The flexible epidermal electrode is fabricated by integrating an LM nanoink-based conductive pattern onto a flexible PET film (Figure 1(a)). Firstly, the eutectic gallium-indium (EGaIn) alloy (melting point at 15°C) is dispersed into nanoparticles (average size 636.9\pm 6.4 nm) (Figure 1(b) and Figure S1) by probe sonication assisted by polyvinyl pyrrolidone (PVP) as surfactant. After filtrating the EGaIn nanoparticles onto filter paper where they form a uniform film, they are mechanically sintered to restore their electric conductivity [30]. This destroys the oxide shell of the LM nanoparticle, and the conductive LM cores are fused together forming a flat film (Figure 1(c)). The sintering reduced the resistivity by nine orders of magnitude to 2.74\times {10}^{-5}  Ωm, a value that is comparable to common alloys and carbon materials (Figure 1(d)) [31–33]. The conductive film is finally laser patterned into electrodes as required and transferred onto transparent ultrathin polyethylene terephthalate (PET) film (0.02 mm) to yield a flexible LM-based electrode for integration into the gas sensor (Figure 1(e)).

To examine how the conductivity of the LM-based electrode varies when exposed to simulated human movement, we bent the electrode from 0° to 180° which resulted in only small changes in electrical resistance of less than 0.5% (Figure 1(f)). We then attached the wire onto the skin of an arm. When attached to human skin, the electrode maintained good conductivity even when bent of compressed (Figure 1(g)). We further measured the resistance change of the liquid metal electrode accurately. The relative change of resistance is only 0.17% (Figure S2). The stable conductivity gives the credit to the high fluidity of liquid metal, which avoid cracks when the circuit deforms. Besides, there is no obvious change for the conductivity of the LM electrode in 15 days (Figure S3), owing to the protection of the dense gallium oxide “skin” on the surface. To further explore its utility in wearable applications for respiratory gas sensing, the as-prepared epidermal electrode was attached to human skin where it showed good resilience to physical straining and maintained good adhesion (Figure 1(h)) which is promising for possible future applications as a wearable gas sensor.

SnS₂ is a relatively abundant and environmentally friendly semiconductor material with a wide bandgap of 2.1 eV [33–35]. It has been reported to be a highly sensitive gas sensing material capable of detecting NO₂ and NH₃ at ppb-level concentrations [36–38]. Here, for the first time, we explore the capabilities of SnS₂ for sensing NO molecules. The SnS₂ gas sensing material was synthesized using a slightly modified version of a previously reported hydrothermal method (Figure 2(a)) [39]. SnO₃^2- and L-cysteine were chosen as source materials for Sn and S, respectively. The microstructure of the obtained SnS₂ nanomaterials was drop-cast onto a substrate (Figure 2(b)) resulting in randomly overlapping hexagonal layers (Figure 2(c)). We measured the lattice fringe spacing as 0.315 nm on both the (100) and (010) lattice planes of hexagonal SnS₂. Interaxial angles were consistently 120° (Figure 2(c), inset) [40]. The fast Fourier transform pattern of this region can be indexed to 2H-SnS₂ along the (100) and (110) zone axes (Figure 2(d)) [41]. The crystal phase was identified using X-ray diffraction (XRD) (Figure 2(e)). The observed diffraction peaks are in agreement with the hexagonal 2H SnS₂ structure (ICDD 23-0677). This structure belongs to the P3m1 space group where three atoms extend over only one monolayer in a unit cell [37]. These characterizations prove the successful fabrication of the SnS₂ 2D nanomaterials.

The 2D nanomaterials of the as-prepared SnS₂ were drop-cast onto the LM electrodes. Figure 3(a) illustrates the I–V curves of the integrated gas sensing device. Due to the higher work function, a commercially available Ag/Pd electrode-based device displayed rectifying characteristics, forming Schottky junctions between metallic electrodes and the sensing material [35]. This potential barrier could hinder the flow of charge carriers and thereby affect sensing performance [42]. In contrast, our epidermal LM-based device exhibited linear characteristics, indicating ohmic contact between the sensing material and the electrode. The band energy diagram is shown in Figure S4. The work function of SnS₂ and EGIA is reported to be 5.1 eV and 4.1-4.2 eV, respectively [40, 43]. Since the work function of EGIA is higher than that of SnS₂, ohmic contact is forming. As a result, our LM electrode would not affect the sensing behavior of SnS₂.

To examine the ability of the LM electrode to sense NO, we fabricated a gas-testing platform using a homemade chamber (Scheme S1). To mimic the actual human respiratory environment, we chose air as background vapor and performed all experiments at room temperature. The target gas concentration was controlled by injecting certain volumes of NO gas into the chamber using high-precision syringes. The sensors ability to detect any change in NO concentration was estimated by measuring changes in the resistance of the sensing material prior (R_0) and after (R_g) having injected the target gas. We defined the sensitivity as \left(R_g-R_0\right)/R_0 and the response time as the time required to reach 90% of the full sensitivity. As an n-type semiconductor, SnS₂ could potentially adsorb NO gas and electrons could then transfer from SnS₂ layers to NO molecules (forming NO^-) based on the charge transfer mechanism [44]. Thus the charge concentration of SnS₂ would decrease while the resistance dramatically increases. Figure 3(b) shows the dynamic performance of the device responding to ppb-level concentrations of NO gas that increased from 15 ppb to 200 ppb, thus covering the typical range of human-exhaled NO concentrations. The device exhibited a 197% sensitivity to 200 ppb with a response time of 223 s and easily recovered its initial resistance once we had stopped injecting NO. In contrast to most other NO sensors, our LM-based electrode performed rather well at low concentrations and could detect NO levels as low as 15 ppb with a sensitivity of 8.7% [45–54]. The sensitivity of our sensor exhibits a strong linear correlation with the gas concentration (Figure 3(c), R^2=0.95). The limit of detection (LOD) is usually defined as the target gas concentration at which the sensor is still capable to produce a signal that is three time higher than the sensor’s noise level. The LOD of our device was determined as 1.32 ppb. When comparing the sensitivity, experimental LOD, and operating temperature of our epidermal device to other NO sensors that had been reported in the literature (Table S1), we found that our sensor outperforms other NO sensors in terms of sensitivity, i.e., detecting lower ppb-level concentrations with higher sensitivity, a lower detection limit, and—more importantly—its good performance at room temperature which is a crucial requirement for NO gas detection in most respiratory applications.

Selectivity is crucial for avoiding false alarms that can be caused by other interfering gases, especially in the monitoring of human exhaled breath which is composed of a variety of gas mixtures. Five other commonly exhaled gas molecules are selected to test the selectivity of the device: H₂ (biomarker of pancreatic diseases), CH₄ (biomarker of gut diseases), NH₃ (biomarker of oral, renal, and liver diseases), and volatile organic compounds (VOCs) such as ethanol (main component of wine which can be exhaled after drinking) and acetone (biomarker of diabetes mellitus) [7]. The concentration of H₂, CH₄, NH₃, and NO is controlled at their clinically relevant concentrations: 20 ppm, 11.4 ppm, 820 ppb, and 25 ppb, respectively [7, 55, 56]. And the concentration of VOC is 100 ppm (much higher than exhaled concentration of healthy people) [57]. Compared with the response to NO, each of these tested interfering gases yielded a negligible sensing response (Figure 3(d)). In order to better simulate the situation of multiple exhaled gases detection, another experiment was carried out by injecting one of these five interfering gases along with NO, but at higher concentrations. While NO was maintained at 50 ppb, each of H₂, CH₄, and NH₃ was injected to yield 500 ppb, while VOC was injected to yield 100 ppm. Thus, although the interfering gas concentration was between one and ten orders of magnitude higher than that of NO, the response to NO remained relatively constant at around 70% with only minor variations (Figure 3(e)). This indicates that our sensor is capable to accurately identify and measure the concentration of NO gas and is relatively immune to the presence of other interfering gas biomarkers.

Water vapor (H₂O) is one of the main components of human exhalation, and the relative humidity in exhaled gas is around 85% [7], i.e., the concentration of water vapor is much higher than the concentration of common gas biomarkers (generally of the order of ppm). Unlike N₂ which is present at high concentrations in exhaled breath and that possess a stable electronic structure, H₂O molecules easily become ionized to H₂O⁺. If the electrons from the ionization of water vapor are passed on to SnS₂, the resistance of SnS₂ would change. In order to assess the impact of this process on the resistance of our electrode, we exposed our prototypes to high relative humidities of 70% to 90% (Figure S5). The resistance of electron-doped SnS₂ was indeed reduced, showing an opposite electronic signal response compared to NO. Since the sensor response is correlated to H₂O concentration, water vapor can act as a marker of respiration state. We also tested whether the sensor response to NO differed for different relative humidities (Figure 3(f)). We found that even when the relative humidity was increased to 90%, there was only a minor decrease in sensor sensitivity (<10%) for NO concentrations of 200 ppb. Higher relative humidities also decreased the response time by 19.2%, which may be due to the fact that the ionized form of SnS₂ is more abundant at higher humidities and can therefore provide its additional electron to the adsorbed NO molecules. Thus, while the detection of NO by our gas sensor would not be affected by the amount of exhaled water vapor, the amount of vapor can be used to identify the respiratory state, which endows the devices with an additional functionality in respiratory monitoring.

O₂ and CO₂ are the two main gases which are inhaled or exhaled by the human body, respectively. As shown in Figures S6 and S7, we investigate the influence of the concentration change of the two gases on the sensing response. The response to 5% concentration of O₂ is lower than 7.5% (Figure S6), much lower than the response to 25 ppb of NO (21.6%). Therefore, the change of oxygen concentration has little effect on the sensing performance. The response to 4% concentration of CO₂ is about 62% (Figure S7). Since both NO and CO₂ are oxidizing gases, the contribution from CO₂ would decrease the threshold for the alarm function of NO sensing (details are discussed in cloud-based remote breath monitoring and diagnosis section). We also calculated the adsorption parameters of SnS₂ layer to CO molecule to evaluate the sensing response to CO biomarker. The adsorption energy and charge transfer are far less than that to NO and H₂O; thus, we can infer that the sensor is not sensitive to CO (Figures S8 and S9).

Considering the temperature difference of ambient environment and body temperature, we further tested the temperature effect on the sensing signal (Figure S10). The relative resistance variation of the sensor is only 1.997% when changing temperature from 24.5°C to 34.3°C, indicating that the temperature change caused by respiration has negligible effect on the performance of the sensor. Besides, like all wearable devices, epidermal sensors often need to operate under conditions of physical strain (e.g., being bent when adapting to the deformation of the skin as the person moves). As mechanical deformations can potentially increase the contact area between SnS₂ nanosheets and NO molecules, the sensor sensitivity can increase by up to 10% when being bent (Figure 3(h)). Furthermore, while mechanical strain can decrease the amount of overlap between the different SnS₂ layers and thereby increase the sensor noise (manifested as increasing jitter in Figure 3(h)) as the bending angle is increased, the sensor still maintains a high signal-to-noise ratio for all tested angles.

Considering the photosensitivity of the SnS₂ nanomaterial, we investigated the capacity of the epidermal sensor for photoelectric sensing [58, 59]. The band gas of the SnS₂ layer was confirmed to be 2.4 eV by UV-vis spectra (Figure S11). Detailed results and a discussion can be found in the supplementary information (Figures S5-S8). In short, our device produced a significant photocurrent (I_p) in response to blue (450 nm), green (532 nm), and red (650 nm) light with blue light triggering the largest response (Figure S12). The photocurrent was highly correlated to the power density, P, as I_p=143.55\times P^{0.51} (R^2=0.9906) (Figures S13 and S14). Previous studies have found that photocurrents can affect the gas sensing performance [40, 60]. We therefore examined the effect of photo-generated currents on the gas sensing behavior of our sensor to 200 ppb of NO gas under different illumination conditions. Red light with an intensity of 3 mW/cm² boosted the sensitivity by a factor 2.14 compared to dark conditions (Figure S15). We expect that future designs of the sensor could include a remote light control mechanism to achieve on-demand regulation of the sensor’s performance.

To elucidate the underlying mechanisms behind the good gas sensing performance of our LM-SnS₂ sensor, we performed some numerical simulations of the molecule-surface binding energy, charge transfer, and adsorption distance employing density functional theory (DFT). The adsorption energy was calculated from \text{Eads}\left[\text{Sn}{\text{S}}_2+\text{gas}\right]=E\left[\text{compound}\right]-E\left[\text{Sn}{\text{S}}_2\right]-E\left[\text{gas}\right], where E\left[\text{compound}\right] is the total energy of the sensor supercell and the target gas molecule after adsorption, E\left[\text{Sn}{\text{S}}_2\right] is the total energy of the SnS₂ layer, and E\left[\text{gas}\right] is the energy of the adsorbed target gas molecule.

Figure 4(a) shows the model of SnS₂ adsorbing NO and the aforementioned interfering gases. The distance between the target gas molecules and the SnS₂ surface ranges from 2.14 to 3.01 Å which is within the typical range for physisorption (Figure S16) [37]. The sensitivity of this process depends on the total charge transfer from the target gas to SnS₂ that leads to the change in resistance. A key factor is the binding strength that determines the number of adsorbed molecules. The two highest binding energies were calculated for NO and H₂O, which demonstrates their strong binding strength to the SnS₂ surface relative to the other interfering gases (Figure 4(b)). The charge transfer between individual molecules is another key factor to explain the high sensitivity of the sensor. A Mulliken population analysis demonstrates how NO has a much higher charge density (0.103 e^-) compared to the interfering gases (Figure 4(c)). This may be due to a more favorable Fermi energy of the SnS₂ layers and the only partially occupied molecular orbitals of NO [37]. Furthermore, for respiratory monitoring applications, the sensors need to be able to complete the response and recovery during a single breathing cycle, i.e., the adsorption speed of the target gas at room temperature is a critical parameter. Our sensor was capable of much faster recovery kinetics for NO and H₂O compared to reported values for other physisorption-based sensors (e.g., graphene, MoS2) [61, 62]. An adsorption distance greater than 2 Å facilitates desorption during thermal vibrations or gas flow disturbance. This effect together with the strong physical affinity of NO gas to SnS₂ ensures the fast reaction and recovery kinetics of SnS₂ layers for NO sensing even at room temperature.

Although the charge transfer for individual H₂O molecule is low, the fact that the concentration of H₂O in human breath is several orders of magnitude higher than the concentration of other biomarker gases means that H₂O vapor can significantly affect the resistance of SnS₂. Figure 4(d) shows the difference in charge density between NO and H₂O adsorption on SnS₂. As the charge transfer occurs in opposite directions, this produces opposite electrical signals in SnS₂ (Figure 3(b)). These differing responses of SnS₂ to the presence of H₂O and NO enable the sensor to perform respiratory monitoring and disease warning simultaneously. That is, the resistance of SnS₂ is reduced when in contact with water vapor and increased with increasing NO concentration.

The strong affinity of SnS₂ for NO gas is also shown by the distortion of the molecular structure of SnS₂ after gas adsorption. NO adsorption induced a deformation of the sandwiched structure of SnS₂ layers. The Raman peaks have shifted after exposure to NO gas (Figure 4(e)). The A_1g (located at 312 cm^-1) and E_g peaks (located at 232 cm^-1) correspond to the vertical and horizontal plane vibrational modes of the Sn-S bonds, respectively [37]. As there was no other peak present after gas adsorption, this confirms the physisorption process. NO adsorption led to a significant increase in the intensity of the A_1g mode and a slight decrease in the E_g mode, which in turn led to an increase in the intensity ratio I\left({\text{A}}_{1\text{g}}/{\text{E}}_{\text{g}}\right) from 9.04 to 17.72 (Figure 4(f)). This behavior corresponds to the SnS₂ layers elongating in the vertical and compressing in the horizontal direction, with the vertical elongation being greater than the horizontal compression. This behavior is confirmed by our calculations regarding the deformations along the X- and Z-axes (Figure 4(g)). We also calculated the changes in position of sulfur atoms at seven different sites, with NO adsorption occurring at the site labeled 3 (Figure 4(h)). As a result of NO adsorption, this third sulfur atom exhibited the greatest displacement in both X- and Z-directions, proving the structural change of the SnS₂ layers (Figures 4(i) and 4(j)).

To test the ability of our epidermal sensor to monitor the human respiratory activity, we recorded several breath patterns (Figure 5(a)). In healthy people, the resistance of the SnS₂ on the electrode is reduced due to the humidity in the exhaled air. Clearly, the different periodicities and amplitudes in the electrical signal mirror the person’s respiratory patterns for different levels of activity, and we could distinguish between normal, rapid, and deep breathing and even identify when a person was holding their breath. These respiratory patterns cover nearly all breathing situations occurring in everyday life like resting, exercising, and even emergency situations like respiratory arrest. In addition, as out sensor simultaneously responds to the level of relative humidity and the concentration of NO gas, the device can be used for asthma patients. To simulate this, we simultaneously injected water vapor and NO gas into the test chamber, raising the relative humidity of the chamber from 60% (ambient level) to 85% (level in exhaled air) while varying the concentration of NO gas from 0 ppb to 200 ppb, to represent different severity levels of asthma. For NO concentrations <50 ppb, the sensitivity of the sensor is negative (Figure 5(b)) and the simulated waveform is consistent with the measured waveform from Figure 5(a). Once the NO concentration reached 75 ppb, the sensitivity changed from negative to positive, indicating the shift of SnS₂ from an electron-doped to a hole-doped state. This change in sensitivity allows potential separation into healthy and unhealthy regions, which may be useful in a telediagnostic setting (Figure 5(c)). The reversal from negative to positive sensitivity occurs at an NO concentration of 58 ppb. In addition, the exhaled CO₂ may also influence dope state of the SnS₂. Since both CO₂ and NO are oxidizing gases which would change the SnS₂ from electron-doped to a hole-doped state. Thus, in practical uses, considering the effect of CO₂, the reversal concentration of NO will be lower than 58 ppb which is closer to the clinical warning level. However, in the real breath monitoring process (Figure 5(a)), we did not observe the reverse of the signal with the presence of exhaled H₂O and CO₂, which further shows that the NO sensing function would not be affected by the presence of these two gases. In future clinical test with healthy people and typical patients, careful calibration should be carried out in order to eliminate the interference from these gases and fully satisfy the clinical use. The fractional exhaled nitric oxide (FeNO) concentration in healthy people typically ranges from 6.7 to 51.1 ppb and can increase to several hundred ppb in patients suffering from asthma [28, 63]. The upper limit of the healthy range (51.1 ppb) is close to the concentration at which the sensitivity changes sign (58 ppb). For practical applications, this means when the sensor is exposed to exhaled gas from people with a lung disease like asthma, the signal is reversed which allows the device to function as an early warning system for certain types of lung disease.

As a long-term and real-time monitoring device, it is important to minimize any adverse effects on a person’s normal daily activities. Wireless capabilities can expand the activity range of people subject to health monitoring using wearable devices. As a proof-of-concept, we developed a graphical user interface for the communication between our device and a laptop or phone, allowing data transmission in real-time for continuous breath monitoring and potential diagnostic and treatment decisions by a clinician (Figure 5(d)). This is illustrated with a block diagram in Figure 5(e) (for the detailed design, see Figure S17). The Microcontroller Unit (MCU) was 32-bit 8-pin packaged, used to design the prototype of the data processing system. The output voltage of the SnS₂-based gas sensor was processed (analog filter) and amplified by an Op amp before being transmitted to a PC/phone via Bluetooth (HJ-131MH) at a rate of 460,800 Baud. The electronic device is powered by lithium batteries mounted on the back of the chip which greatly minimizes the size of the device (shown in Figure S18). The lithium battery is 1.5 V, and the model is CR1220.The real-time data is acquired via a custom-built application at the frequency of 250 Hz. The device weighs less than 11 g and is easy to wear. The epidermal sensing part can be directly attached to the philtrum while the processing unit can be worn around the neck (Figure 5(f)). The complete concept of the wireless communication and cloud-based processing is shown in Figures 5(g) and 5(h). We tested the device on a volunteer (Movie S1) and is showing an example output of eight complete breath cycles in Figure 5(g).

Hospitals with outpatients are one potential scenario for a possible application of the device. The multichannel monitoring interface is convenient for doctors to quickly obtain access to the respiratory status of all their patients. Asthma and respiratory arrest could trigger an automated alert. The patients’ respiratory rate and amplitude can be analyzed in real time from the dynamic waveform. The cloud-based design facilitates applications in modern medical Internet of Things (IoT) systems providing high degrees of accessibility, immediacy, and accuracy.

EGaIn (300 mg) and PVP (100 mg) were added into 10 ml ethanol; then, the mixture was sonicated by Ultrasonic Homogenizer (JY92-IIDN) with the power output setting to 50% (300 W) for 30 min. The temperature of the sample was controlled by using a cold-water bath in case of overheated. The gray liquid metal dispersion was then filtrated onto an organic filter membrane, followed by a mechanical sintering process using a metal roller. A laser of 355 nm (power: 35 W) was utilized to cut the membrane into the electrode shape. And the electrode was pasted to a PET film with thickness of 0.02 mm.

L-Cysteine (0.500 g and 4.0 mmol) and K₂SnO₃·3H₂O (0.300 g and 1.0 mmol) were dissolved in 40 mL DI water. After stirring for 30 min, the mixture was transferred into a 50 mL Teflon-lined stainless steel autoclave and was heated to 200°C for 24 h. When cooling down to room temperature, the product was centrifuged with DI water for several times then vacuum dried at 60°C for 24 h. Finally, the yellowish brown powder was collected and dispersed in ethanol with the concentration of 20 mg/ml by ultrasonicator for 10 min. The sensing material was then drop-casted onto the LM electrode for further use.

The morphology of the SnS₂ was observed by scanning electron microscopy (SEM; HELIOS NanoLab 600i) and transmission electron microscopy (TEM; Tecnai G2 F30). The crystal structure was confirmed by X-ray diffractometer (XRD; Philips X’pert). The Raman spectrum was obtained by Via-Reflex under room temperature with excitation wavelength of 532 nm. UV-vis absorption spectrum was measured using PE Lambda 950. The morphology of LM nanoparticle was observed by SEM (Phenom Scientific).

The resistance of the sensor was recorded by a multimeter. The homemade chamber for simulating the real breath environment was made with a transparent acrylic board, and the volume was 8 L (20\text{cm}\times 20\text{cm}\times 20\text{cm}). The air was acted as the background gas, and the concentration of target gas was controlled by injection certain volume of high concentration gas into the chamber by a high precision syringe. A ceramic heater (3\text{cm}\times 3\text{cm}) was placed in the chamber, and the RH was controlled by evaporating a certain amount of water by the heater. For the investigation of the optoelectronic and light sensitivity performance, the lasers of different wavelengths and power densities, which were bought from Yunxiang Co., Ltd., were vertically irradiated through transparent acrylic plate right onto the sensing material. The tests of the epidermal electrodes and wearable sensors on human skin were approved by all participants and Ethics Board of Shenzhen Institutes of Advanced Technology (approval number: SIAT-IRB-170320-YGS-HJP-A0340).

All the calculation was performed by the package DMol3 in the Material Studio software. The Local Density Approximate (LDA) with Perdew-Wang (PWC) function was chosen for structural relaxations and total energy calculation. A 4\times 4\times 1 supercell was built, and 2\times 2\times 1\Gamma centered Monkhorst–Pack grid for the Brillouin zone sampling was operated, which gave the converged results of all properties. A 15 Å vacuum region was added, which was large enough to weaken the effects of the periodic images. The convergence threshold of the maximum energy changes was 1.0\times {10}^{-5} Ha/per atom, the maximum force was 0.002 Ha/Å, and the maximum displacement was 0.005 Å.