresis error: ~6%). Last, when pressing a flat plate to the tactile sensor and holding the plate at a fixed position, the normal force gradually decreases over time due to the stress relaxation caused by polymer encapsulation. Figure 2J captures the force change measured from a commercial force gauge (green curve) and the ΔR/R₀ of the sensor (black curve) during the pressing and holding of the flat plate. The two curves show a similar trend, verifying the minimal drift of the sensor. The time-domain curve in fig. S33 depicts the ΔR/R₀ of the tactile sensor during continuous application of normal force at various amplitudes.

Arrays and multifunctional systems enabled by the modular feature

Measuring the force at a single point is insufficient in many scenarios, such as compression therapy, human-machine interface, and robotic automation. The modular feature of the flexible tactile sensors creates opportunities for assembly of arrays in the large scale, to reflect spatial distributions of mechanical and thermal stimuli. Here, the four μSGs in the tactile sensor interconnect together but are separated by two row selections and two column selections (top left frame in Fig. 3A). Such configurations yield a 2 by 2 array of μSGs with only four connection pads (bottom left frame in Fig. 3A). The modular feature facilitates further assembly of the 2 by 2 array of μSGs on an FPCB to expand the scale of the array (right frames in Fig. 3A). The FPCB also contains row selections and column selections isolated by a layer of PI. The connection pad of each row selection passes through a vertical interconnect access (VIA) to locate on the same surface with the column selection and to connect with the pads of the modular tactile sensor (fig. S34). This connection scheme minimizes the number of wires but introduces current cross-talk (38), as illustrated in fig. S35A. We suppress the cross-talk by implementing a voltage follower circuit at the input end and an inverse proportional operational amplifier circuit at the output end, thereby balancing the voltage at both ends of the resistive sensors adjacent to the selected sensor (note S4 and figs. S35B and S36).

Fig. 3. Various arrays constructed from the modular tactile sensors.

(A) Schematic diagram of the row and column selections, including the interconnection for an individual tactile sensor and the arrangement of a 4 by 4 array on FPCBs. (B and C) Optical image (B) and pressure mapping results (C) of an 8 by 8 tactile sensor array when subjected to normal force in the shape of the letter P. (D and E) Optical image (D) and temperature mapping results (E) of an 8 by 8 tactile sensor array when a heater approaches. (F) Microscopic image of a tactile sensor array with row and column selections. The linewidth of the Ni₈₀Cr₂₀ strain gauges is 4.8 μm. (G and H) Optical image (G) and the magnified view (H) of the tactile sensor array with a spatial density of 360 cm⁻². (I) Optical image of a hand-shaped FPCB integrated with high-density arrays of tactile sensors at the fingertips and other distributed tactile sensors at the palm region. (J) Optical image of the high-density array of tactile sensors with encapsulation. (K and L) Microscopic image (K) and the magnified view (L) of the high-density array without encapsulation. (M) Pressure mapping results from the high-density tactile sensor array under pressing from external millimeter-scale objects molded in different shapes. (N) Optical image of the stretchable tactile sensor array with serpentine interconnections. (O and P) Testing results of a tactile sensor installed on an FPCB with single-axis serpentines. The results in (O) show the ΔR/R₀ of the sensor under different radii of curvature and normal forces. The inset in (O) illustrates the testing setup. The results in (P) demonstrate the ΔR/R₀ of the sensor in initial, stretching, and bending states during loading and unloading of normal force.

On the basis of the above connection scheme and anti–cross-talk circuit, we fabricate dense arrays of tactile sensors (each tactile sensor contains four μSGs) on a Si wafer, release them to produce flexible, modular devices, and assemble them on an FPCB to form an 8 by 8 array of flexible tactile sensors (containing to 256 μSGs in total). The process explodes the dense array on the Si wafer into a sparse but large-area array on FPCBs. Advantages of this process include the removal of Si substrate to exhibit a flexible form factor, the compatibility with the microelectronic process to produce a large amount of sensors with microscale features in a parallel fashion, and the employment of macroelectronics technique (i.e., FPCBs) to generate tactile sensors in large areas. To demonstrate the capabilities in spatial mapping of the pressure, the experiments use an external object in the shape of a letter “P” to depress some of the tactile sensors (Fig. 3B). The variations in resistance also show a spatial distribution similar to the letter P (Fig. 3C). Each pixel in Fig. 3C represents the average response of four μSGs from each tactile sensor. In addition, the large-area tactile sensor array also consists of a temperature module beneath each pixel. Figure 3 (D and E) demonstrates the spatial distribution of temperature when a heater with a temperature of 200°C approaches the array at a distance of 1 mm (fig. S37).

In applications such as texture sensing (39), digital palpation (40), and dexterous manipulation (41), the spatial resolution of the tactile sensor array is of vital importance. The compatibility with lithographic techniques allows for the reduction of the linewidth and the construction of row and column selections within the tactile sensors (Fig. 3H). For example, we demonstrated an array using this connection scheme with a spatial density of 360 cm⁻² (Fig. 3, G and H, and figs. S38 and S39), higher than the spatial density of the mechanoreceptors in human fingertips (140 cm⁻²) (42). The array exploits 12 row selection lines and 30 column selection lines to address 360 individual 3D μSGs, which also has the capability to connect with FPCBs. It exhibits a good linear response to normal and shear forces and maintains stable performance under 5000 cycles of loading and unloading of normal force (1 N; figs. S40 and S41). It is also possible to construct high-density arrays for triaxial force mapping by adjusting the arrangement of μSGs. As a demonstration, we arrange 64 3D μSGs with different orientations into an 8 by 8 array within an area of 1 cm² (fig. S42). In this configuration, every four 3D μSGs in adjacent rows and columns collectively form a unit for triaxial force sensing. Consequently, the 64 3D μSGs collectively form a 7 by 7 array for triaxial force mapping. The minimum linewidth of the 3D μSGs here is 7 μm. When the linewidth is scaled down to 4.8 and 2 μm, the spatial density of the triaxial force sensor array can increase to 169 cm⁻² (13 by 13 array) and 1156 cm⁻² (34 by 34 array), respectively. Figure S43 (A, B, D, and E) shows the resistance response of the encapsulated 3D μSG array under normal force and shear force. When subjected to normal force, the arrangement of vectors in fig. S43C appears disordered, indicating that the external force acting on the μSGs is primarily in a normal direction. Under shear force, the vectors in fig. S43F exhibit a distinct left-to-right trend, representing the direction of the shear force.

As a demonstration, we design a hand-shaped FPCB, integrate five modular high-density arrays (each array contains 40 μSGs arranged in 4 rows and 10 columns; Fig. 3, J to L, and figs. S44 to S47) on the fingertips, and distribute other modular tactile sensors (each sensor consists of four μSGs and a temperature module) across the palm region (Fig. 3I). The high-density array on fingertips has a lateral dimension 1 by 1 cm², with an elastomer encapsulation of 8 by 8 by 1 mm³. When subjected to normal force from millimeter-scale external objects molded in different shapes (i.e., two triangles and a rectangle), the high-density array exhibits distinct spatial distributions in resistance change, as shown in Fig. 3M and fig. S48. This configuration resembles the tactile sensing capability of the human hand, where the fingertip comprises high-density mechanoreceptors over an area of ~1 cm² and the palm has capabilities in tactile sensing across a large area (~100 cm²) but with relatively low spatial resolution.

The modular tactile sensors can also integrate on an FPCB with serpentine interconnections to endow stretchability and to decouple other types of mechanical stimuli (Fig. 3N). The experimental verification exploits a single unit—a tactile sensor installed on an FPCB with single-axis serpentines—to demonstrate the insensitivity to bending and stretching. As shown in Fig. 3O and figs. S49 and S50, the single unit mounted on cylinders with different radii of curvature (20 to 50 mm) exhibits similar ΔR/R₀ values under external forces due primarily to the low stiffness of the serpentine interconnections. Figure 3P and figs. S51 and S52 demonstrate that the sensor shows similar response curves under stretching (uniaxial strain: 25%) and bending (radius of curvature: 30 mm), with R² > 0.990, nonlinear error < ±6%, and hysteresis error < 7% for all cases. The collective results in Fig. 3 indicate that the modular feature enables a variety of arrays of flexible tactile sensors, including arrays with customizable shapes (e.g., square array and circular array; fig. S53), arrays with adjustable spatial densities (Fig. 3I), and arrays that are insensitive to bending and stretching (Fig. 3N). Last, we attach our large-area tactile sensor array with stretchable interconnections onto a nondevelopable spherical surface. Results in figs. S54 and S55 demonstrate that our large-area tactile sensor array can conform to nondevelopable surfaces and perform triaxial force mapping without the influence of bending and stretching. The above results (figs. S54 and S55) validate the capability of the tactile sensor array in mapping the spatial distribution of the amplitude and direction of external force. Compared with other force sensors, the tactile sensor array achieves high spatial density and sufficient sensing solution (table S2).

The modular feature also offers opportunities in heterogeneous integration of the flexible tactile sensors and other commercial electronic components, including, but not limited to, amplifiers, microcontrollers, MEMS sensors, application-specific ICs (ASICs), flexible cables, and circuits. For instance, the modular tactile sensor can assemble onto an FPCB together with a processing circuit. The shape of the FPCB can be customized to be attached to a robotic finger (Fig. 4A). The serpentine interconnection in the FPCB can withstand stretching during the bending of the robotic finger (Fig. 4, B and C). When the flexible tactile sensor at the robotic fingertip contacts with the human radial artery, the blood pulse periodically deforms the sensor. As shown in Fig. 4D, the flexible tactile sensor demonstrates sufficient sensitivity to monitor the pulse signal. The processing circuit on the FPCB converts the resistance change of the sensor into voltage variations to facilitate amplification and digitization.

Fig. 4. Multifunctional systems constructed from the modular tactile sensors and other electronic components.

(A and B) Optical images of a tactile sensor and an FPCB integrated on a robotic hand (A) for monitoring pulse signals (B). The inset illustrates the magnified view of the tactile sensor. (C) Magnified view of the FPCB in (B). (D) Voltage signals from the circuit (black curve) and converted resistance changes of the tactile sensor (green curve) when pressing the index finger of the robotic hand on the radial artery of a healthy participant. (E) Optical image of a wearable patch that comprises a tactile sensor and a wireless circuit. (F and G) Optical images of the wearable patch on the skin of a lower leg of a healthy participant before (F) and after (G) covering with a compression bandage. (H) Optical image of a multifunctional sensing module consisting of a tactile sensor and various commercially available sensors. (I) Time-domain responses of the proximity sensor and the tactile sensor recorded during three cycles of approaching and pressing of an external object on the multifunctional sensing module. (J and K) Optical image (J) and the magnified view (K) of an FPCB integrated with an array of tactile sensors and multiple ASIC chips. (L) Optical image of the tactile sensor array connected to a wireless anti–cross-talk circuit through an FPC cable.

As another example, Fig. 4 (E to G) highlights a wearable patch comprising the modular tactile sensor, a Bluetooth system on chip, a battery, an operational amplifier, and an antenna. Here, two serpentine traces connect the tactile sensor to the wireless circuit. The wearable patch has a dimension of 55 mm by 13 mm by 1 mm and can tolerate bending and stretching. The flexible and thin form factors allow the patch to be attached onto the human skin (Fig. 4F) and covered with compression bandage (Fig. 4G). The capability in wireless, continuous monitoring of normal force, shear force, and temperature shows promising applications in compression therapy. In addition, a variety of commercial sensors can integrate on FPCBs together with the tactile sensors. As shown in Fig. 4H, an FPCB hub collects seven types of sensors, including the modular, flexible tactile sensor and other commercial sensors, for multifunctional sensing. Figure 4I records the responses of a tactile sensor and an optical proximity sensor during the approaching and pressing of a flat plate. Specifically, the illuminance gradually decreases and the tactile sensor does not respond when the plate approaches. During the pressing process, the illuminance remains at its minimum value and the resistance of the μSGs increases. The collaborative work of these two sensors diversifies the capabilities of the system in perceiving external stimuli. The integration of flexible tactile sensors with other circuits and sensors on FPCBs can leverage mature technologies in electronics and expand the capabilities in related fields by providing new sensing modalities.

Arrays of flexible tactile sensors can also integrate with chips or other circuits for further processing of the signals, such as multiplexing and amplification. Figure 4 (J and K) and fig. S55A demonstrate that a 4 by 4 array of tactile sensors (containing 64 μSGs and 16 temperature modules) can connect with multiple ASIC chips through wire bonding. The ASIC chips used in this study primarily serve as a demonstration, with no function in signal processing. By contrast, sophisticated anti–cross-talk circuits, analog-to-digital conversion modules, and wireless communication modules at the PCB level can connect with the tactile sensor array through a flexible cable, as depicted in Fig. 4L and fig. S56B. The integration permits wireless, continuous measurements of the spatial distributions of normal force, shear force, and temperature, thereby suggesting promise for a variety of applications in robotics, wearable devices, and biomedicine.

Tactile mapping at the skin interface

To apply the tactile sensor array at the skin interface, we encapsulate the sensor with a low-modulus, biocompatible silicone elastomer (Ecoflex 00-30; modulus: ~68 kPa) doped with white dye (Elkem, RAL9003; 5%) and coat a biocompatible adhesive layer (MG-2402) at the bottom surface of the array. Although the encapsulation slightly alters the performance of the sensor due to the changes in the surrounding materials (fig. S57), the sensitivity can be recalibrated to guarantee accuracy. In addition, the encapsulation layer covers the sharp edges of the FPCB, thereby allowing the tactile sensor array to conformally adhere to the human body with improved safety. The adhesive layer ensures a stable interface between the sensor and the skin and prevents the sensor from falling off even during intense exercise (fig. S58). The total thickness and lateral dimension of the tactile sensor array here are ~1.5 mm and 45 by 42 cm, respectively. The resulting ratio between the thickness and side length reaches ~1:30, which is small enough to permit compliance and conformability in many applications. As shown in fig. S12 (D and E), further improvement of the compliance and conformability is possible by adopting thinner tactile sensors and encapsulation layers (thickness: <500 μm). In this study, we connect a 4 by 4 tactile sensor array (each tactile sensor comprises four μSGs and a temperature module) to a wireless anti–cross-talk circuit (figs. S59 and S60) through an FPC cable and demonstrate its application in wireless monitoring of the distributions of normal force, shear force, and temperature at the skin interface (Fig. 5A and fig. S61). After attaching the tactile sensor array onto the skin, the temperature modules exhibit a resistance increase (average ΔR/R₀: ~0.82%) due to the elevation of temperature from ambient temperature (~31°C) to human body temperature (~37°C), as depicted in Fig. 5C. The temperature elevation also induces resistance variation of the μSGs (average ΔR/R₀: ~0.14%). Each unit in Fig. 5D corresponds to the actual position of the μSG, and the response of one temperature module allows for the calibration of four corresponding μSGs (fig. S62).

Fig. 5. Spatial mapping of tactile information at the skin interface.

(A) Optical image of the encapsulated tactile sensor array. (B) Optical image of the tactile sensor array adhered to triceps muscle close to the elbow joint of a healthy human participant. (C and D) ΔR/R₀ of the temperature modules (C) and μSGs (D) caused by body temperature. (E and F) Optical image (E) and ΔR/R₀ of the μSGs (F) when the elbow joint transitions from a flexed position to an extended position. (G and H) Optical image (G) and ΔR/R₀ of the μSGs (H) when the elbow joint transitions from an extended position to a torsion state. (I to K) Optical image (I), ΔR/R₀ of the μSGs (J), and vector diagram illustrating shear components (K) when the elbow joint vertically presses against a wooden board. (L to N) Optical image (L), ΔR/R₀ of the μSGs (M), and vector diagram illustrating shear components (N) when the elbow joint is obliquely pressed against a wooden board.

When the elbow joint transitions from flexion to extension, the triceps muscle contracts, leading to localized skin deformation and a decrease in the radius of curvature. During this process, the ΔR/R₀ in all the μSGs remains constant, indicating the bending-insensitive feature of the array (Fig. 5, E and F). Other motions near the elbow joint, such as torsion, also induce no change in the ΔR/R₀ for all the μSGs (Fig. 5, G and H). The results prove that the array can circumvent the influence of skin deformations to enable the precise measurement of the contact forces at the skin interface. The contact forces applied to the tactile sensor may involve a combination of normal force and shear force. In each tactile sensor, the responses of four orthogonally arranged 3D μSGs can reflect the amplitude and direction of the external force. Specifically, the disparity of ΔR/R₀ between the left and right μSGs indicates the shear component of the external stimuli along x axis (positive x axis points to the right), while the difference of ΔR/R₀ between the top and bottom μSGs reveals the shear component in y axis (positive y axis points to the bottom). Analyzing the differences of ΔR/R₀ in the left, right, top, and bottom μSGs in each tactile sensor can generate the distribution of shear components (note S5). Figure 5 (I to K) demonstrates the ΔR/R₀ of 16 tactile sensors (64 μSGs) when the triceps muscle presses vertically onto a hard board in wood. In this case, the ΔR/R₀ of all μSGs shows a notable increment compared to the initial state (Fig. 5J). The random direction and short length of the arrow vectors in Fig. 5K indicate that the external force acting on the triceps muscle is primarily normal force.

Switching the pressing direction yields different distributions of the ΔR/R₀ and shear components. As shown in Fig. 5L, in a case where only a subset of the tactile sensors on the skin contact with the wood board, the ΔR/R₀ in the first and second columns of Fig. 5M is comparable to the initial resistance drift caused by body temperature (Fig. 5C). The responses of the third and fourth columns, by contrast, exhibit clear shear components from left to right (the length of the shear vector represents the relative magnitude of the shear component; Fig. 5N). These results validate the capability of the tactile sensor array in mapping the spatial distribution of the amplitude and direction of external force. The flexible and bending-insensitive features of the tactile sensor array enable continuous monitoring of normal and shear forces at large-area skin interfaces, with a diverse set of potential biomedical applications in the future, such as the evaluation of compressive therapy, the prevention of skin ulcers, and the continuous monitoring of plantar pressure.

DISCUSSION

The results presented here provide frameworks in device designs and manufacturing approaches to enable seamless integration of a variety of flexible, modular tactile sensors with other electronic components, either on wafers with high density or on FPCBs across large areas. Diverse collections of examples, including the high-density array, the flexible multifunctional system, the large-area bending-insensitive array, the wireless wearable patch, and others, prove the compatibility of the flexible tactile sensors with technologies in microelectronics and macroelectronics. The 2D and 3D metal/alloy thin wires in the tactile sensors can effectively differentiate normal force, shear force, and temperature with high spatial resolution or across large areas and are immune to other types of mechanical stimuli such as bending and stretching. These features provide opportunities to potential applications ranging from robotics (e.g., anthropomorphic dexterous hand for manipulation), to biomedicine (e.g., wearable patch for continuous monitoring at the skin interface), and to consumer electronics (e.g., human-machine interface). The technology approach promotes the sensing performances and integration schemes of flexible tactile sensors and foreshadows promising opportunities in microelectronics and macroelectronics.

MATERIALS AND METHODS

Wafer-scale fabrication of the tactile sensor with an embedded temperature calibration module

Preparation of the tactile sensor with an embedded temperature module began with spin coating (800 rpm for 10 s and 1000 rpm for 30 s) and curing (110°C for 10 min, 140°C for 30 min, 200°C for 30 min, and 300°C for 90 min in nitrogen) of two layers of PI (POME, 3022s) on a 4-inch Si wafer with high conductivity (0.001 to 0.005 ohm·cm). Photolithography of negative photoresist (SUNTIFIC, lift1303; 3.5 μm in thickness) and deposition of Cr (10 nm in thickness) and Au (150 nm in thickness) films through magnetron sputter, followed by lift-off in acetone, defined the pattern of the metal thin wires for temperature calibration. Spin coating (800 rpm for 10 s and 4500 rpm for 30 s) and curing another layer of PI formed an insulating film on these conductive metal wires. Thin layers of Al (100 nm in thickness) and SiO₂ (150 nm in thickness) were sputtered onto PI as the sacrificial layer and stress layer, respectively. Plasma etching (ICP-5000; 300 W, 4 min) using photoresist (RESEMI, S1813; 2 μm in thickness) as the mask patterned the SiO₂ layer. Subsequent photolithography and wet etching of Al in phosphoric acid (H₃PO₄; diluted in water with a weight ratio of 10:1, 80°C) defined the geometry of the sacrificial layer. Sequential formation of lower PI (through spin coating, 800 rpm for 10 s and 2900 rpm for 30 s), Ni₈₀Cr₂₀ (100 nm in thickness; deposited through sputter and patterned by lift-off), Cr/Au (10/150 nm in thickness; deposited through sputter and patterned by lift-off), and upper PI (spin coat, 800 rpm for 10 s and 2900 rpm for 30 s) yielded the structural support, μSGs, and interconnection of the tactile sensor. As the next step, a layer of sputtered Al (100 nm in thickness) served as the hard mask for the dry etching [MPR-6, O₂: 50 standard cubic centimeter per minute (SCCM), 200 W, and 3 hours] of the PI structural support. Immersion in H₃PO₄ (10:1, 80°C) overnight dissolved the Al sacrificial layer, thus separating the SiO₂ layer from the substrate. The internal stress in SiO₂ bent the structural PI layer upward to form 3D structures. The encapsulation process of the 3D structure involved pouring the silicone polymer (Sylgard 184; 10:1) using a mold, curing at 70°C on a hotplate for 120 min, and subsequent demolding.

Fabrication of the tactile sensor array with row and column selections

Preparation of the tactile sensor array with row and column selections began with spin coating (800 rpm for 10 s and 1000 rpm for 30 s) and curing (110°C for 10 min, 140°C for 30 min, 200°C for 30 min, and 300°C for 90 min in nitrogen) two layers of PI (3022 s) on a 4-inch Si wafer with high conductivity (0.001 to 0.005 ohm·cm). Thin layers of Al (100 nm in thickness; deposited through sputter) and SiO₂ (150 nm in thickness; deposited through sputter) served as the sacrificial layer and stress layer, respectively. Plasma etching (ICP-5000; 300 W, 4 min) using photoresist (S1813; 2 μm in thickness) as the mask patterned the SiO₂ layer. Subsequent photolithography and wet etching of Al in H₃PO₄ (10:1, 80°C) defined the geometry of the sacrificial layer. Spin coating (800 rpm for 10 s and 3300 rpm for 30 s) and curing a layer of PI formed the bottom structural layer. Then, we defined the pattern through photolithography and deposited Cr (10 nm in thickness) and Au (100 nm in thickness) films to create the row selections using a lift-off process. Spin coating (800 rpm for 10 s and 6000 rpm for 30 s) and curing another layer of PI formed an insulating film on the row selections. After depositing a layer of Al (100 nm in thickness), the process used photolithography and wet etching techniques (H₃PO₄, 10:1, 80°C) to pattern it as a hard mask. Dry etching (MPR-6, O₂: 50 SCCM, 200 W, and 30 min) of the insulating layer of PI created the VIAs. Sequential formation of Ni₈₀Cr₂₀ (100 nm in thickness; deposited through sputter and patterned by lift-off), Cr/Au (10/150 nm in thickness; deposited through sputter and patterned by lift-off), and upper PI (spin coat, 800 rpm for 10 s and 3300 rpm for 30 s) yielded the μSGs, column selections, and upper structural layer. Subsequently, a layer of patterned Al (100 nm in thickness; deposited through sputter) served as the hard mask for the dry etching (MPR-6, O₂: 50 SCCM, 200 W, and 3 hours) of the PI structural layer. The remaining releasing and encapsulation steps followed similar procedures as mentioned above.

Transfer of 3D μSGs from the liquid environment to the air

Dissolving the Al sacrificial layer in phosphoric acid yielded the 3D μSGs in the liquid environment. After taking the sample out from the phosphoric acid, three to four cycles of immersion of the sample in anhydrous ethanol effectively substituted the water within the phosphoric acid solution with ethanol, thereby reducing the surface tension from 68.9 to 22.1 mN/m (43, 44). During the drying of ethanol, the stress exerted by the SiO₂ overcame the surface tension of the anhydrous ethanol, allowing for the spontaneous formation of the 3D μSGs (movie S2).

Release of the tactile sensors from the Si wafer

The process started by immersing the Si wafer (0.001 to 0.005 ohm·cm), with the encapsulated tactile sensors facing upward, into an aqueous solution of NaCl electrolyte (concentration: 1 M) at a tilting angle of 30°. The wafer substrate connected to a dc voltage of 20 V and worked as the positive electrode, while the NaCl solution served as the ground. The positively polarized Si wafer underwent an anodic reaction, resulting in the generation of air gaps between PI and the Si wafer. The gaps initiated capillary forces to allow for NaCl solution penetration at the interface between the PI film and Si wafer. Continuous application of the voltage for 5 min separated the PI film from the 4-inch wafer.

Assembly of the modular tactile sensors onto circuit boards

After detachment from the Si wafer, the flexible, modular tactile sensors were picked up and sliced into separate units. An adhesive layer (HoldWell, BE-1800) served as the joining agent to affix the sensors onto the FPCB. Appling conductive pastes (MECHANIC, DJ912) using screen printing established electrical connectivity between the contact pads of the tactile sensors and the circuit board.

Calibration and characterization of the tactile sensor

The setup for calibrating the response of the tactile sensor under normal force included a motorized test stand (ESM 303) to apply the external load, a force gauge (model M5-10, Mark-10) to measure the amplitude of the normal force, and a source meter (KEITHLEY 2450) to measure the resistance of the sensor (sampling rate: 10 Hz). A PowerLab computer interface (ADInstruments, model 16/35) allowed for testing the response time of tactile sensors using higher sampling frequencies (2 kHz).

The setup for calibrating the response of the tactile sensor under shear force included PI tape attached to the upper surface of the elastomer packaging of the tactile sensor, with the other end fixed to a motorized test stand (ESM 303) to apply the shear force. Similar with the characterization for normal force, the setup exploited a force gauge (model M5-10, Mark-10) to measure the amplitude of the shear force and a source meter (KEITHLEY 2450) to measure the resistance of the sensor (sampling rate: 10 Hz).

The setup for the characterization of temperature response of the tactile sensor included a thermostatic heater (ET 150) for the adjustment of the temperature and a source meter (KEITHLEY 2450) to measure the resistance of the sensor (sampling rate: 10 Hz).

Test on healthy human participants

Two layers of the silicone elastomer (Ecoflex 00-30, Smooth-On) doped with white dye (Elkem, RAL9003; 5%) served as the packaging material to sandwich the FPCB in the middle. Specifically, uncured silicone poured into a customized mold served as the bottom layer of packaging. After curing, the FPCB was placed on the bottom layer, and another layer of uncured silicone was poured into the mold. The last step of the packaging process was to suck out the excess uncured silicone elastomer until the tactile sensors exposed. The array was subsequently affixed to the triceps muscle close to the elbow joint of the human body using silicone adhesives (MG-2402) and tested under various conditions.

For the pulse wave test through the robotic hand, the setup included a tactile sensor mounted on the customized FPCB. The sensor was located on the front side of the robot’s index finger and connected to the inverse proportional operation circuit located on the back of the index finger through the FPCB. A PowerLab computer interface measured the output voltage of the inverse proportional operation circuit and converted it into the resistance of the μSG. Data processing included baseline removal and low-pass filtering (digital filter; cutoff frequency: 50 Hz).

Application of the tactile sensor array on healthy human participants (ages 20 to 30, males) performed in this study followed informed consent and the protocol approved by the institutional review board at the Peking University Third Hospital, Beijing, China (IRB00001052-23081).

Characterization of tactile sensors under stretching and bending conditions

A customized fixture enabled the adjustment of uniaxial stretching. Wooden blocks with different radii of curvature (2 to 5 cm) served as the substrate to apply bending stimuli. A source meter (KEITHLEY 2450) measured the resistance of the sensor under both stretching and bending (sampling rate: 10 Hz).

Circuit for anti–cross-talk and wireless mapping of tactile information

The anti–cross-talk circuit comprised voltage followers constructed from operational amplifiers (AD8574) at the input ends, analog switches (BL1551) at both the input and output ends, and operational amplifiers (AD8574) configured as the inverse proportional operation circuit at the output ends. The circuit eliminated the cross-talk currents, and a microcontroller (STM32f103) enabled time-division multiplexing at a frequency of 80 Hz. The output voltage from the circuit was digitized through analog-to-digital converters (AS1412), transmitted to a mobile phone through a Bluetooth module (HC-42), and displayed on a customized user interface.

Characterization of the stress in SiO₂

Characterization of the stress in SiO₂ began with subjecting a 4-inch Si wafer in its initial state to stress testing (128NT, Frontier Semi) as a benchmark. After depositing a SiO₂ film (150 nm in thickness) onto the wafer, a second stress test was conducted. Analyzing the disparity between the results of the two tests yielded the stress distribution of the SiO₂ film (fig. S3).

Supplementary Materials

This PDF file includes:

Supplementary Notes S1 to S5

Figs. S1 to S63

Tables S1 and S2

Legends for movies S1 and S2

References

Other Supplementary Material for this manuscript includes the following:

Movies S1 and S2