Introduction

Skin-like flexible wearable or implantable sensors have been extensively applied in health monitoring, therapeutic interventions and human-machine interfaces1,2,3,4,5. To construct such high-performance sensor systems, it is crucial to elaborately design functional nanomaterials (e.g., nanosheets, nanotubes, nanodots) in combination with micro-nano structures (e.g., pores, pillars, pyramids), which however involve intricate physical or chemical synthesis and patterning6,7,8,9. In this regard, laser-induced graphene (LIG) has emerged as a versatile technique, which is typically derived by selectively converting polyimide (PI) into porous graphene via photochemical and photothermal reactions in a mask-free manner10,11. Its physical and chemical properties can be judiciously tailored by adjusting laser irradiation, processing ambient, and precursors12,13. Due to these tunable features, LIG, as a pivotal sensing material or interconnection, has been integrated into diverse categories of physical, chemical and electrophysiological sensors14,15,16,17,18. Nevertheless, owing to relatively high Young’s modulus of precursors (e.g., PI, textiles), it is generally required to transfer LIG onto receiver substrates with superior flexibility or stretchability to extend its application scenarios19,20,21,22.

To realize on-demand applications of LIG transfer, two primary methods have been reported, including the vacuum-assisted mechanical peeling and adhesive surfaces-induced separation15. The former typically relies on the infiltration of elastomer solution into the porous LIG, followed by the curing. This allows the transfer of LIG onto polydimethylsiloxane (PDMS), Ecoflex, polyurethane (PU) and styrene-ethylene-butylene-styrene (SEBS)23,24,25,26,27,28. However, the removal of precursor film necessitates relatively strong stripping force attributed to the strong physical interactions among LIG, precursor film and elastomers. This generally requires elastomers with relatively large modulus and thicknessto prevent film fracture, and otherwise incomplete LIG transfer may occur. On the other hand, the approach of sticky media-assisted LIG transfer tends to remove superficial LIG layers or suffers from unpredictable cracks external forces16,29. This also results in relatively low quality of LIG transfer and poor reproducibility.

In this article, we report a universal cryogenic stripping strategy for transferring LIG onto diverse surfaces without modulus limitations. The high-quality and complete transfer of LIG is driven by thermal expansion-induced interlocking, ease of interfacial separation among multi-layer materials, as well as strong electrostatic interactions within the graphene layers. Using this stripping method, LIG can be facilely transferred onto elastic polymers, water-containing hydrogels and fabrics infused with various solvents, such as water, ethanol, ethylene glycol (EG), and glycerol. Taking PDMS as an instance, the thickness of PDMS substrate can be as low as 6.7 μm with its Young’s modulus ranging from 4.5 MPa to 3.9 kPa. Such broad categories of transfer media for supporting LIG afford customized designs of wearable and implantable sensors.

As one of the examples to demonstrate the merit of this transfer approach, we create large-area (175 mm × 274 mm) electronic skins (E-skins) for humanoid robotic applications. Two E-skins mimicking the functionality of human skins are designed and produced, including one for static mechanical sensing and the other for dynamic tactile detection. After vertical integration, the double-layer E-skin on the robot face enabled emotional interactions between volunteers and humanoid robots by simulating human facial expressions.

Strategy of universally transferring LIG without modulus restrictions

To achieve the universal transfer of LIG for on-demand skin-like soft electronics, it is crucial to first penetrate fluid media into the porous LIG framework at ambient temperature. The fluid materials, such as water, solvents, liquid elastomers and hydrogels, can transform to solid states below their freezing point or glass transition temperature (Tg), which significantly increases their mechanical strength. To create such transformations, liquid N2 is applied, serving as an ideal refrigerant to instantaneously reach ultralow temperatures. At the cryogenic state, these fluid media are able to form mechanically interlocking structures with LIG framework due to the phase transition and thermal expansion effect. Such difference in expansion coefficients also allows the facile removal of precursors (e.g., PI film), leading to the complete and efficient transfer of LIG onto almost arbitrary substrate materials (Fig. 1a–c). As an example, we successfully stripped LIG in the shape of a butterfly pattern from a PI film onto a water-infused nonwoven fabric via soft attachment and immersion in liquid N2 (Fig. 1a).

Fig. 1: Concept of universally transferring LIG inside liquid nitrogen.
Fig. 1: Concept of universally transferring LIG inside liquid nitrogen.The alternative text for this image may have been generated using AI.
Full size image

a A digital photo displaying the transferred LIG onto water-infused nonwoven fabric inside liquid N2. b Schematic illustration depicting solvents or uncured polymers in their liquid and solid states at room temperature and liquid N2 temperature, respectively. c Schematic of the interlocking structure formed between porous LIG and water due to the thermal expansion effect. d Dynamic modulus and transition temperature of PDMS elastomers with varying proportions, measured over the temperature range from 133 K to 298 K. The inset presents Young’s modulus ratios of PDMS with varying compositions at 133 K and 298 K. e Photos of LIG patterns on original PI, transferred LIG on PDMS and left PI after LIG removal with different line widths and corresponding optical images. f Photo showing an ultrathin LIG/PDMS composite that can be easily deformed on a glass rod through pressing and twisting. Scale bar: 50 μm.

To evaluate whether the modulus of transfer media may influence the LIG transfer, we prepared PDMS with varying proportions of curing agents for comparison. The mechanical behavior of cured PDMS samples with different moduli were characterized by dynamic thermomechanical analysis (DMA) from 133 K to 298 K. It was found that all the PDMS samples, regardless of their curing ratios (e.g., 10:1, 15:1, 20:1, 25:1, 30:1, and 50:1), exhibited a similar Tg value of around 153 K (Fig. 1d, Supplementary Fig. 1)30. This is because the Tg of PDMS is primarily determined by the chemical structure of its siloxane backbone, rather than the crosslinking density30,31,32. Consequently, regardless of the curing ratio or even in the case of uncured PDMS precursor, the material undergoes a transition into a glassy state at cryogenic temperatures, reaching a modulus close to that of a rigid plastic. As a result, the maximum difference in storage modulus (E’/E” > 3 × 10⁴) was observed for PDMS (50:1) at temperatures of 133 K (E’) and 298 K (E”), indicating a sufficient peeling strength offered by the glassy PDMS samples during transfer (Fig. 1d). All PDMS samples showed a similar storage modulus of over 103 MPa at 133 K (in liquid N₂ atmosphere). This implies that these PDMS with different moduli can offer sufficient peeling strength during transfer (see details in Fig. 2). In contrast, it is hard to perform such transfer using conventional LIG transfer methods limited by mechanical peeling-off strength.

Fig. 2: Characterizations of LIG before and after transfer.
Fig. 2: Characterizations of LIG before and after transfer.The alternative text for this image may have been generated using AI.
Full size image

a Raman spectra of pristine LIG on PI, embedded LIG on relatively thick PDMS by vacuum-assisted transfer and partially embedded LIG on relatively thin PDMS by cryogenic transfer. Sheet resistances and graphitization quality (ID/IG) of pristine LIG on PI and transferred LIG onto PDMS as functions of PDMS b thickness and c modulus under liquid N2. Data were presented as means ± SD, n = 3. d–f Scanning electron microscope (SEM) images of pristine LIG, transferred LIG on the thicker (63.3 μm) PDMS, and transferred LIG on the thinner (6.7 μm) PDMS by vacuum-assisted and cryogenic transfer methods, respectively. g X-ray diffraction (XRD) results of pristine LIG at varied temperatures. h, i High-resolution spherical aberration-corrected transmission electron microscope (SAC-TEM) images of crystalline and disordered graphene layers of LIG after exfoliation in ethanol before and after cryogenic treatment. The exfoliated pristine LIG was observed with a typical spacing (3.4 Å) of graphene layers. j, k Chemical state analysis of carbon (C) and nitrogen (N) in LIG after cryogenic treatment, determined by X-ray photoelectron spectroscopy (XPS). Photos of transferred LIG micro-line patterns on PDMS (l) and left PI substrate after cryogenic transfer (m). n Transfer yields and resistance variations of LIG micro-line patterns under different transfer methods. Scale bar: 1 cm. Data were presented as means ± SD, n = 100.

The digital and optical images of LIG on PI, transferred LIG on PDMS, and absent LIG residues on PI demonstrated complete LIG transfer using the cryogenic stripping method (Fig. 1e, f). Notably, the linewidth resolution is restricted by the laser processing equipment. There is only slight increase of resistance after LIG transfer (Supplementary Fig. 2). In short, the proposed cryogen-assisted strategy enables the modulus-free transfer of digitally patterned LIG onto diverse material surfaces, dramatically extending its application scenarios.

Characterizations of LIG before and after transfer

After verifying the feasibility of this transfer technique, the results of LIG transfer using PDMS with various thicknesses and moduli were investigated. Raman spectra of LIG transferred onto a 6.7-μm-thick PDMS by cryogenic stripping exhibited clear D (1336.69 cm⁻¹), G (1592.04 cm⁻¹), 2D (2904.09 cm⁻¹), and D + D’ (2962.83 cm⁻¹) peaks, similar to pristine LIG. As the reported thickness of LIG typically exceeds 30 μm, which is significantly greater than 6.7 μm, it presents a larger exposed surface area on the PDMS surface, resulting in a stronger Raman signal19. The transferred LIG on the thinner PDMS also exhibited a clear 2D peak, while it was almost absent on the thicker PDMS. These peaks shifted notably due to N-doping after cryogenic treatment using liquid N₂. However, the spectrum of LIG on 63.3-μm-thick PDMS using the vacuum-assisted transfer displayed broader and weaker D and G peaks due to the highly embedded structure of LIG in PDMS (Fig. 2a)21. Additionally, a significant widening of D and G peaks for both transferred samples indicates a reduction in crystallinity.

To assess transfer quality, the intensity ratio of D to G peaks (ID/IG) was analyzed33. By reducing the PDMS coating thickness from 63.3 μm to 6.7 μm, LIG patterns produced under identical laser parameters were successfully transferred using the cryogenic stripping technique. Due to the greater exposed graphene surface, LIG transferred onto the thinner PDMS films exhibited a lower ID/IG ratio, particularly when the PDMS thickness decreased from 63.3 μm (ID/IG = 0.9) to 9.4 μm (ID/IG = 0.8) (Fig. 2b). This suggests a well-maintained high degree of graphitization (i.e., a high transfer quality of LIG). Meanwhile, the sheet resistance of transferred LIG on 9.4-μm-thick PDMS (47.9 Ω/sq) was lower than that on 63.3-μm-thick PDMS (58.4 Ω/sq), further identifying the better graphitization quality. The resistance increase of partially embedded LIG is mainly caused by the pronounced volume expansion of thicker PDMS during recovery from subzero to room temperature, while the disrupted continuity and partial enclosure of LIG by large PDMS grains further aggravate the loss of conductivity. Further decreasing the thickness of PDMS to 6.7 μm, the ID/IG of LIG still displayed a relatively low value of 0.86 in spite of rapid increase of sheet resistance (around 77.1 Ω/sq), indicating the desired transfer quality34. These results demonstrate the preservation of LIG’s structural quality during this stripping process, regardless of the limitations of transfer material thickness. It should be possible to further reduce the thickness of the PDMS, depending on the limitation of spin-coating speeds.

Furthermore, by adjusting the curing agent ratio, PDMS samples with varying Young’s moduli were applied to transfer LIG patterns (Fig. 2c). Notably, the ID/IG values of LIG transferred by PDMS samples with different moduli were almost identical at liquid N₂ temperatures. As previously mentioned, DMA revealed that these PDMS samples exhibited similar glass transition temperatures, indicating a consistent peeling force applied during the transfer process. Nevertheless, the sheet resistance of LIG fluctuated significantly, ranging from 73.7 Ω/sq to 97 Ω/sq, when the modulus of PDMS decreased from 2.4 MPa to 488.7 kPa. This fluctuation was probably attributed to the unsteady crosslinking structure of PDMS with a lower curing agent content (below 10%). Obviously, further decreasing the modulus of PDMS to about 3.9 kPa led to a rapid growth of sheet resistance over 100 Ω/sq. This is due to the collapse of LIG within the PDMS matrix prepared at a 50:1 ratio.

For most E-skins, conformal contact between sensors and target surfaces is critical for acquiring or transmitting significant signals. Surface morphologies demonstrate that the vacuum-assisted transfer of LIG onto PDMS presents a rougher surface with micro-sized grains (Fig. 2d–f). In contrast, the cryogenic stripping technique leads to more exposed LIG with the higher porosity, indicating the well-preserved configuration of LIG (Supplementary Fig. 3). Furthermore, the proposed cryogenic stripping method is able to transfer LIG with a minimum thickness of PDMS less than 10 μm, implying its more seamless attachment with curved surfaces.

To reveal the change of physical and chemical feature of LIG treated at extremely low temperatures, characterizations of X-ray diffraction (XRD), transmission electron microscopy (TEM) and X-ray photoelectron spectroscopy (XPS) were performed. When the temperature decreased from 298 K to 77 K by liquid N₂, the XRD patterns of LIG indicated lattice distortion due to nitrogen doping (N-doping) (Fig. 2g)35. Furthermore, high-resolution SAC-TEM and TEM images distinctly revealed the crystalline structure of LIG and the disordered structure of N-doped LIG treated by liquid N₂ (Fig. 2h, i, Supplementary Fig. 4). The diffraction pattern in the top right inset of Fig. 2h showed layered spacing consistent with the reported graphene value of 3.4 Å36. This implies that the LIG structure transformed from crystalline to amorphous phase due to N-doping and large-scale distortion. Additionally, the XPS results displayed chemical groups including C-N, C-O, C=O, and N-H, providing additional evidence of N-doping in LIG after immersion in liquid N₂ for 5 min (Fig. 2j, k, Supplementary Fig. 5)37. These findings suggest that one of the reasons for the complete transfer of LIG is attributed to the enhanced interactions among graphene layers.

Unlike adhesive transfer, which only removes the superficial layers of graphene, PDMS in a low curing state (2%) can rapidly and completely peel off the LIG from the PI substrate inside liquid N₂, without the need for additional procedures (Supplementary Fig. 6). In previously reported methods, as listed in Supplementary Table 1, rapidly transferring large-area, full-thickness LIG by thin substrates was challenging due to relatively weak intermolecular forces in ambient conditions. By utilizing the proposed modulus-free transfer method, PDMS with varying thicknesses and moduli enables efficient, large-scale transfer at temperatures below Tg. More specifically, the transfer yield, resistance reproducibility, and throughput of 200-μm-wide LIG micro-line patterns were quantitatively assessed using 100-sample experiments for the adhesive-based, vacuum-assisted, and modulus-free transfer methods, respectively (Supplementary Figs. 7–9, Supplementary Table 2). In the adhesive-based method using 3M tape, only superficial LIG was transferred, yielding discontinuous patterns and only 2 of 100 samples were conductive (>1.5 MΩ). Vacuum-assisted transfer with PDMS increased yield to 55%, but peeling-induced cracks led to resistance fluctuations of up to ~400%. In contrast, the modulus-free approach, combining enhanced interfacial interlocking, cryogenic stiffening of PDMS, and stronger interlayer interactions of LIG, achieved 96% yield with resistance values close to the original (~12.6% increase in transferred samples) and left minimal LIG residues on PI, with fewer cracks observed in SEM images (Fig. 2l–n, Supplementary Fig. 9). Moreover, three groups of 100 samples peeled-off at a tensile speed of 150 mm/min by modulus-free transfer obtained 96, 96, and 100 successes with averaging ~97 cm²/min throughput. Benefiting from the high transfer efficiency, large-area LIG patterns such as “E-skin” were also successfully transferred onto thin PDMS (Supplementary Fig. 10).

Mechanism investigation for cryogenic-induced LIG transfer

To reveal the mechanism of cryogenic-induced LIG transfer, first, 180° peeling tests were performed using both uncured PDMS and cured PDMS at 298 K and 77 K, respectively (Fig. 3a). Both the uncured and cured PDMS failed to peel off LIG at 298 K. However, when immersed in liquid N₂, LIG was successfully transferred from PI to cured PDMS. After warming back to 298 K, the PDMS layer recovered from the glassy state to its viscoelastic state, resulting in peeling failure due to a lack of enough mechanical strength.

Fig. 3: Mechanism investigation for cryogenic-induced universal LIG transfer.
Fig. 3: Mechanism investigation for cryogenic-induced universal LIG transfer.The alternative text for this image may have been generated using AI.
Full size image

a Photos illustrating the stripping process of LIG from PI using uncured and cured PDMS samples at temperatures of 298 K and 77 K, respectively. b Experimental results of the stripping process showing the peeling force as a function of displacement for the aforementioned samples performed via 180° peeling tests. c Finite element simulation results revealing the interfacial expansion-induced separation of interlocked LIG and cured PDMS from the PI substrate without external mechanical forces by decreasing the temperature from 298 K to 77 K. Snapshots of the Molecular dynamics (MD) simulation results of (d–g) interlayer structural changes of porous defective graphene and (h–k) lattice distortion of the first layer of graphene crystal faces during cooling from 298 K to 77 K in 10 ns. l Mean squared displacements (MSDs) diffusion coefficient result of N2 molecules during the cooling process. m Calculated interaction energy between the first and second graphene layers, as well as the first graphene layer and N2 molecules during the cooling process.

To investigate the interfacial adhesion strength of LIG under ambient and low-temperature conditions, three groups of 180° peel tests were conducted, and the maximum peeling forces were recorded. At 77 K, the peeling force of cured PDMS increased transiently and was nearly six times higher than that at 298 K (Fig. 3b, Supplementary Fig. 11). This behavior is probably attributed to the enhanced interlocking structure between the porous LIG and PDMS under liquid N2, where LIG and PDMS exhibit opposite thermal expansion and contraction behaviors.

Meanwhile, the significant difference in thermal expansion coefficients led to the ease of interfacial separation between PDMS and PI. Finite element analysis showed that as the temperature decreased from 298 K to 77 K, the thermal expansion stress increased, resulting in spontaneous interfacial separation without external force (Fig. 3c). As described in Fig. 2g, the transfer process was accompanied by N-doping and lattice distortion of LIG. This was also verified by molecular dynamics (MD) simulations. A seven-layer graphene structure with five defects was applied to simulate porous LIG. The system temperature was reduced from 298 K to 77 K over 10 ns. Snapshots of the cross-sectional and frontal structures of multi-layered LIG were captured. The cross-sectional view showed lattice distortion and decreased distance between hydroxyl groups in the defective areas of adjacent graphene layers as the temperature decreased (Fig. 3d–g). Additionally, N₂ molecules diffused more effectively through these distorted and defective areas of LIG at sub-zero temperatures (Fig. 3h–k). The mean square displacement (MSD) results confirmed that the diffusion coefficient of N₂ molecules in the LIG framework decreased as the temperature dropped (Fig. 3l).

Specifically, the interaction energy between adjacent graphene layers increased as the temperature decreased from 298 K to 77 K, while the interaction energy between N₂ molecules and the first graphene layer exhibited ignorable changes (Fig. 3m). This indicates enhanced electrostatic interactions between the defective areas, contributing to the successful monolithic transfer of LIG. In short, the complete cryogenic stripping of LIG from PI is driven by thermal expansion-induced interlocking, interfacial separation, as well as strong physical interactions within the graphene layers.

Universal transfer of LIG onto various soft solids and liquids

Based on these analyses, high-efficiency stripping of LIG presents no restrictions on the mechanical properties of the transfer materials by using phase transition under their Tg or freezing point. In addition to PDMS, other supporting materials like UV-curable resins, Ecoflex, and hydrogels can also be used to transfer LIG with different patterns (e.g., buildings) at 77 K (Fig. 4a–c). Remarkably, only the coating and cryogenic stripping processes are required for the transfer.

Fig. 4: Transferring LIG onto various soft solids and liquids without mechanical limitations.
Fig. 4: Transferring LIG onto various soft solids and liquids without mechanical limitations.The alternative text for this image may have been generated using AI.
Full size image

Photos of transferred LIG patterns in the shape of (a–c) buildings on UV-curable resin, Ecoflex, and hydrogel through cryogenic stripping as well as (d, e) a text and interdigital electrodes on tofu and ice, respectively. f Photo of a large-scale LIG pattern depicting a basketball player transferred onto the surface of a cotton vest soaked with hydrogel solution by cryogenic stripping. g–k Various flower-shaped LIG patterns transferred onto nonwoven fabrics absorbed with uncured PDMS, water, ethanol, ethylene glycol (EG), and glycerol via cryogenic stripping, respectively. l Photo of a personalized medical smart glove with transferred LIG electrodes on nonwoven fabric, using honey-based hydrogel solution as the transfer medium. m, n Photos of the LIG-based quick response (QR) code pattern on the glove before and after one hour of ultrasonication to evaluate the mechanical stability. o, p Top-view and cross-sectional SEM images of the suspended LIG on the fabric after removal of the transfer medium. q Photo image of the smart glove with LIG patterns after removal of the transfer medium to verify gas permeation. Scale bar: 2 cm.

In fact, the ultralow temperature (i.e., 77 K) is not essential for certain transfer media, such as water, whose freezing point is 273 K. Therefore, the stripping process of LIG can be achieved inside the freezer of a refrigerator. LIG patterns, including letters and interdigital electrodes, were successfully transferred onto water-containing tofu and ice (Fig. 4d, e). To further demonstrate the versatility of this technique, a hydrogel solution was applied to a T-shirt, onto which a large-scale LIG pattern on PI was transferred by rapid cryogenic stripping (Fig. 4f, Supplementary Fig. 12, Supplementary Video 1). Considering the efficiency of future commercialization, the modulus-free transfer can be implemented industrially using deep-freeze chambers (–120 °C, up to ~1000 L) to stiffen transfer media (e.g. PDMS) instead of liquid N2. Additionally, LIG can be first deposited on water-soluble films or absorbent fabrics and then transferred via adhesive peel-off or water-assisted methods, enabling scalable, high-throughput production with pattern integrity.

In particular, the nonwoven fabric was applied as a flexible supporting substrate capable of absorbing both solvents (e.g., water, ethanol, ethylene glycol, glycerol, etc.) and uncured elastomers such as PDMS. These liquids readily diffused into the porous architecture of the LIG due to its high surface area and capillarity, forming numerous infiltration points within the graphene network. This diffusion-mediated interfacial contact guarantees effective adhesion between the LIG and solvent-absorbed fabric under frozen conditions. Simultaneously, immersion in liquid N2 temporarily stiffens the solvent-filled fabric. This results in an instantly enhanced strength of the transfer medium for peeling flower-patterned LIG off from PI substrate. (Fig. 4g–k, Supplementary Fig. 13).

In a potential wearable application, LIG electrodes were transferred onto a nonwoven glove infused with a honey-based hydrogel to assist clinical personnel with electrophysiological detection. A customized quick response (QR) code was also introduced to prevent cross-infection (Fig. 4l, Supplementary Fig. 14). By immersing the glove with LIG patterns inside warm water (60 °C), the partially crosslinked hydrogel was able to be removed except the LIG areas. The smart glove was dried, ensuring that the LIG remained firmly attached to the hydrogel-wrapped fabric (Supplementary Fig. 15). Even after one hour of ultrasonication inside water at the room temperature, the QR code pattern remained fully intact on the nonwoven fabric (Fig. 4m, n). SEM images confirmed the continuous patterns of suspended LIG on the fabric (Fig. 4o, p), while the removal of hydrogel from other areas ensured that the fabric still maintained its superior gas permeability (Fig. 4q). In short, this approach offers a promising solution for real-time recording of electrophysiological signals and personalized QR code reporting of medical information (Supplementary Fig. 16).

Large-scale and double-layer LIG-based E-skins for humanoid robots

A primary merit of the proposed transfer technology lies in its capability to efficiently create large-scale E-skins with multiple functionalities, which are suitable for industry-level applications, including wearable and implantable health monitoring or therapy as well as human-robot interactions. For instance, in near future, humanoid robots will assist human beings across various fields, such as domestic service, medical care, and emergency response. Nevertheless, challenges still exist in autonomous decision-making, emotional understanding, multitasking, and social interaction.

To enhance emotional interaction between robots and humans, we designed and created two types of customizable, full-face E-skins for humanoid robotic faces, namely static mechanical E-skin (SME) and high-density dynamic E-skin (HDDE) (Fig. 5a, Supplementary Fig. 17). These two layers, equipped with static mechanical and dynamic tactile sensing functions, were designed to mimic the Ruffini endings and Meissner corpuscles in human skin, with SME detecting static tactile information and HDDE capturing dynamic tactile signals, respectively.

Fig. 5: Double-layer LIG-based E-skin for human-robot tactile and emotional interactions.
Fig. 5: Double-layer LIG-based E-skin for human-robot tactile and emotional interactions.The alternative text for this image may have been generated using AI.
Full size image

a Concept of human-robot emotional interactions through touching the double-layer E-skin (i.e., HDDE and SME) of a robot face. b, c Photos of the mechanical sensor and shear oscillation sensor on the forehead before and after transfer onto PDMS, with thicknesses of ~12 μm and ~14 μm, respectively. The line widths of the sensors are 750 μm and 150 μm, respectively. Scale bar: 1 cm. d Mechanical sensor for strain sensing, with gauge factors measured in both vertical and horizontal stretchings. e, f Photo (left) and optical (right) images of intrinsically sticky and laser-textured surfaces of PDMS for shear oscillation sensor on the chin. Scale bar: 1 cm. g Voltage outputs of an interdigital electrode on the chin of robot face with a viscous PDMS surface and a rough surface after one or two cycles of laser texturing. h Voltage outputs of HDDE for distinguishing the tactile signals on Montessori touch boards made of different materials and with varying surface roughness. i Photos of Montessori touch boards, including metal, marble, wood, cotton cloth, and touch boards with increased levels of roughness ranging from 1 to 9 in sequence from top to bottom and from left to right. j, k The confusion matrix classification results of tactile signals on Montessori touch boards with different materials and surface roughness. l, m Outputs of SME and HDDE under static and dynamic stimulations, respectively. n Real-time static and dynamic responses of SME (three channels) and HDDE (four channels) on the inner and outer sides of the robot face during different interactions such as touching, squeezing, slapping, and brushing, respectively.

The programmed LIG patterns were cryogenically transferred onto PDMS, enabling batch fabrication of large-scale E-skins with high reproducibility (Supplementary Fig. 18). Notably, the PDMS thicknesses were set as low as 12 and 14 μm for SME and HDDE, respectively (Fig. 5b, c). Nearly no LIG residues were captured on the remaining PI substrates for either SME or HDDE by using the cryogenic stripping technique (Supplementary Fig. 19). The LIG electrodes for mechanical and shear oscillation sensors were designed in straight lines, with widths of 750 μm and 150 μm, respectively. Serpentine-shaped connecting electrodes were applied to mitigate strain-induced signal variations. Furthermore, elastic-flexible (E-F) connections were employed to ensure steady data communication between the rigid circuit and flexible LIG pins (Supplementary Fig. 20). During 20,000 cycles of strain testing, the mechanical sensor with E-F connections exhibited a smoother baseline than that with elastic-elastic (E-E) connections. The gauge factors of the mechanical sensor were also measured along two stretching directions, with the horizontal stretching yielding a higher value (30.5) compared to the vertical stretching (11.4) due to the faster crack formation of LIG (Fig. 5d).

Distinguished from the resistive mechanical sensor, the shear oscillation sensor constructed into cross-distributed and high-density electrodes, served as a freestanding-mode triboelectric nanogenerator (TENG) for collecting the alternating voltage outputs between the adjacent electrodes under dynamic stimulations, such as sliding, high-frequency flapping and frictional vibration (Supplementary Fig. 21). The amplitude of sensor response is significant for the mechanical sensors on SME to perceive force magnitude, while the subtle and sequential signals with recognizable shapes are the primary basis for the shear oscillation sensors on HDDE to distinguish the dynamic contact modalities. Due to relatively high sensitivity of shear oscillation sensor for small vibrations led by touching, the surface viscosity of sensor may result in unwanted noise and affect the further data analyses. To minimize viscosity-induced noise, laser texturing was applied to the sensor surface of HDDE, improving its roughness and hydrophobicity. After laser treatment, the surface of the sensor on chin was observed with high roughness (Fig. 5e, f). After one-step surface texturing, the viscosity of shear oscillation sensor was greatly reduced when in contact with a PTFE film. However, further texturing (i.e., 2nd texturing) caused a decrease in signal amplitude due to increase of LIG resistance (Fig. 5g). Therefore, laser texturing was applied only once to balance performance.

The shear oscillation sensor, working as a TENG, can classify dynamic contact modalities by generating alternating sequential signals under different contact frequencies, locations, and sliding speeds. In interdigitated electrode-based TENGs, tactile signals arise from contact electrification coupled with electrostatic induction. The higher electrode density lowers the total resistance on each side, enhancing effective signal generation and improving sensitivity to contact events. Moreover, when the size of the contacting object approaches the electrode spacing, more distinct temporal peaks appear in the high-density fingertip-like electrodes, further enhancing the resolution and fidelity of dynamic tactile signals (Supplementary Fig. 22). For example, the sensor on the robot’s chin produced distinct signals when subjected to single-point contact and three-point sliding with a 20 g weight and a water-filled nitrile glove (Supplementary Fig. 23). Due to the high-density electrode design, the sensor also exhibited clear signals when stimulated by a brush, with the voltage duration inversely proportional to the brushing speed. Additionally, this sensor demonstrated response to various materials and surface roughness, enabling the collection of morphologic signals from metal, marble, wood, cotton, and Montessori touchpads with different roughness levels (Fig. 5h, i, Supplementary Fig. 24). The average classification accuracy of these materials and touchpads with diverse roughness levels can reach 96.3% and 98.6%, respectively (Fig. 5j, k).

Touch perception is a key form of emotional interaction between humans and humanoid robots. This presents a challenge for integrating large-area E-skins with versatile touch-sensing functionalities. Under static stimulation, the SME exhibited a continuous, regular resistance change with varying pressures, while the HDDE only showed transient signals during contact and separation (Fig. 5l). Under dynamic stimuli such as sliding or quick slaps, the HDDE displayed fast and accurate voltage responses, whereas the SME exhibited significant delay and signal overlap of resistance due to the slower response speeds (Fig. 5m).

Depending on their distinct performances, the two large-area E-skins were tailored and attached to the inner and outer sides of a humanoid robot face (fabrication processes in Supplementary Figs. 25, 26). The responses of two vertically distributed E-skins to static and dynamic interactions at different locations of a robot face after integration were recorded. The SME and HDDE consisted of three and four sensor channels, respectively, distributed on the forehead, cheeks and chin. A self-designed circuit integrated with wireless transmission and on-chip data filtering functions was applied to collect these seven channels of signals in real time (Supplementary Fig. 27). With the integration of a data filtering function in the circuit, the tactile signals became clearer and exhibited an improved signal-to-noise ratio. When a volunteer touched the left face or slapped the right cheek, the two E-skins displayed distinct responses. In addition, for static stimulation like squeezing, the SME was measured with evident resistance changes (Fig. 5n). In contrast, the HDDE effectively detected fine touch modalities such as slapping and brushing, even with high-frequency and low-contact forces. By mimicking human emotional expressions, the humanoid robot, for instance, could respond to gentle touches with a happy voice and unfriendly slaps with an angry tone (Supplementary Video 2). In sum, the proposed transfer technique can promote the development of macroscale E-skins for humanoid robots or wearable/implantable bioelectronics.

Discussions

We have reported a universal cryogenic stripping approach to transfer LIG from PI films onto various surfaces like elastic polymers, hydrogels, and fluids-infused fabrics. By cooling the transfer media to liquid nitrogen temperatures (77 K), the process enhances the interlocking between LIG and transfer surfaces, facilitateing facile separation from PI without leaving residues. This strategy allows the transfer of LIG onto PDMS (minimum thickness of 6.7 μm) across three orders of magnitude moduli and even onto modulus-free fluids as long as the supporting substrate is absorbent. This technique enables large-scale creation of LIG-based E-skins, overcoming the limitations of traditional methods like vacuum-assisted mechanical peeling and adhesive-induced separation.

The SME and HDDE were created by transferring LIG onto thin and soft PDMS films, sensitively responding to static forces and dynamic touch modalities, effectively mimicking human skin functionalities. The double-layer E-skin integrated on a humanoid robot’s face enables effective human-robot interactions, simulating human emotional expressions. This cryogenic transfer technique offers a promising approach for the production of large-scale, multifunctional LIG-based E-skins, with potential applications in wearable electronics, medical implantable electronics, and humanoid robots. Although the proposed dual-mode tactile devices conform well to humanoid faces, localized wrinkling or edge lifting may compromise comfort and aesthetics on complex or irregular surfaces. Incorporating additional sensing modalities, such as temperature, humidity, or chemical signals, may also introduce interference. These challenges may be addressed through structure topological optimization, substrate modulus control and combined use of perception mechanisms.

Methods

Fabrication of conductive LIG patterns

The customized LIG patterns were designed by a software named CorelDRAW and produced by carbonizing a clean 50-μm-thick polyimide (PI) (50 μm, Dupont) film using a CO2 laser system (VLS2.50, UNIVERSAL Laser System Inc.) with a wavelength of 10.6 μm in raster mode. The system was available for operation with two types of lenses (i.e., 1.5-inch and HPDFO) with corresponding beam diameters of 76 μm and 25.4 μm, respectively. Compared to the 1.5-in lens, this laser system equipped with HPDFO lens was able to produce the high-density LIG patterns with a smaller line width. Additional parameters, such as laser fluence of 7.24 J/cm2, scanning speed of 209 mm/s and PPI (pulses per inch) value of 500, were also utilized to achieve high-quality LIG patterns.

Universally transfer LIG onto diverse soft materials

Transfer LIG onto PDMS, Ecoflex, SEBS and TPU

First, the PDMS precursors (Sylgard 1844, Dow Corning) with different mixing ratios (10:1, 15:1, 20:1, 25:1, 30:1 and 50:1) were prepared. The Ecoflex (0050, Smooth-on) precursor was prepared by mixing the component A with B in a weight ratio of 1:1. The TPU (Elec-LT506, Shenzhen Yilai Technology Co., Ltd.) solution was directly purchased for use. Before the transfer process of LIG, the as-prepared precursors of these soft materials were spin-coated on the LIG/PI surface in two steps at different speeds depending on the desired thickness. To be specific, the first spin-coating speed of 300 rpm for 10 s was applied to spread precursors over the substrate, while the second step of coating at the higher speed (>1000 rpm) for 30 s was to reduce the thickness and improve the uniformity of soft materials. For example, the PDMS with thicknesses of 63.3 μm, 38.3 μm, 35.6 μm, 25 μm and 20.3 μm were fabricated under high coating speeds of 1000 rpm, 2000 rpm, 3000 rpm, 4000 rpm and 5000 rpm, respectively. This step was repeated for one more cycle. The thicknesses of 9.4 μm- and 6.7 μm-PDMS were fabricated by high-speed coating at 4000 and 5000 rpm, respectively for only one time.

Next, the LIG/PI samples covered with precursors were put into a vacuum container to immerse the precursors into the LIG porous structure under gravity force. Note that this step was optional but recommended. After coating of PDMS, the elastomers were cured at 90 °C for 30 min. For transfer layers thinner than 10 μm, a supporting layer (i.e., PET film) coated with ultrathin PDMS was used. To be specific, as a supporting layer, the PET film was cut and spin-coated with a layer of ultrathin PDMS under speed of 5000 rpm for 60 s. Then, the surface of PDMS on PET was attached onto the LIG/PI. The air bubbles between the two surfaces were rolled out by a rubber roller. Then, the PDMS on the supporting layer was cured at 90 °C for 10 min. After that, the obtained samples were immersed in liquid N2 and the PI films on the top side were easily peeled off using tweezers. Finally, the LIG was almost completely transferred onto soft materials. This method is universally applicable to soft materials with a glass transition temperature (Tg) higher than liquid nitrogen temperature.

Transfer of LIG onto UV-curable resins and honey-based hydrogel

UV-curable resins were purchased for immediate use. The honey-based hydrogel consists of polyvinyl alcohol (PVA, Aladdin), phytic acid (PA, 50% wt in water, Macklin), and natural honey. The presence of PA and monosaccharides from honey contributes to the formation of a highly stretchable polymer network. PA, with its six phosphate groups, can form ester bonds with PVA, allowing for thermo-crosslinking without additional curing agents. To prepare the hydrogel, a 10 wt% PVA solution was made by dissolving PVA in deionized water at 80 °C with continuous stirring for 10 h. The mixture of PVA, PA, and honey was combined in a mass ratio of 5:2:4, and the precursor solution was left to cool and defoam at room temperature. The crosslinking process was facilitated by water-bath heating at 80 °C, turning it from colorless to orange.

After preparing the hydrogel solution and the resin, they were poured into an acrylic mold, and the LIG/PI was placed on top of the mold. The samples were then frozen at −20 °C for 3 hours or immersed in liquid N₂ for 2–5 min. Once the temperature dropped below the Tg of the resin or hydrogel, the PI films were easily peeled off, leaving the LIG transferred onto the soft material.

Transfer of LIG onto solvent-absorbed nonwoven fabrics

Common fluids like water (Tf = 273 K), ethanol (Tf = 159 K), ethylene glycol (EG) (Tf = 153 K), and glycerol were capable of forming crystalline structures at temperatures below their freezing points (Tf). Due to the enhanced interactions between adjacent graphene layers at liquid nitrogen temperature (77 K), multi-layered LIG could be transferred to solvents at their solid states. Nonwoven fabrics were used as supporting layers to absorb liquids, preventing cracks formed in the transferred LIG pattern during temperature recovery process. Specifically, the nonwoven fabric was cut to a suitable size and immersed in the target liquid. After removal, a piece of LIG/PI with a designed pattern was softly attached to the fabric’s surface. The samples were then immersed in liquid N₂, where the LIG was transferred onto the fabric by peeling off the PI substrate. Capillary action promoted this transfer process by facilitating rapid infiltration of liquids into LIG’s porous structure, thus improving transfer speed and efficiency.

Characterizations

The glass transition temperature and rheological properties of PDMS prepared in different ratios were analyzed by a dynamic thermo-mechanical analyzer (DMA) (TAQ800, TA instruments). The surface morphology of LIG-based samples was captured by a scanning electron microscope (SEM) (G300, Zeiss). The crystalline and disordered structure of LIG were observed by a spherical aberration-corrected transmission electron microscope (SACTEM) (Titan Cubed Themis G2300, Thermo Fisher Scientific). The graphitization quality corresponding to the ID/IG was evaluated using a Raman spectrometer ((LabRAM Soleil, HORIBA). The real-time peeling force of LIG from PI to PDMS was measured by a tensile apparatus (ZQ-990B, Zhiqu). The resistance change of LIG-based mechanical sensors was recorded by a multimeter (34661A, Keysight). The fundamental study of high-density dynamic E-skin was carried out using an electrometer ((Keithley 6514, Tektronix) for data recording and a dispenser position shift platform for dynamic contact stimuli (HL-331, Xiaotao Zidonghua Co., Ltd.). The thermal expansion stress for interfacial separation of PI and PDMS under 298 K and 77 K was calculated by using a simulation software (Comsol Multiphysics 6.1).

Signal processing and wireless data transmission for E-skins

A custom-designed circuit system was developed, consisting of a server module and a client module. The server was responsible for multimodal signal acquisition, preprocessing, and wireless transmission, while the client received the transmitted data and forwarded them to a host computer.

Specifically, the server was designed to collect four channels of TENG signals at a sampling rate of 1000 Hz and three channels of strain signals at 50 Hz, respectively. The sampled signals were preprocessed using an on-chip mean filter to reduce noise before further processing. The server adopted a dual-MCU architecture. An STM32H503CBT6 MCU, chosen for its high operating frequency of up to 250 MHz, performed signal acquisition, preprocessing, and data packaging. Four channels of differential voltage signals were acquired using four ADS7044 differential-input ADCs and transmitted to the STM32H503 via Serial Peripheral Interface (SPI) buses. For the strain signals, they were first converted to analog voltages through a voltage divider circuit, buffered using an AD8244 unity-gain amplifier, and then digitized by the built-in 12-bit ADC of the MCU. The packaged multimodal data were subsequently transmitted through the built-in Universal Asynchronous Receiver-Transmitter (UART) to an STM32WB55CGU6 MCU, which was dedicated to wireless communication. The STM32WB55CGU6 then forwarded the data to the client device via Bluetooth Low Energy (BLE) in notify mode.

On the client side, the built-in BLE of the STM32WB55 MCU received the data and transmitted them to a CH340N USB-to-TTL chip through the MCU’s built-in UART, which forwarded the data to the host computer.

The server operated at 3.3 V, powered by a 3.7 V lithium battery, while the client operated at 3.3 V, powered by a 5 V USB input. All sensor connections were implemented with a 1-mm pitch flexible printed circuit (FPC) cable.

Molecular dynamics (MD) simulation

All MD simulations were performed with the large-scale atomic molecular massively parallel simulator (LAMMPS)38. A visual molecular dynamics (VMD) software was applied for trajectory visualization and analysis. Each graphene layer was modeled as porous graphene with intrinsic defects, and the carbon atoms in the graphene sheet were treated as uncharged Lennard–Jones (L-J) particles. The simulation box had a size of 4.10 × 5.10 × 5.20 nm3, which was filled with N2. The optimized potentials for liquid simulations-all atoms (OPLS-AA) force field was used for graphene. The parameters for N2 molecules were taken from a relevant literature, while the well-known Lennard–Jones potential was adopted to model the neutral N239. The non-bonded van der Waals interactions were molded using the 12-6 Lennard–Jones potential, while the electrostatic interactions, such as long-range Coulomb interactions, were addressed by the particle-particle-particle mesh (P3M) technique40. The potential of mean force (PMF) was calculated by numerically integrating the constraining forces, which were conducted from a series of successive constraint MD simulations as a function of interlayer separation, d.

In the production run, a time step of 1 fs was used, and the data were collected every 10 ps. The system was minimized (atomic positions and cell sizes), keeping the box length isotropic. For this system, all production runs were first performed under the canonical ensemble (NVT) at 298 K and 0.1 MPa for 10 ps to achieve equilibration. Subsequently, the system was cooled from 298 K to 77 K over 10 ns under the isothermal-isobaric ensemble (NPT) at 0.1 MPa, with a cooling rate of 22.1 K/ns. Periodic boundary conditions were applied in all directions.

In the MD simulation, the diffusion coefficients could be calculated by the average Mean Squared Displacement (MSD) of the molecule i, which was based on the Einstein equation as reported in Eq. (1),

$${D}_{s}=\mathop{lim}\limits_{t\to \infty }\frac{1}{6t}\left(\frac{1}{N}\sum\limits_{i=1}^{N}{\Vert {r}_{i}(t)-{r}_{i}(0)\Vert }^{2}\right)$$
(1)

Where \((\frac{1}{N}{\sum }_{i=1}^{N}{\Vert {r}_{i}(t)-{r}_{i}(0)\Vert }^{2})\) was an ensemble average, ri (t) denoted the position vector of molecule i at time t and N corresponded to the number of molecules i in system.

Statistics and reproducibility

All the experimental data were statistically analyzed and the results were expressed as a mean standard deviation (s.d.), n ≥ 3. The softwares of Excel, Origin (2018) and Matlab (R2023a) were used for data analyses.